Advances on Broad-Band Wireless Computing, Communication and Applications: Proceedings of the 14th International Conference on Broad-Band Wireless Computing, Communication and Applications (BWCCA-2019) [1st ed. 2020] 978-3-030-33505-2, 978-3-030-33506-9

Table of contents :
Front Matter ....Pages i-xlvi
Front Matter ....Pages 1-1
A Fuzzy-Based Decision System for Sightseeing Spots Considering Hot Spot Access as a New Parameter (Yi Liu, Kevin Bylykbashi, Leonard Barolli)....Pages 3-11
A Multi-sensor Based Physical Condition Estimator for Home Healthcare (Toshiyuki Haramaki, Hiroaki Nishino)....Pages 12-21
Performance Evaluation of WMNs WMN-PSOHC System Considering Constriction and Linearly Decreasing Inertia Weight Replacement Methods (Shinji Sakamoto, Seiji Ohara, Leonard Barolli, Shusuke Okamoto)....Pages 22-31
A Fuzzy-Based Simulation System for IoT Node Selection in Opportunistic Networks and Testbed Implementation (Miralda Cuka, Donald Elmazi, Keita Matsuo, Makoto Ikeda, Leonard Barolli)....Pages 32-43
Consensus Based Mechanism Using Blockchain for Intensive Data of Vehicles (Tehreem Ashfaq, Muhammad Ahmed Younis, Shahzad Rizwan, Zahid Iqbal, Shahid Mehmood, Nadeem Javaid)....Pages 44-55
Block-VN: A Distributed Blockchain-Based Efficient Communication and Storage System (Hassan Farooq, Muhammad Usman Arshad, Muhammad Faraz Akhtar, Shahid Abbas, Bilal Zahid, Nadeem Javaid)....Pages 56-66
Electric Vehicles Privacy Preserving Using Blockchain in Smart Community (Omaji Samuel, Nadeem Javaid, Faisal Shehzad, Muhammad Sohaib Iftikhar, Muhammad Zohaib Iftikhar, Hassan Farooq et al.)....Pages 67-80
A Nodes Selection Algorithm for Fault Recovery in the GTBFC Model (Ryuji Oma, Shigenari Nakamura, Tomoya Enokido, Makoto Takizawa)....Pages 81-92
A TBOI (Time-Based Operation Interruption) Protocol to Prevent Late Information Flow in the IoT (Shigenari Nakamura, Tomoya Enokido, Makoto Takizawa)....Pages 93-104
Enhancement of Binary Spray and Wait Routing Protocol for Improving Delivery Probability and Latency in a Delay Tolerant Network (Evjola Spaho, Klodian Dhoska, Leonard Barolli, Vladi Kolici, Makoto Takizawa)....Pages 105-113
Data Exchange Algorithm at Aggregate Level in the TWTBFC Model (Yinzhe Guo, Ryuji Oma, Shigenari Nakamura, Tomoya Enokido, Makoto Takizawa)....Pages 114-124
Trust-Based Game-Theoretical Decision Making for Food-Energy-Water Management (Suleyman Uslu, Davinder Kaur, Samuel J. Rivera, Arjan Durresi, Meghna Babbar-Sebens)....Pages 125-136
Energy-Efficient Purpose Ordering Scheduler (Tomoya Enokido, Makoto Takizawa)....Pages 137-149
NFC-Based Commissioning of Adaptive Sensing Applications for the 5G IIoT (Hadil Abukwaik, Christian Groß, Markus Aleksy)....Pages 150-161
SCHC-Based Solution for Roaming in LoRaWAN (Wael Ayoub, Mohamad Mroue, Abed Ellatif Samhat, Fabienne Nouvel, Jean-Christophe Prévotet)....Pages 162-172
Reputation System for IoT Data Monetization Using Blockchain (Atia Javaid, Maheen Zahid, Ishtiaq Ali, Raja Jalees Ul Hussen Khan, Zainib Noshad, Nadeem Javaid)....Pages 173-184
Blockchain Based Balancing of Electricity Demand and Supply (Maheen Zahid, Ishtiaq Ali, Raja Jalees Ul Hussen Khan, Zainib Noshad, Atia Javaid, Nadeem Javaid)....Pages 185-198
Data Replication Based on Cuckoo Search in Mobile Ad-Hoc Networks (Takeru Kurokawa, Naohiro Hayashibara)....Pages 199-209
Trusted, Decentralized and Blockchain-Based M2M Application Service Provision (Besfort Shala, Ulrich Trick, Armin Lehmann, Bogdan Ghita, Stavros Shiaeles)....Pages 210-221
A New Mobile Agent System for Sharing Disaster Information Under Unstable Network Conditions (Natsuki Matsumoto, Tetsuya Shigeyasu)....Pages 222-230
A Deep Hybrid Collaborative Filtering Based on Multi-dimension Analysis (Chunyan Zeng, Songnan Lv, Shangli Zhou, Zhifeng Wang)....Pages 231-240
An Energy Efficient Mechanism for Downlink and Uplink Decoupling in 5G Networks (Christos Bouras, Georgios Diles, Rafail Kalogeropoulos)....Pages 241-252
Efficient 5G Network Decoupling Using Dynamic Modulation and Coding Scheme Selection (Christos Bouras, Vasileios Kokkinos, Evangelos Michos)....Pages 253-265
A Probabilistic Offloading Approach in Mobile Edge Computing (Bhed Bahadur Bista, Jiahong Wang, Toyoo Takata)....Pages 266-278
Fuzzy Geocasting in Opportunistic Networks (Sanjay K. Dhurandher, Jagdeep Singh, Isaac Woungang, Makoto Takizawa, Geetanshu Gupta, Raghav Kumar)....Pages 279-292
Digital Content Refinement by Collecting Partly Unreliable Attributes over a Network (Shinji Sugawara)....Pages 293-302
Web Version of IntelligentBox (WebIB) and Its Extension for Web-Based VR Applications - WebIBVR (Yoshihiro Okada)....Pages 303-314
Enemy Attack Management Algorithm for Action Role-Playing Games (Tianhan Gao, Qingwei Mi)....Pages 315-326
Apply Lagrange Interpolation Based Access Control Mechanism in Personal Health Record Medical System (Kuang-Yen Tai, Dai-Lun Chiang, Chun-Yen Chuang, Tzer-Shyong Chen, Frank Yeong-Sung Lin)....Pages 327-337
Analysis of the Relationship Between Psychological Manipulation Techniques and Personality Factors in Targeted Emails (Kota Uehara, Hiroki Nishikawa, Takumi Yamamoto, Kiyoto Kawauchi, Masakatsu Nishigaki)....Pages 338-351
Gait-Based Authentication Using Anomaly Detection with Acceleration of Two Devices in Smart Lock (Kazuki Watanabe, Makoto Nagatomo, Kentaro Aburada, Naonobu Okazaki, Mirang Park)....Pages 352-362
Accurate Online Energy Consumption Estimation of IoT Devices Using Energest (Adnan Sabovic, Carmen Delgado, Jan Bauwens, Eli De Poorter, Jeroen Famaey)....Pages 363-373
Comparison of LoRa Simulation Environments (Christos Bouras, Apostolos Gkamas, Spyridon Aniceto Katsampiris Salgado, Vasileios Kokkinos)....Pages 374-385
Proactive Network Slices Management Algorithm Based on Fuzzy Logic System and Support Vector Regression Model (Amal Kammoun, Nabil Tabbane, Gladys Diaz, Nadjib Achir, Abdulhalim Dandoush)....Pages 386-397
An Optimal Route Recommendation Method for a Multi-purpose Travel Route Recommendation System (Chen Yuan, Minoru Uehara)....Pages 398-408
Artificial Intelligence Technique for Optimal Allocation of Renewable Energy Based DGs in Distribution Networks (Zia Ullah, M. R. Elkadeem, Shaorong Wang)....Pages 409-422
Impact of Sharing Algorithms for Cloud Services Management (Lidia Ogiela, Makoto Takizawa, Urszula Ogiela)....Pages 423-427
Application of Cognitive Protocols in Transformative Computing (Marek R. Ogiela, Lidia Ogiela)....Pages 428-432
Analyzing Mobile Cycling Applications for Monitoring Workouts (Fabricio Landero Cristobal, Miguel A. Wister, Pablo Payro Campos)....Pages 433-444
Road State Information Platform Based on Multi-sensors and Bigdata Analysis (Yoshitaka Shibata, Goshi Sato, Noriki Uchida)....Pages 445-454
A New Discounting Approach to Conflict Information Fusion Using Multi-criteria of Reliability in Dempster-Shafer Evidence Theory (Jin Zhu)....Pages 455-467
Front Matter ....Pages 469-469
The Group-Based Linear Time Causally Ordering Protocol in a Scalable P2PPS System (Takumi Saito, Shigenari Nakamura, Tomoya Enokido, Makoto Takizawa)....Pages 471-482
Algorithm for Detecting Implicitly Faulty Replicas Based on the Power Consumption Model (Hazuki Ishii, Ryuji Oma, Shigenari Nakamura, Tomoya Enokido, Makoto Takizawa)....Pages 483-493
Parallel Data Transmission Protocols in the Mobile Fog Computing Model (Kosuke Gima, Ryuji Oma, Shigenari Nakamura, Tomoya Enokido, Makoto Takizawa)....Pages 494-503
Recovery of Fiber Networks C/M-Plane via an IoT-Based Narrow-Band Links System Based on LoRa Mesh Platform (Goshi Sato, Yoshitaka Shibata, Noriki Uchida)....Pages 504-511
Clustering Analysis and Visualization of TCM Patents Based on Deep Learning (Na Deng, Xu Chen, Caiquan Xiong)....Pages 512-520
Efficient Resource Utilization Using Blockchain Network for IoT Devices in Smart City (Muhammad Zohaib Iftikhar, Muhammad Sohaib Iftikhar, Muhammad Jawad, Annas Chand, Zain Khan, Abdul Basit Majeed Khan et al.)....Pages 521-534
Recommendation System Based on Deep Learning (Tianhan Gao, Lei Jiang, Xibao Wang)....Pages 535-543
Routing Method Based on Data Transfer Path in DTN Environments (Kazuma Ikenoue, Kazunori Ueda)....Pages 544-552
Front Matter ....Pages 553-553
A Hybrid Intelligent Simulation System for Node Placement in WMNs Considering Load Balancing: A Comparison Study for Exponential and Normal Distribution of Mesh Clients (Seiji Ohara, Heidi Durresi, Admir Barolli, Shinji Sakamoto, Leonard Barolli)....Pages 555-569
Multi-dimensional Contract Incentive Design for Mobile Crowdsourcing Networks (Nan Zhao, Menglin Fan, Chao Tian, Pengfei Fan, Xiao He)....Pages 570-578
Evaluation and Comparison of CO2 and Fuel Consumption for Different Car Following Models (Ningling Jiang, Elis Kulla)....Pages 579-588
Individually Separated Wireless Access Point to Protect User’s Private Information (Myoungsu Kim, Kangbin Yim)....Pages 589-596
Long-Term Care (LTC) Monitoring System for Caregivers Based on Wireless Sensing Technology (Hsing-Chung Chen, Mei-He Jiang, Tzu-Ya Chen)....Pages 597-605
Front Matter ....Pages 607-607
Concatenated Path Domain for Dijkstra’s Algorithm Based Ray Tracing to Enhance Computational Areas (Kazunori Uchida, Leonard Barolli)....Pages 609-620
Routing of Optical Baseband Signal Depending on Wavelength in Periodic Structure (Naoki Higashinaka, Hiroshi Maeda)....Pages 621-629
Two-Stage Dynamic Contract Design Under Asymmetric Information in Cooperative Communication (Nan Zhao, Pengfei Fan, Xiao He, Menglin Fan, Chao Tian)....Pages 630-637
Minimizing Control Overhead of Routing Protocols in Wireless Multihop Networks: Simulation Evaluation (Soushi Morita, Elis Kulla)....Pages 638-645
Effect of Parasitic Element on Communication Performance of 13.56 MHz RFID System (Kiyotaka Fujisaki, Yuki Yoshigai)....Pages 646-654
Front Matter ....Pages 655-655
Perception Mining of Network Protocol’s Stealth Attack Behaviors (Yan-Jing Hu, Xu An Wang)....Pages 657-669
Digital Image Anti-counterfeiting Technology (Chin-Ling Chen, Chin-Feng Lee, Fang-Wei Hsu, Yong-Yuan Deng, Ching-Cheng Liu)....Pages 670-677
System Implementation of AUSF Fault Tolerance (Wei-Sheng Chen, Fang-Yie Leu, Heru Susanto)....Pages 678-687
News Collection and Analysis on Public Political Opinions (Zhi-Qian Hong, Fang-Yie Leu, Heru Susanto)....Pages 688-697
Mobile Physiological Sensor Cloud System for Long-Term Care (Ping-Jui Chiang, Heru Susanto, Fang-Yie Leu, Hui-Ling Huang)....Pages 698-707
Front Matter ....Pages 709-709
A Message Relaying Method with Enhanced Dynamic Timer Considering Decrease Rate of Neighboring Nodes for Vehicular-DTN (Shogo Nakasaki, Makoto Ikeda, Leonard Barolli)....Pages 711-720
Prediction of RSSI by Scikit-Learn for Improving Position Detecting System of Omnidirectional Wheelchair Tennis (Keita Matsuo, Leonard Barolli)....Pages 721-732
Decentralized Mechanism for Hiring the Smart Autonomous Vehicles Using Blockchain (Zain Abubaker, Muhammad Usman Gurmani, Tanzeela Sultana, Shahzad Rizwan, Muhammad Azeem, Muhammad Zohaib Iftikhar et al.)....Pages 733-746
An Intelligent Approach for Resource Management in SDN-VANETs Using Fuzzy Logic (Ermioni Qafzezi, Kevin Bylykbashi, Evjola Spaho, Leonard Barolli)....Pages 747-756
Tutorial Educating Developer of Reinforcement Learning Agent Using IDEAL (Takahiro Uchiya, Kodai Shimano, Ichi Takumi)....Pages 757-762
Front Matter ....Pages 763-763
Trusted Remote Patient Monitoring Using Blockchain-Based Smart Contracts (Hafiza Syeda Zainab Kazmi, Faiza Nazeer, Sahrish Mubarak, Seemab Hameed, Aliza Basharat, Nadeem Javaid)....Pages 765-776
A Survey of Malicious HID Devices (Songyin Zhao, Xu An Wang)....Pages 777-786
Power Consumption Attack Based on Improved Principal Component Analysis (Zeyu Wang, Wei Zhang, Peng Ma, Xu An Wang)....Pages 787-799
How Securely Are OAuth/OpenID Connect Implemented in Japan? (Takamichi Saito, Tsubasa Kikuta, Rikita Koshiba)....Pages 800-811
Front Matter ....Pages 813-813
Enhanced Decentralized Management of Patient-Driven Interoperability Based on Blockchain (Asad Ullah Khan, Affaf Shahid, Fatima Tariq, Abdul Ghaffar, Abid Jamal, Shahid Abbas et al.)....Pages 815-827
Design and Construction of Intelligent Decision-Making System for Marine Protection and Law Enforcement (Na Deng, Xu Chen, Caiquan Xiong)....Pages 828-837
Data Authenticity Analysis for Online O2O Data: A Case Study of Second-Hand Houses Posting Data (Xu Chen, Deliang Zhong, Yingzhou Zheng, Shudong Liu, Yipeng Li, Na Deng)....Pages 838-845
A Brief Survey: 3D Face Reconstruction (Tianhan Gao, Hui An)....Pages 846-854
A Feasibility Study on Wrist Rehabilitation Using the Leap Motion (Linlin Zhang, Kin Fun Li)....Pages 855-864
Classification of Cotton and Flax Fiber Images Based on Inductive Transfer Learning (Yuhan Jiang, Song Cai, Chunyan Zeng, Zhifeng Wang)....Pages 865-871
Back Matter ....Pages 873-875

Citation preview

Lecture Notes in Networks and Systems 97

Leonard Barolli Peter Hellinckx Tomoya Enokido   Editors

Advances on Broad-Band Wireless Computing, Communication and Applications Proceedings of the 14th International Conference on Broad-Band Wireless Computing, Communication and Applications (BWCCA-2019)

Lecture Notes in Networks and Systems Volume 97

Series Editor Janusz Kacprzyk, Systems Research Institute, Polish Academy of Sciences, Warsaw, Poland Advisory Editors Fernando Gomide, Department of Computer Engineering and Automation—DCA, School of Electrical and Computer Engineering—FEEC, University of Campinas— UNICAMP, São Paulo, Brazil Okyay Kaynak, Department of Electrical and Electronic Engineering, Bogazici University, Istanbul, Turkey Derong Liu, Department of Electrical and Computer Engineering, University of Illinois at Chicago, Chicago, USA; Institute of Automation, Chinese Academy of Sciences, Beijing, China Witold Pedrycz, Department of Electrical and Computer Engineering, University of Alberta, Alberta, Canada; Systems Research Institute, Polish Academy of Sciences, Warsaw, Poland Marios M. Polycarpou, Department of Electrical and Computer Engineering, KIOS Research Center for Intelligent Systems and Networks, University of Cyprus, Nicosia, Cyprus Imre J. Rudas, Óbuda University, Budapest, Hungary Jun Wang, Department of Computer Science, City University of Hong Kong, Kowloon, Hong Kong

The series “Lecture Notes in Networks and Systems” publishes the latest developments in Networks and Systems—quickly, informally and with high quality. Original research reported in proceedings and post-proceedings represents the core of LNNS. Volumes published in LNNS embrace all aspects and subfields of, as well as new challenges in, Networks and Systems. The series contains proceedings and edited volumes in systems and networks, spanning the areas of Cyber-Physical Systems, Autonomous Systems, Sensor Networks, Control Systems, Energy Systems, Automotive Systems, Biological Systems, Vehicular Networking and Connected Vehicles, Aerospace Systems, Automation, Manufacturing, Smart Grids, Nonlinear Systems, Power Systems, Robotics, Social Systems, Economic Systems and other. Of particular value to both the contributors and the readership are the short publication timeframe and the world-wide distribution and exposure which enable both a wide and rapid dissemination of research output. The series covers the theory, applications, and perspectives on the state of the art and future developments relevant to systems and networks, decision making, control, complex processes and related areas, as embedded in the fields of interdisciplinary and applied sciences, engineering, computer science, physics, economics, social, and life sciences, as well as the paradigms and methodologies behind them. ** Indexing: The books of this series are submitted to ISI Proceedings, SCOPUS, Google Scholar and Springerlink **

More information about this series at http://www.springer.com/series/15179

Leonard Barolli Peter Hellinckx Tomoya Enokido •

•

Editors

Advances on Broad-Band Wireless Computing, Communication and Applications Proceedings of the 14th International Conference on Broad-Band Wireless Computing, Communication and Applications (BWCCA-2019)

123

Editors Leonard Barolli Department of Information and Communication Engineering Fukuoka Institute of Technology Fukuoka, Japan

Peter Hellinckx Department of Electronics University of Antwerp Antwerp, Belgium

Tomoya Enokido Rissho University Tokyo, Japan

ISSN 2367-3370 ISSN 2367-3389 (electronic) Lecture Notes in Networks and Systems ISBN 978-3-030-33505-2 ISBN 978-3-030-33506-9 (eBook) https://doi.org/10.1007/978-3-030-33506-9 © Springer Nature Switzerland AG 2020 This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, expressed or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. This Springer imprint is published by the registered company Springer Nature Switzerland AG The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland

Welcome Message of BWCCA-2019 International Conference Organizers

Welcome to the 14th International Conference on Broadband and Wireless Computing, Communication and Applications (BWCCA-2019), which will be held in conjunction with the 14th 3PGCIC-2019 International Conference from November 7 to 9, 2019, at University of Antwerp, Antwerp, Belgium. This International Conference is a forum for sharing ideas and research work in the emerging areas of broadband and wireless computing. Information networks of today are going through a rapid evolution. Different kinds of networks with different characteristics are emerging and they are integrating into heterogeneous networks. For these reasons, there are many interconnection problems which may occur at different levels of the hardware and software design of communicating entities and communication networks. These kinds of networks need to manage an increasing usage demand, provide support for a significant number of services, guarantee their QoS and optimize the network resources. The success of all-IP networking and wireless technology has changed the ways of living people around the world. The progress of electronic integration and wireless communications is going to pave the way to offer people the access to the wireless networks on the fly, based on which all electronic devices will be able to exchange the information with each other in a ubiquitous way whenever necessary. The aim of this conference is to present the innovative research and technologies as well as developments related to broadband networking and mobile and wireless communications. This edition BWCCA-2019 received 142 paper submissions, and based on review results, we accepted 41 papers (about 29% acceptance ratio) for presentation in the conference and publication by Springer in Lecture Notes in Networks and Systems Proceedings. The organization of an International Conference requires the support and help of many people. A lot of people have helped and worked hard to produce a successful BWCCA-2019 technical program and conference proceedings. First, we would like to thank all authors for submitting their papers, Program Committee Members and reviewers who carried out the most difficult work by carefully evaluating the submitted papers.

v

vi

Welcome Message of BWCCA-2019 International Conference Organizers

This year in conjunction with BWCCA-2019, we have seven International Workshops that complemented BWCCA-2019 program with contributions for specific topics. We would like to thank the Workshop Co-Chairs and all workshops organizers for organizing these workshops. We thank Web Administrators Co-Chairs and Finance Chair for their excellent work. We would like to express our gratitude to Honorary Co-Chairs of BWCCA-2019 for their support and help. We give special thanks to Keynote Speakers of BWCCA-2019 and Local Arrangement Team of University of Antwerp for making excellent local arrangement for the conference. We hope you will enjoy the conference and have a great time in Antwerp, Belgium. Peter Hellinckx Leonard Barolli BWCCA-2019 General Co-chairs Maarten Weyn Tomoya Enokido BWCCA-2019 Program Committee Co-chairs

Welcome Message from BWCCA-2019 Workshops Co-chairs

Welcome to the Workshops of the 14th IEEE International Conference on Broadband and Wireless Computing, Communication and Applications (BWCCA-2019), which will be held in conjunction with the 14th 3PGCIC-2019 International Conference from November 7 to 9, 2019, at University of Antwerp, Antwerp, Belgium. This year seven workshops will be held in conjunction with BWCCA-2019 International Conference. The workshops are very important part of the main conference and they cover specific topics related to next-generation networks, network traffic analysis, sensor technologies, smart environments, complex systems, wireless communication, mobile networks and multimedia networking. BWCCA-2019 workshops are listed in following: 1. The 21st International Symposium on Multimedia Network Systems and Applications (MNSA-2019) 2. The 12th International Workshop on Next Generation of Wireless and Mobile Networks (NGWMN-2019) 3. The 10th International Workshop on Methods, Analysis and Protocols for Wireless Communication (MAPWC-2019) 4. The 10th International Workshop on Cloud, Wireless and e-Commerce Security (CWECS-2019) 5. The 8th International Workshop on Robot and Vehicle Interaction, Control, Communication and Cooperation (RVI3C-2019) 6. The 5th International Workshop on Advanced Techniques and Algorithms for Security and Privacy (ATASP-2019) 7. The 2nd International Workshop on Bio-Sensing, Processing, Application and Networking (BioSPAN-2019) These workshops bring to the researchers conducting research in specific themes and the opportunity to learn from this rich multi-disciplinary experience.

vii

viii

Welcome Message from BWCCA-2019 Workshops Co-chairs

The Workshop Chairs would like to thank the workshop organizers for their great efforts and hard work in proposing the workshop, selecting the papers, the interesting programs and for the arrangements of the workshop during the conference days. We hope you enjoy the workshop’s programs and proceedings. Bart Lannoo Ben Bellekens Keita Matsuo Farookh Hussain BWCCA-2019 Workshops Co-chairs

BWCCA-2019 Organizing Committee

Honorary Co-chairs Makoto Takizawa Walter Sevenhans

Hosei University, Japan University of Antwerp, Belgium

General Co-chairs Peter Hellinckx Leonard Barolli

University of Antwerp, Belgium Fukuoka Institute of Technology, Japan

Program Committee Co-chairs Maarten Weyn Tomoya Enokido

University of Antwerp University, Belgium Rissho University, Japan

Workshop Co-chairs Bart Lannoo Ben Bellekens Keita Matsuo Farookh Hussain

University of Antwerp, Belgium University of Antwerp, Belgium Fukuoka Institute of Technology, Japan University of Technology Sydney, Australia

Finance Chair Makoto Ikeda

Fukuoka Institute of Technology, Japan

ix

x

BWCCA-2019 Organizing Committee

Web Administrator Co-chairs Kevin Bylykbashi Donald Elmazi Miralda Cuka

Fukuoka Institute of Technology, Japan Fukuoka Institute of Technology, Japan Fukuoka Institute of Technology, Japan

Local Organizing Co-chairs Thomas Huybrechts Jens De Hoog

University of Antwerp, Belgium University of Antwerp, Belgium

Steering Committee Co-chair Leonard Barolli

Fukuoka Institute of Technology, Japan

Track Areas 1. Next Generation Wireless Networks Track Co-chairs Bhed Bista Szu-Yin Lin Sriram Chellappan

Iwate Prefectural University, Japan Chung Yuan Christian University, Taiwan University of South Florida, USA

PC Members Jiahong Wang Shigetomo Kimura Chotipat Pornavalai Danda B. Rawat Gongjun Yan Vamsi Paruchuri Arjan Durresi Shih-Yi James Chien Pei-Ju Lee Chih-Hao Lin Hao-Hsiang Ku Jung-Bin Li Thoshitha Gamage Mukundan Sridharan Brijesh Chejerla Srinivas Chakravarthi Thandu

Iwate Prefectural University, Japan University of Tsukuba, Japan King Mongkut’s Institute of Technology Ladkrabang, Thailand Howard University, USA University of Southern Indiana, USA University of Central Arkansas, USA IUPUI, USA National Sun Yat-sen University, Taiwan National Chung Cheng University, Taiwan Chung Yuan Christian University, Taiwan National Taiwan Ocean University, Taiwan Fu Jen Catholic University, Taiwan Southern Illinois University, USA Samraksh Company, USA Florida Blue, USA Amazon, USA

BWCCA-2019 Organizing Committee

xi

2. Cloud and Service Computing Track Co-chairs Hwamin Lee Ramesh C. Hansdah Baojiang Cui

Soonchunhyang University, Korea Indian Institute of Science, Bangalore, India Beijing University of Posts and Telecommunications, China

PC Members Gang Wang Jianxin Wang Jie Cheng Shaoyin Cheng Yan Zhang Willy Susilo Kamil Kluczniak Francesco Palmieri Jian Shen Jin Li Fangguo Zhang Xinyi Huang Shengli Liu Zhenjie Huang Joseph K. Liu Yong Yu Ding Wang Tao Jiang Jianfeng Wang

Nankai University, China Beijing Forestry University, China Shandong University, China University of Science And Technology of China, China Hubei University, China University of Wollongong, Australia Wroclaw University of Technology, Poland University of Salerno, Italy Nanjing University of Information Science and Technology, China Guangzhou University, China Sun Yat-sen University, China Fujian Normal University, China Shanghai Jiaotong University, China Zhangzhou City University, China Institute for Infocomm Research, Australia University of Wollongong, China Peking University, China Xidian University, China Xidian University, China

3. Multimedia and Web Applications Track Co-chairs Yoshihiro Okada Chuan-Yu Chang Salem Alkhalaf

Kyushu University, Japan National Yunlin University of Science and Technology, Taiwan Qassim University, Saudi Arabia

xii

BWCCA-2019 Organizing Committee

PC Members Hiroaki Nishino Kaoru Sugita Tomoyuki Ishida Makoto Nakashima Nobukazu Iguchi Kenzi Watanabe Nobuo Funabiki Shinji Sugawara Li-Wei Kang Chia-Hung Yeh Jun-Wei Hsieh Wu-Chih Hu Chien-Cheng Lee Muhammad Hussain Umair Azfar Khan Shigeru Takano Kosuke Kaneko Akira Haga Wei Shi

Oita University, Japan Fukuoka Institute of Technology, Japan Fukuoka Institute of Technology, Japan Oita University, Japan Kinki University, Japan Hiroshima University, Japan Okayama University, Japan Chiba Institute of Technology, Japan National Yunlin University of Science and Technology, Taiwan National Taiwan Normal University, Taiwan National Taiwan Ocean University, Taiwan National Penghu University of Science and Technology, Taiwan Yuan-Ze University, Taiwan King Saud University, Saudi Arabia Habib University, Pakistan Kyushu University, Japan Kyushu University, Japan Kyushu University, Japan Kyushu University, Japan

4. Security and Privacy Track Co-chairs Tianhan Gao Masakatsu Nishigaki Mohamed Abdur Rahman

Northeastern University, China Shizuoka University, Japan University of Prince Mugrin, Saudi Arabia

PC Members Nan Guo Zhenhua Tan Jian Xu Hiroaki Kikuchi Takamichi Saito Rashid Tahir Syed Sadiq

Northeastern University, China Northeastern University, China Northeastern University, China Meiji University, Japan Meiji University, Japan University of Prince Mugrin, Madinah, Saudi Arabia University of Prince Mugrin, Madinah, Saudi Arabia

BWCCA-2019 Organizing Committee

Md. Mamunur Rashid (Mamun) Akhlaq Ahmad Shyhtsun Felix Wu Zhen-Yu Wu Tsung-Chih Hsiao Kuo-Kun Tseng Akira Otsuka Naonobu Okazaki Masaki Shimaoka

xiii

King’s Business School, UK Umm Al-Qura University, Makkah, Saudi Arabia University of California, Davis, USA Penghu University of Science and Technology, Taiwan Southeast University, China Harbin Institute of Technology, China Institute of Information Security, Japan University of Miyazaki, Japan Secom Co., Ltd., Japan

5. Network Protocols and Performance Analysis Track Co-chairs Tetsuya Shigeyasu Ching-Feng Liang Vamsi Paruchuri

Prefectural University of Hiroshima, Japan Industrial Technology Research Institute, Taiwan University of Central Arkansas, USA

PC Members Xiaoyi Wang Yu Sun Qiang Duan Han-Chieh Wei Masaaki Yamanaka Misako Urakami Tomoya Kawakami Masaaki Noro Nobuyoshi Sato Phone Lin Ray-Guang Cheng Shun-Ren Yang Whai-En Chen

Nokia Solutions and Networks, USA University of Central Arkansas, USA Pennsylvania State University, USA Dallas Baptist University, USA Japan Coast Guard Academy, Japan Tokuyama College of Technology, Japan Nara Institute of Science and Technology, Japan Fujitsu Corp., Japan Iwate Prefectural University, Japan National Taiwan University, Taiwan National Taiwan University of Science and Technology, Taiwan National Tsing Hua University, Taiwan National Ilan University, Taiwan

6. Intelligent and Cognitive Computing Chairs Lidia Ogiela Takahiro Uchiya Hai Dong

Pedagogical University of Krakow, Poland Nagoya Institute of Technology, Japan RMIT University, Australia

xiv

BWCCA-2019 Organizing Committee

PC Members Atsuko Mutoh Shinsuke Kajioka Ryota Nishimura Shohei Kato Francesco Pascale Jan Platoš Pavel Krömer Urszula Ogiela Jana Nowaková Chang, Choi Hoon Ko Hae-Duck Joshua Jeong Pengcheng Zhang Sajib Mistry Tooba Aamir Wei Du Wei Zhang Shang-Pin Ma

Nagoya Institute of Technology, Japan Nagoya Institute of Technology, Japan Tokushima University, Japan Nagoya Institute of Technology, Japan University of Salerno, Italy VŠB-Technical University of Ostrava, Czech Republic VŠB-Technical University of Ostrava, Czech Republic AGH University of Science and Technology, Poland VŠB-Technical University of Ostrava, Czech Republic Chosun University, Korea Chosun University, Korea Korean Bible University, Korea Hohai University, China University of Sydney, Australia RMIT University, Australia Wuhan University of Technology, China Macquarie University, Australia National Taiwan Ocean University, Taiwan

7. Distributed and Parallel Computing Track Co-chairs Naohiro Hayashibara Omar Khadeer Hussain

Kyoto Sangyo University, Japan University of New South Wales, Canberra, Australia

PC Members Sazia Parvin Naeem Janjua Alireza Faed Adil Hammadi Lucian Prodan Kanwalinderjit Kaur Gagneja Rohaya Latip Makoto Takizawa Tomoya Enokido

Melbourne Polytechnic, Australia Edith Cowan University, Australia Ryerson University, Canada Curtin University, Australia Polytechnic University, Timisoara, Romania Florida Polytechnic University Universiti Putra Malaysia, Malaysia Hosei University, Japan Rissho University, Japan

BWCCA-2019 Organizing Committee

xv

8. IoT and Smart Environment Track Co-chairs Nadeem Javaid Chun-Wei Tsai Osama AlFarraj

COMSATS University Islamabad, Pakistan National Chung Hsing University, Taiwan King Saud University (KSA), Saudi Arabia

PC Members Zahoor Ali Khan Umar Qasim Farookh Hussain Elis Kulla Keita Matsuo Hsin-Hung Cho Fan-Hsun Tseng Hsin-Te Wu

Higher Colleges of Technology, UAE University of Alberta, Canada University of Technology Sydney, Australia Okayama University of Science, Japan Fukuoka Institute of Technology, Japan National Ilan University, Taiwan National Taiwan Normal University, Taiwan National Penghu University of Science and Technology, Taiwan

9. Database, Data Mining and Big Data Track Co-chairs Antonio Esposito Yao-Chung Fan Morteza Saberi

University of Campania Luigi Vanvitelli, Italy National Chung Hsing University, Taiwan University of New South Wales, Australia

PC Members Mehran Samavati Farshid Hajati Jinnie Hee Yoon Elena Sitnikova Chen-Yi Lin Lun-Chi Chen Huan Chen Luca Tasquier Stefania Nacchia Salvatore Augusto Maisto Salvatore D’Angelo

University of Sydney, Australia Griffith University, Australia Sejong University, Korea UNSW, Australia National Taichung University of Science and Technology, Taiwan National Center for High-performance Computing (NCHC), Taiwan National Chung Hsing University, Taiwan University of Campania Luigi Vanvitelli, Italy University of Campania Luigi Vanvitelli, Italy University of Campania Luigi Vanvitelli, Italy University of Campania Luigi Vanvitelli, Italy

xvi

BWCCA-2019 Organizing Committee

10. Ubiquitous and Pervasive Computing Track Co-chairs Isaac Woungang Asm Kayes Chyi-Ren Dow

Ryerson University, Canada La Trobe University, Australia Feng Chia University, Taiwan

PC Members Evjola Spaho Makoto Ikeda Elis Kulla Admir Barolli Donald Elmazi Alan Colman Iqbal H. Sarker Eric Pardede Syed Mahbub Patrick Hung Pei-Chun Lin Zhang Kejun Duc-Binh Nguyen

Wei Lu Luca Caviglione Hamed Aly Danda B. Rawat Marcelo Luis Brocardo Glaucio Carvalho

Polytechnic University of Tirana, Albania Fukuoka Institute of Technology, Japan Okayama University of Science, Japan Aleksander Moisiu University of Durres, Albania Fukuoka Institute of Technology, Japan Swinburne University of Technology, Australia Swinburne University of Technology, Australia La Trobe University, Australia La Trobe University, Australia The University of Ontario Institute of Technology, Canada Feng Chia University, Taiwan ZheJiang University, China Thai Nguyen University of Information and Communications Technology (ICTU), Vietnam Keene State College, USA CNIT, Italy Acadia University, Canada Howard University, USA University of Victoria, Canada Ryerson University, Canada

BWCCA-2019 Organizing Committee

xvii

BWCCA-2019 Reviewers Ali Khan Zahoor Barolli Admir Barolli Leonard Bista Bhed Caballé Santi Chellappan Sriram Chen Hsing-Chung Cui Baojiang Di Martino Beniamino Durresi Arjan Enokido Tomoya Ficco Massimo Fun Li Kin Funabiki Nobuo Gao Tianhan Gotoh Yusuke Hachaj Tomasz Hussain Farookh Hussain Omar Javaid Nadeem Jeong Joshua Ikeda Makoto Ishida Tomoyuki Izu Tetsuya Kanzaki Akimitsu Kayes Asm Kikuchi Hiroaki Kolici Vladi Koyama Akio Kulla Elis Lee Kyungroul Leu Fang-Yie

Loia Vincenzo Matsuo Keita Moore Philip Koyama Akio Kryvinska Natalia Nishigaki Masakatsu Nishino Hiroaki Ogiela Lidia Ogiela Marek Okada Yoshihiro Paruchuri Vamsi Krishna Rahayu Wenny Rawat Danda Sakamoto Shinji Shibata Yoshitaka Shigeyasu Tetsuya Sato Fumiaki Saito Takamichi Spaho Evjola Sugawara Shinji Takizawa Makoto Taniar David Uchida Kazunori Uehara Minoru Uda Ryuya Venticinque Salvatore Vitabile Salvatore Waluyo Agustinus Borgy Wang Xu An Woungang Isaac Xhafa Fatos Yim Kangbin

Welcome Message from MNSA-2019 International Symposium Co-chairs

It is our great pleasure to welcome you to the 21st International Symposium on Multimedia Network Systems and Applications (MNSA-2019), which will be held in conjunction with the 14th International Conference on Broadband and Wireless Computing, Communication and Applications (BWCCA-2019) from November 7 to 9, 2019, at University Antwerp, Antwerp, Belgium. This international symposium is a forum for sharing ideas and research work in the emerging areas of multimedia networking and systems. Networks of today are going through rapid evolution and the growing popularity of wired and wireless networks, multimedia network systems and applications is changing our daily life. In the last few years, we have observed an explosive growth of multimedia computing, communication and applications. This revolution is transforming the way people lives, works and interacts with each other and is impacting the way business, education, entertainment and health care are operating. Presently, a lot of research on high-speed networks and multimedia communication is going on. The papers included in this symposium cover aspects of Cloud Computing, Fog Computing VANETs, multimedia applications, network protocols, distributed computing systems and wireless networks. Many people contributed to the success of MNSA-2019. First, we would like to thank the organizing committee of BWCCA-2019 International Conference for giving us the opportunity to organize the symposium. We would like to thank all authors for submitting their research work and for their participation. We are looking forward to meet them again in the forthcoming editions of the workshop. We would like to express our appreciation to MNSA-2019 reviewers who carefully evaluated the submitted papers. Finally, we would like to thank the Local Arrangement Chairs for the local arrangement of the workshop.

xix

xx

Welcome Message from MNSA-2019 International Symposium Co-chairs

We hope you will enjoy the symposium and have a great time in Antwerp, Belgium. Makoto Takizawa Leonard Barolli MNSA-2019 Symposium Organizers Tomoya Enokido MNSA-2019 Program Co-chairs

MNSA-2019 Organizing Committee Symposium Co-chairs Makoto Takizwa Leonard Barolli

Hosei University, Japan Fukuoka Institute of Technology, Japan

Symposium PC Chair Tomoya Enokido

Rissho University, Japan

Program Committee Members Testuya Shigeyasu Shintaro Imai Takuya Yoshihiro Motoi Yamagiwa Kazunori Ueda Markus Aleksy Irfan Awan Bhed Bahadur Bista Yusuke Gotoh Hui-Huang Hsu Rei Itsuki Satoru Izumi Akio Koyama Tomotaka Kozuki Toshiaki Osada Fumiaki Sato Takuo Suganuma Hideyuki Takahashi Atsushi Takeda Noriki Uchida Misako Urakami Masaaki Yamanaka Muhammad Younas Fatos Xhafa

Prefectural University of Hiroshima, Japan Iwate Prefectural University, Japan Wakayama University, Japan University of Yamanashi, Japan Kochi University of Technology, Japan ABB AG, Germany University of Bradford, UK Iwate Prefectural University, Japan Okayama University, Japan Tamkang University, Taiwan Hiroshima International University, Japan Tohoku University, Japan Yamagata University, Japan Hiroshima International University, Japan Tohoku Bunka Gakuen University, Japan Toho University, Japan Tohoku University, Japan Tohoku University, Japan Tohoku Gakuin University, Japan Fukuoka Institute of Technology, Japan Oshima National College of Maritime Technology, Japan Hiroshima International University, Japan Oxford Brookes University, UK Technical University of Catalonia, Spain

Welcome Message from NGWMN-2019 International Workshop Co-chairs

Welcome to the 12th International Workshop on Next Generation of Wireless and Mobile Networks (NGWMN-2019), which will be held in conjunction with the 14th International Conference on Broadband and Wireless Computing, Communication and Applications (BWCCA-2019) from November 7 to 9, 2019, at University Antwerp, Antwerp, Belgium. The aim of this workshop is to present the innovative researches, methods and algorithms for wireless networks, sensor networks and ubiquitous computing. The next generation of wireless and mobile networks is expected to allow a single mobile user to access heterogeneous wireless and mobile networks. Therefore, this workshop will provide a timely technical forum for the dissemination of new results in this exciting research area and is devoted to the architectures, protocols, resource management, mobility management and scheduling in integrated wireless and mobile networks. Many people have kindly helped us to prepare and organize the NGWMN-2019 workshop. First, we would like to thank the authors who submitted high-quality papers and reviewers who carefully evaluated the submitted papers. We would like to give our special thanks to General Co-Chairs of BWCCA-2019 for their strong encouragement and guidance to organize this workshop. We would like to thank all of the PC members for their serious review works in order to make a successful organization of NGWMN-2019. Finally, we would like to thanks the Local Organizing Committee of BWCCA-2019 for excellent arrangement. We hope you will enjoy the conference and have a great time in Antwerp, Belgium. Leonard Barolli Hsing-Chung Chen (Jack Chen) Kangbin Yim NGWMN-2019 Co-chairs

xxi

xxii

Welcome Message from NGWMN-2019 International Workshop Co-chairs

NGWMN-2019 Organizing Committee Workshop Co-chairs Leonard Barolli Hsing-Chung Chen (Jack Chen) Kangbin Yim

Fukuoka Institute of Technology, Japan Asia University, Taiwan Soonchunhyang University, Korea

Program Committee Members Makoto Ikeda David Taniar Kin Fun Li Fatos Xhafa Vamsi Paruchuri Neng-Yih Shih Yeong-Chin Chen Akio Koyama Ming-Shiang Huang Isaac Woungang Arjan Durresi Jyh-Horng Wen Cheng-Ying Yang Tzu-Liang Kung Muhammad Younas Awan Irfan Yung-Fa Huang Chia-Hsin Cheng Neng-Yih Shih Jyu-Wei Wang

Fukuoka Institute of Technology, Japan Monash University, Australia University of Victoria, Canada Technical University of Catalonia, Spain University of Central Arkansas, USA Asia University, Taiwan Asia University, Taiwan Yamagata University, Japan Asia University, Taiwan Ryerson University, Canada Indiana University–Purdue University Indianapolis, USA Tunghai University, Taiwan Department of Computer Science, University of Taipei, Taiwan Asia University, Taiwan Oxford Brookes University, UK University of Bradford, UK Chaoyang University of Technology, Taiwan National Formosa University, Yunlin County, Taiwan Asia University, Taiwan Asia University, Taiwan

Message from MAPWC-2019 International Workshop Organizers

Welcome to the 10th International Workshop on Methods, Analysis and Protocols for Wireless Communication (MAPWC-2019), which will be in conjunction with the 14th International Conference on Broadband and Wireless Computing, Communication and Applications (BWCCA-2019) from November 7 to 9, 2019, at University Antwerp, Antwerp, Belgium. Wireless communications are characterized by high bit error rates and burst errors, which arise due to interference fading, shadowing, terminal mobility and so on. Since the traditional design of the algorithms, methods and protocols of the wired Internet did not take wireless networks into account, the performance over wireless networks is largely degraded. Especially, multi-hop communication aggravates the problem of wireless communication even further. To solve these problems, there has been increased interest to propose and design new algorithms and methodologies for wireless communication. The aim of this workshop is to present the innovative researches, methods and numerical analysis for wireless communications and wireless networks. The workshop contains high-quality research papers, which were selected carefully by Program Committee Members. It is impossible to organize such a successful program without the help of many individuals. We would like to express our appreciation to the authors of the submitted papers and to the program committee members, who provided timely and significant review. We hope all of you will enjoy MAPWC-2019 and find this a productive opportunity to exchange ideas with many researchers. Makoto Ikeda Elis Kulla Hiroshi Maeda MAPWC-2019 Workshop Co-chairs

xxiii

xxiv

Message from MAPWC-2019 International Workshop Organizers

MAPWC-2019 Organizing Committee Workshop Chair Makoto Ikeda Elis Kulla Hiroshi Maeda

Fukuoka Institute of Technology, Japan Okayama University of Science, Japan Fukuoka Institute of Technology, Japan

Program Committee Members Arjan Durresi Leonard Barolli Kazunori Uchida Koki Watanabe Shinichi Ichitsubo Zhi Qi Meng Irfan Awan Tsuyoshi Matsuoka Fatos Xhafa Kiyotaka Fujisaki Donald Elmazi

Indiana University–Purdue University Indianapolis (IUPUI), USA Fukuoka Institute of Technology, Japan Fukuoka Institute of Technology, Japan Fukuoka Institute of Technology, Japan Kyushu Institute of Technology, Japan Fukuoka University, Japan Bradford University, UK Kyushu Sangyo University, Japan Technical University of Catalonia, Spain Fukuoka Institute of Technology, Japan Fukuoka Institute of Technology, Japan

Message from CWECS-2019 International Workshop Organizers

Welcome to the 10th International Workshop on Cloud, Wireless and e-Commerce Security (CWECS-2019), which is held in conjunction with the 14th International Conference on Broadband and Wireless Computing, Communication and Applications (BWCCA-2019) from November 7 to 9, 2019, at University Antwerp, Antwerp, Belgium. Computer network and communication have been a part of our everyday life. People use them to contact others almost anytime anywhere. However, hackers due to business benefits, enjoying their skill/professional achievement or some other reasons very often attack, intrude or penetrate our systems. This is the key reason why computer/network security has been one of the important issues in computer research. Many researchers have tried to do their best to develop system security techniques and the methods to protect a system. But system attacks still occur worldwide every day. In fact, current system security technology is far away from perfect and should be continuously improved. This workshop aims to present the innovative researches, methods and applications for cloud, wireless and e-commerce security. Other network related papers are also welcomed. The workshop contains high-quality research papers, which were selected carefully by Program Committee Members. It is impossible to organize such a successful program without the help of many individuals. We would like to express our appreciation to the authors of the submitted papers and to the program committee members, who provided timely and significant reviews. We hope all of you will enjoy CWECS-2019 and find this a productive opportunity to exchange ideas with many researchers. Fang-Yie Leu Aniello Castiglione Kun-Lin Tsai Jia-Chun Lin CWECS-2019 Workshop Co-chairs

xxv

xxvi

Message from CWECS-2019 International Workshop Organizers

CWECS-2019 Organizing Committee Workshop Co-chairs Fang-Yie Leu Aniello Castiglione Kun-Lin Tsai Jia-Chun Lin

Tunghai University, Taiwan University of Naples Parthenope, Italy Tunghai University, Taiwan University of Oslo, Norway

Program Committee Members Alessandra Sala Agung Budi Prasetijo Antonio Colella Chin-Cheng Lien Chin-Ling Chen Chiu-Ching Tuan Claudio Soriente Francesco Palmieri Fuw-Yi Yang Heru Susanto I-Long Lin Ilsun You Jinn-Ke Jan Jung-Chun Liu Lein Harn Sen-Tang Lai Sergio Ricciardi Shyhtsun Felix Wu Tzung-Her Chen Ugo Fiore

University of California, Santa Barbara, USA Diponegoro University, Indonesia Italian Army, Italy Soochow University, Taiwan Chaoyang University of Technology, Taiwan National Taipei University of Technology, Taiwan Universidad Politecnica de Madrid, Spain University of Salerno, Italy Chaoyang University of Technology, Taiwan The Indonesian Institute of Sciences, Indonesia Central Police University, Taiwan Soonchunhyang University, Korea National Chung Hsing University, Taiwan Tunghai University, Taiwan University of Missouri-Kansas City, USA Shih Chien University, Taiwan Technical University of Catalonia, Spain University of California, Davis, USA National Chiayi University, Taiwan University of Naples Parthenope, Italy

Message from RVI3C-2019 International Workshop Organizers

Welcome to the 8th International Workshop on Robot and Vehicle Interaction, Control, Communication and Cooperation (RVI3C-2019), which will be held in conjunction with the 14th International Conference on Broadband and Wireless Computing, Communication and Applications (BWCCA-2019) from November 7 to 9, 2019, at University Antwerp, Antwerp, Belgium. Robots and vehicles are being steadily introduced into modern everyday life and are expected to play a key role in the near future. Typically, the robots are deployed in situations where it is too dangerous, expensive, tedious and complex for humans to operate, while vehicles are used in the roads and highways. Although many of the real-life applications may only need a single robot or vehicle, a large number of them require the cooperation, coordination and communication of a team of robots and vehicles to accomplish a certain task. The use of multiple robots and vehicles of overlapping capabilities offers a cost-effective and more robust solution. This redundancy in the robots and vehicles capabilities makes the overall system more flexible and fault-tolerant. This workshop focuses on the emerging field of robot and vehicles interaction, communication and cooperation bringing together research and application of methodology from robotics, Vehicular Ad Hoc Networks (VANETs), human factors, human–computer interaction, interaction design, cognitive psychology, education and other fields to enable robots and vehicles to have more natural and more rewarding interactions, communication and cooperation with humans throughout their spheres of functioning. The design of an efficient collaborative framework that ensures the autonomy and the individual requirements of the involved robots and vehicles is a very challenging task. Developing operational multi-robot and multi-vehicle teams involves research on a number of topics such as fault-tolerant cooperative control, adaptive action selection, distributed control, robot and vehicle awareness of team member actions, improving efficiency through learning, inter-robot and inter-vehicle communication, action recognition, local versus global control and metrics for measuring the success.

xxvii

xxviii

Message from RVI3C-2019 International Workshop Organizers

The aim of this workshop is to present the innovative researches, technologies and new concepts, services and application software of robotic systems and VANETs. The organization of an International Workshop needs the help of many people. We would like to express our appreciation to the authors of the submitted papers and to the program committee members. We hope all of you will enjoy RVI3C-2019 program and your stay in Taichung, Taiwan. Keita Matsuo Takahiro Uchiya Leonard Barolli RVI3C-2019 Workshop Organizers

RI3C-2019 Organizing Committee Workshop Co-chairs Keita Matsuo Takahiro Uchiya Leonard Barolli

Fukuoka Institute of Technology, Japan Nagoya Institute of Technology, Japan Fukuoka Institute of Technology, Japan

Program Committee Members Tatsushi Tokuyasu Hiroyuki Fujioka Akio Koyama Kaoru Fujioka Tetsuya Morizono Junpei Arai Arjan Durresi Fatos Xhafa Makoto Ikeda Evjola Spaho

Fukuoka Institute of Technology, Japan Fukuoka Institute of Technology, Japan Yamagata University, Japan Fukuoka Women’s University, Japan Fukuoka Institute of Technology, Japan Yamagata College of Industry and Technology, Japan Indiana University–Purdue University at Indianapolis (IUPUI), USA Technical University of Catalonia, Spain Fukuoka Institute of Technology, Japan Polytechnic University of Tirana, Albania

Web Administrator Co-chairs Donald Elmazi Miralda Cuka

Fukuoka Institute of Technology, Japan Fukuoka Institute of Technology, Japan

Message from ATASP-2019 International Workshop Organizers

Welcome to the 5th International Workshop on Advanced Techniques and Algorithms for Security and Privacy (ATASP-2019) which will be in conjunction with the 14th International Conference on Broadband and Wireless Computing, Communication and Applications (BWCCA-2019) from November 7 to 9, 2019, at University Antwerp, Antwerp, Belgium. As cloud computing becomes prevalent, more and more organizations outsource the expensive computing and storage into the cloud servers. It brings appealing benefits including relief of the burden for storage management, universal data access with independent geographical locations and avoidance of capital expenditure on hardware, software and personnel maintenances, etc. Despite the tremendous benefits, outsource storage inevitably suffers from some new security challenges, such as security and privacy of outsourced data. To address these issues, there has been increased interest to propose and design new algorithms and methodologies for secure cloud computing. This workshop covers the latest advances in securing cloud storage, cloud computing, secure systems and privacy issues that lead to gain competitive advantages in business and academia scenarios. Industry and academic researchers, professionals and practitioners are invited to exchange their experiences and present their ideas in this field. The workshop contains high-quality research papers, which were selected carefully by Program Committee Members. The main topics of interest of ATASP-2019 include but are not limited to the following: Security infrastructure and framework of cloud computing Coding and cryptography for secure cloud Remote data integrity and possession Distributed computation and access control on encrypted data Privacy-preserving technologies in cloud computing Security and privacy in outsourcing data and computation Dependability, availability and forensics in cloud Secure data sharing and secure data replication Security in cloud and grid systems

xxix

xxx

Message from ATASP-2019 International Workshop Organizers

It is impossible to organize such a successful program without the help of many individuals. We would like to express our appreciation to the authors of the submitted papers and to the program committee members, who provided timely and significant review. We hope all of you will enjoy ATASP-2019 and find this a productive opportunity to exchange ideas with many researchers. Leonard Barolli Takamichi Saito Workshop Co-chairs Mirang Park Tetsuya Izu Workshop PC Co-chairs

ATASP-2019 International Workshop Organizers Workshop Co-chairs Leonard Barolli Takamichi Saito

Fukuoka Institute of Technology, Japan Meiji University, Japan

Workshop PC Co-chairs Mirang Park Tetsuya Izu

Kanagawa Institute of Technology, Japan Fujitsu Laboratories Ltd., Japan

Program Committee Members Dai Yamamoto Yukie Unno Toshihiro Ohigashi Yuichi Nakamura Yasuyuki Nogami Hiroaki Kikuchi Xu An Wang Naonobu Okazaki Hisaaki Yamaba Masakatsu Nishigaki

Fujitsu Laboratories Ltd., Japan Fujitsu Laboratories Ltd., Japan Tokai University, Japan Waseda University, Japan Okayama University, Japan Meiji University, Japan Engineering University of CAPF, China University of Miyazaki, Japan University of Miyazaki, Japan Shizuoka University, Japan

Web Administrator Co-chairs Donald Elmazi Miralda Cuka Kevin Bylykbashi

Fukuoka Institute of Technology, Japan Fukuoka Institute of Technology, Japan Fukuoka Institute of Technology, Japan

Message from BioSPAN-2019 International Workshop Organizers

Welcome to the 2nd International Workshop on Bio-Sensing, Processing, Application and Networking (BioSPAN-2019), which will be held in conjunction with the 14th International Conference on Broadband and Wireless Computing, Communication and Applications (BWCCA-2019) from November 7 to 9, 2019, at University Antwerp, Antwerp, Belgium. As an international workshop, BioSPAN aims to share the recent developments in the biomedical domain of broadband and wireless computing and communication. This workshop will bring together leading researchers, engineers and scientists from around the world to disseminate their work in bio-sensing, bio-signal processing, biomedical application and bio-networking technologies. Topics of interest of BioSPAN-2019 are related with: Bioelectronics and diagnostics, Biomedical image analysis, Biosensors and biosensor networks, Data mining and pattern recognitions in bio-data, Validation of biosensors in practical environment, Bio-system integration, Signal transduction technology, Commercial developments, manufacturing and markets, Acoustic wave biosensors, Optical and spectral-analysis-based biosensors, Advanced bio-signal processing, Bio-sensing applications and case studies, Wireless technology for medical biology and the life sciences, Body area networks and body sensor networks. It is impossible to organize such a successful program without the help of many individuals. We would like to express our appreciation to the authors of the submitted papers and to the program committee members, who provided timely and significant review.

xxxi

xxxii

Message from BioSPAN-2019 International Workshop Organizers

We hope all of you will enjoy BioSPAN-2019 and find this a productive opportunity to exchange ideas with many researchers. Kin Fun Li Kosuke Takano BioSPAN-2019 International Workshop Organizers

BioSPAN-2019 Organizing Committee Workshop Co-chairs Kin Fun Li Kosuke Takano

University of Victoria, Canada Kanagawa Institute of Technology, Japan

Program Committee Members Yvonne Coady Watheq El-Kharashi Michael Horie Wei Li Wei Lu Ana-Maria Sevcenco Makoto Ikeda Nadeem Javaid Fatos Xhafa Tomoya Enokido Makoto Takizawa Evjola Spaho Elis Kulla Shinji Sakamoto Admir Barolli Hiroaki Nishino Keita Matsuo Akio Koyama

University of Victoria, Canada Ain Shams University, Egypt British Columbia Institute of Technology, Canada Beijing University of Posts and Telecommunication, China Keene University, USA University of Victoria, Canada Fukuoka Institute of Technology, Japan COMSATS University Islamabad, Pakistan Technical University of Catalonia, Spain Rissho University, Japan Hosei University, Japan Polytechnic University of Tirana, Albania Okayama University of Science, Japan Seikei University, Japan Aleksander Moisiu University of Durres, Albania Oita University, Japan Fukuoka Institute of Technology, Japan Yamagata University, Japan

BWCCA-2019 Keynote Talks

Wireless Experimentation with SDR: The Way to Drive Innovation Ingrid Moerman Ghent University, Ghent, Belgium

Abstract. There exist many ways for researching and developing innovative solutions: from theoretical analysis, simulations, small-scale setup to large-scale experimentation. This first part of this talk will discuss the benefits and pitfalls of different approaches and illustrate them with some concrete examples. While experimentation seems to be the most challenging approach, the second part of this talk will present how the software-defined radio (SDR) facility offered in the H2020 ORCA project is capable to accelerate wireless innovation. The advantage of SDR over “off-the-shelf” technology is that it enables full and open implementation of all network functionality, also the lower physical and medium access control (MAC) layers. The ultimate goal of the ORCA project is to enable wireless experimenters to unlock the potential of reconfigurable radio technology by setting up advanced experiments involving end-to-end applications that require control of novel wireless technologies or cooperation between multiple networked SDR platforms within extreme and/or diverging communication needs in terms of latency, reliability or throughput, well before new radio technologies become available on the market in commercial off-the-shelf products. In the third and last part of the talk, the ORCA vision toward orchestrating next-generation services through end-to-end network slicing will be presented. Network slicing (also known as network virtualization) allows network resources to be used in a flexible, dynamic and customized manner, and most crucially, provides isolation between different virtual networks. ORCA believes that each network segment should have their own orchestrator, tailored to the segment’s particularities. The use of a separate orchestrator for each network segment reduces complexity and breaks down the larger E2E network orchestration problem into smaller parts. In this way, each segment orchestrator can focus on a limited number of well-defined tasks, reducing the software complexity, both in terms of design and implementation. The ORCA vision is expected to foster innovation for everyone (not only big industrial players but also smaller companies and the research community), to reduce development life cycle, to simplify standardisation and to stimulate multi-disciplinary experimentation.

xxxv

2020: The AI Decade Deevid De Meyer Cronos Group, Leuven, Belgium

Abstract. By now, it should be clear to everyone that AI has had a significant impact over the past decade. Thanks to the rise of deep learning, applications are being released almost every week that were previously deemed impossible. Chatbots, deepfakes, self-driving cars, intelligent cameras, digital authors and these technologies have been made feasible in the past 10 thanks to machine learning, and breakthroughs are still happening on almost a weekly basis. Where 2010 was the decade where AI broke through, many people think that 2020 will be the decade where it reaches maturity and widespread adoption. In this presentation, we will look at today’s frontier of artificial intelligence and predict how the field of AI will evolve in the coming decade.

xxxvii

Contents

The 14th International Conference on Broad-Band andd Wireless Computing, Communication and Applications (BWCCA-2019) A Fuzzy-Based Decision System for Sightseeing Spots Considering Hot Spot Access as a New Parameter . . . . . . . . . . . . . . . . . . . . . . . . . . . Yi Liu, Kevin Bylykbashi, and Leonard Barolli

3

A Multi-sensor Based Physical Condition Estimator for Home Healthcare . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Toshiyuki Haramaki and Hiroaki Nishino

12

Performance Evaluation of WMNs WMN-PSOHC System Considering Constriction and Linearly Decreasing Inertia Weight Replacement Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Shinji Sakamoto, Seiji Ohara, Leonard Barolli, and Shusuke Okamoto A Fuzzy-Based Simulation System for IoT Node Selection in Opportunistic Networks and Testbed Implementation . . . . . . . . . . . . Miralda Cuka, Donald Elmazi, Keita Matsuo, Makoto Ikeda, and Leonard Barolli Consensus Based Mechanism Using Blockchain for Intensive Data of Vehicles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Tehreem Ashfaq, Muhammad Ahmed Younis, Shahzad Rizwan, Zahid Iqbal, Shahid Mehmood, and Nadeem Javaid Block-VN: A Distributed Blockchain-Based Efficient Communication and Storage System . . . . . . . . . . . . . . . . . . . . . . . . . . . Hassan Farooq, Muhammad Usman Arshad, Muhammad Faraz Akhtar, Shahid Abbas, Bilal Zahid, and Nadeem Javaid

22

32

44

56

xxxix

xl

Contents

Electric Vehicles Privacy Preserving Using Blockchain in Smart Community . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Omaji Samuel, Nadeem Javaid, Faisal Shehzad, Muhammad Sohaib Iftikhar, Muhammad Zohaib Iftikhar, Hassan Farooq, and Muhammad Ramzan

67

A Nodes Selection Algorithm for Fault Recovery in the GTBFC Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Ryuji Oma, Shigenari Nakamura, Tomoya Enokido, and Makoto Takizawa

81

A TBOI (Time-Based Operation Interruption) Protocol to Prevent Late Information Flow in the IoT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Shigenari Nakamura, Tomoya Enokido, and Makoto Takizawa

93

Enhancement of Binary Spray and Wait Routing Protocol for Improving Delivery Probability and Latency in a Delay Tolerant Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105 Evjola Spaho, Klodian Dhoska, Leonard Barolli, Vladi Kolici, and Makoto Takizawa Data Exchange Algorithm at Aggregate Level in the TWTBFC Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114 Yinzhe Guo, Ryuji Oma, Shigenari Nakamura, Tomoya Enokido, and Makoto Takizawa Trust-Based Game-Theoretical Decision Making for Food-Energy-Water Management . . . . . . . . . . . . . . . . . . . . . . . . . . . 125 Suleyman Uslu, Davinder Kaur, Samuel J. Rivera, Arjan Durresi, and Meghna Babbar-Sebens Energy-Efficient Purpose Ordering Scheduler . . . . . . . . . . . . . . . . . . . . 137 Tomoya Enokido and Makoto Takizawa NFC-Based Commissioning of Adaptive Sensing Applications for the 5G IIoT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150 Hadil Abukwaik, Christian Groß, and Markus Aleksy SCHC-Based Solution for Roaming in LoRaWAN . . . . . . . . . . . . . . . . . 162 Wael Ayoub, Mohamad Mroue, Abed Ellatif Samhat, Fabienne Nouvel, and Jean-Christophe Prévotet Reputation System for IoT Data Monetization Using Blockchain . . . . . . 173 Atia Javaid, Maheen Zahid, Ishtiaq Ali, Raja Jalees Ul Hussen Khan, Zainib Noshad, and Nadeem Javaid Blockchain Based Balancing of Electricity Demand and Supply . . . . . . 185 Maheen Zahid, Ishtiaq Ali, Raja Jalees Ul Hussen Khan, Zainib Noshad, Atia Javaid, and Nadeem Javaid

Contents

xli

Data Replication Based on Cuckoo Search in Mobile Ad-Hoc Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199 Takeru Kurokawa and Naohiro Hayashibara Trusted, Decentralized and Blockchain-Based M2M Application Service Provision . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210 Besfort Shala, Ulrich Trick, Armin Lehmann, Bogdan Ghita, and Stavros Shiaeles A New Mobile Agent System for Sharing Disaster Information Under Unstable Network Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . 222 Natsuki Matsumoto and Tetsuya Shigeyasu A Deep Hybrid Collaborative Filtering Based on Multi-dimension Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231 Chunyan Zeng, Songnan Lv, Shangli Zhou, and Zhifeng Wang An Energy Efficient Mechanism for Downlink and Uplink Decoupling in 5G Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241 Christos Bouras, Georgios Diles, and Rafail Kalogeropoulos Efficient 5G Network Decoupling Using Dynamic Modulation and Coding Scheme Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253 Christos Bouras, Vasileios Kokkinos, and Evangelos Michos A Probabilistic Offloading Approach in Mobile Edge Computing . . . . . 266 Bhed Bahadur Bista, Jiahong Wang, and Toyoo Takata Fuzzy Geocasting in Opportunistic Networks . . . . . . . . . . . . . . . . . . . . . 279 Sanjay K. Dhurandher, Jagdeep Singh, Isaac Woungang, Makoto Takizawa, Geetanshu Gupta, and Raghav Kumar Digital Content Refinement by Collecting Partly Unreliable Attributes over a Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293 Shinji Sugawara Web Version of IntelligentBox (WebIB) and Its Extension for Web-Based VR Applications - WebIBVR . . . . . . . . . . . . . . . . . . . . . 303 Yoshihiro Okada Enemy Attack Management Algorithm for Action Role-Playing Games . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315 Tianhan Gao and Qingwei Mi Apply Lagrange Interpolation Based Access Control Mechanism in Personal Health Record Medical System . . . . . . . . . . . . . . . . . . . . . . 327 Kuang-Yen Tai, Dai-Lun Chiang, Chun-Yen Chuang, Tzer-Shyong Chen, and Frank Yeong-Sung Lin

xlii

Contents

Analysis of the Relationship Between Psychological Manipulation Techniques and Personality Factors in Targeted Emails . . . . . . . . . . . . 338 Kota Uehara, Hiroki Nishikawa, Takumi Yamamoto, Kiyoto Kawauchi, and Masakatsu Nishigaki Gait-Based Authentication Using Anomaly Detection with Acceleration of Two Devices in Smart Lock . . . . . . . . . . . . . . . . . . 352 Kazuki Watanabe, Makoto Nagatomo, Kentaro Aburada, Naonobu Okazaki, and Mirang Park Accurate Online Energy Consumption Estimation of IoT Devices Using Energest . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 363 Adnan Sabovic, Carmen Delgado, Jan Bauwens, Eli De Poorter, and Jeroen Famaey Comparison of LoRa Simulation Environments . . . . . . . . . . . . . . . . . . . 374 Christos Bouras, Apostolos Gkamas, Spyridon Aniceto Katsampiris Salgado, and Vasileios Kokkinos Proactive Network Slices Management Algorithm Based on Fuzzy Logic System and Support Vector Regression Model . . . . . . . . . . . . . . . 386 Amal Kammoun, Nabil Tabbane, Gladys Diaz, Nadjib Achir, and Abdulhalim Dandoush An Optimal Route Recommendation Method for a Multi-purpose Travel Route Recommendation System . . . . . . . . . . . . . . . . . . . . . . . . . 398 Chen Yuan and Minoru Uehara Artificial Intelligence Technique for Optimal Allocation of Renewable Energy Based DGs in Distribution Networks . . . . . . . . . . 409 Zia Ullah, M. R. Elkadeem, and Shaorong Wang Impact of Sharing Algorithms for Cloud Services Management . . . . . . . 423 Lidia Ogiela, Makoto Takizawa, and Urszula Ogiela Application of Cognitive Protocols in Transformative Computing . . . . . 428 Marek R. Ogiela and Lidia Ogiela Analyzing Mobile Cycling Applications for Monitoring Workouts . . . . . 433 Fabricio Landero Cristobal, Miguel A. Wister, and Pablo Payro Campos Road State Information Platform Based on Multi-sensors and Bigdata Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 445 Yoshitaka Shibata, Goshi Sato, and Noriki Uchida A New Discounting Approach to Conflict Information Fusion Using Multi-criteria of Reliability in Dempster-Shafer Evidence Theory . . . . . 455 Jin Zhu

Contents

xliii

The 21th International Symposium on Multimedia Network Systems and Applications (MNSA-2019) The Group-Based Linear Time Causally Ordering Protocol in a Scalable P2PPS System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 471 Takumi Saito, Shigenari Nakamura, Tomoya Enokido, and Makoto Takizawa Algorithm for Detecting Implicitly Faulty Replicas Based on the Power Consumption Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 483 Hazuki Ishii, Ryuji Oma, Shigenari Nakamura, Tomoya Enokido, and Makoto Takizawa Parallel Data Transmission Protocols in the Mobile Fog Computing Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 494 Kosuke Gima, Ryuji Oma, Shigenari Nakamura, Tomoya Enokido, and Makoto Takizawa Recovery of Fiber Networks C/M-Plane via an IoT-Based Narrow-Band Links System Based on LoRa Mesh Platform . . . . . . . . . 504 Goshi Sato, Yoshitaka Shibata, and Noriki Uchida Clustering Analysis and Visualization of TCM Patents Based on Deep Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 512 Na Deng, Xu Chen, and Caiquan Xiong Efficient Resource Utilization Using Blockchain Network for IoT Devices in Smart City . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 521 Muhammad Zohaib Iftikhar, Muhammad Sohaib Iftikhar, Muhammad Jawad, Annas Chand, Zain Khan, Abdul Basit Majeed Khan, and Nadeem Javaid Recommendation System Based on Deep Learning . . . . . . . . . . . . . . . . 535 Tianhan Gao, Lei Jiang, and Xibao Wang Routing Method Based on Data Transfer Path in DTN Environments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 544 Kazuma Ikenoue and Kazunori Ueda The 12th International Workshop on Next Generation of Wireless and Mobile Networks (NGWMN-2019) A Hybrid Intelligent Simulation System for Node Placement in WMNs Considering Load Balancing: A Comparison Study for Exponential and Normal Distribution of Mesh Clients . . . . . . . . . . . 555 Seiji Ohara, Heidi Durresi, Admir Barolli, Shinji Sakamoto, and Leonard Barolli

xliv

Contents

Multi-dimensional Contract Incentive Design for Mobile Crowdsourcing Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 570 Nan Zhao, Menglin Fan, Chao Tian, Pengfei Fan, and Xiao He Evaluation and Comparison of CO2 and Fuel Consumption for Different Car Following Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . 579 Ningling Jiang and Elis Kulla Individually Separated Wireless Access Point to Protect User’s Private Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 589 Myoungsu Kim and Kangbin Yim Long-Term Care (LTC) Monitoring System for Caregivers Based on Wireless Sensing Technology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 597 Hsing-Chung Chen, Mei-He Jiang, and Tzu-Ya Chen The 10th International Workshop on Methods, Analysis and Protocols for Wireless Communication (MAPWC-2019) Concatenated Path Domain for Dijkstra’s Algorithm Based Ray Tracing to Enhance Computational Areas . . . . . . . . . . . . . . . . . . . . . . . 609 Kazunori Uchida and Leonard Barolli Routing of Optical Baseband Signal Depending on Wavelength in Periodic Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 621 Naoki Higashinaka and Hiroshi Maeda Two-Stage Dynamic Contract Design Under Asymmetric Information in Cooperative Communication . . . . . . . . . . . . . . . . . . . . . 630 Nan Zhao, Pengfei Fan, Xiao He, Menglin Fan, and Chao Tian Minimizing Control Overhead of Routing Protocols in Wireless Multihop Networks: Simulation Evaluation . . . . . . . . . . . . . . . . . . . . . . 638 Soushi Morita and Elis Kulla Effect of Parasitic Element on Communication Performance of 13.56 MHz RFID System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 646 Kiyotaka Fujisaki and Yuki Yoshigai The 10th International Workshop on Cloud, Wireless and e-Commerce Security (CWECS-2019) Perception Mining of Network Protocol’s Stealth Attack Behaviors . . . . 657 Yan-Jing Hu and Xu An Wang Digital Image Anti-counterfeiting Technology . . . . . . . . . . . . . . . . . . . . 670 Chin-Ling Chen, Chin-Feng Lee, Fang-Wei Hsu, Yong-Yuan Deng, and Ching-Cheng Liu

Contents

xlv

System Implementation of AUSF Fault Tolerance . . . . . . . . . . . . . . . . . 678 Wei-Sheng Chen, Fang-Yie Leu, and Heru Susanto News Collection and Analysis on Public Political Opinions . . . . . . . . . . 688 Zhi-Qian Hong, Fang-Yie Leu, and Heru Susanto Mobile Physiological Sensor Cloud System for Long-Term Care . . . . . . 698 Ping-Jui Chiang, Heru Susanto, Fang-Yie Leu, and Hui-Ling Huang The 8th International Workshop on Robot and Vehicle Interaction, Control, Communication and Cooperation (RVI3C-2019) A Message Relaying Method with Enhanced Dynamic Timer Considering Decrease Rate of Neighboring Nodes for Vehicular-DTN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 711 Shogo Nakasaki, Makoto Ikeda, and Leonard Barolli Prediction of RSSI by Scikit-Learn for Improving Position Detecting System of Omnidirectional Wheelchair Tennis . . . . . . . . . . . . 721 Keita Matsuo and Leonard Barolli Decentralized Mechanism for Hiring the Smart Autonomous Vehicles Using Blockchain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 733 Zain Abubaker, Muhammad Usman Gurmani, Tanzeela Sultana, Shahzad Rizwan, Muhammad Azeem, Muhammad Zohaib Iftikhar, and Nadeem Javaid An Intelligent Approach for Resource Management in SDN-VANETs Using Fuzzy Logic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 747 Ermioni Qafzezi, Kevin Bylykbashi, Evjola Spaho, and Leonard Barolli Tutorial Educating Developer of Reinforcement Learning Agent Using IDEAL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 757 Takahiro Uchiya, Kodai Shimano, and Ichi Takumi The 5th International Workshop on Advanced Techniques and Algorithms for Security and Privacy (ATASP-2019) Trusted Remote Patient Monitoring Using Blockchain-Based Smart Contracts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 765 Hafiza Syeda Zainab Kazmi, Faiza Nazeer, Sahrish Mubarak, Seemab Hameed, Aliza Basharat, and Nadeem Javaid A Survey of Malicious HID Devices . . . . . . . . . . . . . . . . . . . . . . . . . . . . 777 Songyin Zhao and Xu An Wang Power Consumption Attack Based on Improved Principal Component Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 787 Zeyu Wang, Wei Zhang, Peng Ma, and Xu An Wang

xlvi

Contents

How Securely Are OAuth/OpenID Connect Implemented in Japan? . . . 800 Takamichi Saito, Tsubasa Kikuta, and Rikita Koshiba The 2nd International Workshop on Bio-Sensing, Processing, Application and Networking (BioSPAN-2019) Enhanced Decentralized Management of Patient-Driven Interoperability Based on Blockchain . . . . . . . . . . . . . . . . . . . . . . . . . . . 815 Asad Ullah Khan, Affaf Shahid, Fatima Tariq, Abdul Ghaffar, Abid Jamal, Shahid Abbas, and Nadeem Javaid Design and Construction of Intelligent Decision-Making System for Marine Protection and Law Enforcement . . . . . . . . . . . . . . . . . . . . . 828 Na Deng, Xu Chen, and Caiquan Xiong Data Authenticity Analysis for Online O2O Data: A Case Study of Second-Hand Houses Posting Data . . . . . . . . . . . . . . . . . . . . . . . . . . . 838 Xu Chen, Deliang Zhong, Yingzhou Zheng, Shudong Liu, Yipeng Li, and Na Deng A Brief Survey: 3D Face Reconstruction . . . . . . . . . . . . . . . . . . . . . . . . 846 Tianhan Gao and Hui An A Feasibility Study on Wrist Rehabilitation Using the Leap Motion . . . 855 Linlin Zhang and Kin Fun Li Classification of Cotton and Flax Fiber Images Based on Inductive Transfer Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 865 Yuhan Jiang, Song Cai, Chunyan Zeng, and Zhifeng Wang Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 873

The 14th International Conference on Broad-Band andd Wireless Computing, Communication and Applications (BWCCA-2019)

A Fuzzy-Based Decision System for Sightseeing Spots Considering Hot Spot Access as a New Parameter Yi Liu1(B) , Kevin Bylykbashi2 , and Leonard Barolli1 1 Department of Information and Communication Engineering, Fukuoka Institute of Technology (FIT), 3-30-1 Wajiro-Higashi, Higashi-Ku, Fukuoka 811-0295, Japan [email protected], [email protected] 2 Graduate School of Engineering, Fukuoka Institute of Technology (FIT), 3-30-1 Wajiro-Higashi, Higashi-Ku, Fukuoka 811-0295, Japan [email protected]

Abstract. Discovering and recommending points of interest are drawing more attention to meet the increasing demand from personalized tours. In this paper, we propose and evaluate a new fuzzy-based system for decision of sightseeing spots considering different conditions. In our system, we considered four input parameters: Ambient Temperature (AT), Air Quality (AQ), Noise Levle (NL) and Hot Spot Access (HSA) to decide the sightseeing spots Visit or Not Visit (VNV). We evaluate the proposed system by computer simulations. From the simulations results, we conclude that when the AT is normal, the VNN is the best. But when AQ and NL are increased, the VNV is decreased. Considering the effect of HSA parameter, we found that when HSA is increased, the VNV is increased. The simulation results have shown that the proposed system has a good performance and can choose good sightseeing spots.

1

Introduction

Social image hosting websites have recently become very popular. On these sites, users can upload and tag images for sharing their travelling experiences. The geotagged images are widely used in landmark recognitions and trip recommendations. Large amount of information generated from these location-based social services covers not only popular locations but also obscure ones. Since personalized tours are becoming popular, more attention is focusing on obscure sightseeing locations that are less well-known while still worth visiting. In Fig. 1 are show two dimensions of diverse sightseeing resources. The evaluation can be done using the sightseeing quality and popularity [1–5]. In this work, we use Fuzzy Logic (FL) for decision of sightseeing spots. The FL is the logic underlying modes of reasoning which are approximate rather then exact. The importance of FL derives from the fact that most modes of human reasoning and especially common sense reasoning are approximate in nature [6]. FL uses linguistic variables to describe the control parameters. By c Springer Nature Switzerland AG 2020  L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 3–11, 2020. https://doi.org/10.1007/978-3-030-33506-9_1

4

Y. Liu et al.

Fig. 1. Two dimensions of diverse sightseeing resources.

using relatively simple linguistic expressions it is possible to describe and grasp very complex problems. A very important property of the linguistic variables is the capability of describing imprecise parameters. The concept of a fuzzy set deals with the representation of classes whose boundaries are not determined. It uses a characteristic function, taking values usually in the interval [0, 1]. The fuzzy sets are used for representing linguistic labels. This can be viewed as expressing an uncertainty about the clear-cut meaning of the label. But important point is that the valuation set is supposed to be common to the various linguistic labels that are involved in the given problem. The fuzzy set theory uses the membership function to encode a preference among the possible interpretations of the corresponding label. A fuzzy set can be defined by examplification, ranking elements according to their typicality with respect to the concept underlying the fuzzy set [7]. In this paper, we propose and evaluate a fuzzy-based system for decision of sightseeing spots considering hot spot access as a new parameter. In our system, we considered four input parameters: Ambient Temperature (AT), Air Quality (AQ), Noise Levle (NL) and Hot Spot Access (HSA) to decide the output parameter Visit or Not Visit (VNV). The structure of this paper is as follows. In Sect. 2, we introduce FL used for control. In Sect. 3, we present the proposed fuzzy-based system. In Sect. 4, we discuss the simulation results. Finally, conclusions and future work are given in Sect. 5.

A Fuzzy-Based System for Decision of Sightseeing Spots

2

5

Application of Fuzzy Logic for Control

The ability of fuzzy sets and possibility theory to model gradual properties or soft constraints whose satisfaction is matter of degree, as well as information pervaded with imprecision and uncertainty, makes them useful in a great variety of applications [8–16]. The most popular area of application is Fuzzy Control (FC), since the appearance, especially in Japan, of industrial applications in domestic appliances, process control, and automotive systems, among many other fields. In the FC systems, expert knowledge is encoded in the form of fuzzy rules, which describe recommended actions for different classes of situations represented by fuzzy sets. In fact, any kind of control law can be modeled by the FC methodology, provided that this law is expressible in terms of “if ... then ...” rules, just like in the case of expert systems. However, FL diverges from the standard expert system approach by providing an interpolation mechanism from several rules. In the contents of complex processes, it may turn out to be more practical to get knowledge from an expert operator than to calculate an optimal control, due to modeling costs or because a model is out of reach. A concept that plays a central role in the application of FL is that of a linguistic variable. The linguistic variables may be viewed as a form of data compression. One linguistic variable may represent many numerical variables. It is suggestive to refer to this form of data compression as granulation. The same effect can be achieved by conventional quantization, but in the case of quantization, the values are intervals, whereas in the case of granulation the values are overlapping fuzzy sets. The advantages of granulation over quantization are as follows: • it is more general; • it mimics the way in which humans interpret linguistic values; • the transition from one linguistic value to a contiguous linguistic value is gradual rather than abrupt, resulting in continuity and robustness. FC describes the algorithm for process control as a fuzzy relation between information about the conditions of the process to be controlled, x and y, and the output for the process z. The control algorithm is given in “if ... then ...” expression, such as: If x is small and y is big, then z is medium; If x is big and y is medium, then z is big. These rules are called FC rules. The “if” clause of the rules is called the antecedent and the “then” clause is called consequent. In general, variables x and y are called the input and z the output. The “small” and “big” are fuzzy values for x and y, and they are expressed by fuzzy sets. Fuzzy controllers are constructed of groups of these FC rules, and when an actual input is given, the output is calculated by means of fuzzy inference.

6

Y. Liu et al.

AT AQ FVS

VNV

NL HSA Fig. 2. FVS structure.

3

Proposed Fuzzy-Based System

The proposed system stucture is show in Fig. 2. We call this system: Fuzzybased Visiting Spots (FVS) system. In this work, we consider four parameters: Ambient Temperature (AT), Air Quality (AQ), Noise Level (NL) and Hot Spot Access (HSA) to decide the sightseeing spots Visit or Not Visit (VNV). The AT is the temperature at the sightseeing spots. We use the air pollution data around sightseeing spots to decide the AQ. The NL is the amplitude level of the noise. For HSA, we consider the access by walk, train, bus, car and airplane. These four parameters are not correlated with each other, for this reason we use fuzzy system. The membership functions for our system are shown in Fig. 3. In Table 1, we show the fuzzy rule base of our proposed system, which consists of 135 rules. The input parameters for FVS are: AT, AQ, NL and HSA. The output linguistic parameter is VNV. The term sets of AT, AQ, NL and HSA are defined respectively as: AT = {Very Cold , Cold , Normal , Hot, Very Hot} = {V C, Co, N o, Ho, V H}; AQ = {Good , Normal , Bad} = {Good , N or, Bad}; N L = {Low, Middle, High} = {Lo, M i, Hi}; HSA = {Bad, Normal , Good } = {Bd, N, Gd}.

(1)

A Fuzzy-Based System for Decision of Sightseeing Spots Table 1. FRB. Rule AQ

AT NL HSA VHV Rule AQ AT NL HSA VHV Rule AQ AT NL HSA VHV

1

Good VC Lo Bd

VL2

46

Nor VC Lo Bd

VL1

91

Bad VC Lo Bd

VL1

2

Good VC Lo N

VL3

47

Nor VC Lo N

VL2

92

Bad VC Lo N

VL1

3

Good VC Lo Gd

VL5

48

Nor VC Lo Gd

VL4

93

Bad VC Lo Gd

VL3

4

Good VC Mi Bd

VL1

49

Nor VC Mi Bd

VL1

94

Bad VC Mi Bd

VL1

5

Good VC Mi N

VL2

50

Nor VC Mi N

VL1

95

Bad VC Mi N

VL1

6

Good VC Mi Gd

VL3

51

Nor VC Mi Gd

VL2

96

Bad VC Mi Gd

VL1

7

Good VC Hi Bd

VL1

52

Nor VC Hi Bd

VL1

97

Bad VC Hi Bd

VL1

8

Good VC Hi N

VL1

53

Nor VC Hi N

VL1

98

Bad VC Hi N

VL1

9

Good VC Hi Gd

VL2

54

Nor VC Hi Gd

VL1

99

Bad VC Hi Gd

10

Good C

Lo Bd

VL4

55

Nor C

Lo Bd

VL3

100 Bad C

Lo Bd

VL2

11

Good C

Lo N

VL6

56

Nor C

Lo N

VL5

101 Bad C

Lo N

VL3

12

Good C

Lo Gd

VL7

57

Nor C

Lo Gd

VL6

102 Bad C

Lo Gd

VL5

13

Good C

Mi Bd

VL3

58

Nor C

Mi Bd

VL2

103 Bad C

Mi Bd

VL1

14

Good C

Mi N

VL4

59

Nor C

Mi N

VL3

104 Bad C

Mi N

VL2

15

Good C

Mi Gd

VL6

60

Nor C

Mi Gd

VL5

105 Bad C

Mi Gd

VL3

16

Good C

Hi Bd

VL2

61

Nor C

Hi Bd

VL1

106 Bad C

Hi Bd

VL1

17

Good C

Hi N

VL3

62

Nor C

Hi N

VL2

107 Bad C

Hi N

VL1

18

Good C

Hi Gd

VL4

63

Nor C

Hi Gd

VL3

108 Bad C

Hi Gd

VL2

19

Good No Lo Bd

VL6

64

Nor No Lo Bd

VL5

109 Bad No Lo Bd

VL3

20

Good No Lo N

VL7

65

Nor No Lo N

VL6

110 Bad No Lo N

VL5

21

Good No Lo Gd

VL7

66

Nor No Lo Gd

VL7

111 Bad No Lo Gd

VL6

22

Good No Mi Bd

VL4

67

Nor No Mi Bd

VL3

112 Bad No Mi Bd

VL2

23

Good No Mi N

VL6

68

Nor No Mi N

VL5

113 Bad No Mi N

VL3

24

Good No Mi Gd

VL7

69

Nor No Mi Gd

VL6

114 Bad No Mi Gd

VL5

25

Good No Hi Bd

VL3

70

Nor No Hi Bd

VL2

115 Bad No Hi Bd

VL1

26

Good No Hi N

VL4

71

Nor No Hi N

VL3

116 Bad No Hi N

VL2

27

Good No Hi Gd

VL6

72

Nor No Hi Gd

VL5

117 Bad No Hi Gd

VL3

28

Good H

Lo Bd

VL4

73

Nor Hi Lo Bd

VL3

118 Bad H

Lo Bd

VL2

29

Good H

Lo N

VL6

74

Nor H

Lo N

VL5

119 Bad H

Lo N

VL3

30

Good H

Lo Gd

VL7

75

Nor H

Lo Gd

VL6

120 Bad H

Lo Gd

VL5

31

Good H

Mi Bd

VL3

76

Nor H

Mi Bd

VL2

121 Bad H

Mi Bd

VL1

32

Good H

Mi N

VL4

77

Nor H

Mi N

VL3

122 Bad H

Mi N

VL2

33

Good H

Mi Gd

VL6

78

Nor H

Mi Gd

VL5

123 Bad H

Mi Gd

VL3

34

Good H

Hi Bd

VL2

79

Nor H

Hi Bd

VL1

124 Bad H

Hi Bd

VL1

35

Good H

Hi N

VL3

80

Nor H

Hi N

VL2

125 Bad H

Hi N

VL1

36

Good H

Hi Gd

VL4

81

Nor H

Hi Gd

VL3

126 Bad H

Hi Gd

VL2

37

Good VH Lo Bd

VL2

82

Nor H

Lo Bd

VL1

127 Bad VH Lo Bd

VL1

38

Good VH Lo N

VL3

83

Nor VH Lo N

VL2

128 Bad VH Lo N

VL1

39

Good VH Lo Gd

VL5

84

Nor VH Lo Gd

VL4

129 Bad VH Lo Gd

VL3

40

Good VH Mi Bd

VL1

85

Nor VH Mi Bd

VL1

130 Bad VH Mi Bd

VL1

41

Good VH Mi N

VL2

86

Nor VH Mi N

VL1

131 Bad VH Mi N

VL1

42

Good VH Mi Gd

VL3

87

Nor VH Mi Gd

VL2

132 Bad VH Mi Gd

VL1

43

Good VH Hi Bd

VL1

88

Nor VH Hi Bd

VL1

133 Bad VH Hi Bd

VL1

44

Good VH Hi N

VL1

89

Nor VH Hi N

VL1

134 Bad VH Hi N

VL1

45

Good VH Hi Gd

VL2

90

Nor VH Hi Gd

VL1

135 Bad VH Hi Gd

VL1

VL1

7

8

Y. Liu et al. µ(AT)

VC

Co

No

Ho

VH

1

AT -50 µ(AQ)

-40

-30

-20

-10

0

Good

10

20

30

Nor

40

50

Bad

1

AQ 0 µ(NL)

10

20

30

40

50

60

Lo

70

80

90

100

Mi

Hi

1

NL 0

10

20

30

40

Bd

µ(HSA)

50

60

70

80

90

N

100 Gd

HSA 0

10

µ(VNV) VL1 1

20 VL2

30

40

VL3

50 VL4

60

70

VL5

80

90

VL6

100 VL7

VNV 0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Fig. 3. Membership functions.

and the term set for the output VNV is defined as: ⎞ ⎞ ⎛ ⎛ V L1 VisitLevel 1 ⎜ VisitLevel 2 ⎟ ⎜ V L2 ⎟ ⎟ ⎟ ⎜ ⎜ ⎜ VisitLevel 3 ⎟ ⎜ V L3 ⎟ ⎟ ⎟ ⎜ ⎜ ⎟ ⎟ ⎜ V NV = ⎜ ⎜ VisitLevel 4 ⎟ = ⎜ V L4 ⎟ . ⎜ VisitLevel 5 ⎟ ⎜ V L5 ⎟ ⎟ ⎟ ⎜ ⎜ ⎝ VisitLevel 6 ⎠ ⎝ V L6 ⎠ V L7 VisitLevel 7

A Fuzzy-Based System for Decision of Sightseeing Spots

4

9

Simulation Results

In this section, we present the simulation results for our proposed fuzzy-based system. In our system, we decided the number of term sets by carrying out many simulations. From Fig. 4, 5 and 6, we show the relation of VNV with AT, AQ, NL and HSA. In these simulations, we consider the NL and HSA as constant parameters. In Fig. 4, we consider NL value 10 units. We change the HSA value from 20 to 80 units. When the HSA increases, the VNV is increased. By increaseing AQ, the VNV is decreased. And when AT is normal, the VNV is the best. In Fig. 5 and Fig. 6, we change NL value to 50 and 90 units, respectively. We see that, when the NL increases, the VNV is decreased. NL=10,HSA=20

NL=10,HSA=80

1

1

AT=Ho AT=No AT=Co AT=VC AT=VH

0.9 0.8 0.7

0.9 0.8 0.7 0.6 VNV

VNV

0.6 0.5

0.5

0.4

0.4

0.3

0.3

AT=Ho 0.2 AT=No AT=Co 0.1 AT=VC AT=VH 0 0 10

0.2 0.1 0

0

10

20

30

40

50 AQ

60

70

80

90

100

20

(a) HSA=20

30

40

50 AQ

60

70

80

90

100

(b) HSA=80

Fig. 4. Relation of VNV with AT and AQ for different HSA when NL = 10. NL=50,HSA=20

NL=50,HSA=80

1

1

AT=Ho 0.9 AT=No AT=Co 0.8 AT=VC AT=VH 0.7

0.8 0.7 0.6 VNV

VNV

0.6 0.5

0.5

0.4

0.4

0.3

0.3

0.2

0.2

0.1

0.1

0

AT=Ho AT=No AT=Co AT=VC AT=VH

0.9

0

10

20

30

40

50 AQ

60

(a) HSA=20

70

80

90

100

0

0

10

20

30

40

50 AQ

60

70

80

90

100

(b) HSA=80

Fig. 5. Relation of VNV with AT and AQ for different HSA when NL = 50.

10

Y. Liu et al. NL=90,HSA=20

NL=90,HSA=80

1

1

AT=Ho 0.9 AT=No AT=Co 0.8 AT=VC AT=VH 0.7

0.8 0.7 0.6 VNV

VNV

0.6 0.5

0.5

0.4

0.4

0.3

0.3

0.2

0.2

0.1

0.1

0

AT=Ho AT=No AT=Co AT=VC AT=VH

0.9

0

10

20

30

40

50

60

70

80

90

100

0

AQ

(a) HSA=20

0

10

20

30

40

50

60

70

80

90

100

AQ

(b) HSA=80

Fig. 6. Relation of VNV with AT and AQ for different HSA when NL = 90.

5

Conclusions and Future Work

In this paper, we proposed a fuzzy-based system to decide the sightseeing spots. We took into consideration four parameters: AT, AQ, NL and HSA. We evaluated the performance of proposed system by computer simulations. From the simulations results, we conclude that when AQ and NL are increased, the VNV is decreased. When the AT is normal, the VNV is the best. But by increasing HSA, the VNV is increased. In the future, we would like to make extensive simulations to evaluate the proposed system and compare the performance of our proposed system with other systems.

References 1. Luo, J., Joshi, D., Yu, J., Gallagher, A.C.: Geotagging in multimedia and computer vision a survey. Multimedia Tools Appl. 51(1), 187–211 (2011) 2. Chenyi, Z., Qiang, M., Xuefeng, L., Masatoshi, Y.: An obscure sightseeing spots discovering system. In: 2014 IEEE International Conference on Multimedia and Expo (ICME), pp. 1–6. IEEE (2014) 3. Cheng, Z., Ren, J., Shen, J., Miao, H.: Building a large scale test collection for effective benchmarking of mobile landmark search. In: Advances in Multimedia Modeling. Springer, Berlin, , pp. 36–46 (2013) 4. Chen, W., Battestini, A., Gelfand, N., Setlur, V.: Visual summaries of popular landmarks from community photo collections. In: 2009 Conference Record of the Forty-Third Asilomar Conference on Signals, Systems and Computers, pp. 1248– 1255. IEEE (2009) 5. Cao, X., Cong, G., Jensen, C.S.: Mining significant semantic locations from GPS data. Proc. VLDB Endowment 3(1–2), 1009–1020 (2010) 6. Inaba, T., Obukata, R., Sakamoto, S., Oda, T., Ikeda, M., Barolli, L.: Performance evaluation of a QoS-aware fuzzy-based CAC for LAN access. Int. J. SpaceBased Situated Comput. 6(4), 228–238 (2016). https://doi.org/10.1504/IJSSC. 2016.082768

A Fuzzy-Based System for Decision of Sightseeing Spots

11

7. Terano, T., Asai, K., Sugeno, M.: Fuzzy Systems Theory and Its Applications. Academic Press, Inc. Harcourt Brace Jovanovich, Cambridge (1992) 8. Spaho, E., Kulla, E., Xhafa, F., Barolli, L.: P2P solutions to efficient mobile peer collaboration in MANETs. In: Proceedings of 3PGCIC 2012, pp. 379–383, November 2012 9. Kandel, A.: Fuzzy Expert Systems. CRC Press, Boca Raton (1992) 10. Zimmermann, H.J.: Fuzzy Set Theory and Its Applications. Kluwer Academic Publishers, Dordrecht (1991). Second Revised Edition 11. McNeill, F.M., Thro, E.: Fuzzy Logic. A Practical Approach. Academic Press Inc., Cambridge (1994) 12. Zadeh, L.A., Kacprzyk, J.: Fuzzy Logic for the Management of Uncertainty. Wiley, Hoboken (1992) 13. Procyk, T.J., Mamdani, E.H.: A linguistic self-organizing process controller. Automatica 15(1), 15–30 (1979) 14. Klir, G.J., Folger, T.A.: Fuzzy Sets, Uncertainty, and Information. Prentice Hall, Englewood Cliffs (1988) 15. Munakata, T., Jani, Y.: Fuzzy systems: an overview. Commun. ACM 37(3), 69–76 (1994) 16. Yi, L., Kouseke, O., Keita, M., Makoto, I., Leonard, B.: A fuzzy-based approach for improving peer coordination quality in mobilepeerdroid mobile system. In: Proceedings of IMIS 2018, pp. 60–73, July 2018

A Multi-sensor Based Physical Condition Estimator for Home Healthcare Toshiyuki Haramaki(&) and Hiroaki Nishino Division of Computer Science and Intelligent Systems, Faculty of Science and Technology, Oita University, Oita, Japan {haramaki,hn}@oita-u.ac.jp

Abstract. According to the WHO(World Health Organization) and UNSD (United Nations Statistics Division) definition, when the percentage of elderly people (65 years of age or older) in the population exceeds 7%, it becomes an “aging society”, if it exceeds 14%, it becomes an “aged society”, and if it exceeds 21%, it becomes a “super-aged society”. Some developed countries are becoming super-aged societies. In a super-aged society, there are various problems in medical services for health management. To solve these problems, it is desirable for all generations, including the elderly, to take the initiative to maintain their own health. In this paper, we propose a system aimed at every one of them actively managing their health. The system always monitors and accumulates the biological information of the subject using various contact or non-contact sensors. By analyzing these data in an integrated manner, the subject can easily recognize changes in the physical condition. And also, it promotes the provision of information to remote healthcare professionals when people receive healthcare at home.

1 Introduction In recent years, the world’s population is increasing year by year. At the same time, the proportion of the aging population is increasing in more and more countries. As society ages, the burden of medical expenses increases. According to the Universal Health Coverage (UHC) study, global healthcare costs for 2015 are estimated to be around $10 trillion and expected to increase to $20 trillion by 2040. The burden of medical expenses is painful for individuals as well, but it may be considered for the country to control the rise in medical expenses. In order to reduce the burden of medical expenses, it is important for the general public, especially elderly people, to actively maintain their health. Maintenance of health leads to reducing various burdens of medical treatment and brings benefits to both individuals and society. One of the means to achieve that is to review lifestyle habits that lead to illness. To date, many researchers and healthcare professionals have suggested lifestyles to maintain good health. However, ordinary people find it difficult to assess whether their lifestyle habits are appropriate for maintaining good health. For example, there is a method of taking a meal to check whether the eating habits are correct. It will be useful to measure how much energy and nutrients can be obtained. However, I think that it is necessary to observe whether the person’s age or daily © Springer Nature Switzerland AG 2020 L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 12–21, 2020. https://doi.org/10.1007/978-3-030-33506-9_2

A Multi-Sensor Based Physical Condition Estimator for Home Healthcare

13

activity level is also good with the energy and nutrients. To that end, we consider one of the important approaches to maintaining health, to be aware of both the daily diet and life. In addition, another approach to maintaining health is to work on treatment and life improvement while the symptoms are mild by detecting the signs of the disease in advance before progressing to a serious illness. This approach also has the advantage of reducing the economic, physical and psychological burden of medical treatment by detecting and treating the disease early. However, in order to realize these, it is necessary to continue to check vital data that is an indicator of health on a regular basis. That could even increase the burden on consumers. However, even if the burden on consumers increases slightly, I think that maintaining the health by continuing to check my condition on a regular basis has great benefits for both individuals and society. As it turns out, the best way to reduce the risk of causing illness is to continue the lifestyle of maintaining good health every day and to detect early whether it is a sign of illness or not from unusual circumstances Think. As a method of detecting signs of illness, for example, the state of accumulation and distraction of the fatigue of the consumer, change in physical condition or change in mood, rapid fluctuation of biological data, etc. are important in determining the health status of the consumer. I think that it is a judgment material. In order to maintain health and detect signs of illness, we believe that it is necessary to comprehensively detect the status of consumers in all areas and judge the degree of risk based on them. In these researches on health maintenance and disease detection, various researches and developments have been conducted for the purpose of monitoring the condition of patients who are already suffering from a disease. Many medical institutions have already reduced the burden on medical professionals by actually using this equipment to monitor patients. At present, research and development is also underway on devices that monitor biological information of patients in remote locations in real time. However, most of the sensing devices used in these systems need to be worn directly on the human body. For healthy people who are not yet ill, it is considered that they are resistant to wearing such devices in their lives. In addition, there is a risk that the quality of life (quality of life) will be degraded by the equipment of those devices. Therefore, it is necessary to develop a method to acquire the behavior and vital data of the living person without attaching the device directly, so that the health condition of the person living normally can be monitored indirectly. In order to solve those various problems at once, we propose to use non-contact bio-information sensing comprehensively. Our proposed system prepares a smart home where multiple sensors are installed inside the house. When the subject lives in the interior, the subject’s body movements, facial expressions, voice, etc., non-contact biological information that can be acquired without contact, indoor temperature changes, humidity changes, air component changes, etc. It is hoped that they will be able to estimate their health status with a certain degree of accuracy from such information. Although these partial studies have already been conducted, I believe that they have not achieved a reliable system. It is thought that there is a high novelty in searching for a method to use for estimating health conditions by integrating and using multiple observation data and exploring possibilities

14

T. Haramaki and H. Nishino

2 Related Work Many research projects have been conducted on how to look at people to realize home health care. In research on methods of acquiring human activity data and biological data using sensors, various methods are used as means for acquiring data. Heretofore, sensors generally worn on the subject’s body have been used. Those sensors have the function to connect to the Internet, and by using such IoT devices, sensor data is collected in the server. Research is also in progress to determine the state of the observation target by analyzing those data. In the study of home health care, it is divided into four parts considered to be important. The first is research on how to acquire human activity data and biological data using sensors. There are many studies and practical applications of wearable devices that can be worn on the human body. There are also various researches on how to present information to observation objects. Next is the research on how to collect and analyze those acquired data. Furthermore, the question is who and how to present the information obtained as a result of analyzing the data. Finally, it is about the research to realize the reuse of knowledge using the big data and information obtained by those mechanisms. So far, we have been working on research to provide useful information for safe driving by keeping watch over the people driving the car by various sensors [1–3]. Figure 1 is an image of a study in which a sensor was installed in the car to observe the movement of a driver or a car.

Fig. 1. The sensors and a controller installed in the vehicle.

There are various researches to realize human health by observing people and the environment. Gao et al. propose to use wearable accelerometer data to predict human personality characteristics [4]. Lowe et al. survey technologies for monitoring human health behavior in the living environment [5]. Arshad et al. introduce recent advances in health monitoring systems [6]. Mshali et al. survey on health monitoring system using health smart home mechanism [7]. Chaaraoui et al. present related technologies

A Multi-Sensor Based Physical Condition Estimator for Home Healthcare

15

and services on how to implement Health Monitoring System (HMS) with Health Smart Home [8]. Dias et al. introduce vital sign monitoring techniques and systematization using wearable health devices (WHDs) [9]. Majumder et al. survey several low-cost and non-invasive health and activity monitoring systems and textile-based sensors [10]. Abuella et al. propose a wireless vital sign monitoring system using visible light sensing (VLS) [11]. Li et al. propose a way to detect lifestyle by wearing a wearable device that monitors the acoustic activity of the oral cavity [12]. There are many researches on telemedicine that aim to realize advanced services by using sensors and networks. Kakria et al. propose an online interactive telemedicine system that monitors the heart in real time [13]. Malasinghe et al. provide a review of recent advances in remote health care and monitoring in both contactless and contactless methods. [14]. Ha et al. proposes wearable sensors based on functional materials with a unique sensing function for detecting vital signs [15]. Tamura et al. introduce various types of wearable thermometers such as touch, patch and invisible (radiometric analysis) [16]. Wang et al. propose a non-contact infrared thermometer (NCIT) that measures the temperature of children quickly and accurately. [17]. AlHamry et al. health monitoring of human respiratory health with graphene oxide-based sensors [18]. Adib et al. propose a smart home sensor that monitors the respiration and heart rate of indoor people using radio waves. [19]. Güntner et al. study the current status and selection of respiratory markers and respiratory sampling to realize advanced respiratory sensors. [20]. Khan et al. propose a method for accurately measuring the heart rate of people in the room using passive Wi-Fi sensing [21]. We will use the findings from these studies to build a system to monitor people living and dining in the house. The method to watch is the observation data obtained from various sensors, and the goal is to construct a system that detects slight changes that may impair health by analyzing those data. Then, based on the detected results, provide advice and warnings to the target person. If this system detects an indoor accident, contact not only the person but also the family, caregivers, security companies and medical institutions. The mechanism for observing the health condition of the target person and presenting advice to promote a healthy life by observing the living space of the target person proposed in this study is called Ambient Assisted Living (AAL) in recent years. Bygholm et al. direct methodological behavior that should be considered in AAL studies [22]. Nawaz et al. propose sensor edge computing that realizes sensor level security and privacy for use in AAL [23]. Lloret et al. introduce intelligent communication architecture use in AAL [24]. Almeida et al. propose an inconspicuous system to monitor the behavior of the elderly and detect changes [25]. Shulman et al. introduce the Pregnancy Risk Assessment Monitoring System (PRAMS) and how to use it [26]. Polsky et al. recommend continuous glucose monitoring (CGM) for pregnant women [27]. Wang et al. show that using remote cardiac monitoring of fetuses in late pregnancy increases the detection rate of neonatal asphyxia [28]. El Murabet et al. report the results of a survey of AAL system requirements and implementation issues [29]. Julia et al. survey on the potential needs of the technologies that will enable AAL and the technologies [30]. With the initiative, researchers and companies around the world are working towards realization. Figure 2 is a diagram showing components of the AAL. In this system, the daily life performed by the target person in the living space is measured by a smart home using various sensors. As a result, data on daily lifestyles, movements

16

T. Haramaki and H. Nishino

and voices of the target person is acquired. Those data are sent to a hospital, a family, a carer, etc. through a network, and not only the system but also the mechanism that people of various positions and various devices watch together.

Fig. 2. Image of ambient assist living

3 System Design In this chapter, we introduce the design of a system that has the function of observing consumers with passive sensors and the function of advising consumers using the observation results. As an observation by a passive sensor, a method of analyzing the shape of a person photographed by a camera and detecting the motion of the hand and foot has already been put to practical use. In our research, we use sensors to track how people move and how they are changing. Multiple microphones are used to accurately grasp the movement of the living person. By using this, it is possible to obtain the distance between the sensor and the observation object, the positional relationship, and the relative velocity. Furthermore, in recent years, research on vital sensing that measures human heart rate and respiration using radar sensors is underway. Using an optical camera to detect human movement and facial expression, to detect temperature with a far-infrared sensor, to obtain voice data and living sound using a microphone, and to analyze components in the air in the room with a gas detection sensor.

A Multi-Sensor Based Physical Condition Estimator for Home Healthcare

3.1

17

Data Collection Using Non-wearable Sensors with No Waves

In the present system, the sensor used to acquire action data and biological data of the subject is not worn by the subject. And those sensors do not emit waves and observe their reflected waves. The waves are sound, light, electromagnetic waves, etc., and they are not used at all in this system. All sensors used in this system are only passive sensors that capture what originates from the subject. Components that make up the system include a camera and a microphone placed in the living space, a thermography, a sensor that observes the component of air in the living room, a pressure sensor, a potential sensor, and the on/off of the room. And all kinds of data such as body movement and sounds, temperature and humidity of each living room. In this system, sensors are installed in all rooms except private limited areas. The system aggregates observations obtained by those sensors into one information integration server. The information integration server analyzes the obtained data in real time to determine whether it is in a state of high urgency such as a serious accident or disease onset. If there is a high possibility of a high degree of urgency, immediately notify in advance (Fig. 3).

Fig. 3. Home healthcare support system by non-wearable sensors with no waves.

3.2

Get Information Using Collected Sensor Data

This system consolidates the observation values obtained by the sensors described in the previous section into one information integration server. The information integration server analyzes the obtained data in real time. As a result of analysis, it is judged whether it is in a state of high urgency such as a serious accident or the onset of a disease. If it is determined that the system is highly likely to be in a state of high urgency, a preset notification is issued immediately. If the level of urgency is not high, acquired data is sent to the database on the cloud through the network after securing anonymity. By analyzing sensor data acquired by this system collected on a database

18

T. Haramaki and H. Nishino

using a general rule base, we aim to detect the occurrence of an accident immediately and also detect signs of illness. In addition, by accumulating daily life, by reading unusual movements, voices, and atmospheres, and aiming at detecting changes in the body and mind by noticing minute changes, it is also aimed. Based on the analysis results, this system provides the subject with appropriate advice. In addition, in the case of an emergency such as an accident or a sudden illness, a system is also planned to realize notification immediately to an appropriate emergency contact. The system also analyzes the data and estimates the health hazards. And we aim to control the possibility of getting sick by improving the lifestyle based on the estimation. The system also observes whether the subject has changed their lifestyle after the advice. The system learns the results by evaluating on its own whether the advice is appropriate and continues to improve to give better advice. 3.3

Safe Driving Support Including Driver Monitoring by Radar

Figure 4 shows the components of the system that observes the safety of consumers and provides appropriate advice for a healthy life by observing the observation target with the sensor installed in the smart home, and the processing flow. This system determines whether the subject is in a dangerous state based on data acquired from a plurality of sensors installed in each room of the smart home. This determination uses a rule-based algorithm or a trained neuron. If it is determined to be in danger, notify appropriate contacts. If it is judged that the condition is not dangerous, the observation data is output to the server, and the server side predicts signs of poor health and illness. At the same time, provide advice for the subject to lead a healthy life. The advice is given to the user and the family in another home through an interface device or device installed at home. We improve the effectiveness of our advice through these processes.

Fig. 4. System components and processing flow

A Multi-Sensor Based Physical Condition Estimator for Home Healthcare

19

4 Preliminary Experiments In order to realize this system, we performed preliminary experiments. The sensors used in this experiment were two common webcams and two microphones. The videos acquired from the webcam detected human body movements and expressions. Two microphones were used to capture voice and life sounds. The data observed by these sensors was continuously transmitted to the information aggregation server, and experiments were conducted while observing the data in real time. As sensors, Raspberry Pi 3B and Arduino Uno were used. We used Nvidia ADG Jetson Xavier as the information aggregation server. We connected Grove Base HAT to the Raspberry Pi. And we connected various Grove sensors to the HAT. Figure 5 shows the HAT with Grove sensors. It connects Light sensor, Sound sensor, Multichannel Gas Sensor (CO, NO2, H2, NH3, CH4), Ultrasonic Ranger, PIR Motion Sensor, Temperature & Humidity Sensor, LiDAR (Light Detection and Ranging) Sensor, GSR (Galvanic Skin Response) Sensor, and ear clip type heart rate monitor. We have created a program that continuously acquires data from these sensors. We also created a program that displays the acquired data on the console and a program that writes it to a file. In addition, we have created a simple crisis detection routine and data analysis and advice routine program as shown in the design. As an interface device for advice, we created a program to output to the communication robot Sota, in-vehicle display, speakers, etc.

Fig. 5. Raspberry Pi & Grove Base HAT with Grove sensors

5 Conclusions In this paper, healthy and safe life support aimed at reducing the number of fatalities due to household accidents and sudden illness by monitoring the target person in the living space and providing appropriate support and information presentation according

20

T. Haramaki and H. Nishino

to the situation. I proposed a system. In this research, by using non-contact and nonwave generating sensors as a monitoring method of the target person, it is ensured that the quality of life of the target person is not degraded and the maximum safety. Compared with conventional monitoring by wearable sensors and radars, it is difficult to precisely identify the physical condition of the body because it can’t be expected for accurate acquisition of biological data. However, if the subject is normally healthy and does not have chronic illness, we think that loose observation without wearable sensors is preferable. For example, when it is judged that there is a problem in health condition, it is desirable to shift to acquiring and observing detailed biological data. By observing the subject’s life for a long time using these sensors, it is possible to identify the subject’s health and mental condition and provide appropriate support for improving the quality of life. Think of it. Furthermore, I think that being able to usually acquire the health condition of the target person using sensors leads to leading correlation of life cycle and health. In addition, I think that it leads to the improvement of the accuracy which predicts the possibility that the accident and sudden illness occur in the life.

References 1. Haramaki, T., Yatsuda, A., Nishino, H.: A robot assistant in an edge-computing-based safe driving support system. In: Proceedings of the 21st International Conference on NBiS 2018, pp. 144–155 (2018) 2. Okazaki, S., Haramaki, T., Nishino, H.: A safe driving support method using olfactory stimuli. In: Proceedings of the 12th International Conference on CISIS 2018, pp. 958–967 (2018) 3. Haramaki, T., Nishino, H.: An improved safe driving training system based on learning of driving behaviors. In: Proceedings of IEEE International Conference on 2019 ICCE-TW (2019) 4. Gao, N., Shao, W., Salim, F.D.: Predicting personality traits from physical activity intensity. Int. J. IEEE Comput. 52(7), 47–56 (2019) 5. Lowe, S.A., Ólaighin, G.: Monitoring human health behaviour in one’s living environment: a technological review. Int. J. Elsevier Med. Eng. Phys. 36(2), 147–168 (2014) 6. Arshad, A., Khan, S., Alam, A.Z., Ahmad, F.I., Tasnim, R.: A study on health monitoring system: recent advancements. IIUM Eng. J. 15(2), 87–99 (2014) 7. Mshali, H., Lemlouma, T., Moloney, M., Magoni, D.: A survey on health monitoring systems for health smart homes. Int. J. Ind. Ergonomics 66, 26–56 (2018) 8. Chaaraoui, A.A., Padilla-López, J.R., Ferrández-Pastor, F.J., Nieto-Hidalgo, M., FlórezRevuelta, F.: A vision-based system for intelligent monitoring: human behaviour analysis and privacy by context. Sensors 14(5), 8895–8925 (2014). (Basel, Switzerland) 9. Dias, D., Paulo Silva Cunha, J.: Wearable health devices-vital sign monitoring, systems and technologies. Sensors 18(8), 2414 (2018). (Basel, Switzerland) 10. Majumder, S., Mondal, T., Deen, M.J.: Wearable sensors for remote health monitoring. Sensors 17(1), 130 (2017) 11. Abuella, H., Ekin, S.: Wireless vital signs monitoring system using visible light sensing (VLS). Signal Processing (arXiv.org), arXiv:1807.05408v1 (2018) 12. Li, Y., Lu, W., He, Y., Dang, J., Chen, S.: Chewing monitoring with bone-conduction microphone for body area network. IEEE Access 10, 1109 (2019)

A Multi-Sensor Based Physical Condition Estimator for Home Healthcare

21

13. Kakria, P., Tripathi, N.K., Kitipawang, P.: A real-time health monitoring system for remote cardiac patients using smartphone and wearable sensors. Int. J. Telemed. Appl. 2015(373474) (2015) 14. Malasinghe, L.P., Ramzan, N., Dahal, K.: Remote patient monitoring: a comprehensive study. J. Ambient Intell. Humanized Comput. 10(1), 57–76 (2019) 15. Ha, M., Lima, S., Ko, H.: Wearable and flexible sensors for user-interactive healthmonitoring devices. J. Mater. Chem. B R. Soc. Chem. 2018(6), 4043–4064 (2018) 16. Tamura, T., Huang, M., Togawa, T.: Current developments in wearable thermometers. J. Adv. Biomed. Eng. 2018(7), 88–99 (2018) 17. Wang, K., Gill, P., Wolstenholme, J., Price, C.P., Heneghan, C., Thompson, M., Plüddemann, A.: Non-contact infrared thermometers for measuring temperature in children: primary care diagnostic technology update. Br. J. Gen. Pract.: J. R. College Gen. Pract. 64(627), e681–e683 (2014) 18. Al-Hamry, A., Panzardi, E., Mugnaini, M., Kanoun, O.: Health monitoring of human breathing by graphene oxide based sensors. In: Proceedings of the 19th ITG/GMASymposium, Sensors and Measuring Systems (2018) 19. Adib, F., Mao, H., Kabelac, Z., Katabi, D., Miller, R.C.: Smart homes that monitor breathing and heart rate. In: Proceedings of the 33rd Annual ACM Conference on Human Factors in Computing Systems, pp. 837–846 (2015) 20. Güntner, A.T., Abegg, S., Königstein, K., Gerber, P.A., Schmidt-Trucksäss, A., Pratsinis, S.E.: Breath sensors for health monitoring. J. ACS Sens. 4, 268–280 (2019) 21. Khan, U.M., Kabir, Z., Hassan, S.A.: Wireless health monitoring using passive wifi sensing. Computer Science (arXiv.org) arXiv:1704.00620v1 (2017) 22. Bygholm, A., Kanstrup, A.M.: The living challenge of ambient assisted living - a literature review. In: Proceedings of the 13th Scandinavian Conference on Health Informatics, pp. 89– 92 (2015) 23. Nawaz, T., Rinner, B., Ferryman, J.: User-centric, embedded vision-based human monitoring: a concept and a healthcare use case. In: Proceedings of the 10th International Conference on Distributed Smart Cameras (2016) 24. Lloret, J., Canovas, A., Sendra, S., Parra, L.: A smart communication architecture for ambient assisted living. IEEE Commun. Mag. 53(1), 26–33 (2015) 25. Almeida, A., Mulero, R., Ramett, P., Urošević, V., Andrić, M., Patrono, L.: A critical analysis of an IoT - aware AAL system for elderly monitoring. J. Future Gener. Comput. Syst. 97, 598–619 (2019) 26. Shulman, H.B., D’Angelo, V.D., Harrison, L., Smith, R.A., Warner, L.: The pregnancy risk assessment monitoring system (PRAMS): overview of design and methodology. Am. J. Public Health 108(10), 1305–1313 (2018) 27. Polsky, S., Garcetti, R.: CGM, pregnancy, and remote monitoring. J. Diab. Technol. Ther. 19(3), S49–S59 (2017) 28. Wang, Q., Yang, W., Li, L., Yan, G., Wang, H., Li, J.: Late pregnancy analysis with Yunban’s remote fetal monitoring system. Int. J. Distrib. Sens. Netw. 15(3), 1–7 (2019) 29. El Murabet, A., Abtoy, A., Touhafi, A., Tahiri, A.: Ambient assisted living system’s models and architectures: a survey of the state of the art. J. King Saud Univ. – Comput. Inf. Sci. (2018) 30. Julia, O.H., Schomakers, E.M., Ziefle, M.: Bare necessities? How the need for care modulates the acceptance of ambient assisted living technologies. Int. J. Med. Inf. 127, 147– 156 (2019)

Performance Evaluation of WMNs WMN-PSOHC System Considering Constriction and Linearly Decreasing Inertia Weight Replacement Methods Shinji Sakamoto1(B) , Seiji Ohara2 , Leonard Barolli3 , and Shusuke Okamoto1 1

Department of Computer and Information Science, Seikei University, 3-3-1 Kichijoji-Kitamachi, Musashino-shi, Tokyo 180-8633, Japan [email protected], [email protected] 2 Graduate School of Engineering, Fukuoka Institute of Technology, 3-30-1 Wajiro-Higashi, Higashi-Ku, Fukuoka 811-0295, Japan [email protected] 3 Department of Information and Communication Engineering, Fukuoka Institute of Technology, 3-30-1 Wajiro-Higashi, Higashi-Ku, Fukuoka 811-0295, Japan [email protected]

Abstract. Wireless Mesh Networks (WMNs) have many advantages such as low cost and increased high-speed wireless Internet connectivity, therefore WMNs are becoming an important networking infrastructure. In our previous work, we implemented a Particle Swarm Optimization (PSO) based simulation system for node placement in WMNs, called WMN-PSO. Also, we implemented a simulation system based on Hill Climbing (HC) for solving node placement problem in WMNs, called WMN-HC. Then, we implemented a hybrid simulation system based on PSO and HC, called WMN-PSOHC. In this paper, we analyze the performance of WMNs by using WMN-PSOHC considering Constriction Method (CM) and Linearly Decreasing Inertia Weight Method (LDIWM). Simulation results show that a good performance is achieved for CM compared with LDIWM.

1

Introduction

The wireless networks and devices are becoming increasingly popular and they provide users access to information and communication anytime and anywhere [3,4,6,9,10,13,15,16,21,24,26]. Wireless Mesh Networks (WMNs) are gaining a lot of attention because of their low cost nature that makes them attractive for providing wireless Internet connectivity. A WMN is dynamically self-organized and self-configured, with the nodes in the network automatically establishing and maintaining mesh connectivity among them-selves (creating, in effect, an ad hoc network). This feature brings many advantages to WMNs such as low up-front cost, easy network maintenance, robustness and reliable service c Springer Nature Switzerland AG 2020  L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 22–31, 2020. https://doi.org/10.1007/978-3-030-33506-9_3

Performance Evaluation of WMNs by WMN-PSOHC System

23

coverage [1]. Moreover, such infrastructure can be used to deploy community networks, metropolitan area networks, municipal and corporative networks, and to support applications for urban areas, medical, transport and surveillance systems. We consider the version of the mesh router nodes placement problem in which we are given a grid area where to deploy a number of mesh router nodes and a number of mesh client nodes of fixed positions (of an arbitrary distribution) in the grid area. The objective is to find a location assignment for the mesh routers to the cells of the grid area that maximizes the network connectivity and client coverage. Node placement problems are known to be computationally hard to solve [11,12,30]. In some previous works, intelligent algorithms have been recently investigated [5,8,14,19,22,23]. In [20], we implemented a Particle Swarm Optimization (PSO) based simulation system, called WMN-PSO. Also, we implemented a simulation system based on Hill Climbing (HC) for solving node placement problem in WMNs, called WMN-HC [18]. In our previous work, we implemented a hybrid simulation system based on PSO and HC. We called this system WMN-PSOHC. In this paper, we analyze the performance of hybrid WMN-PSOHC system considering Constriction Method (CM) and Linearly Decreasing Inertia Weight Method (LDIWM). The rest of the paper is organized as follows. The mesh router nodes placement problem is defined in Sect. 2. We present our designed and implemented hybrid simulation system in Sect. 3. The simulation results are given in Sect. 4. Finally, we give conclusions and future work in Sect. 5.

2

Node Placement Problem in WMNs

For this problem, we have a grid area arranged in cells we want to find where to distribute a number of mesh router nodes and a number of mesh client nodes of fixed positions (of an arbitrary distribution) in the considered area. The objective is to find a location assignment for the mesh routers to the area that maximizes the network connectivity and client coverage. Network connectivity is measured by Size of Giant Component (SGC) of the resulting WMN graph, while the user coverage is simply the number of mesh client nodes that fall within the radio coverage of at least one mesh router node and is measured by Number of Covered Mesh Clients (NCMC). An instance of the problem consists as follows. • N mesh router nodes, each having its own radio coverage, defining thus a vector of routers. • An area W × H where to distribute N mesh routers. Positions of mesh routers are not pre-determined and are to be computed. • M client mesh nodes located in arbitrary points of the considered area, defining a matrix of clients.

24

S. Sakamoto et al.

It should be noted that network connectivity and user coverage are among most important metrics in WMNs and directly affect the network performance. In this work, we have considered a bi-objective optimization in which we first maximize the network connectivity of the WMN (through the maximization of the SGC) and then, the maximization of the NCMC. In fact, we can formalize an instance of the problem by constructing an adjacency matrix of the WMN graph, whose nodes are router nodes and client nodes and whose edges are links between nodes in the mesh network. Each mesh node in the graph is a triple v = representing the 2D location point and r is the radius of the transmission range. There is an arc between two nodes u and v, if v is within the transmission circular area of u.

3 3.1

Proposed and Implemented Simulation System WMN-PSOHC Hybrid Simulation System

3.1.1 Particle Swarm Optimization In Particle Swarm Optimization (PSO) algorithm, a number of simple entities (the particles) are placed in the search space of some problem or function and each evaluates the objective function at its current location. The objective function is often minimized and the exploration of the search space is not through evolution [17]. However, following a widespread practice of borrowing from the evolutionary computation field, in this work, we consider the bi-objective function and fitness function interchangeably. Each particle then determines its movement through the search space by combining some aspect of the history of its own current and best (best-fitness) locations with those of one or more members of the swarm, with some random perturbations. The next iteration takes place after all particles have been moved. Eventually the swarm as a whole, like a flock of birds collectively foraging for food, is likely to move close to an optimum of the fitness function. Each individual in the particle swarm is composed of three D-dimensional vectors, where D is the dimensionality of the search space. These are the current position xi , the previous best position pi and the velocity vi . The particle swarm is more than just a collection of particles. A particle by itself has almost no power to solve any problem; progress occurs only when the particles interact. Problem solving is a population-wide phenomenon, emerging from the individual behaviors of the particles through their interactions. In any case, populations are organized according to some sort of communication structure or topology, often thought of as a social network. The topology typically consists of bidirectional edges connecting pairs of particles, so that if j is in i’s neighborhood, i is also in j’s. Each particle communicates with some other particles and is affected by the best point found by any member of its topological neighborhood. This is just the vector pi for that best neighbor, which we will denote with pg . The potential kinds of population “social networks” are hugely varied, but in practice certain types have been used more frequently.

Performance Evaluation of WMNs by WMN-PSOHC System

25

In the PSO process, the velocity of each particle is iteratively adjusted so that the particle stochastically oscillates around pi and pg locations. 3.1.2 Hill Climbing Hill Climbing (HC) algorithm is a heuristic algorithm. The idea of HC is simple. In HC, the solution s  is accepted as the new current solution if δ ≤ 0 holds, where δ = f (s ) − f (s). Here, the function f is called the fitness function. The fitness function gives points to a solution so that the system can evaluate the next solution s  and the current solution s. The most important factor in HC is to define the neighbor solution, effectively. The definition of the neighbor solution affects HC performance directly. In our WMN-PSOHC system, we use the next step of particle-pattern positions as the neighbor solutions for the HC part. 3.1.3 WMN-PSOHC System Description In following, we present the initialization, particle-pattern, fitness function and router replacement methods. Initialization Our proposed system starts by generating an initial solution randomly, by ad hoc methods [31]. We decide the velocity of particles by a random process considering the area size. For√instance, when √ the area size is W × H, the velocity is decided randomly from − W 2 + H 2 to W 2 + H 2 . Particle-Pattern A particle is a mesh router. A fitness value of a particle-pattern is computed by combination of mesh routers and mesh clients positions. In other words, each particle-pattern is a solution as shown is Fig. 1. Therefore, the number of particle-patterns is a number of solutions. Fitness Function One of most important thing is to decide the determination of an appropriate objective function and its encoding. In our case, each particle-pattern has an own fitness value and compares other particle-patterns fitness value in order to share information of global solution. The fitness function follows a hierarchical approach in which the main objective is to maximize the SGC in WMN. Thus, we use α and β weight-coefficients for the fitness function and the fitness function of this scenario is defined as: Fitness = α × SGC(x i j , y i j ) + β × NCMC(x i j , y i j ). Router Replacement Methods A mesh router has x, y positions and velocity. Mesh routers are moved based on velocities. There are many router replacement methods in PSO field [7,27–29]. In this paper, we consider CM and LDIWM.

26

S. Sakamoto et al.

Fig. 1. Relationship among global solution, particle-patterns and mesh routers.

Constriction Method (CM) CM is a method which PSO parameters are set to a week stable region (ω = 0.729, C1 = C2 = 1.4955) based on analysis of PSO by Clerc et. al. [2,7]. Linearly Decreasing Inertia Weight Method (LDIWM) In LDIWM, C1 and C2 are set to 2.0, constantly. On the other hand, the ω parameter is changed linearly from unstable region (ω = 0.9) to stable region (ω = 0.4) with increasing of iterations of computations [25,29].

3.2

WMN-PSOHC Web GUI Tool

The Web application follows a standard Client-Server architecture and is implemented using LAMP (Linux + Apache + MySQL + PHP) technology (see Fig. 2). We show the WMN-PSOHC Web GUI tool in Fig. 3. Remote users (clients) submit their requests by completing first the parameter setting. The parameter values to be provided by the user are classified into three groups, as follows. • Parameters related to the problem instance: These include parameter values that determine a problem instance to be solved and consist of number of router nodes, number of mesh client nodes, client mesh distribution, radio coverage interval and size of the deployment area. • Parameters of the resolution method: Each method has its own parameters.

Fig. 2. System structure for web interface.

Performance Evaluation of WMNs by WMN-PSOHC System

27

Fig. 3. WMN-PSOHC web GUI tool.

• Execution parameters: These parameters are used for stopping condition of the resolution methods and include number of iterations and number of independent runs. The former is provided as a total number of iterations and depending on the method is also divided per phase (e.g., number of iterations in a exploration). The later is used to run the same configuration for the same problem instance and parameter configuration a certain number of times.

Table 1. Parameter settings. Parameters

Values

Clients distribution

Normal distribution

Area size

32.0 × 32.0

Number of mesh routers

16

Number of mesh clients

48

Total iterations

800

Iteration per phase

4

Number of particle-patterns

9

Radius of a mesh router

2.0

Fitness function weight-coefficients (α, β) 0.7, 0.3 Replacement method

CM, LDIWM

S. Sakamoto et al.

100

Distribution of Clients=Normal distribution Replacement Method=CM Iteration per Phase=4 Total Iterations=800 Number of Particle-patterns=9

Best Solutions Average of Best Solutions

Giant Component[%]

Giant Component[%]

28

100

75 50 25

0

50

100

Number of Phases

150

Distribution of Clients=Normal distribution Replacement Method=LDIWM Iteration per Phase=4 Total Iterations=800 Number of Particle-patterns=9

Best Solutions Average of Best Solutions

75 50 25

200

0

50

(a) CM

100

Number of Phases

150

200

(b) LDIWM

100

Distribution of Clients=Normal distribution Replacement Method=CM Iteration per Phase=4 Total Iterations=800 Number of Particle-patterns=9

Best Solutions Average of Best Solutions

Number of Covered Mesh Clients[%]

Number of Covered Mesh Clients[%]

Fig. 4. Simulation results of WMN-PSOHC for SGC.

75 50 25

0

50

100

Number of Phases

150

(a) CM

200

100

Distribution of Clients=Normal distribution Replacement Method=LDIWM Iteration per Phase=4 Total Iterations=800 Number of Particle-patterns=9

Best Solutions Average of Best Solutions

75 50 25

0

50

100

Number of Phases

150

200

(b) LDIWM

Fig. 5. Simulation results of WMN-PSOHC for NCMC.

4

Simulation Results

In this section, we show simulation results using WMN-PSOHC system. In this work, we consider Normal distributions of mesh clients. The number of mesh routers is considered 16 and the number of mesh clients 48. The total number of iterations is considered 800 and the iterations per phase is considered 4. We consider the number of particle-patterns 9. We conducted simulations 100 times, in order to avoid the effect of randomness and create a general view of results. We show the parameter setting for WMN-PSOHC in Table 1. We show the simulation results in Figs. 4 and 5. For SGC, both replacement methods reach the maximum (100%). However, CM converges faster than LDIWM. Also, for the NCMC, CM performs better than LDIWM. Therefore, we conclude that the performance for CM is better compared with LDIWM.

5

Conclusions

In this work, we evaluated the performance of a hybrid simulation system based on PSO and HC (called WMN-PSOHC) considering CM and LDIWM. Simulation results show that the performance is better for CM compared with LDIWM.

Performance Evaluation of WMNs by WMN-PSOHC System

29

In our future work, we would like to evaluate the performance of the proposed system for different parameters and scenarios.

References 1. Akyildiz, I.F., Wang, X., Wang, W.: Wireless mesh networks: a survey. Comput. Netw. 47(4), 445–487 (2005) 2. Barolli, A., Sakamoto, S., Ozera, K., Ikeda, M., Barolli, L., Takizawa, M.: Performance evaluation of WMNs by WMN-PSOSA simulation system considering constriction and linearly decreasing Vmax methods. In: International Conference on P2P, pp. 111–121. Parallel, Grid, Cloud and Internet Computing, Springer (2017) 3. Barolli, A., Sakamoto, S., Barolli, L., Takizawa, M.: A hybrid simulation system based on particle swarm optimization and distributed genetic algorithm for WMNs: performance evaluation considering normal and uniform distribution of mesh clients. In: International Conference on Network-Based Information Systems, pp. 42–55. Springer (2018) 4. Barolli, A., Sakamoto, S., Barolli, L., Takizawa, M.: Performance analysis of simulation system based on particle swarm optimization and distributed genetic algorithm for WMNs considering different distributions of mesh clients. In: International Conference on Innovative Mobile and Internet Services in Ubiquitous Computing, pp. 32–45. Springer (2018) 5. Barolli, A., Sakamoto, S., Ozera, K., Barolli, L., Kulla, E., Takizawa, M.: Design and implementation of a hybrid intelligent system based on particle swarm optimization and distributed genetic algorithm. In: International Conference on Emerging Internetworking, Data & Web Technologies, p. 79–93. Springer (2018) 6. Barooli, A., Sakamoto, S., Barolli, L., Takizawa, M.: Performance evaluation of WMN-PSODGA system for node placement problem in WMNs considering four different crossover methods. In: The 32nd IEEE International Conference on Advanced Information Networking and Applications (AINA-2018), pp. 850–857. IEEE (2018) 7. Clerc, M., Kennedy, J.: The particle swarm-explosion, stability, and convergence in a multidimensional complex space. IEEE Trans. Evol. Comput. 6(1), 58–73 (2002) 8. Girgis, M.R., Mahmoud, T.M., Abdullatif, B.A., Rabie, A.M.: Solving the wireless mesh network design problem using genetic algorithm and simulated annealing optimization methods. Int. J. Comput. Appl. 96(11), 1–10 (2014) 9. Inaba, T., Obukata, R., Sakamoto, S., Oda, T., Ikeda, M., Barolli, L.: Performance evaluation of a QoS-aware fuzzy-based CAC for LAN access. Int. J. Space-Based Situated Comput. 6(4), 228–238 (2016) 10. Inaba, T., Sakamoto, S., Oda, T., Ikeda, M., Barolli, L.: A testbed for admission control in WLAN: a fuzzy approach and its performance evaluation. In: International Conference on Broadband and Wireless Computing, Communication and Applications, pp. 559–571. Springer (2016) 11. Lim, A., Rodrigues, B., Wang, F., Xu, Z.: k-center problems with minimum coverage. In: Computing and Combinatorics, pp. 349–359 (2004) 12. Maolin, T., et al.: Gateways placement in backbone wireless mesh networks. Int. J. Commun. Netw. Syst. Sci. 2(1), 44 (2009) 13. Matsuo, K., Sakamoto, S., Oda, T., Barolli, A., Ikeda, M., Barolli, L.: Performance analysis of WMNs by WMN-GA simulation system for two WMN architectures and different TCP congestion-avoidance algorithms and client distributions. Int. J. Commun. Netw. Distrib. Syst. 20(3), 335–351 (2018)

30

S. Sakamoto et al.

14. Naka, S., Genji, T., Yura, T., Fukuyama, Y.: A hybrid particle swarm optimization for distribution state estimation. IEEE Trans. Power Syst. 18(1), 60–68 (2003) 15. Ohara, S., Barolli, A., Sakamoto, S., Barolli, L.: Performance analysis of WMNs by WMN-PSODGA simulation system considering load balancing and client uniform distribution. In: International Conference on Innovative Mobile and Internet Services in Ubiquitous Computing, pp. 25–38. Springer (2019) 16. Ozera, K., Inaba, T., Bylykbashi, K., Sakamoto, S., Ikeda, M., Barolli, L.: A WLAN triage testbed based on fuzzy logic and its performance evaluation for different number of clients and throughput parameter. Int. J. Grid Util. Comput. 10(2), 168–178 (2019) 17. Poli, R., Kennedy, J., Blackwell, T.: Particle swarm optimization. Swarm Intell. 1(1), 33–57 (2007) 18. Sakamoto, S., Lala, A., Oda, T., Kolici, V., Barolli, L., Xhafa, F.: Analysis of WMN-HC simulation system data using friedman test. In: The Ninth International Conference on Complex, Intelligent, and Software Intensive Systems (CISIS-2015), pp. 254–259. IEEE (2015) 19. Sakamoto, S., Oda, T., Ikeda, M., Barolli, L., Xhafa, F.: An integrated simulation system considering WMN-PSO simulation system and network simulator 3. In: International Conference on Broadband and Wireless Computing, Communication and Applications, pp. 187–198. Springer (2016) 20. Sakamoto, S., Oda, T., Ikeda, M., Barolli, L., Xhafa, F.: Implementation and evaluation of a simulation system based on particle swarm optimisation for node placement problem in wireless mesh networks. Int. J. Commun. Netw. Distrib. Syst. 17(1), 1–13 (2016) 21. Sakamoto, S., Obukata, R., Oda, T., Barolli, L., Ikeda, M., Barolli, A.: Performance analysis of two wireless mesh network architectures by WMN-SA and WMN-TS simulation systems. J. High Speed Netw. 23(4), 311–322 (2017) 22. Sakamoto, S., Ozera, K., Oda, T., Ikeda, M., Barolli, L.: Performance evaluation of intelligent hybrid systems for node placement in wireless mesh networks: a comparison study of WMN-PSOHC and WMN-PSOSA. In: International Conference on Innovative Mobile and Internet Services in Ubiquitous Computing, pp. 16–26. Springer (2017) 23. Sakamoto, S., Ozera, K., Oda, T., Ikeda, M., Barolli, L.: Performance evaluation of WMN-PSOHC and WMN-PSO simulation systems for node placement in wireless mesh networks: a comparison study. In: International Conference on Emerging Internetworking, Data & Web Technologies, pp. 64–74. Springer (2017) 24. Sakamoto, S., Ozera, K., Ikeda, M., Barolli, L.: Implementation of intelligent hybrid systems for node placement problem in WMNs considering particle swarm optimization, hill climbing and simulated annealing. Mobile Netw. Appl. 23(1), 27–33 (2018) 25. Sakamoto, S., Ohara, S., Barolli, L., Okamoto, S.: Performance evaluation of WMNs by WMN-PSOHC system considering random inertia weight and linearly decreasing inertia weight replacement methods. In: International Conference on Innovative Mobile and Internet Services in Ubiquitous Computing, pp. 39–48. Springer (2019) 26. Sakamoto, S., Ozera, K., Barolli, A., Ikeda, M., Barolli, L., Takizawa, M.: Implementation of an intelligent hybrid simulation systems for WMNs based on particle swarm optimization and simulated annealing: performance evaluation for different replacement methods. Soft. Comput. 23(9), 3029–3035 (2019) 27. Schutte, J.F., Groenwold, A.A.: A study of global optimization using particle swarms. J. Global Optim. 31(1), 93–108 (2005)

Performance Evaluation of WMNs by WMN-PSOHC System

31

28. Shi, Y.: Particle swarm optimization. IEEE Connections 2(1), 8–13 (2004) 29. Shi, Y., Eberhart, RC.: Parameter selection in particle swarm optimization. In: Evolutionary programming VII, pp. 591–600 (1998) 30. Wang, J., Xie, B., Cai, K., Agrawal, D.P.: Efficient mesh router placement in wireless mesh networks. In: Proceedings of IEEE International Conference on Mobile Adhoc and Sensor Systems (MASS-2007), pp. 1–9 (2007) 31. Xhafa, F., Sanchez, C., Barolli, L.: Ad hoc and neighborhood search methods for placement of mesh routers in wireless mesh networks. In: Proceedings of 29th IEEE International Conference on Distributed Computing Systems Workshops (ICDCS2009), pp. 400–405 (2009)

A Fuzzy-Based Simulation System for IoT Node Selection in Opportunistic Networks and Testbed Implementation Miralda Cuka1(B) , Donald Elmazi2 , Keita Matsuo2 , Makoto Ikeda2 , and Leonard Barolli2 1

Graduate School of Engineering, Fukuoka Institute of Technology (FIT), 3-30-1 Wajiro-Higashi, Higashi-Ku, Fukuoka 811-0295, Japan [email protected] 2 Department of Information and Communication Engineering, Fukuoka Institute of Technology (FIT), 3-30-1 Wajiro-Higashi, Higashi-Ku, Fukuoka 811-0295, Japan [email protected],[email protected],{kt-matsuo,barolli}@fit.ac.jp Abstract. In opportunistic networks the communication opportunities (contacts) are intermittent and there is no need to establish an endto-end link between the communication nodes. The enormous growth of nodes having access to the Internet, along the vast evolution of the Internet and the connectivity of objects and nodes, has evolved as Internet of Things (IoT). There are different issues for these networks. One of them is the selection of IoT nodes in order to carry out a task in opportunistic networks. In this work, we implement a Fuzzy-Based System for IoT node selection in opportunistic networks. For our proposed system, we use four input parameters: Node’s Distance from Task (NDT), Node’s Remaining Energy (NRE), Node’s Buffer Occupancy (NBO) and Node Inter Contact Time (NICT). The output parameter is Node Selection Decision (NSD). We also implemented a testbed with the same input and output parameters and compared its results with the simulation results. The results show that the proposed system makes a proper selection decision of IoT nodes in opportunistic networks. The IoT node selection is increased up to 40% and decreased 38% by decreasing NBO and increasing NICT, respectively.

1

Introduction

Future communication systems will be increasingly complex, involving thousands of heterogeneous nodes with diverse capabilities and various networking technologies interconnected with the aim to provide users with ubiquitous access to information and advanced services at a high quality level, in a cost efficient manner, any time, any place, and in line with the always best connectivity principle. The Opportunistic Networks (OppNets) can provide an alternative way to support the diffusion of information in special locations within a city, particularly in crowded spaces where current wireless technologies can exhibit congestion issues. The efficiency of this diffusion relies mainly on user mobility. In fact, c Springer Nature Switzerland AG 2020  L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 32–43, 2020. https://doi.org/10.1007/978-3-030-33506-9_4

A Fuzzy-Based System for Selection of IoT Nodes in OppNets

33

mobility creates the opportunities for contacts and, therefore, for data forwarding [1]. OppNets have appeared as an evolution of the MANETs. They are also a wireless based network and hence, they face various issues similar to MANETs such as frequent disconnections, highly variable links, limited bandwidth etc. In OppNets, nodes are always moving which makes the network easy to deploy and decreases the dependence on infrastructure for communication [2]. In Internet of Things (IoT), the traffic is going through different networks. The IoT can seamlessly connect the real world and cyberspace via physical objects embedded with various types of intelligent sensors. A large number of Internet-connected machines will generate and exchange an enormous amount of data that make daily life more convenient, help to make a tough decision and provide beneficial services. The IoT probably becomes one of the most popular networking concepts that has the potential to bring out many benefits [3,4]. OppNets are the variants of Delay Tolerant Networks (DTNs). It is a class of networks that has emerged as an active research subject in the recent times. Owing to the transient and un-connected nature of the nodes, routing becomes a challenging task in these networks. Sparse connectivity, no infrastructure and limited resources further complicate the situation [5,6]. Routing methods for such sparse mobile networks use a different paradigm for message delivery. These schemes utilize node mobility by having nodes carry messages, waiting for an opportunity to transfer messages to the destination or the next relay rather than transmitting them over a path [7]. Hence, the challenges for routing in OppNet are very different from the traditional wireless networks and their utility and potential for scalability makes them a huge success. In mobile OppNet, connectivity varies significantly over time and is often disruptive. Examples of such networks include interplanetary communication networks, mobile sensor networks, vehicular ad hoc networks (VANETs), terrestrial wireless networks, and under-water sensor networks. While the nodes in such networks are typically delay-tolerant, message delivery latency still remains a crucial metric, and reducing it is highly desirable [8]. The Fuzzy Logic (FL) is unique approach that is able to simultaneously handle numerical data and linguistic knowledge. The fuzzy logic works on the levels of possibilities of input to achieve the definite output. Fuzzy set theory and FL establish the specifics of the nonlinear mapping. In this paper, we propose and implement a Fuzzy-based system for selection of IoT nodes in OppNet considering four parameters: Node’s Distance from Task (NDT), Node’s Remaining Energy (NRE), Node’s Buffer Occupancy (NBO) and Node Inter Contact Time (NICT) for IoT node selection. We show the simulation results for different values of parameters. The remainder of the paper is organized as follows. In the Sect. 2, we present IoT and OppNet. In Sect. 3, we introduce the Fuzzy-based simulator system and testbed implementation. The evaluation results are shown in Sect. 4. Finally, conclusions and future work are given in Sect. 5.

34

2 2.1

M. Cuka et al.

IoT and OppNets IoT

IoT allows to integrate physical and virtual objects. Virtual reality, which was recently available only on the monitor screens, now integrates with the real world, providing users with completely new opportunities: interact with objects on the other side of the world and receive the necessary services that became real due the wide interaction [9]. The IoT will support substantially higher number of end users and nodes. In Fig. 1, we present an example of an IoT network architecture. The IoT network is a combination of IoT nodes which are connected with different mediums using IoT Gateway to the Internet. The data transmitted through the gateway is stored, processed securely within cloud server. These new connected things will trigger increasing demands for new IoT applications that are not only for users. The current solutions for IoT application development generally rely on integrated service-oriented programming platforms. In particular, resources (e.g., sensory data, computing resource, and control information) are modeled as services and deployed in the cloud or at the edge. It is difficult to achieve rapid deployment and flexible resource management at network edges, in addition, an IoT system’s scalability will be restricted by the capability of the edge nodes [10]. 2.2

OppNets

In Fig. 2 we show an OppNet scenario. OppNets comprises a network where nodes can be anything from pedestrians, vehicles, fixed nodes and so on. The data is sent from the sender to receiver by using communication opportunity that can be Wi-Fi, Bluetooth, cellular technologies or satellite links to transfer the message to the final destination. In such scenario, IoT nodes might roam and opportunistically encounter several different statically deployed networks and perform either data collection or dissemination as well as relaying data

Fig. 1. An Iot network architecture.

A Fuzzy-Based System for Selection of IoT Nodes in OppNets

35

between these networks, thus introducing further connectivity for disconnected networks. For example, as seen in Fig. 2, a car could opportunistically encounter other IoT nodes, collect information from them and relay it until it finds an available access point where it can upload the information. Similarly, a person might collect information from home-based weather stations and relay it through several other people, cars and buses until it reaches its intended destination [11].

Fig. 2. OppNets scenario.

OppNets are not limited to only such applications, as they can introduce further connectivity and benefits to IoT scenarios. In an OppNet, due to node mobility network partitions occur. These events result in intermittent connectivity. When there is no path existing between the source and the destination, the network partition occurs. Therefore, nodes need to communicate with each other via opportunistic contacts through store-carry-forward operation.

3

Proposed Fuzzy-Based Simulator and Testbed Implementation

In this work, we use fuzzy logic to implement the proposed system. Fuzzy sets and fuzzy logic have been developed to manage vagueness and uncertainty in a reasoning process of an intelligent system such as a knowledge based system, an expert system or a logic control system [12–25].

36

3.1

M. Cuka et al.

Proposed Fuzzy-Based Simulation System

The structure of the proposed system for the node selection is shown in Fig. 3. Based on OppNets characteristics and challenges, we consider the following parameters for implementation of our proposed system: Node’s Distance to Task (NDT): The distance of a node from the task is an important parameter. An IoT node will be selected to carry out a task with high possibility if the node is close to the task. Node’s Remaining Energy (NRE): The IoT nodes are active and can perform tasks and exchange data in different ways from each other. Consequently, some IoT nodes may have a lot of remaining power and other may have very little, when an event occurs. Node’s Buffer Occupancy (NBO): In an network that consists of diverse IoT nodes with different resources, buffer occupancy at a certain time is very important. Some IoT nodes are in more advantageous position than others, making them more likely to deliver messages thus making them busier than others. Due to high amount of traffic, these nodes’s buffer may overflow affecting the average throughput and the dropping ratio. Node Inter Contact Time (NICT): The inter-contact time measures the time between the end of previous contact and the beginning of a new one between two IoT nodes. Shorter inter-contact time means having more opportunities to forward the message to the next IoT node. Our proposed system consists of one Fuzzy Logic Controller (FLC), which is the main part of our system and its basic elements which are shown in Fig. 4. They are the fuzzifier, inference engine, Fuzzy Rule Base (FRB) and defuzzifier. The FRB forms a fuzzy set of dimensions |T (N DT )| × |T (N RE)| × |T (N BO)| × |T (N ICT )|, where |T (x)| is the number of terms on T (x). We have four input parameters, so our system has 81 rules. The term sets for these parameters are shown in Table 1. The control rules which are shown in Table 2 have the form: IF “conditions” THEN “control action”. These parameters will be represented from numerical form into linguistic variables. We use fuzzy membership functions to quantify the linguistic term. The fuzzy membership functions of our system our shown in Fig. 5. We use triangular and trapezoidal membership functions for FLC, because they are suitable for real-time operations [26]. 3.2

Testbed Implementation

In order to evaluate the simulation system, we have implemented a Testbed as shown in Fig. 6. The testbed setup consists of the hardware and software part. Different data sensing sensors, are mounted on Arduino Uno via IoT Tab Shield 4. This sensed data gets collected by a processing device which is connected to Arduino Uno via USB cable. The processing device consists of Raspberry Pi 3 model B+ which operates on an optimized Debian based system, or a Mac os laptop. For the software part, we used Arduino IDE to collect the sensed data, Processing language to read this data and FuzzyC [12] to evaluate which

A Fuzzy-Based System for Selection of IoT Nodes in OppNets

37

NDT

NRE

FLC

NSD

NBO

NICT

Fig. 3. Proposed system model. Input

Inference Engine

Fuzzifier

Defuzzifier

Output

Fuzzy Rule Base

Fig. 4. FLC structure. Table 1. Parameters and their term sets for FLC. Parameters

Term sets

Node’s Distance to Task (NDT)

Near (Nr), Close (Cl), Far (Fr)

Node’s Remaining Energy (NRE) Low (Lo), Medium (Md), High (Hg) Node’s Buffer Occupancy (NBO)

Minimum (Min), Medium (Med), Maximum (Max)

Node Inter Contact Time (NICT) Short (Sh), Medium (Mdm), Long (Lng) Node Selection Decision (NSD)

Extremely Low Selection Possibility (ELSP), Very Low Selection Possibility (VLSP), Low Selection Possibility (LSP), Medium Selection Possibility (MSP), High Selection Possibility (HSP), Very High Selection Possibility (VHSP), Extremely High Selection Possibility (EHSP)

of the nodes based on the data is more likely to be selected for a certain task. The hardware is mounted on different IoT nodes to mimic a real life scenario. In Fig. 6(a) and (b) are shown static and mobile IoT nodes, respectively. In static IoT nodes, the data is sensed by the sensor mounted in Arduino with IoT Tab Shield 4, read and processed using the laptop. For mobile IoT nodes, we use Raspberry Pi 3 model B+ for data reading and processing, which is power supplied by a 24000 mAh battery with a lcd display for battery level reading.

38

M. Cuka et al. Table 2. FRB.

No. NDT NRE NBO NICT NSD

No. NDT NRE NBO NICT NSD

No. NDT NRE NBO NICT NSD

1

Nr

Lo

Min Sh

EHSP 28

Cl

Lo

Min Sh

Fr

Lo

Min Sh

2

Nr

Lo

Min Mdm VHSP 29

Cl

Lo

Min Mdm MSP

56

Fr

Lo

Min Mdm LSP

3

Nr

Lo

Min Lng

VHSP 30

Cl

Lo

Min Lng

MSP

57

Fr

Lo

Min Lng

LSP

4

Nr

Lo

Med Sh

EHSP 31

Cl

Lo

Med Sh

HSP

58

Fr

Lo

Med Sh

MSP

5

Nr

Lo

Med Mdm HSP

32

Cl

Lo

Med Mdm VLSP 59

Fr

Lo

Med Mdm VLSP

6

Nr

Lo

Med Lng

HSP

33

Cl

Lo

Med Lng

VLSP 60

Fr

Lo

Med Lng

VLSP

7

Nr

Lo

Max Sh

HSP

34

Cl

Lo

Max Sh

LSP

61

Fr

Lo

Max Sh

VLSP

8

Nr

Lo

Max Mdm LSP

35

Cl

Lo

Max Mdm ELSP

62

Fr

Lo

Max Mdm ELSP

9

Nr

Lo

Max Lng

LSP

36

Cl

Lo

Max Lng

ELSP

63

Fr

Lo

Max Lng

ELSP

10

Nr

Md

Min Sh

EHSP 37

Cl

Md

Min Sh

EHSP 64

Fr

Md

Min Sh

VHSP

11

Nr

Md

Min Mdm EHSP 38

Cl

Md

Min Mdm HSP

65

Fr

Md

Min Mdm MSP

12

Nr

Md

Min Lng

EHSP 39

Cl

Md

Min Lng

HSP

66

Fr

Md

Min Lng

MSP

13

Nr

Md

Med Sh

EHSP 40

Cl

Md

Med Sh

VHSP 67

Fr

Md

Med Sh

HSP

14

Nr

Md

Med Mdm HSP

41

Cl

Md

Med Mdm LSP

68

Fr

Md

Med Mdm VLSP

15

Nr

Md

Med Lng

HSP

42

Cl

Md

Med Lng

LSP

69

Fr

Md

Med Lng

VLSP

16

Nr

Md

Max Sh

VHSP 43

Cl

Md

Max Sh

MSP

70

Fr

Md

Max Sh

LSP

17

Nr

Md

Max Mdm MSP

44

Cl

Md

Max Mdm VLSP 71

Fr

Md

Max Mdm ELSP

18

Nr

Md

Max Lng

MSP

45

Cl

Md

Max Lng

VLSP 72

Fr

Md

Max Lng

ELSP

19

Nr

Hg

Min Sh

EHSP 46

Cl

Hg

Min Sh

EHSP 73

Fr

Hg

Min Sh

EHSP

20

Nr

Hg

Min Mdm EHSP 47

Cl

Hg

Min Mdm EHSP 74

Fr

Hg

Min Mdm VHSP

21

Nr

Hg

Min Lng

EHSP 48

Cl

Hg

Min Lng

EHSP 75

Fr

Hg

Min Lng

VHSP

22

Nr

Hg

Med Sh

EHSP 49

Cl

Hg

Med Sh

EHSP 76

Fr

Hg

Med Sh

EHSP

23

Nr

Hg

Med Mdm EHSP 50

Cl

Hg

Med Mdm VHSP 77

Fr

Hg

Med Mdm HSP

24

Nr

Hg

Med Lng

EHSP 51

Cl

Hg

Med Lng

VHSP 78

Fr

Hg

Med Lng

HSP

25

Nr

Hg

Max Sh

EHSP 52

Cl

Hg

Max Sh

EHSP 79

Fr

Hg

Max Sh

VHSP

26

Nr

Hg

Max Mdm VHSP 53

Cl

Hg

Max Mdm MSP

80

Fr

Hg

Max Mdm LSP

27

Nr

Hg

Max Lng

Cl

Hg

Max Lng

81

Fr

Hg

Max Lng

4 4.1

VHSP 54

VHSP 55

MSP

VHSP

LSP

Proposed System Evaluation Simulation Results

We present the simulation results in Fig. 7. We show the relation between the possibility of an IoT node to be selected (NSD) to carry out a task, versus NDT, NRE, NBO and NICT. In Fig. 7(a) and (b), we show how the output parameter NSD is affected by NRE. IoT nodes with more remaining energy, have a higher possibility to be selected for carrying out a job. To show how remaining energy affects the selection of an IoT node, we compare Fig. 7(a) with Fig. 7(b) for NICT = 0.4, NBO = 0.9. We see that NSD is increased 37%. In Fig. 7(c) and (d) are shown the simulation results for NDT = 0.5. Comparing Fig. 7(c) with (a), when NICT = 0.4 and NBO = 0.1, we see that that NDS is decreased 16%. This means that nodes which are far from task, are less likely to be selected since these IoT nodes will need more resources to reach this task. In Fig. 7(e) and (f), the NDT is increased to 0.9. We have a further decrease of NSD with the increase of NDT. In Fig. 7(e), for NICT = 0.2 to NICT = 0.4 and NBO = 0.1, we see that NSD is decreased 38%. IoT nodes that take a longer

A Fuzzy-Based System for Selection of IoT Nodes in OppNets

39

Fig. 5. Fuzzy membership functions.

Fig. 6. Testbed Implementation.

time to come in contact with other nodes will create less connections, thus the possibility that the IoT node be selected decreases. To see the effect that buffer occupancy has on NSD, we take NICT = 0.4 for NBO = 0.9 and NBO = 0.1 in Fig. 7(f). We see that NSD is increased 40% with the decrease of NBO from NBO = 0.9 to NBO = 0.1. The buffer of some IoT nodes may be occupied or fully

40

M. Cuka et al.

Fig. 7. Simulation results.

occupied. Since these networks use store-carry-forward mechanism, an occupied buffer will cause a congestion due to buffer overflow. 4.2

Experimental Results

The experimental results are shown in Fig. 8. In Fig. 8(a) and (b) are shown the results for NDT = Near, NRE = Low and NDT = Near, NRE = High, respectively. During the testbed implementation we gathered a lot of data from the sensors. The simulation results in Fig. 7(a) and (b) are close with experimental results in Fig. 8(a) and (b). However, there are some variations from point to point which represent the different outside factors that affect experimental results. In Fig. 8(c) and (d), are shown results for NDT = Close, NRE = Low and NDT = Close, NRE = High. In Fig. 8(e) and (f), are shown results for NDT = Far, NRE = Low and NDT = Far, NRE = High. For all the above results, we can see that the simulation results are close to the experimental results.

A Fuzzy-Based System for Selection of IoT Nodes in OppNets

41

Fig. 8. Experimental results.

5

Conclusions and Future Work

In this paper, we proposed a fuzzy-based IoT node selection system for OppNets considering four parameters: NDT, NRE, NBO, NICT. We implemented a testbed and compared experimental results with the simulation results for the selection of IoT nodes in an Oppnet scenario. The simulation results and experimental results are close, but in experiment there are some variations. In the future work, we will also consider other parameters for IoT node selection and make extensive simulations and experiments to evaluate the proposed system.

42

M. Cuka et al.

References 1. Mantas, N., Louta, M., Karapistoli, E., Karetsos, G.T., Kraounakis, S., Obaidat, M.S.: Towards an incentive-compatible, reputation-based framework for stimulating cooperation in opportunistic networks: a survey. IET Netw. 6(6), 169–178 (2017) 2. Sharma, D.K., Sharma, A., Kumar, J., et al.: KNNR: K-nearest neighbour classification based routing protocol for opportunistic networks. In: 10-th International Conference on Contemporary Computing (IC3), pp. 1–6. IEEE (2017) 3. Kraijak, S., Tuwanut, P.: A survey on internet of things architecture, protocols, possible applications, security, privacy, real-world implementation and future trends. In: 16th International Conference on Communication Technology (ICCT), pp. 26– 31. IEEE (2015) 4. Arridha, R., Sukaridhoto, S., Pramadihanto, D., Funabiki, N.: Classification extension based on iot-big data analytic for smart environment monitoring and analytic in real-time system. Int. J. Space-Based Situated Comput. 7(2), 82–93 (2017) 5. Dhurandher, S.K., Sharma, D.K., Woungang, I., Bhati, S.: HBPR: history based prediction for routing in infrastructure-less opportunistic networks. In: 27th International Conference on Advanced Information Networking and Applications (AINA), pp. 931–936. IEEE (2013) 6. Spaho, E., Mino, G., Barolli, L., Xhafa, F.: Goodput and PDR analysis of AODV, OLSR and DYMO protocols for vehicular networks using cavenet. Int. J. Grid Util. Comput. 2(2), 130–138 (2011) 7. Abdulla, M., Simon, R.: The impact of intercontact time within opportunistic networks: protocol implications and mobility models. TechRepublic White Paper (2009) 8. Patra, T.K., Sunny, A.: Forwarding in heterogeneous mobile opportunistic networks. IEEE Commun. Lett. 22(3), 626–629 (2018) 9. Popereshnyak, S., Suprun, O., Suprun, O., Wieckowski, T.: IoT application testing features based on the modelling network. In: The 14-th International Conference on Perspective Technologies and Methods in MEMS Design (MEMSTECH), pp. 127–131. IEEE (2018) 10. Chen, N., Yang, Y., Li, J., Zhang, T.: A fog-based service enablement architecture for cross-domain IoT applications. In: 2017 IEEE Fog World Congress (FWC), pp. 1–6. IEEE (2017) 11. Pozza, R., Nati, M., Georgoulas, S., Moessner, K., Gluhak, A.: Neighbor discovery for opportunistic networking in internet of things scenarios: a survey. IEEE Access 3, 1101–1131 (2015) 12. Inaba, T., Sakamoto, S., Kolici, V., Mino, G., Barolli, L.: A CAC scheme based on fuzzy logic for cellular networks considering security and priority parameters. In: The 9-th International Conference on Broadband and Wireless Computing, Communication and Applications (BWCCA-2014), pp. 340–346 (2014) 13. Spaho, E., Sakamoto, S., Barolli, L., Xhafa, F., Barolli, V., Iwashige, J.: A fuzzybased system for peer reliability in JXTA-overlay P2P considering number of interactions. In: The 16th International Conference on Network-Based Information Systems (NBiS-2013), pp. 156–161 (2013) 14. Matsuo, K., Elmazi, D., Liu, Y., Sakamoto, S., Mino, G., Barolli, L.: FACS-MP: a fuzzy admission control system with many priorities for wireless cellular networks and its performance evaluation. J. High Speed Netw. 21(1), 1–14 (2015)

A Fuzzy-Based System for Selection of IoT Nodes in OppNets

43

15. Grabisch, M.: The application of fuzzy integrals in multicriteria decision making. Eur. J. Oper. Res. 89(3), 445–456 (1996) 16. Inaba, T., Elmazi, D., Liu, Y., Sakamoto, S., Barolli, L., Uchida, K.: Integrating wireless cellular and ad-hoc networks using fuzzy logic considering node mobility and security. The 29th IEEE International Conference on Advanced Information Networking and Applications Workshops (WAINA-2015), pp. 54–60 (2015) 17. Kulla, E., Mino, G., Sakamoto, S., Ikeda, M., Caball´e, S., Barolli, L.: FBMIS: a fuzzy-based multi-interface system for cellular and ad hoc networks. In: International Conference on Advanced Information Networking and Applications (AINA2014), pp. 180–185 (2014) 18. Elmazi, D., Kulla, E., Oda, T., Spaho, E., Sakamoto, S., Barolli, L.: A comparison study of two fuzzy-based systems for selection of actor node in wireless sensor actor networks. J. Ambient Intell. Humaniz. Comput. 6(5), 635–645 (2015) 19. Zadeh, L.: Fuzzy logic, neural networks, and soft computing. ACM Commun. 37, 77–85 (1994) 20. Spaho, E., Sakamoto, S., Barolli, L., Xhafa, F., Ikeda, M.: Trustworthiness in P2P: performance behaviour of two fuzzy-based systems for JXTA-overlay platform. Soft. Comput. 18(9), 1783–1793 (2014) 21. Inaba, T., Sakamoto, S., Kulla, E., Caballe, S., Ikeda, M., Barolli, L.: An integrated system for wireless cellular and ad-hoc networks using fuzzy logic. In: International Conference on Intelligent Networking and Collaborative Systems (INCoS-2014), pp. 157–162 (2014) 22. Matsuo, K., Elmazi, D., Liu, Y., Sakamoto, S., Barolli, L.: A multi-modal simulation system for wireless sensor networks: a comparison study considering stationary and mobile sink and event. J. Ambient Intell. Humaniz. Comput. 6(4), 519–529 (2015) 23. Kolici, V., Inaba, T., Lala, A., Mino, G., Sakamoto, S., Barolli, L.: A fuzzy-based CAC scheme for cellular networks considering security. In: International Conference on Network-Based Information Systems (NBiS-2014), pp. 368–373 (2014) 24. Liu, Y., Sakamoto, S., Matsuo, K., Ikeda, M., Barolli, L., Xhafa, F.: A comparison study for two fuzzy-based systems: improving reliability and security of JXTAoverlay P2P platform. Soft. Comput. 20(7), 2677–2687 (2015) 25. Matsuo, K., Elmazi, D., Liu, Y., Sakamoto, S., Mino, G., Barolli, L.: FACS-MP: a fuzzy admission control system with many priorities for wireless cellular networks and its perforemance evaluation. J. High Speed Netw. 21(1), 1–14 (2015) 26. Mendel, J.M.: Fuzzy logic systems for engineering: a tutorial. Proc. IEEE 83(3), 345–377 (1995)

Consensus Based Mechanism Using Blockchain for Intensive Data of Vehicles Tehreem Ashfaq1 , Muhammad Ahmed Younis2 , Shahzad Rizwan3 , Zahid Iqbal4 , Shahid Mehmood4 , and Nadeem Javaid1(B) 1

3

COMSATS University Islamabad, Islamabad, Pakistan [email protected], [email protected] 2 University of Agriculture Faisalabad, Faisalabad, Pakistan [email protected] COMSATS University Islamabad (CUI), Attock Campus, Attock, Pakistan [email protected] 4 Bejing University of Post’s and Telecommunication, Beijing, China zahid [email protected], [email protected]

Abstract. The explosive development of Intelligent Vehicles (IVs) has led to a complex network, which is difficult to manage due to the extensive communication of vehicles and storage of vehicles’ data. Due to increase in number of vehicles, IVs come up with large difficulties. Huge data generated by IVs is very difficult to be handled due to limited storage and lack of intelligent management. Many security and privacy problems are also related to the IV networks. Traditional centralized approaches are used to deal with limited storage and security issues. Increasing number of vehicles expand the number of links in network and also leads to the intensive data. Lack of coordination of vehicles, reliability of the network and traffic among vehicles are some of the major issues. These issues hinder the performance of the vehicle industry. We propose a consensus based mechanism using blockchain technology to manage the intensive data and authenticate the data of vehicles in the EV industry. This mechanism also ensures data privacy, security and also promotes data immutability. The transactions are stored in distributed ledger to provide facility of transparency. In a nutshell, blockchain technology is incorporated in EVs sector to revolutionize the World.

Keywords: Blockchain technology data · Intelligent vehicles

1

· Consensus mechanism · Intensive

Introduction

In the modern era, vehicle sector is getting modernized with the advancement in the infrastructure and the communication sector. With the immense increase in the population over the past few years, number of vehicles running on the roads has also increased. With this advancement, the transport sector has undergone a whole new driving experience, comprising of autonomous cars, self-driven cars, c Springer Nature Switzerland AG 2020  L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 44–55, 2020. https://doi.org/10.1007/978-3-030-33506-9_5

Consensus Based Mechanism Using Blockchain

45

etc. Not just the infrastructure of the vehicles is changing, also new services are being introduced, particularly the communication services [1]. Blockchain is an emerging technology. It is a distributed ledger that can record the transactions between two nodes/vehicles managed by a Peer-to-Peer (P2P) network. Blockchain is about the transparency, security and trust without any third party. Third party is like a centralized system. Blocks are connected to each other and each block has its own hash. Blockchain is also about transparency; distributed ledger is managed by network nodes, each having the copy of ledger. Nowadays, blockchain based P2P smart contracts for Internet Vehicles (IVs) are one of the recent applications of blockchain which is growing rapidly. With the increase in IVs, many problems arise, e.g., huge amount of data generated, limited storage capacity, lack of management, etc. Privacy and security issues also exist. According to [1], number of IVs will increase to 140 million by 2030. These figures show that it is just a matter of time when the conventional vehicles will totally be replaced by the IVs. Even now, the conventional vehicles are being equipped with latest technologies. IVs communicate in a P2P manner, which removes the involvement of the third party. Still, the security issues exist in the vehicle sector and the vehicle users are hesitant to interact and communicate with each other. The users personal data can be hacked by the malicious users, as this data is quite easily accessible and can be tampered. To overcome the security issues, blockchain technology has been proposed. Blockchain is an emerging technology which promotes decentralization along with security, data immutability and transparency [2]. Therefore, people are migrating towards blockchain technology. In blockchain, data is stored in a distributed ledger, copy of which is available with all the participants of the network. All the transactions taking place in a blockchain network are stored in blocks; each having a unique hash [3]. The blockchain based IV scenario is proposed, which deals with the issues of: lack of coordination between vehicles and generation of intensive data. It also manages the storage and channel reliability. The vehicles are added in the network and assigned unique cryptographic identities. The consensus mechanism is used whenever a new vehicle is added in the network and it requires 51% positive response. The proposed blockchain network is divided in two separate branches. i.e., Integrity Chain (I-Chain) and Fraud chain (F-Chain). The validated vehicles and the associated transactions are stored in integrity I-Chain whereas the malicious users are added to the F-Chain [4]. Figure 1 shows the indepth communication between the vehicles and also the following of the smart contract. 1.1

Motivation

After reviewing the past work done by many authors for secure communication between vehicles, sellers and buyers in [1,2,4], we have the following motivations for this paper:

46

T. Ashfaq et al.

Fig. 1. V2V communication using blockchain

• to deal with the issue of unsecured and trustless communication, blockchain technology should be used, • to deal with the issue of lack of privacy of IVs, crypto IVTP is used and • a new secure technique is required to remove the unauthenticated vehicles from the network. 1.2

Problem Statement

As we are living in a world of automation and latest technologies are being introduced, each field of life is getting automated and inter connected. Intelligent Vehicle (IV) is an entity, which is connected with other IVs through internet for communication purposes. It enables a better vehicular network, still it has many issues which needs to be addressed like security and privacy issues. The explosive development of IVs has led to a complex network, resulting in difficulty in communication and storage of vehicles. Huge amount of data generated by IVs is very difficult to handle due to limited storage and lack of intelligent management. Many information security and privacy problems are also related to the IoV networks [3].

Consensus Based Mechanism Using Blockchain

47

For promoting distributed storage and security purposes, blockchain technology is used. However, channel reliability and coordination among the vehicles is not considered. Increasing the number of vehicles expand the number of links and nodes. Ultimately, single loop hole can open the way for attackers. For efficient working of vehicular network, it is necessary that proper coordination exists between the vehicles. It minimizes the number of malicious vehicles. Range anxiety also exists in the vehicular network, i.e., vehicles are reluctant to go on long travels because of limited resources [4]. To tackle these issues, the blockchain based IV scenario is proposed, which deals with the issue of: lack of coordination between vehicles, intensive data storage. Proposed work manages the data storage and promotes channel reliability.

2

Related Work

In recent times, researchers have applied BC technology is various areas. Following are some related studies; 2.1

Network Communication with Blockchain

Blockchain is an emerging technology so, different authors used this technology with different domains. There are some works which focused on blockchain based WSNs. The authors consider the issues of security, data storage constraints of sensor nodes, computational capability and node failure. In [3], authors identify the problem of user access control to optimize the network. Proposed solution considers authenticity of Channel State Information (CSI) using blockchain consensus and deep learning. In [5], authors deal with the nodes failure issue during data transmission. There are two main reasons for node failure: mobility and selfishness of nodes. Firstly, they set a threshold value to check the node failure. In the second step, a multi-link concurrent tree is built using greedy approach. In this way, transmitting capacity of node is maximized while validating a block transaction time is decreased. However, they did not consider selection of failure nodes, which leads to transmission delay and security issues. In [6] and [7], authors proposed an incentive mechanism for location privacy protection of users. The proposed structure is divided into sub parts. The data is sensed by crowd sensing network and sent to the confusion mechanism. Confusion mechanism protects the data from attacks. Blockchain makes the data temper proof and maintain its integrity. They compared the non-encrypted traditional method with proposed encrypted method. Results showed that females are less concerned for privacy as compared to males. In the paper [8–11], they identified the problems of data sharing. They proposed the solution on secure data sharing and data rights management. In [12], the authors proposed novel hybrid network architecture. The proposed architecture consists of two sub parts: (i) core network (ii) edge network. The proposed network is based on both distributed network and centralized architecture. They also proposed a scheme based on

48

T. Ashfaq et al.

Proof of Work (PoW) to ensure the privacy and security of a network. In the proposed architecture, there is no privacy encryption and no user involvement. In [13], authors used Provable Data Possession (PDP) technique instead of PoW to obtain better results. They also applied preserving hash function to compare the existing data of nodes with the new one. The only problem with PDP is that it can identify the damaged data on nodes, but is unable to recover it. 2.2

IoT with Blockchain

In complex IoT networks, some technical challenges are being faced. There are many scenarios and models have proposed for security purposes. However, the centralized nature of IoT networks create many issues. It is difficult to deal with the centralized IoT networks. Blockchain provides a secure decentralized IoT network which manages the IoT devices in a network. In [14], authors proposed a solution for IoT network which provides scalability. More throughput and minimum delay in access management framework. In [15], authors proposed a Distributed BC based Network (DistBlockNet) for IoT networks. The authors use the advantage of blockchain and Software Defined Network (SDN) technologies, which solve the issues of scalability, efficiency, availability and security. System is also able to provide threat detection and data protection. However, distributed architecture for data storage is still missing. 2.3

Vehicular Communication with Blockchain

The concept of Electric Vehicle, i.e., EV brings new concepts in the market. Road congestion has increased manifold due to vast increase in the number of vehicles. To reduce this huge amount of energy, scientific and research community has focused on the EVs as a source of clean energy [16–20]. In [21], blockchain is integrated with IVs to provide large and secure data storage. The authors designed multiple blockchain model which consists of five blockchain based on different data blocks. Results show that this integration provides large and secure data storage. They achieved high throughput with increasing data. However, delay also increases. Branch based blockchain technology for Intelligent Vehicles (IVs) was proposed in [4]. Branching is done at Locally Dynamic blockchain (LDB). It is to handle the large amount of data generated by IVs. Blockchain is used to keep track of the data generated by IVs and to verify it. Additionally, the concept of IV Trust Point (IVTP) is also introduced to build trust. Problem with branching is that duplicate state changes increase with increasing load.

3

System Model

In smart cities, communication of smart vehicles is not secure. So, we proposed a secure ITS scenario for the secure communication of the intelligent smart vehicles

Consensus Based Mechanism Using Blockchain

49

using blockchain which can keep data of smart vehicles and their transactions. When vehicles communicate with each other, they use proof of work for mining of blocks. In the proposed model, there is an IVTP which assigns the unique ID to each vehicle. This IVTP is generated by crypto mechanism. Vehicles use this IVTP during communication for building the trust and for using the network services. Each vehicle has its own IVTP known as IVTP-ID of that particular vehicle. The details of the vehicles are stored in separate blocks and when these blocks are joined together, a blockchain is formed. Each block has its current hash and previous hash, time stamp, nonce number and targeted address of the previous block. When more vehicles are added to the network, the computational power is increased and the data of vehicles becomes more intensive. From the user’s perspective, it becomes difficult to deal with such type of data. To deal with intensive data, consensus based scheme for authentication of data is used. This scheme is used to add the user to the network on the basis of authenticity while enhancing the efficiency and performance of the network.

Fig. 2. System model

50

T. Ashfaq et al.

The blockchain based proposed IV scenario deals with the issues of: lack of coordination between vehicles and intensive data. Moreover, it also deals with the issues of limited storage and channel reliability. The vehicles are added in the network and are assigned unique crypto identities, respectively. The consensus mechanism is used to achieve 51% positive response. The validated vehicles and the associated transactions are stored in I-chain; whereas, the malicious users are added to the F-chain [4]. Whenever a new user joins the network, the ledger broadcasts this on the mobile gateway for all the users. When the broadcast message is received by all other users then the consensus mechanism is applied to check the authenticity. Once a user is verified to be authentic then a message is sent to the I-chain; otherwise, if the user is not authentic and a message is sent to the F-chain. Figure 2 shows the proposed system model in which the V2V communication, V2X communication and the central trust point is shown. The idea of this proposed model is taken from [4].

4

Reasoning of Graphs

In this section, the results are displayed and their discussion is also given. Initially, when the smart contracts are made and deployed, the transaction and execution costs are calculated. Simulations are performed on Solidity and RemixIDE (online platform); Solidity is used for writing the smart contracts and RemixIDE is used to deploy the smart contracts. For the testing of smart contracts, we use Ganache and for the validation of transactions, we use MetaMask. Simulations are run on the system having specifications: Intel Core i5, 4 GB RAM and 500 GB storage. Presently, the conversion rate of gas and ether is as follows, taken from [22]. There are some performance parameters for the proposed system. • Gas consumption of smart contracts. • Execution cost for the processing power. • Transaction cost of operation being performed. 4.1

Execution Cost

The execution cost of a smart contract depends upon the processing power of the system that how many tasks are performed. The execution cost is directly proportional to the processing power. The execution cost is the cost of executing only a certain function. 4.2

Transaction Cost

The transaction cost depends on: the cost of data being sent, operations being performed and the storage of contract. It can be said that transaction cost is the sum of the execution cost and the deployment cost.

Consensus Based Mechanism Using Blockchain

Fig. 3. Transaction cost

Fig. 4. Execution cost

51

52

T. Ashfaq et al.

Transaction cost is determined by Transaction Cost = Gas Used × Gas Price Figures 3 and 4 shows the transaction and execution costs for the smart contracts and the functions are deployed in terms of gas. These values are taken from RemixIDE. Fluctuations can be observed in the gas values for different functions. The function “transferring balance” has the maximum cost. It can be observed that the function “transferring balance” has the maximum cost because it is a function that used more processing power. Similar is the case with the function “assigning plate number”. This function is intended to add new IVs to the blockchain network. The transaction costs are always higher than the execution costs. The reason is because the transaction costs are the costs of deploying the contract whereas the execution costs are the costs of executing only a certain function. The six functions shown in these figures are: assigning the plate number, activating the car, signing a car, sharing request, transferring balance and assigning value to car. When the smart contract is deployed, then less gas consumption is used for the activation function that is 21650 gwei. The gas consumption also shows the complexity of different functions; the computational power decreases with the decrease in gas consumption.

Fig. 5. Computational time

In Fig. 5, the increasing trend between the amount of data generated and the computational time is shown. When the number of vehicles to be added in the network increases, the data associated with the vehicles also increases. This in turn increases the computational time which is required to access that data. It is seen that the increasing trend is exponential in nature. The reason is that

Consensus Based Mechanism Using Blockchain

53

the computational time is an exponential function of the amount of data stored, which increases exponentially when more vehicles are added in the network.

Fig. 6. Users’ status

Figure 6 shows the number of total users, authentic users, unauthentic users and the total number of requests made. By users we mean vehicles, which are to be added in the network. The authentic users will be added to the I-chain; whereas, the unauthentic users will be added to the F-chain. The graph is plotted between the number of participants (vehicles) and the requests made by them. It is seen that, the proposed model behaves in an efficient manner and is able to respond to large number of authentic requests made and only a few are marked as unauthentic.

5

Conclusion

In this paper, the blockchain technology is proposed in the EV sector to solve the existing issues. The proposed work helped to solve the trust issue among the users, ensured the data immutability and distinction among the authentic and unauthentic data. The vehicles are validated by the use of unique crypto identities being assigned to all the vehicles. All vehicles communicate through these crypto identities. Each vehicle is connected with its neighbors. When two vehicles want to communicate, they broadcast a message on network and then a smart contract is deployed for their communication. The consensus mechanism is employed to ensure the transparency and the transaction data is stored in the form of distributed ledger, i.e., copy of data which is available at all nodes. The proposed work surpasses the existing work as it involves the concept of branching the vehicles in two different branches instead of keeping the data in a

54

T. Ashfaq et al.

single blockchain. This branching mechanism helps reducing the computational time and the storage requirement. The concept of IVTP also helps in assigning the vehicles according to respective trust value. The simulation results prove the claim that the proposed work is better than the existing works.

References 1. Bahga, A., Madisetti, V.K.: Blockchain platform for industrial Internet of Things. J. Softw. Eng. Appl. 9(10), 533 (2016) 2. Nakamoto, S.: Bitcoin: a peer-to-peer electronic cash system (2008) 3. Lin, D., Tang, Y.: Blockchain consensus based user access strategies in D2D networks for data-intensive applications. IEEE Access 6, 72683–72690 (2018) 4. Khan, R.J.H., Javaid, N., Iqbal, S.: Blockchain based node recovery scheme for wireless sensor networks. MS thesis, COMSATS University Islamabad (CUI), Islamabad 44000, Pakistan, July 2019 5. Li, J.: Data transmission scheme considering node failure for blockchain. Wireless Pers. Commun. 103(1), 179–194 (2018) 6. Jia, B., Zhou, T., Li, W., Liu, Z., Zhang, J.: A blockchain-based location privacy protection incentive mechanism in crowd sensing networks. Sensors 18(11), 3894 (2018) 7. Ali, I., Javaid, N., Iqbal, S.: An incentive mechanism for secure service provisioning for lightweight clients based on blockchain. MS thesis, COMSATS University Islamabad (CUI), Islamabad 44000, Pakistan, July 2019 8. Naz, M., Javaid, N., Iqbal, S.: Research based data rights management using blockchain over ethereum network. MS thesis, COMSATS University Islamabad (CUI), Islamabad 44000, Pakistan, July 2019 9. Noshad, Z., Javaid, N., Imran, M.: Analyzing and securing data using data science and blockchain in smart networks. MS thesis, COMSATS University Islamabad (CUI), Islamabad 44000, Pakistan, July 2019 10. Samuel, O., Javaid, N., Awais, M., Ahmed, Z., Imran, M., Guizani, M.: A blockchain model for fair data sharing in deregulated smart grids. In: IEEE Global Communications Conference (GLOBCOM 2019) (2019) 11. Rehman, M., Javaid, N., Awais, M., Imran, M., Naseer, N.: Cloud based secure service providing for IoTs using blockchain. In: IEEE Global Communications Conference (GLOBCOM 2019) (2019) 12. Novo, O.: Scalable access management in IoT using blockchain: a performance evaluation. IEEE IoT J. https://doi.org/10.1109/JIOT.2018.2879679(2018) 13. Jiang, T., Fang, H., Wang, H.: Blockchain-based Internet of vehicles: distributed network architecture and performance analysis. IEEE Internet Things J. 6(3), 4640–4649 (2019) 14. Ren, Y., Liu, Y., Ji, S., Sangaiah, A.K., Wang, J.: Incentive mechanism of data storage based on blockchain for wireless sensor networks. Mobile Inf. Syst. 2018, 10 p. (2018). Article no. 6874158 15. Sharma, P.K., Singh, S., Jeong, Y.-S., Park, J.H.: Distblocknet: distributed blockchains-based secure SDN architecture for iot network. IEEE Commun. Mag. 55(9), 78–85 (2017) 16. Awais, M., Javaid, N., Imran, M.: Energy efficient routing with void hole alleviation in underwater wireless sensor networks. MS thesis, COMSATS University Islamabad (CUI), Islamabad 44000, Pakistan, July 2019

Consensus Based Mechanism Using Blockchain

55

17. Mateen, A., Javaid, N., Iqbal, S.: Towards energy efficient routing in blockchain based underwater WSNs via recovering the void holes. MS thesis, COMSATS University Islamabad (CUI), Islamabad 44000, Pakistan, July 2019 18. Javaid, A., Javaid, N., Imran, M.: Ensuring analyzing and monetization of data using data science and blockchain in loT devices. MS thesis, COMSATS University Islamabad (CUI), Islamabad 44000, Pakistan, July 2019 19. Kazmi, H.S.Z., Javaid, N., Imran, M.: Towards energy efficiency and trustfulness in complex networks using data science techniques and blockchain. MS thesis, COMSATS University Islamabad (CUI), Islamabad 44000, Pakistan, July 2019 20. Zahid, M., Javaid, N., Rasheed, M.B.: Balancing electricity demand and supply in smart grids using blockchain. MS thesis, COMSATS University Islamabad (CUI), Islamabad 44000, Pakistan, July 2019 21. Dai, M., Zhang, S., Wang, H., Jin, S.: A low storage room requirement framework for distributed ledger in blockchain. IEEE Access 6, 22970–22975 (2018) 22. https://currencio.co/gas/eth/. Accessed 10 May 2019

Block-VN: A Distributed Blockchain-Based Efficient Communication and Storage System Hassan Farooq1 , Muhammad Usman Arshad1 , Muhammad Faraz Akhtar2 , Shahid Abbas1 , Bilal Zahid2 , and Nadeem Javaid1(B) 1

2

Department of Computer Science, COMSATS University, Islamabad 44000, Pakistan [email protected] Computer Science, Government College University, Faisalabad, Pakistan http://www.njavaid.com/

Abstract. Internet of vehicles (IoVs) are connected with each other through Internet. In recent years, IoV provides security mechanisms and quick information sharing schemes, etc. The rapid growth of IoV causes various challenges including data storage, intelligent transport system, selfishness of nodes, distrusted nodes and sensor’s data leakage of information. To overcome data storage and delay in services, a decentralized, distributed, secure, transparent and scalable management system is proposed using blockchain technology. Provable data possession (PDP) scheme is used to validate the new data blocks. Message transfer process argon2 (MTP-Argon2) technique is used for data filtration. Using this technique, the raw data are filtered to remove the duplicate and unnecessary data which are obtained from different nodes. Data storage, less delay in service request or response, more sensor’s data sharing and secure communication channel are achieved using proposed system. Keywords: Blockchain vehicular network · Internet of Things Internet of vehicle · Scalable · Provable data possession

1

·

Introduction

Nowadays, smart vehicles (SVs) are efficient, faster and secure due to different features, such as automatic braking system, low fuel consumption, automatic driving system and electronic control units (ECUs), etc. The smart vehicle manufacturers are introducing not only physical vehicle design, they are also introducing the new functionalities which facilitate the drivers. These functionalities are achieved using vehicle’s ECUs having hundreds of megabytes of code. When these SVs are connected with each other through the Internet, it becomes an internet of vehicles (IoVs) network. There are numerous new protocols and safety devices in SVs such as automatic emergency braking, forward collision warnings and vast communication network, etc. Intelligent transport system (ITS) uses c Springer Nature Switzerland AG 2020  L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 56–66, 2020. https://doi.org/10.1007/978-3-030-33506-9_6

Block-VN: Efficient Communication and Storage System

57

ad-hoc networks for communication of SVs, such as wireless access vehicular environment (WAVE), dedicated short range communication (DSRC) and cellular network, etc., however, these networks do not ensure data transmission [1]. ITS also provide different sensory information, traffic road safety information and low to high congestion information for services. SVs gather information from different sources and process the information either by sending it to central server or locally. A huge amount of data is generated through SVs, road side units (RSUs), sensors and Internet of things (IoT) devices, which is stored on central servers causes the traffic load and low storage. As the traffic load is increased with the increase in large amount of SV’s data, it is difficult to handle it on a centralized server. It is also difficult to secure and store all information obtained from different information providers like SVs, RSUs and cellular devices, etc. When artificial intelligence is combined with IoV, it can produce infinite possibilities. When numerous SVs share their services with each other, the traditional centralized system will encounter different challenges like data storage, data security and communication delay. Therefore, distributed and decentralized data storage management is required in the future to overcome these issues. The high security is required for data storage and communication with other vehicles or service providers when the decentralized system is adopted. Blockchain technology is the decentralized approach used to solve the aforementioned problems. Blockchain is a distributed ledger that contains all information of nodes. Every node contains a duplicate copy of the ledger and when any transaction occurs, it first validated by all miner nodes then added in the blockchain. In a distributed system, nodes do not have to trust each other because all nodes have a shared copy of the ledger. Each information of data is stored in a block which is validated by miner nodes having high computation power. Each block contains a previous hash, current hash, nonce and timestamp. Blockchain technology solves high computational problem, data encryption, distributed consensus and insecure data storage which are common problems in centralized systems. With the rapid growth of blockchain technology, it is used in wireless sensor networks, intelligent vehicles, smart grids and deep underwater sensor networks, etc. The main challenging issues in the rapid growth of data are lack of distrust and insecure communication among SVs. In IoV network, SVs cannot communicate with each other without compromise of trust. Yang et al. use blockchain in a distributed network as it is being considered as more secure for the data credibility vehicular network [2]. It provides data security mechanism based on ITS standards. Due to increase in data packet size of SVs, the latency of message and waiting time is increased which affects the whole vehicular network. Some SVs receive the data from other vehicles and become a selfish node and do not share the sensor data. To overcome all the above mention issues, a scalable, secure, transparent and distributed system is proposed for data storage mechanism based on blockchain vehicular network (Block-VN). Provable data possession (PDP) scheme is used to validate the new data blocks. Message transfer process argon2 (MTP-Argon2)

58

H. Farooq et al.

technique is used for data filtration. Using this technique, the raw data are filtered to remove the duplicate and unnecessary data which are obtained from different nodes. Data storage, less delay in the service request or response and secure communication channel are achieved using a proposed system. Local blockchain stores vehicles sensory data for a limited time. The main blockchain contains all local blockchain data, RSUs data and all types of service records. The rest of the paper is structured as follows: Sect. 2 reviews the related work and literature. Section 3 presents the motivation. Section 4 presents the problem statement. Section 5 presents the distributed Block-VN system model. Finally, Sect. 6 concludes this work.

2 2.1

Related Work Blockchain in Smart Grid

The main challenging issues in smart grid are scheduling and optimization of manageable devices. They have not enough power production to fulfill the requirement of consumers. Electricity trading is distribution of energy with wholesale energy market which is centralized. In distributed renewable energy system, energy supply and trading are based on trusted third party. The authors in [3] used centralized approaches which are costly in computation and communication with different devices. The authors in [4] used smart metering and IoT to collects transfer data between different entities using different communication techniques. In [5], authors used concept of game theory on base of incentives and penalties. The authors in [6] proposed decentralized market based on blockchain. It fulfills the needs of energy consumption for end users in transparent and user-friendly applications. In [7], the authors proposed fair data sharing in deregulated smart gird using blockchain. In [8], the proposed blockchain-based electricity demand and supply balancing. In [9], the authors proposed decentralized, secure and trusted electricity trading using consortium blockchain technology. To overcome aforementioned problems, a decentralized blockchain is proposed for authorization, authentication and monitoring of different power voltages. Hyper-ledger consensus algorithm is used to manage the accounts of consumer to done efficient power trading. However, there are some limitations in above proposed models: traditional power trading, power transaction cost and supervisory utility constraint mechanisms are not considered, which can decide the pricing scheme of power supply and demand. 2.2

Blockchain in Wireless Sensor Network

The main challenging issue in wireless sensor network is that a huge amount of data is produced which is stored at central network and very difficult to manage. Device-to-device (D2D) network is used to manage the user access control and dynamic power with cellular users. These IoT devices have limited storage so they

Block-VN: Efficient Communication and Storage System

59

cannot manage their own resources and single point of failure can breakdown the network. The mobile operator can only trust on narrow-band IoT (NB-IoT) network which has high cost of deployment and licensed spectrum resources. In [10], IoT virtual resources using different IoT components is proposed in wireless sensor network. In [11], the authors described user friendly network which processes, collects, and maintains data easily and also provides incentive mechanism for data storage. In [12], the authors elaborated dynamic power control mechanism to control D2D data transmissions. In [13,14], the authors proposed node recovery scheme in WSN using blockchain technology. In [15], the authors proposed blockchain-based smart networks to analyze and secure data. In [16], the authors proposed blockchain-based complex networks to achieve energy efficiency and trustfulness. In [17], the authors proposed blockchain based analyzing and monetization of data. In [18], virtual extensible LAN based blockchain is used in tunnels to enhance the security and performance mechanism. In [19], constrained application protocol (CoAP) is proposed that manages the data of these devices on centralized server. To overcome these issues, long range wide area network (LoRaWAN) provides long range and secure network for mobile operators at low cost using blockchain based trusted servers. D2D mechanism is used for authentication of each user by channel state information (CSI) using consensus blockchain technology. Software defined networks (SDNs) is based on blockchain which improves performance based scalable access management for IoT devices. However, there are some limitations in aforementioned models. They cannot provide a fully-scaled network to link servers, gateways and customers. The recovery of damaged nodes with failure rate is also a challenging issue in request of data. They cannot be applied and achieved in reality to get spectral highest efficiency due to large consumption of resources. Shared network asset approaches realized the multi-signature smart contract which integrates the consistency issue. 2.3

Blockchain in Intelligent of Vehicle

The main challenging issue in IoV and IoT devices is generation of huge data rapidly. So, it builds insecure and distrust among vehicles. In [2], the authors used blockchain in distributed network as it is being considered as more secure for the data credibility vehicular network. In [20], peer-to-peer (P2P) mechanism to communicate among vehicles. In intelligent vehicles, they cannot communicate with each other without compromise of trust. To overcome these problems, intelligent vehicle network is proposed based on blockchain. It also provides security mechanism based on ITS standard. Authors used combination of proof of work (PoW) and proof of authority (PoA) consensus mechanism to find the malicious vehicles and build the trust among them. However, there are some limitations in existing model for exchanging of traffic information services. Due to increase in data packet size of vehicles, the latency of message and rating are increased that affects the whole vehicular network.

60

3

H. Farooq et al.

Motivation

IoV technology is increasing rapidly with the increase in population and development. It has capability to share and exchange information with different IoVs using IoT devices [21]. Due to increase in IoV, it is expected that there are lack of wide roads, inefficient communication channel and low storage mechanisms to tackle rapid increase of data problem. IoV cannot trust with each other due to selfishness of nodes. To overcome these issues, authors in [22] proposed an architecture of distributed scalable access management system based on blockchain. It stores all distributed data on decentralized servers which reduce the storage burden. In [23], authors proposed decentralized, secure and reliable distributed transport management system based on blockchain for vehicular network. It stores all type of information and services from sensors, vehicles and computing devices. The stored information manages scalability, security and transparency. In [24], authors proposed secure blockchain-enabled IoV using reputation and contract theory. It encourages the IoV to share data with each other. The IoV nodes gets more incentive which provide valid sensor’s data while the IoV nodes have to pay penalty, which do not share valid sensor’s data. Reputation of IoV nodes are evaluated on the basis of historical interaction either valid or invalid response and recommended opinions either positive and negative rating from other IoV nodes. In [25–28], authors proposed blockchain-based secure network service provisioning for lightweight clients. They introduce service providers and service codes to be validated on-chain transaction which helps to reduce the overhead of data from different IoT devices. They also introduce off-chain transaction to maintain the validity states of services.

4

Problem Statement

The development of IoV network is originated from smart vehicles such as driverless cars. Nowadays, IoV have been applied in many smart cities to control traffic, enhance driving aids and environment. Due to increase in population, IoV networks are increasing day-to-day. To manage the large network, it faces different challenges such as communication delay, low data storage, selfishness of nodes, distrusted nodes and high congestion, etc. To overcome these challenges, authors in [23] proposed transport management system based on blockchain. In this system, ordinary nodes collect local sensor’s data and upload on controller node through miner nodes. The proposed system can easily resolves the high congestion problem. However, due to constantly uploading duplicated or unnecessary ordinary node’s data on controller node, the system can face low storage management issue which causes high communication delay in request and response of services. To overcome these issues, we propose a scalable, secure, transparent and distributed system for data storage mechanism based on Block-VN. The new data blocks are validated using PDP scheme. For data filtration, MTP-Argon2 technique is used. It filters the obtained raw data from different nodes to remove the

Block-VN: Efficient Communication and Storage System

61

duplicated and unnecessary data. Data storage, less delay in the service request or response, more sensor’s data sharing and secure communication channel are achieved using proposed system.

5

Proposed System Model of Distributed Block-VN

By getting motivation from existing model in [23], a Block-VN for smart IoV is proposed. In Fig. 1, there are three types of nodes: ordinary nodes are SVs, miner nodes having high computational power shown in red circle; and controller nodes are static base stations. Ordinary nodes have low storage, low battery life, low computing power and less resources. These nodes are connected with other ordinary and miner nodes to request services. These nodes can directly request for services with controller node, if there is no ordinary or miner node near to request services. Miner nodes have high storage, high computing power and high resources as compared to ordinary node. The purpose of miner nodes is to get request from ordinary nodes and provide valid services. All miner nodes are connected with each other to share services. They store all road side information, traffic information, ordinary nodes information, environmental information and performance of vehicles through different sensors like global positioning system (GPS), event data recorder (EDR), radio handset, small-scale radar, cameras and various kind of detection devices. Controller nodes are connected with cloud server and miner nodes. These nodes are connected P2P with other controller nodes. They provide the necessary information to all miners and ordinary nodes on large scale. They store the information of whole area provided by miner nodes. Cloud server gathers filtered information and services from all controller nodes. 5.1

Proposed Model of Miner Node

Each miner node has three main types of classification assets: computing, sensors and data storage. Each miner node has public local blockchain which contains all these assets shown in Fig. 2. It shares the duplicate copy of stored services to other miner nodes. IoV authority has all records of miner and ordinary nodes. When new service request is generated, it must be validated before stored in local blockchain. The service request is also sent to IoV authority which checks from stored record that either the given request is valid or not. There are different types of services which are provided by miner nodes like road accident, traffic blockage, sensor data of ordinary nodes and service sharing of data, etc. For example, miner nodes searches the services from local blockchain, if ordinary node wants the data of weather or traffic. If the required service is not found, then it requests the controller node to give response back to ordinary node. Miner nodes have low storage, low computing power and less sensor capabilities as compared to controller node. At some point, the storage of miner nodes will become full due constant storing of data from other ordinary and miner nodes.

62

H. Farooq et al.

Fig. 1. Proposed system model for blockchain based vehicular network.

To overcome this issue, we got motivation from existing flowchart in [29] and propose a storage mechanism which sets the time limit and blocks limit on local blockchain. After this step, new data blocks are created and stored on local blockchain. When the storage of local blockchain is full, minor node stores the previous blocks’ data on controller node and deletes all blocks’ data of local blockchain. In flowchart of local blockchain in Fig. 2, we assumed that the total time limit (M) of stored blocks is 24-hours. The total number of blocks (N) created in one hour is 100. The first block of local blockchain is genesis block (B0). When SV shares information with miner node, new data block is created by miner node and stored in memory cell of local blockchain. After one hour, the 100 memory-cell slot is full, then it shifts all data blocks to controller node and delete previous data blocks from its memory cells. In this way, the data storage issue of IoV is optimized and 24-hours of data is stored in the controller nodes. 5.2

Proposed Model of Controller Node

Each controller node is connected in distributed and static manner to provide services such as road traffic, routes, charging stations, etc. These services are stored and provided to all miners and ordinary nodes. Each controller node has public main blockchain ledger as shown in Fig. 3. Each controller node shares duplicate main blockchain ledger’s data to all other controller nodes. All received

Block-VN: Efficient Communication and Storage System

63

Fig. 2. Proposed system model of miner node and flowchart of local blockchain.

raw data from miner node can process and necessary data is stored in main blockchain. Controller nodes are responsible to provides all necessary data to miner or ordinary nodes. Cloud server gathers filtered information and services from all controller nodes. When ordinary nodes are moving from one place to another, they require necessary information like traffic signal data, traffic blockage, which path is shortest path and less time consuming, number of vehicles in their path, how much far the charging station if they have low battery, etc. All these types of informations are stored in miner nodes and controller nodes. IoV simply requests to its neighbor nodes which is either miner or ordinary node. If it cannot respond to its request, then it direct request to controller node. Controller node provides services to the IoV node. If IoV can request public data (which is: traffic on road, charging station and environment condition) then it cannot pays any cryptocurrency fee. However, if it requires private data (which is: vehicle moving position, video of road condition) it must have to pays the fee of specific data which is sets on the basis of smart contract. All the set of rules are written in programmatic form which are validated by all controller nodes. When smart contract is triggered, the new block is created in main blockchain. In flowchart of main blockchain, the work-flow is as follow. First, IoV sends the sensor node’s data to miner node for storage. Miner node checks the received data. If the data is same as previous one, then it will send “unnecessary data” message back to IoV node. If the data is new, necessary and important for other nodes, it starts processing and combining the data in blocks. When miner node’s storage is full, it transfers all raw data to controller node and delete all the previous memory slot data. The raw data is then filtered using MTP-Argon2 technique, which reduces the data by removing

64

H. Farooq et al.

duplicate and unnecessary data. In this technique, we defined the threshold value which contains the required information that is set on the basis of our scenario. For example, if only accident detections traffic state on road is required, then a threshold value is assigned and than compared with the raw data. If the compared data matches with threshold value then it adds the information and continue comparing. At last, we have filtered data that is transfered for validation to controller node. For validation, we use PDP consensus technique which ensure the integrity of stored data. It reduces the mining and computing power to validate the new data block. By using this technique, request or response delay of services are efficiently decreased. Due to decrease in delay, vehicles can efficiently communicate with each other and gain low latency. After validation process, blocks are created which contain previous hash, current hash, timestamp, Merkel root data and signature value. After that, blocks are uploaded in main blockchain and provide services to requester nodes and also uploaded in cloud server for further use in future.

Fig. 3. Proposed system model of controller node and flowchart of main blockchain.

6

Conclusion

In this paper, we observed that as the number of IoV increases, an efficient communication channel is required to share information with each other. There is a need of secure and improved storage mechanism to store all sensors information. A Block-VN model for smart IoV is proposed. We improved the storage and communication mechanism of existing system which increases the storage optimization and capacity using local blockchain and decreases the communication delay using cloud server. Due to these improvements, our Block-VN is secure, distributed, decentralized, scalable, trusted and transparent using blockchain.

Block-VN: Efficient Communication and Storage System

65

Future Work In future work, we will increase the size of vehicular network in different areas. We will improve the delay time and response of services. The storage problem remains a big issue because the number of IoV nodes or SVs are increasing dayto-day. We need an efficient detection and security mechanism for malicious IoV nodes or SVs that share invalid or wrong information to other IoV nodes.

References 1. Rodrigues, J.J.P.C., Hsing, R., Chen, M., Jiao, B., Vaidya, B.: Guest editorial on vehicular communications and applications. J. Netw. Comput. Appl. 5(36), 1273– 1274 (2013) 2. Yang, Z., Zheng, K., Yang, K., Leung, V.C.M.: A blockchain-based reputation system for data credibility assessment in vehicular networks. In: 2017 IEEE 28th Annual International Symposium on Personal, Indoor, and Mobile Radio Communications (PIMRC), pp. 1–5. IEEE (2017) 3. Meng, W., Wang, X.: Distributed energy management in smart grid with wind power and temporally coupled constraints. IEEE Trans. Industr. Electron. 64(8), 6052–6062 (2017) 4. Zafar, R., Mahmood, A., Razzaq, S., Ali, W., Naeem, U., Shehzad, K.: Prosumer based energy management and sharing in smart grid. Renew. Sustain. Energy Rev. 82, 1675–1684 (2018) 5. Fadlullah, Z.M., Quan, D.M., Kato, N., Stojmenovic, I.: GTES: an optimized gametheoretic697 demand-side management scheme for smart grid. IEEE Syst. J. 8, 588–597 (2014) 6. Mengelkamp, E., Notheisen, B., Beer, C., Dauer, D., Weinhardt, C.: A blockchainbased smart grid: towards sustainable local energy markets. Comput. Sci.-Res. Dev. 33(1–2), 207–214 (2018) 7. Samuel, O., Javaid, N., Awais, M., Ahmed, Z., Imran, M., Guizani, M.: A blockchain model for fair data sharing in deregulated smart grids. In: IEEE Global Communications Conference (GLOBCOM 2019) (2019) 8. Zahid, M., Javaid, N., Rasheed, M.B.: Balancing electricity demand and supply in smart grids using blockchain. MS thesis, COMSATS University Islamabad (CUI), Islamabad 44000, Pakistan, July 2019 9. Kang, J., Rong, Y., Huang, X., Maharjan, S., Zhang, Y., Hossain, E.: Enabling localized peer-to-peer electricity trading among plug-in hybrid electric vehicles using consortium blockchains. IEEE Trans. Industr. Inf. 13(6), 3154–3164 (2017) 10. Samaniego, M., Deters, R.: Blockchain as a service for IoT. In: 2016 IEEE International Conference on Internet of Things (iThings) and IEEE Green Computing and Communications (GreenCom) and IEEE Cyber, Physical and Social Computing (CPSCom) and IEEE Smart Data (SmartData), pp. 433–436. IEEE (2016) 11. Wang, J., Cao, Y., Li, B., Kim, H., Lee, S.: Particle swarm optimization based clustering algorithm with mobile sink for WSNs. Future Gener. Comput. Syst. 76, 452–457 (2017) 12. Lin, D., Tang, Y., Vasilakos, A.V.: User-priority-based power control in D2D networks for mobile health. IEEE Syst. J. 12(4), 3142–3150 (2017) 13. Raja, J.H.K., Javaid, N., Iqbal, S.: Blockchain based node recovery scheme for wireless sensor networks. MS thesis, COMSATS University Islamabad (CUI), Islamabad 44000, Pakistan, July 2019

66

H. Farooq et al.

14. Mateen, A., Javaid, N., Iqbal, S.: Towards energy efficient routing in blockchain based underwater WSNs via recovering the void holes. MS thesis, COMSATS University Islamabad (CUI), Islamabad 44000, Pakistan, July 2019 15. Noshad, Z., Javaid, N., Imran, M.: Analyzing and securing data using data science and blockchain in smart networks. MS thesis, COMSATS University Islamabad (CUI), Islamabad 44000, Pakistan, July 2019 16. Kazmi, H.S.Z., Javaid, N., Imran, M.: Towards energy efficiency and trustfulness in complex networks using data science techniques and blockchain. MS thesis, COMSATS University Islamabad (CUI), Islamabad 44000, Pakistan, July 2019 17. Javaid, A., Javaid, N., Imran, M.: Ensuring analyzing and monetization of data using data science and blockchain in loT devices. MS thesis, COMSATS University Islamabad (CUI), Islamabad 44000, Pakistan, July 2019 18. Sharma, P.K., Singh, S., Jeong, Y.-S., Park, J.H.: DistBlockNet: a distributed blockchains-based secure SDN architecture for IoT networks. IEEE Commun. Mag. 55(9), 78–85 (2017) 19. Shelby, Z., Hartke, K., Bormann, C.: The constrained application protocol (CoAP). RFC 7252 (proposed standard), RFC Editor, Fremont, CA, USA, pp. 1–112, June 2014. Updated by RFC 7959 20. Gerla, M., Kleinrock, L.: Vehicular networks and the future of the mobile internet. Comput. Netw. 55(2), 457–469 (2011) 21. Bahga, A., Madisetti, V.: Internet of Things: a hands-on approach. Vpt (2014) 22. Novo, O.: Blockchain meets IoT: an architecture for scalable access management in IoT. IEEE IoT J. 5(2), 1184–1195 (2018) 23. Sharma, P.K., Moon, S.Y., Park, J.H.: Block-VN: a distributed blockchain based vehicular network architecture in smart city. J. Inf. Process. Syst. 13(1), 184–195 (2017) 24. Kang, J., Xiong, Z., Niyato, D., Ye, D., Kim, D.I., Zhao, J.: Towards secure blockchain-enabled internet of vehicles: optimizing consensus management using reputation and contract theory. IEEE Trans. Veh. Technol. 68, 2906–2920 (2019) 25. Xu, Y., Wang, G., Jidian Yang, J., Ren, Y.Z., Zhang, C.: Towards secure network computing services for lightweight clients using blockchain. Wirel. Commun. Mobile Comput. (2018) 26. Rehman, M., Javaid, N., Awais, M., Imran, M., Naseer, N.: Cloud based secure service providing for IoTs using blockchain. In: IEEE Global Communications Conference (GLOBCOM 2019) (2019) 27. Ali, I., Javaid, N., Iqbal, S.: An incentive mechanism for secure service provisioning for lightweight clients based on blockchain. MS thesis, COMSATS University Islamabad (CUI), Islamabad 44000, Pakistan, July 2019 28. Naz, M., Javaid, N., Iqbal, S.: Research based data rights management using blockchain over ethereum network. MS thesis, COMSATS University Islamabad (CUI), Islamabad 44000, Pakistan, July 2019 29. Kushch, S., Castrillo, F.P.: A rolling blockchain for a dynamic WSNs in a smart city. CoRR arXiv:abs/1806.11399 (2018)

Electric Vehicles Privacy Preserving Using Blockchain in Smart Community Omaji Samuel1 , Nadeem Javaid1(B) , Faisal Shehzad1 , Muhammad Sohaib Iftikhar1 , Muhammad Zohaib Iftikhar1 , Hassan Farooq1 , and Muhammad Ramzan2,3 1

Department of Computer Science, COMSATS University, Islamabad 44000, Pakistan [email protected], [email protected] 2 Department of Computer Science and IT, University of Sargodha, Sargodha, Pakistan 3 Pakistan School of Systems & Technology, University of Management and Technology, Lahore, Pakistan

Abstract. During the process of charging, electric vehicle’s location is usually revealed when making payment. This brings about the potential risk to privacy of electric vehicle. We observe that the trade information recorded on blockchain may raise privacy concern and therefore, we propose a blockchain oriented approach to resolve the privacy issue without restricting trading activities through (, δ)-differential privacy. The proposed scheme does not only preserve the electric vehicle’s location; however, prevents semantic, linking and data mining based attacks. Simulation results show that as the privacy level increases, the risk revealing decreases as well. Keywords: Blockchain · Demand side management Energy trading and privacy preserving

1

· Electric vehicle ·

Introduction

Presently, there has been a tremendous advancement in the development of electric vehicles (EVs). EVs as part of demand-side management provide more benefits and environmental advantages [1]. Several countries of the world have started adopting EVs for de-carbonization and mobile energy storage to achieve a green city [2]. As the number of EV continues to increase, there is a need to create a charging infrastructure. Authors in [3] and [4] have proposed an optimal settings of charging station (CS) and optimal scheduling to minimize vehicular resources and time. However, authors do not give emphasis on privacy related issues of EV such as location, price and consumption. Traditionally, EV is controlled and monitored by a centralized system [5]. Besides, the centralized system also faces issues of privacy and security like other known centralized schemes [6]. Also, the centralized system lacks the ability to enforce the decision-making process c Springer Nature Switzerland AG 2020  L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 67–80, 2020. https://doi.org/10.1007/978-3-030-33506-9_7

68

O. Samuel et al.

on autonomous EVs. Solutions for aforementioned problem include peer-to-peer and decentralization via blockchain [7]. The Table 1 provides description of the parameters or variables used throughout this paper. Table 1. Parameters and variables Notations

Descriptions

Apmin

Minimum acceptance probability

Apk

Apn

The kth charging station’s (CS) assignment probability The nth electric vehicle’s (EV) acceptance probability

bi,j and zj,i Row and column stochastic matrices P rb

The bth blockchain offered price by CS

CSksel

The kth CS’s selection probability based on P rb and dkn

dkn

Distances of nth EV from the kth CS

gb and prb

The broadcast parameters of distance and offered price, respectively

lap(y)

Cumulative Laplace distribution for the given input y

N − and N + Cardinality of the out-bound and in-bound flow for ith nodes and jth vertices Pnreq

Energy the nth EV required from CS

The concept of blockchain is introduced in 2008 by Satoshi Nakamoto [8] and Bitcoin is its first application. Blockchain is a shared ledger that facilitates the process of recording transaction and tracking assets in a distributed network. Within the last decade, blockchain is now the focus of many researchers, stakeholders and industries spanning from voting, healthcare, finance, real estate, utilities [9], Internet of Things [10,11], wireless sensor network [12,13]. Blockchain provides decentralization, immutability, trustfulness [14], traceability, secure environment and data storage. Advantages of blockchain include real-time transaction and payment; quick response time; avoids duplication; prevents fraud and cyber attacks; minimizes time-consuming vetting process and provides transparency. Several studies in [15–21] used blockchain as a privacy-preserving mechanism for data aggregation; privacy protection and energy storage; secure classification of multiple data; incentive announcement network for a smart vehicle; crowdsensing applications; dynamic tariff decision, payment mechanism for vehicle-to-grid, data right management [22], and incentive for lightweight clients [23]. However, blockchain solution is inefficient to tackle data mining and linking attacks [24]. These attacks take advantage of exposed information stored in a block and privacy is disclosed by linking records of other datasets. From the literature above and the inspiration obtained from the work of [25], we derive our problem statement based on the following analogies: assuming we have a setup of centralized server coordinating the trading between EVs and CSs. The server publishes CSs with offered prices and locations and EVs autonomously choose the preferred CSs. The benefit is that the EVs do not need to disclose their exact locations and the server does not know the CSs which EVs

Electric Vehicles Privacy Preserving Using Blockchain in Smart Community

69

have selected. The disadvantage is that the server has no control over the assignment of CSs and the EVs can select CSs based on their distances and offered prices. In contrast to the centralized approach, we have a setup of blockchainbased energy trading between EVs and CSs. The EVs send their locations and the required quantity of energy to the blockchain. The blockchain controls and allocates nearby CSs to the EVs while maximizes EVs’ acceptance rates. However, EVs’ private information such as locations are revealed to the blockchain during the payment process, which raise privacy concerns to the owners of EV. In a privacy-preserving perspective, information recorded on blockchain may raise privacy concern [26]. Nevertheless, the traditional system cannot protect EVs’ information within this scenario. Hence, we propose a system that protects EVs’ location while ensuring fair energy trading. The proposed system will prevent re-identification attack via private blockchain since EVs’ transaction records are stored across different networks. Thus, honest-but-curious EVs cannot infer the identity of EVs through observational studies. The organization of the paper is as follows: Sect. 2 provides the paper contributions while Sect. 3 discusses the proposed system model as well as problem formulations. Simulation results are discussed in Sects. 4 and 5 provides the conclusion and future work.

2

Contributions

In this section, the contributions of this paper are as follows. 1. We protect EV’s privacy from future blockchain based data transmission by defending EV against a possible breach. Our proposed scheme ensures complete accuracy since it is implemented using real dataset and it is efficiently adoptable since all computations are done off-chain, thereby reducing the number of computing resources on the chain. 2. Differential privacy is proposed by using the consensus energy management algorithm [27] to conceal the broadcast information. 3. Two types of blockchain are proposed: private blockchain located at rural area achieves the following: prevents re-identification and data mining attacks due to membership restrictions and provides subsidy for charging; and public blockchain located in urban area resolves the scalability issue.

3 3.1

Proposed System Model and Problem Formulations System Overview

In the proposed system in Fig. 1, three fundamental entities with distinct functionalities are studied. Firstly, the EV as an entity that requires energy for charging, secondly, CS as an entity that acts as an energy provider. However, CS gets charged by the main grid if its internal generated energy is insufficient. In addition, the CS charged EV on the basis of the offered price [1]. Lastly, the

70

O. Samuel et al.

aggregator (blockchain) acts as a broker between the EV and CS for fair energy transactions. EVs send charging request and location to the aggregator; aggregator broadcasts this information to the blockchain network. CSs who meet this requirement response back with offered price and location to the aggregator. Aggregator reports this information to the requesting EV and CS is assigned to EV on the basis of price and location.

Fig. 1. Proposed system. EV: electric vehicle, and CS: charging station.

3.2

Fig. 2. Illustration of the system network.

Blockchain Based Location Privacy Preserving with Differential Privacy

In energy trading, the EV’s charging request task is denoted as RDT, while CS’s discharging response task given as RST. Thus, the rationality of RDT and RST are as follows: RDT: In the blockchain, EVs addresses are anonymous; hence, the blockchain receives all RDT from EVs and broadcast them. However, blockchain is unaware of the locations and charging request of EVs. In addition, EVs choose charging locations based on reduced P rb and dkn , to minimize traveling costs. Thus, blockchain has no control over the activities of EVs [25]. RST: CSs send lk and P rb to the blockchain. Blockchain assigns CS to EV based on dkn . Thus, the blockchain controls activities of EVs. Since RDT and RST are known to the blockchain, which may raise privacy concerns [25]. A blockchain knowledge base (BKB) that stores all records of CSs and EVs, respectively is proposed. BKB = {EVn , CSk , dkn , Apk , CSksel , ln , lk , Hn {ln , Pnreq }, Hk {lk , P rb }},

(1)

where EVn and CSk are lists of EVs and CSs, respectively. Hn and Hk are the histories of EVs and CSs; while, ln and lk are the locations of EVs and CSs, respectively.

Electric Vehicles Privacy Preserving Using Blockchain in Smart Community

71

3.2.1 Adversary Model We assume that there are honest-but-curious aggregators on the blockchain network. These curious aggregators disclose information of EVs for selfish interest or financial benefits. Also, the curious aggregator known as CurAg can join the public or private blockchain to gain information [25]. Moreover, the EV’s current, past, and future location can be leaked by CurAg during charging and payment process. The attacker can be any participant in the blockchain network. Although, an attacker in the public blockchain can access transactional records of EVs, while attacker as EV can join the private blockchain to get transaction records of other EVs. Besides, access to other private blockchain is hindered due to membership restrictions [25]. Attacker as an aggregator may have access to transactional records of his own dataset. However, it is impossible to access records of other aggregators [25]. 3.2.2 Privacy-Preserving in Blockchain The use of blockchain provides anonymization, ensures that EV fulfilled an agreement with the CSs and decentralized the system to prevent a single point of failure. Also, private blockchain prevents the re-identification attack since each aggregator has distinct transactional history. Thus, it is infeasible for an attacker to access transactional records of all aggregators without poisoning their records [25]. Process of blockchain: 1. Registration: EVs and CSs are required to register with their private sk and public pk key for verification and authentication. 2. CS price mechanism: the price offered to EV is determined by CS. 3. Smart contract: CSs and EVs are required to make an initial token deposit which prevents double spending and false declaration of information. 4. EV’s assignment: EV prefers CS on the basis of ln and P rb , and make requests accordingly. However, EV is validated based on uploaded ln in the urban area; thereby, granting access to a specific CS. 5. CS’s selection: Blockchain ensures that CSs have the available discharging capacities from the urban area to charge EVs. Otherwise, a new block is created with deduction of the deposited token from CS’s account. 6. Consensus: EVs make charging request to the blockchain. Miner validates the authenticity of the request. In this paper, proof of authority (PoA) is used [28]. If requests are accepted, then payment transfer is made to CS’s wallet account. Otherwise, if the claim is falsified, the token deposit is used as a penalty. Payment process: EVs wish to get charged at the closest possible distance to their locations. Assuming all CSs sell energy at a fixed price, the acceptance probability of EV will drop. Thus, the acceptance of EV is enhanced if CSs discharge at different offered prices. Hence, acceptance probability of EV is calculated in Eq. (2) [25].

72

O. Samuel et al.

Apn =

p dmax n,k − Amin ; 0 ≤ Apk ≤ 1, dmax n,k

Apk = 1 − (1 − Apn )R .

(2) (3)

We assume CSs covers all ln of EVs, while some CSs do not cover EV’s ln . This scenario is depicted in Fig. 1. Thus, the acceptance probability of EV is proportional to the lk of CS. However, from Fig. 1, the CS enclosed in green circle gets the highest acceptance by EVs since it covers all locations. The CS’s assignment probability is calculated in Eq. (3); where R = 3 is the number of regions. While the minimum distance of EV from CS is calculated in Eq. (4) [25]. Apmin = 2r,

(4)

We consider the isolated CS, i.e., CS that covers only few EVs’ location; hence, the average distance AV Gdiso is calculated by counting R within EV’s maximum travel distance to CS as given in Eq. (5) [25]; where r = 2 is a constant value. AV Gdiso = dmax n,k − r.

(5)

The CS’s selection probability is solved as the hyperbolic function of the P rb and dkn and given in Eq. (6) [25].  x x e −e k max x −x , if dn ≤ dn,k sel CSk = e +e (6) 0, if otherwise, where, x=α

P rb ; 0 < CSksel ≤ 1, dkn

(7)

where α is a constant value. Assumptions: from Eq. (6), CS with lower distance and minimum offered price is selected with high probability; CS with higher distance and minimum offered price is selected with low probability, whereas, CS whose distance is more than the maximum distance of the concerned EV with higher offered price is not selected. To further protect EV’s location as well as the amount paid to CS, {, δ}differential privacy is proposed in this paper. The communication between EVs and CSs formed a directed graph G, such that G = {V, E}, where V is a set of nodes and E is set of edges. V = N ∪ K and lets {j, i} ∈ E if and only if node i communicates with node j [27]. Node i is the out-bound of node j; however, self loop, i.e., {j, j} is not considered in this paper [27]. We derive the in-bound and out-bound values from Fig. 2 as given in Table 2.

Electric Vehicles Privacy Preserving Using Blockchain in Smart Community

73

Table 2. Cardinality of in-bound and out-bound derived from Fig. 2. A B C D 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 N

−

5 7 7 5 1 1 2 2 1 1 2 2 2 1

3

1

2

2

1

3

3

1

1

3

N+ 5 7 7 5 1 2 4 2 2 4 3 4 1 1

3

1

3

1

2

2

1

1

2

2

In Table 2, stochastic row and column matrices are generated using Eqs. (8) and (9), respectively [27].

bi,j

zj,i

⎧ 1 i ∈ N+ ⎪ ⎨ |N + |+1 , if |N + | = 1 − i=1 bi,j , if i = j ⎪ ⎩ 1 if i = j, |N + | , ⎧ 1 if i ∈ N + ⎪ ⎨ |N − |+1 , − |N | = 1 − i=1 zj,i , if i = j ⎪ ⎩ 1 if i = j. |N − | ,

(8)

(9)

We generate the blockchain broadcast information about the dkn and P rb using Eqs. (10) and (11), respectively [27]. ⎧ min + ⎪ ⎨dn,k , if i ∈ N − (10) gb = dmax n,k , if i ∈ N ⎪ ⎩|N − | if i = j, i=1 bi,j gb + ηprb ,  P rbmin , if i ∈ N + prb = (11) P rbmax , if i ∈ N + , max where, dmin n,k and dn,k are minimum and maximum distances of EVs from CSs; whereas, P rbmin and P rbmax are minimum and maximum offered prices and η = 0.8 is scaling factor. The broadcast information is modified by adding a cumulative Laplace noise as given in Eqs. (12) and (13). Thus, Eq. (1) is updated with the new broadcast information as given in Eq. (16).  gb+1 bi,j + lap(y), if i ∈ N + (12) gb+1 = gb bi,j + lap(y), if i ∈ N − ,  zj,i prb+1 + lap(y), if i ∈ N + prb+1 = (13) zj,i prb + lap(y), if i ∈ N − ,

where

74

O. Samuel et al.

 lap(y) =

σ 2y √ e , if y < 0.5 2 −σ √ e2(1−y) , if y ≥ 2

0.5,

(14)

where max(y) − min(y) ,  BKB(b + 1) = {EVn , CSk , gb+1 , Apk , CSksel , ln , lk , Hn {ln , Pnreq }, Hk {lk , prb+1 }}. σ=

(15)

(16)

BKB(b + 1) is broadcast to the blockchain network. Even if an attacker has the broadcast information, it will be impossible to infer the ownership of information. val Thus, we define the privacy risk of EVs Ri,n over their private information BKB(b + 1) as [29]: val Ri,n (BKB(b + 1)) = P C(BKB(b + 1)).SL(BKB(b + 1)),

(17)

where the privacy concern P C(BKB(b + 1)) ∈ {0, 1} and sensitivity level SL(BKB(b + 1)) ∈ {0, 1}. Using (, δ)-differential privacy, the SL(BKB(b + 1)) is obtained by finding their differences (f (G1 ) − f (G2 )), i.e., the set G1 and G2 differing on at most one element [29]. However,  and δ are privacy levels of price and location with given values of 1, 2, 3, 4, 5 and 6, respectively. 3.3

Blockchain Smart Contract

Figure 3 shows smart contract for the proposed scheme. Blockchain is unaware of when and where EV will go; hence, EV’s exact location is preserved. Since CS status in public blockchain differs from that of a private blockchain. Thus, blockchain ensures CS is available in the urban context before assigning EV to prevent void contract [25]. Similarly, private blockchain must verify if CS is assigned to public blockchain or not before assigning EV to prevent void contract. For EV to make a charge request, its credit value (CR) is verified and authenticated with the sk and pk to ensure EV has been registered. If CR is not empty, EV can make a charge request by uploading its region and Pnreq to the aggregator. The aggregator verifies region via region identity Rid. The Rid is used to determine if EV is in a rural area (private blockchain) or urban area (public blockchain) for which the specified offered prices are determined. Also, the offered prices for types of EV are verified via EV identity EV id. Once CS supplied the required charging, payment is made to CS’s wallet account by concerned EV. If the current time of CS is more than the agreed due time CSdueT ime to verify the payment, a token deduction is made against such CS.

Electric Vehicles Privacy Preserving Using Blockchain in Smart Community

75

Fig. 3. Smart contract.

4

Simulation Results

Simulation results and discussions are presented in this section. 4.1

Experimental Setup

We develop our blockchain using the ethereum platform [30] with the following dependencies; Truffle v5.0.8 (core: 5.0.8), Solidity v0.5.0 (solc-js), Node v10.13.0 and Web3.js v1.0.0-beta.37. Also, we customize our codes using JavaScript. The hash operations are performed using the solidity keccak256 library and some of the data used are randomly generated, if not specified. Simulation results are generated using MATLAB2018. The hardware platform is a Dell i5, with 8 GB ram and CPU of 1.60 Hz and 1.80 GHz. 4.2

Simulation Dataset

In this section, simulation results describe the evaluation of the proposed blockchain based privacy preserving for EV’s location. In this paper, 20 EVs and 4 CSs are used. The offered prices by the four CSs and the real distance between EVs from CSs are taken from [1]. The EV’s battery capacity and CSs’ specifications are also taken from [1] (Figs. 4 and 5).

76

O. Samuel et al.

Fig. 4. Price offered by four CSs [1].

4.3

Fig. 5. Distance of EVs from CSs [1].

Evaluation of EV’s Selection and CS’s Assignment Probability

This section discusses the EV’s acceptance and CS’s assignment probability. EV accepts CS with the closest distance from its location. By assumption, if all CSs announce the same offered for charging of EV, then EV’s selection probability will be reduced. Using Eqs. (6) and (7), the Fig. 6 shows the CS’s selection probability is close to the maximum limit. The results further show that the EV’s acceptance of CS can only be achieved if the number of counted regions fall within the EV’s maximum distance to the CS. Thus, the probabilities of all CSs either as an edge or as isolated for being selected will be increased. However, the offered price by CS also determines its acceptance by EV. The CS with the closest distance and the lowest offered price has a high probability of being accepted. Also, the CS with the longest distance and lowest offered price is accepted with a low probability. Nevertheless, if the distance to CS is more than the maximum distance of EV, CS may be rejected even if it offers the lowest price. Using Eq. (2), the probability of CS being assigned to EV is based on distance and is proportional to regions where the distance is covered.

Fig. 6. Various probabilities of CSs and EVs.

Electric Vehicles Privacy Preserving Using Blockchain in Smart Community

4.4

77

Privacy Preserving Evaluations Using the Proposed Blockchain and Differential Privacy

This section discusses the (, δ)-differential privacy-preserving for the proposed blockchain scheme. In Figs. 7 and 8, the individual EV privacy is protected against set theory attack [26]. The results further explained that as the privacy level increases, the risk revealing decreases as well. The proposed scheme also prevents linking based attack via (, δ)-differential privacy which hindered adversary activities [26]. The private blockchain approach of the scheme prevents data mining attack since transaction records of EVs are scattered across different private network which is strengthen by membership restriction.

Fig. 7. Risk revealing versus privacy level for the offered price.

4.5

Fig. 8. Risk revealing versus privacy level for the distance.

Computational Blockchain Cost Analysis

Creating a new block in blockchain requires strict verification process from an authorized node. In this paper, PoA adopted from our previous work [28] where Pagerank rank mechanism is used to select the node as the authorized node on the basis of its reputation score. Hence, the latency of confirmation time is reduced since only authorized node is allowed to create a block and computes the assignment and selection probability off-chain, thereby reducing the number of computing resources needed on the chain. From Fig. 3, the time complexity of the smart contract is less than O(n) [25]. Hence, the computational burden has no influence on the blockchain.

5

Conclusion

This paper examines that transactional record on blockchain may raise privacy concern such as disclosing private information like location and price. Three ways locations of EV are disclosed such as current, previous and future are examined. To preserve the location privacy of EVs, a private blockchain is incorporated

78

O. Samuel et al.

which prevent re-identification attack due to membership restrictions. Thus, the transactional record histories of EVs cannot be inferred by the attacker since records are spread across the network. To further preserve the records, differential privacy is exploited to conceal the records against observational studies. The CS’s assignment and EV’s selection probability are derived based on the offered price and location of EVs. Simulation results demonstrate that privacy is achieved through risk revealing metric. Also, the proposed approach prevents semantic based attack since private blockchain is involved; data mining and linking based attack since differential privacy is used. In the future, the neighboring energy trading where dynamic pricing is an issue for charging the EVs in a smart community will be explored. Furthermore, we intend to consider the initial state as the possible privacy breach, such that even if an attacker has the exact knowledge about the initial state of other EVs, it will be difficult to breach their privacy.

References 1. Aujla, G.S., Kumar, N., Singh, M., Zomaya, A.Y.: Energy trading with dynamic pricing for electric vehicles in a smart city environment. J. Parallel Distrib. Comput. 127, 169–183 (2019) 2. De Hoog, J., Alpcan, T., Brazil, M., Thomas, D.A., Mareels, I.: Optimal charging of electric vehicles taking distribution network constraints into account. IEEE Trans. Power Syst. 30, 365–375 (2014) 3. Lin, C.-C., Deng, D.-J., Kuo, C.-C., Liang, Y.-L.: Optimal charging control of energy storage and electric vehicle of an individual in the internet of energy with energy trading. IEEE Trans. Industr. Inf. 14, 2570–2578 (2017) 4. Aujla, G.S., Jindal, A., Kumar, N.: EVaaS: electric vehicle-as-a-service for energy trading in SDN-enabled smart transportation system. Comput. Netw. 143, 247– 262 (2018) 5. Liu, H., Zhang, Y., Yang, T.: Blockchain-enabled security in electric vehicles cloud and edge computing. IEEE Netw. 32, 78–83 (2018) 6. Aitzhan, N.Z., Svetinovic, D.: Security and privacy in decentralized energy trading through multi-signatures, blockchain and anonymous messaging streams. IEEE Trans. Dependable Secure Comput. 15, 840–852 (2016) 7. Kang, J., Rong, Y., Huang, X., Maharjan, S., Zhang, Y., Hossain, E.: Enabling localized peer-to-peer electricity trading among plug-in hybrid electric vehicles using consortium blockchains. IEEE Trans. Ind. Inf. 13, 3154–3164 (2017) 8. Nakamoto, S.: Bitcoin: a peer-to-peer electronic cash system, vol. 1, pp. 1–9 (2008) 9. Zahid, M., Javaid, N., Rasheed, M.B.: Balancing electricity demand and supply in smart grids using blockchain. M.S. thesis, COMSATS University Islamabad (CUI), Islamabad 44000, Pakistan, July 2019 10. Rehman, M., Javaid, N., Awais, M., Imran, M., Naseer, N.: Cloud based secure service providing for IoTs using blockchain. In: 2019 IEEE Global Communications Conference: Communication & Information Systems Security, USA, pp. 1–7 (2019) 11. Javaid, A., Javaid, N., Imran, M.: Ensuring analyzing and monetization of data using data science and blockchain in loT devices. M.S. thesis, COMSATS University Islamabad (CUI), Islamabad 44000, Pakistan, July 2019

Electric Vehicles Privacy Preserving Using Blockchain in Smart Community

79

12. Mateen, A., Javaid, N., Iqbal, S.: Towards energy efficient routing in blockchain based underwater WSNs via recovering the void holes. M.S. thesis, COMSATS University Islamabad (CUI), Islamabad 44000, Pakistan, July 2019 13. Khan, R.J.U.H., Javaid, N., Iqbal, S.: Blockchain based node recovery scheme for wireless sensor networks. M.S. thesis, COMSATS University Islamabad (CUI), Islamabad 44000, Pakistan, July 2019 14. Kazmi, S.Z., Javaid, N., Imran, M.: Towards energy efficiency and trustfulness in complex networks using data science techniques and blockchain, M.S. thesis, COMSATS University Islamabad (CUI), Islamabad 44000, Pakistan, July 2019 15. Noshad, Z., Javaid, N., Imran, M.: Analyzing and securing data using data science and blockchain in smart networks. M.S. thesis, COMSATS University Islamabad (CUI), Islamabad 44000, Pakistan, July 2019 16. Guan, Z., Si, G., Zhang, X., Longfei, W., Guizani, N., Xiaojiang, D., Ma, Y.: Privacy-preserving and efficient aggregation based on blockchain for power grid communications in smart communities. IEEE Commun. Mag. 56, 82–88 (2018) 17. Shen, M., Tang, X., Zhu, L., Xiaojiang, D., Guizani, M.: Privacy-preserving support vector machine training over blockchain-based encrypted IoT data in smart cities. IEEE Internet Things J. 1, 1–11 (2019) 18. Li, L., Liu, J., Cheng, L., Qiu, S., Wang, W., Zhang, X., Zhang, Z.: CreditCoin: a privacy-preserving blockchain-based incentive announcement network for communications of smart vehicles. IEEE Trans. Intell. Transp. Syst. 19(7), 2204–2220 (2018) 19. Wang, J., Li, M., He, Y., Li, H., Xiao, K., Wang, C.: A blockchain based privacypreserving incentive mechanism in crowdsensing applications. IEEE Access 6, 17545–17556 (2018) 20. Gao, F., Zhu, L., Shen, M., Sharif, K., Wan, Z., Ren, K.: A blockchain-based privacy-preserving payment mechanism for vehicle-to-grid networks. IEEE Netw. 99, 1–9 (2018) 21. Feng, Q., Debiao, H., Zeadally, S., Khan, M.K., Kumar, N.: A survey on privacy protection in blockchain system. J. Netw. Comput. Appl. 1, 1–14 (2018) 22. Naz, M., Javaid, N., Iqbal, S.: Research based data rights management using blockchain over Ethereum network. M.S. thesis, COMSATS University Islamabad (CUI), Islamabad 44000, Pakistan, July 2019 23. Ali, I., Javaid, N., Iqbal, S.: An incentive mechanism for secure service provisioning for lightweight clients based on blockchain. M.S. thesis, COMSATS University Islamabad (CUI), Islamabad 44000, Pakistan, July 2019 24. Craig, D.W.: Understanding the links between privacy and public data sharing. Nat. Methods 13, 211 (2016) 25. Yang, M., Zhu, T., Liang, K., Zhou, W., Deng, R.H.: A blockchain-based location privacy-preserving crowdsensing system. Future Gener. Comput. Syst. 94, 408–418 (2019) 26. Gai, K., Yulu, W., Zhu, L., Qiu, M., Shen, M.: Privacy-preserving energy trading using consortium blockchain in smart grid. IEEE Trans. Ind. Inf. 1, 1–10 (2019) 27. Zhao, C., Chen, J., He, J., Cheng, P.: Privacy-preserving consensus-based energy management in smart grids. IEEE Trans. Sig. Process. 66, 6162–6176 (2018) 28. Samuel, O., Javaid, N., Awais, M., Ahmed, Z., Imran, M., Guizani, M.: A blockchain model for fair data sharing in deregulated smart grids. In: 2019 IEEE Global Communications Conference: Communication & Information Systems Security, USA, pp. 1–7 (2019)

80

O. Samuel et al.

29. Yassine, A., Shirehjini, A.A.N., Shirmohammadi, S.: Smart meters big data: game theoretic model for fair data sharing in deregulated smart grids. IEEE Access 3, 2743–2754 (2015) 30. Wood, G.: Ethereum: a secure decentralised generalised transaction ledger. Ethereum project yellow paper, vol. 151, pp. 1–32 (2014)

A Nodes Selection Algorithm for Fault Recovery in the GTBFC Model Ryuji Oma1(B) , Shigenari Nakamura1 , Tomoya Enokido2 , and Makoto Takizawa1 1

Hosei University, Tokyo, Japan [email protected], [email protected], [email protected] 2 Rissho University, Tokyo, Japan [email protected]

Abstract. In order to increase the performance of the IoT (Internet of Things), the FC (Fog Computing) model is proposed. Here, subprocesses of an application process to handle sensor data are performed on fog nodes in addition to servers. In this paper, we newly introduce join subprocesses with multiple input parameters. If a node gets faulty, the child nodes are disconnected and have to be reconnected to new parent nodes. In our previous studies, new parent nodes are selected at the same level as the faulty node. In this paper, we newly propose a GTBFC (General TBFC) algorithm by which disconnected nodes are reconnected to new parent nodes so that the data obtained from the output data of every disconnected node is processed by an ancestor join node of the faulty node. In the evaluation, we show the energy consumption and execution time of a new parent node selected by the GTBFC algorithm. Keywords: Process tree · Equivalent nodes Join nodes · GTBFC algorithm

1

· Faults of fog nodes ·

Introduction

In the IoT (Internet of Things) [10], not only computers like servers and clients but also millions of sensor and actuator devices are interconnected in networks. In the cloud computing model [2], data collected by sensors is processed by application processes on servers. Networks are congested to transmit the huge volume of sensor data and servers are also overloaded to handle the sensor data in realtime manner. The FC (Fog Computing) model [17] is proposed to reduce the communication and processing traffic to handle sensor data in the IoT. Sensor data is processed and the output data is sent to another fog node. Thus, servers in clouds finally receive data processed by fog nodes. It is critical to reduce the electric energy consumed by fog nodes and servers. The power consumption models of a computer are proposed [3–7,11]. In order to reduce the energy consumption and execution time of fog nodes and servers, c Springer Nature Switzerland AG 2020  L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 81–92, 2020. https://doi.org/10.1007/978-3-030-33506-9_8

82

R. Oma et al.

the TBFC (Tree-Based FC) [8,9,12] model is proposed. Here, fog nodes are hierarchically structured in a height-balanced tree, where root node is a server and leaf node is an edge node. Each node receives data from child nodes and sends processed data to a parent node. If a node gets faulty, the child nodes are disconnected and have to be reconnected to a new parent node. In order to be tolerant of node faults, the FTBFC (Fault-tolerant TBFC) [15], MFTBFC (Modified FTBFC) [13], and GTBFC (General TBFC) [14] models and the subprocess transmission strategy [16] are proposed. In these models, each subprocess is assumed to receive one type of data, i.e. one input parameter. In this paper, we consider more general types of subprocesses which have multiple types of input data, i.e. subprocesses with multiple parameters. These nodes are referred to as join nodes. If a join node receives different data of the input parameters, the node outputs different data. Hence, if a join node gets faulty, every disconnected node is required to send output data to a join node which supports the same join subprocess as the faulty node. The equivalent node is an alternate node. In another way, a subprocess of a faulty node is sent to a parent or child nodes of a faulty node. A node to which the subprocess is sent is a surrogate node. In this paper, we propose a GTBFC algorithm by which one recovery algorithm is selected in types of data transmission and subprocess transmission ones so that the energy consumption of alternate or surrogate node is minimized. In the evaluation, we show the energy consumption of a new parent node selected by GTBFC algorithm. In Sect. 2, we propose the GTBFC model. In Sect. 3, we discuss the energy consumption and execution time of a fog node. In Sect. 4, we propose the GTBFC algorithm. In Sect. 5, we evaluate the GTBFC algorithm.

2 2.1

GTBFC Model Tree Structure of an Application Process

[Example1]. We consider a GTBFC (General TBFC) model composed of eight nodes [Fig. 1]. A root node f is a server. The sensors s1 and s2 send a pair of temperature and time data to edge nodes f111 and f112 and the other sensors s3 and s4 send a pair of humidity and time data to edge nodes f121 and f122 every one second. A subprocess t-aggregate of the edge nodes f111 and f112 calculates an average value of temperature data collected every one second for one minute. Thus, t-aggregate has one type of input temperature data. Another subprocess haggregate of the edge nodes f121 and f122 calculates an average value of humidity data collected every one second for one minute. A parent fog node f1i receives input data from a pair of child nodes f1i1 and f1i2 . A pair of subprocesses tmerge and h-merge of fog nodes f11 and f12 , respectively, sort and merge a pair of temperature data and a pair of humidity data in the same minute and sends the merged data to its parent fog node f1 . A subprocess join of the fog node f1 joins temperature and humidity data from f11 and f12 into one data, respectively. A subprocess store of the root node f receives joined data from f1 and stores

A Nodes Selection Algorithm in the GTBFC Model

83

the data in the database DB. Thus, the application process p is a hierarchically composed of subprocesses t-aggregate, h-aggregate, t-merge, h-merge, join, and store as shown in Fig. 1(1). Figure 1(2) shows a GTBFC tree for the process tree P of Fig. 1(1).

Fig. 1. GTBFC model.

An application process P is hierarchically composed of subprocesses. A subprocess p is a root node of a process tree P . The root subprocess p has child subprocesses p1 , . . . , pc (c ≥ 1). Then, each subprocess pi has child subprocesses pi1 , . . . , pi,ci (ci ≥ 1). Thus, each subprocess pF i receives cF i types of input data dF i1 , . . . , dF i,cF i from child subprocesses pF i1 , . . . , pF i,cF i (cF i ≥ 1), respectively, i.e. pF i (dF i1 , . . . , dF i,cF i ). Then, the subprocess pF i sends processed data to a parent subprocess pF . A subprocess pF which has multiple child subprocesses is a join subprocess, cF i ≥ 2. On the other hand, a subprocess which has one child subprocess is simple. Here, the label F of a subprocess pF shows a path from a root subprocess p to the subprocess pF in the process tree P . Each non-root subprocess in a process tree P is supported by one or more than one fog node. A leaf subprocess pF is supported by an edge node. Thus, each subprocess pF is installed in a fog node fR . Let p(fR ) be a subprocess supported by a node fR . Fog nodes are also structured in a fog node tree F . A root node f shows a cloud of servers. The root node f has child nodes f1 , . . . , fl . Each node fi has child nodes fi1 , . . . , fi,li . Thus, a node fR has child nodes fR1 , . . . , fR,lR (lR ≥ 0). Thus, the label R of a node fR shows a path from the root node f to fR . An edge node fR receives sensor data from child sensors and sends actions to child actuators. Let st(fR ) show a subtree whose root node is a node fR in a GTBFC tree F . Suppose a node fR supports a subprocess pF which has child subprocesses pF 1 , . . . , pF,cF . Child nodes fR1 , . . . , fR,lR of the node fR are partitioned into

84

R. Oma et al.

subgroups CR1 , . . . , CR,cF . Each subgroup CRi is composed of fog nodes supporting a same subprocess pF i , i.e. CRi = {fRj | p(fRj ) = pF i } (i = 1, . . . , cF ). Let CDRi be a collection of input data from child nodes in CRi . The fog node fR receives cF types CDR1 , . . . , CDR,cF of input data from subgroups CR1 , . . . , CR,cF , respectively, and sends one type of output data dR as shown in Fig. 2. DR is a collection {CDR1 , . . . , CDR,cF } of input data of a fog node fR . A fog node fR generates output data dR obtained by processing the input data CDR1 , ..., CDR,cF . A notation |d| shows the size [KB] of data d. The output ratio ρR of a node fR is |dR |/|DR |.

Fig. 2. Join node.

A node fR implies a node fU (fR  fU ) iff (1) p(fR ) = p(fU ), i.e. the nodes fR and fU support a same subprocess and (2) for every child node fRi of the node fR , there is a child node fU j of the node fU such that p(fRi ) = p(fU j ). Nodes fR and fU are locally equivalent (fR ∼ = fU ) iff fR  fU and fU  fR . Nodes fR and fU are independent (fR | fU ) iff neither fR  fU nor fU  fR . fR ⊆ fS iff fR = fS , fS = pt(fR ), or fR ⊆ fT ⊆ fS for some node fT . That is, fS is an ancestor of fR . A least upper bound (lub) fR ∪ fS of nodes fR and fS is a node fT such that fR ⊆ fT and fS ⊆ fT but there is no node fV such that fR ⊆ fV ⊆ fT and fS ⊆ fV ⊆ fT . For a pair of nodes fS and fT where fS ≡ fR and fT ≡ fR , fS is closer to fR than fT (fS ≤fR fT ) iff fR ∪ fS ⊆ fR ∪ fT . For a pair of nodes fR and fS , a least upper join (luj) fR fS is a join node fJ where fR ∪ fS ⊆ fJ and there is no join node fV such that fR ∪ fS ⊆ fV ⊆ fJ . For a pair of child nodes fRi and fRj of a node fR , fRi fRj shows a common ancestor join node which is a nearest to the nodes fRi and fRj . A pair of nodes fR and fS are equivalent (fR ≡ fS ) iff: 1. fR ∼ = fS . 2. If there is an luj fJ = fR fS , fJR ∈ ch(fJ ), fJS ∈ ch(fJ ), and fJR ∼ = fJS , else fJR ∈ ch(f ), fJS ∈ ch(f ), and fJR ∼ = fJS . Let en(fR ) be a set of nodes equivalent to a node fR , i.e. {fS | fS ≡ fR }. For a node fR , a node fS is an alternate node of fR iff fR ≡ fS . A(fR ) is a set of

A Nodes Selection Algorithm in the GTBFC Model

85

alternate nodes of a node fR . An alternate node fS of a node fR is closer to the node fR than another alternate node fV (≡ fR ) if fS ∪ fR ⊆ fV ∪ fR for every alternate node fV . In Fig. 1(2), the nodes f111 and f112 are equivalent (f111 ≡ f112 ). The nodes f121 and f122 are also equivalent (f121 ≡ f122 ). However, f11 and f12 are not equivalent since the nodes f11 and f12 support different subprocesses t-merge and h-merge, respectively. In Fig. 1(2), f11 f12 = f1 and f111 f112 = f1 while f111 ∪ f112 = f11 .

3

Computation and Energy Consumption Models of a Fog Node

A node fR supporting a subprocess pF (= p(fR )) takes input data DR = {CDR1 , . . . , CDR,cF } of size iR (= |DR |) from child nodes and sends output data dR of size oR (= |dR |) [12] to a parent node pt(fR ), where oR = ρR · iR for the output ratio ρR [Fig. 2]. A node fR is realized by modules to receive and process input data DR and send output data dR [14]. T IR (x), T CR (x), and T OR (x) show the execution time [sec] of a node fR to receive, process, and send data of size x, respectively. T CR (x) depends on the computation complexity of a subprocess pF (= p(fR )). In this paper, T CR (x) is ctR · CR (x) where CR (x) = x or CR (x) = x2 . Here, ctR is a constant. T IR (x) and T OR (x) are proportional to the data size x, i.e. T IR (x) = rcR + rtR · x and T OR (x) = scR + stR · x. Here, rcR , rtR , scR , and stR are constants. Suppose a node fR is a join node supporting a subprocess pF of cF (≥ 2) input parameters. Let xi show the size |CDRi | of the input data CDRi from CRi . The total size x of input data DR is x1 + · · · + xcF . It takes T T IR (x) ( = T IR (x1 ) + · · · + T IR (xcF ) = cF · rcR + rtR · (x1 + · · · + xcF ) = cF · rcR + rtR · x) [sec] to receive all the input data CDR1 , . . . , CDR,cF . It takes T FR (x) (= T T IR (x) + T CR (x) + δR · T OR (ρR · x)) [sec] to process input data DR of size x in each node fR . Here, if fR is a root, δR = 0, else δR = 1. In a Raspberry Pi 3 Model B [1] node fR which is connected with a 10 Gbps network, rcR = 0, scR = stR · 7, rtR = 9 · stR , rtR = ctR , stR = 0.00001, and rtR = 0.00009 [14]. EIR (x), ECR (x), and EOR (x) show the electric energy [J] consumed by a node fR to receive, process, and send input data of size x, respectively. In this paper, we assume each node fR follows the SPC (Simple Power Consumption) model [5–7]. The power consumption of a node fR to process the data is maxER [W]. The energy consumption ECR (x) [J] to process input data of size x (> 0) is maxER · T CR (x). The power consumption P IR and P OR [W] to receive and send data are reR · maxER and seR · maxER , respectively. In the Raspberry Pi 3 Model B node fR , reR = 0.729, seR = 0.676, and maxER = 3.7 [W] [14]. The energy consumption EIR (x) and EOR (x) [J] to receive and send data of size x (> 0), respectively, are EIR (x) = P IR · T T IR (x) = reR · maxER · (cF · rcR + rtR · x) and EOR (x) = P OR · T OR (x) = seR · maxER · (scR + stR · x).

86

R. Oma et al.

A node fR consumes the energy EFR (x) to receive and process cF types of input data CDR1 , . . . , CDR,cF of size x1 , . . . , xcF (x = x1 + · · · + xcF ) and send the output data dR of size ρ · x to a parent node: EFR (x) = EIR (x) + ECR (x) + δR · EOR (ρR · x). = (reR · T T IR (x) + T CR (x) + δR · seR · T OR (ρR · x)) · maxER . = ((reR · (cF · rcR + rtR · x)) + ctR · CR (x) + δR · seR · (scR + stR · ρR · x)) · maxER .

4

(1)

Recovery from Node Faults

If a fog node fR might stop by fault, every child node fRi is disconnected. No disconnected node fRi can deliver the output data dRi to any ancestor node of the node fR . As shown in Fig. 2, a child node in each set CRi supports a same subprocess pF i (i = 1, . . . , cF ). An alternate node fU equivalent to the faulty node fR (fU ≡ fR ) can be a new parent node to which disconnected nodes can be reconnected. If there are multiple alternate nodes, a closest one is taken as a new parent node. For example, a sibling node of fR which supports the same subprocess p(fR ) is a closest alternate node of fR . If there is no alternate node of fR , a subprocess p(fR ) (= pF ) is transferred to some node fS from the root node f . The output data DR = {dR1 , . . . , dR,lR } of all the disconnected nodes is sent to fS . The node fS is referred to as surrogate node of fR , where the input data DR from the disconnected nodes is processed and the output data dR is obtained. Then, the output data dR is sent to the parent node pt(fR ). In this paper, a parent node or a child node of fR is taken as a surrogate node. We assume each child node fRi knows every equivalent node en(fR ) of fR . First, suppose a node fR gets faulty in the GTBFC tree as shown in Fig. 3. Suppose a parent node fG of the faulty node fR is a join node, i.e. fG = fRi fRj for every pair of disconnected nodes fRi and fRj . In one way, one alternate node fU in A(fR ) is selected, whose energy consumption is minimum. Then, every disconnected node fRi is reconnected to fU . All the disconnected nodes are reconnected to one alternate node. This is an OAN (One Alternate Node) algorithm. If fR is a join node, the OAN algorithm is taken. In another way, one alternate node fU is selected for each disconnected node fRi where the energy consumption of fU is minimum. Here, some pair of different disconnected nodes may be reconnected to different alternate nodes. This is an MAN (Multiple Alternate Nodes) algorithm. The MAN algorithm can be taken only if fR is simple. In this paper, we consider closest alternate nodes of fR , i.e. sibling nodes of fR which support the subprocess p(fR ). The energy consumption and the execution time of each new parent node fU are EFU (iU + iR ) and T FU (iU + iR ) where iR is the size of the output data DR from the disconnected nodes to be reconnected to fU . Next, we present a subprocess transmission strategy. One node is selected as a surrogate node to which the subprocess p(fR ) is sent. Suppose a node fR is faulty and a node fS is a surrogate node. In this paper, we assume a disconnected

A Nodes Selection Algorithm in the GTBFC Model

87

node or the parent node fG of fR can be a surrogate node. |p(fR )| shows the size of the subprocess p(fR ). A root node f sends the subprocess p(fR ) to the surrogate node fS . Every disconnected node fRi sends the output data dRi to fS . The data dR1 , . . . , dR,lR are processed by the subprocess p(fR ) in fS . If a disconnected node fRi is a surrogate node fS , fS sends the output data dR to its grandparent node fG . If the parent node fG of fR is a surrogate node fS , fS processes the input data dR1 , . . . , dR,lR from the child nodes and the data dR obtained by processing the input data in the subprocess p(fR ). Let S(fR ) be a set of surrogate nodes of a node fR , i.e. S(fR ) = pt(fR ) ∪ ch(fR ). One surrogate node fS is selected in S(fR ), whose energy consumption is minimum. This is an OSN (One Surrogate Node) algorithm. In the MSN (Multiple Surrogate Nodes) algorithm, one surrogate node is selected for each disconnected node as the MAN algorithm. Suppose a surrogate node fS is selected in disconnected nodes fRi . The node fS processes the input data DS of size iS from its own child nodes fS1 , . . . , fS,lS and additionally processes the output data DR of size iR from the disconnected nodes fR1 , . . . , fR,lR . The energy consumption DEFS [J] and execution time DT FS [sec] of the surrogate node fS which is a disconnected node are given as follows: DEFS (iS , iR ) = EIS (iS ) δS

+ ECS (iS ) + EIS (iR − ρS · iS ) + ECS (iR ) + (2) · EOS (ρR · iR ) + EIS (|p(fR )|).

DT FS (iS , iR ) = T T IS (iS ) + T CS (iS ) + T T IS (iR − ρS · iS ) + T CS (iR ) + (3) · T OS (ρR · iR ) + T IS (|p(fR )|). δS Next, suppose the parent node fG of a faulty node fR is a surrogate node fS . Every disconnected node fRi sends the output data dRi to fS . If the parent node fS is a surrogate node, the energy consumption P EFS [J] and execution time P ETS [sec] of the surrogate node fS for input data DS of size iS from its child nodes and input data of size iR from the disconnected nodes are given as follows: P EFS (iS , iR ) = EIS δS

(iR ) + ECS (iR ) + EIS (iS − ρR · iR ) + ECS (iS ) + (4) · EOS (ρS · iS ) + EIS (|p(fR )|).

P T FS (iS , iR ) = T T IS (iR ) + T CS (iR ) + T T IS (iS − ρR · iR ) + T CS (iS ) + (5) δS · T OS (ρS · iS ) + T IS (|p(fR )|). We newly propose a GTBFC algorithm to select a new parent node of the disconnected nodes for a faulty node fR [Algorithm 1]. Let A(fR ) and S(fR ) be sets of closest alternate nodes and surrogate nodes of a node fR . First, suppose the faulty node fR is a join node. If there is an alternate node of fR , i.e. A(fR ) = φ, one alternate node is selected to be a new parent node in A(fR ). Since fR is a join node, every disconnected node fRi has to be reconnected to a same new parent node in the OAN algorithm. That is, one node fU is selected in A(fR ) where EFU (iU + iR ) is minimum. If A(fR ) = φ, a surrogate node fS is selected in set S(fR ) in the OSN algorithm. That is, one surrogate node fS is selected as a parent of fR or a child node of fR where energy P EFS (iS , iR ) is minimum.

88

R. Oma et al.

Fig. 3. Faulty node.

Next, suppose the faulty node fR is simple. Here, A(fR ) is a set of sibling nodes of fR , i.e. A(fR ) = ch(pt(fR )) − {fR }. If A(fR ) = φ, the OAN and MAN algorithms can be adopted. Since MAN is better than OAN [14], MAN is taken. In the OAN algorithm, one alternate node fU is selected in the set A(fR ) where EFU (iU + iR ) is minimum. In the MAN algorithm, one alternate node fU is selected for each disconnected node fRi . If A(fR ) = φ, a surrogate node fS is selected in S(fR ) where P EFS (iS , iR ) is minimum as a parent of fR and DEFS (iS .iR ) is minimum as a child node of fR .

5

Evaluation

We evaluate the GTBFC algorithm in terms of the energy consumption and execution time of new parent nodes in the GTBFC tree. We consider a heightbalanced process tree P of tree with height 10. Each subprocess pF has child subprocesses pF i . Each node fR has one parent fS and four child nodes fR1 , . . . , fR4 . Hence, the number of edge nodes are 49 at the edge level in the node tree F . The subprocesses p(fS ) and p(fR ) are join and simple ones, respectively. Types of the subprocess p(fS ) of a parent fS of a node fR are join 2, join 4, or simple one and types of the subprocess p(fR ) of a node fR are join or simple one, respectively. The join 2 subprocess joins two types of data into one data and the join 4 subprocess joins four types of data into one data, respectively. The output ratio ρR of each node fR is assumed to be the same, i.e. ρR = ρ and the output ratio ρ = 0.5. In the evaluation, we assume one node fR is selected to be faulty for each level l (1 ≤ l < 9). We assume neither a root node f nor every edge node is faulty for simplicity. That is, nodes at levels 0 and 9 are not faulty. If a node fR is faulty, four child nodes fR1 , . . . , fR4 of fR are disconnected in the GTBFC tree. The algorithms to select a new parent node for disconnected

A Nodes Selection Algorithm in the GTBFC Model

89

Algorithm 1. Selection algorithm

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

Input : fR = faulty node, A(fR ) = set of alternate nodes of fR , S(fR ) = set of surrogate nodes of fR ; Output: N P = a set of new parent nodes; if fR is join node then /* the faulty node fR is a join node. */ if |A(fR )| == 0 then /* there is no alternate node and one surrogate node is selected. */ N P = OSN algorithm; else /* there are multiple alternate nodes. */ N P = OAN algorithm; else /* the faulty node is a simple node. */ if |A(fR )| == 0 then /* there is no alternate node and multiple surrogate nodes are selected. */ N P = MSN algorithm; else /* there are multiple alternate nodes. */ N P = MAN algorithm;

nodes are decided as shown in Table 1. Let SD be the total size [GB] to 49 edge nodes. In this paper, we assume the size of sensor data SD which is collected by all the sensors are 10 [GB]. The size of input data of each edge node is uniformly decided. Hence, each edge node receives the size SD/49 [GB] of input data. The size of a subprocess p(fR ) of fR is 1.5 [KB]. Table 1. Algorithms to select a new parent node. Type of the subprocess p(fS ) Type of the subprocess p(fR ) Algorithm Join 2

Join Simple

OAN algorithm MAN algorithm

Join 4

Join Simple

OSN algorithm MSN algorithm

Simple

Join Simple

OAN algorithm MAN algorithm

Each fog node fR is assumed to be realized by a Raspberry Pi 3 Model B [1]. Here, the maximum power consumption maxER of a node fR is 3.7 [W]. In this paper, we assume a pair of the power ratios reR and seR of a node fR are 0.729 and 0.676, respectively. Hence, the power P IR and P OR of a node fR are 0.729 ∗ 3.7 = 2.7 [W] and 0.676 ∗ 3.7 = 2.5 [W], respectively. In this paper, we assume the execution time ratios rtR , rcR , stR , scR , and ctR of a node fR

90

R. Oma et al.

Fig. 4. Energy consumption of new parent nodes with CR (x) = x for level l at the faulty node fR .

are 0.00009, 0, 0.00001, 0.00007 and 0.00009, respectively. In the evaluation, the execution time to process the data of size x is ctR · CR (x) where CR (x) = x. In Figs. 4 and 5, (p(fS ), p(fR )) = (join 2, simple) denotes the type of a subprocess p(fS ) of a parent node fS of a faulty node fR is join 2 and a subprocess p(fR ) of a faulty node fR is simple. Figure 4 shows the energy consumed by new parent nodes selected by the GTBFC algorithm for CR (x) = x for level l at the faulty node fR . In Fig. 4, the energy consumption of new parent nodes in the type of the subprocesses (p(fS ), p(fR )) with (join 4, simple) is minimum. Then, multiple new parent nodes are selected by the MSN algorithm in the disconnected nodes as the surrogate nodes. In Fig. 4, the energy consumption of new parent nodes selected by the MAN algorithm in the types of the subprocesses (p(fS ), p(fR )) with (simple, simple) is smaller than the energy consumption of one new parent node selected by the OSN algorithm in the type of the subprocesses (p(fS ), p(fR )) with (join 2, join). If the types of subprocesses (p(fS ), p(fR )) are (join 2, join), (join 2, simple), and (join 4, join), the energy consumption of one new parent node in the alternate nodes is maximum. Then, the sibling node fU of the faulty node fR is selected as a new parent node for every disconnected node. The output data dRi from every disconnected node fRi is centralized to the new parent node fU . Figure 5 shows the execution time of new parent nodes selected by the GTBFC algorithm for CR (x) = x for level l at the faulty node fR . Here, the energy consumption of new parent nodes selected in the type of the subprocesses (p(fS ), p(fR )) with (join 4, simple) is minimum. In Fig. 5, the execution time of new parent nodes selected by the MAN algorithm in the type of the subprocesses (p(fS ), p(fR )) with (simple, simple) is smaller than the execution time of one new parent node selected by the OSN algorithm in the type of the subprocesses

A Nodes Selection Algorithm in the GTBFC Model

91

Fig. 5. Execution time of new parent nodes with CR (x) = x for level l at the faulty node fR .

(p(fS ), p(fR )) with (join 2, join). The energy consumption of the new parent node selected in the alternate nodes is maximum if the types of subprocesses (p(fS ), p(fR )) with (join 2, join), (join 2, simple), and (join 4, join).

6

Concluding Remarks

In this paper, an application process is hierarchically structured in a tree. In this paper, we newly consider a join subprocess which has multiple input parameters. Thus, fog nodes are also structured in a node tree according to the process tree. If a node is faulty, a join node has to receive the input data from ancestor nodes. In this paper, we proposed the GTBFC algorithm to select a new parent node for each disconnected node. In the evaluation, we showed the energy consumption of new parent nodes selected by the GTBFC algorithm.

References 1. Raspberry pi 3 model b. https://www.raspberrypi.org/products/raspberry-pi-3model-b/ 2. Creeger, M.: Cloud computing: an overview. Queue 7(5), 3–4 (2009) 3. Duolikun, D., Enokido, T., Takizawa, M.: Dynamic migration of virtual machines to reduce energy consumption in a cluster. Int. J. Grid Util. Comput. (IJGUC) 9(4), 357–366 (2018) 4. Duolikun, D., Kataoka, H., Enokido, T., Takizawa, M.: Simple algorithms for selecting an energy-efficient server in a cluster of servers. Int. J. Commun. Netw. Distrib. Syst. (IJCNDS) 21(1), 1–25 (2018)

92

R. Oma et al.

5. Enokido, T., Ailixier, A., Takizawa, M.: A model for reducing power consumption in peer-to-peer systems. IEEE Syst. J. 4(2), 221–229 (2010) 6. Enokido, T., Ailixier, A., Takizawa, M.: Process allocation algorithms for saving power consumption in peer-to-peer systems. IEEE Trans. Ind. Electron. 58(6), 2097–2105 (2011) 7. Enokido, T., Ailixier, A., Takizawa, M.: An extended simple power consumption model for selecting a server to perform computation type processes in digital ecosystems. IEEE Trans. Ind. Inf. 10(2), 1627–1636 (2014) 8. Guo, Y., Oma, R., Nakamura, S., Duolikun, D., Enokido, T., Takizawa, M.: Data and subprocess transmission on the edge node of TWTBFC model. In: Proceedings of the 11th International Conference on Intelligent Networking and Collaborative Systems (INCoS-2019), pp. 80–90 (2019) 9. Guo, Y., Oma, R., Nakamura, S., Duolikun, D., Enokido, T., Takizawa, M.: Evaluation of a two-way tree-based fog computing (TWTBFC) model. In: Proceedings of the 13th International Conference on Innovative Mobile and Internet Services in Ubiquitous Computing (IMIS-2019), pp. 72–81 (2019) 10. Hanes, D., Salgueiro, G., Grossetete, P., Barton, R., Henry, J.: IoT Fundamentals: Networking Technologies, Protocols, and Use Cases for the Internet of Things. Cisco Press (2018) 11. Kataoka, H., Nakamura, S., Duolikun, D., Enokido, T., Takizawa, M.: Multi-level power consumption model and energy-aware server selection algorithm. Int. J. Grid Util. Comput. (IJGUC) 8(3), 201–210 (2017) 12. Oma, R., Nakamura, S., Duolikun, D., Enokido, T., Takizawa, M.: An energyefficient model for fog computing in the internet of things (IoT). Internet Things 1–2, 14–26 (2018) 13. Oma, R., Nakamura, S., Duolikun, D., Enokido, T., Takizawa, M.: Energy-efficient recovery algorithm in the fault-tolerant tree-based fog computing (FTBFC) model. In: Proceedings of the 33rd International Conference on Advanced Information Networking and Applications (AINA-2019), pp. 132–143 (2019) 14. Oma, R., Nakamura, S., Duolikun, D., Enokido, T., Takizawa, M.: Evaluation of data and subprocess transmission strategies in the tree-based fog computing (TBFC) model. In: Proceedings of the 22nd International Conference on NetworkBased Information Systems (NBiS-2019), pp. 15–26 (2019) 15. Oma, R., Nakamura, S., Duolikun, D., Enokido, T., Takizawa, M.: A fault-tolerant tree-based fog computing model. Int. J. Web Grid Serv. (IJWGS) 15(3), 219–239 (2019) 16. Oma, R., Nakamura, S., Duolikun, D., Enokido, T., Takizawa, M.: Subprocess transmission strategies for recovering from faults in the tree-based fog computing (TBFC) model. In: Proceedings of the 13th International Conference on Complex, Intelligent, and Software Intensive Systems (CISIS-2019), pp. 50–61 (2019) 17. Rahmani, A., Liljeberg, P., Preden, J.S., Jantsch, A.: Fog Computing in the Internet of Things. Springer, Heidelberg (2018)

A TBOI (Time-Based Operation Interruption) Protocol to Prevent Late Information Flow in the IoT Shigenari Nakamura1(B) , Tomoya Enokido2 , and Makoto Takizawa1 1

Hosei University, Tokyo, Japan [email protected], [email protected] 2 Rissho University, Tokyo, Japan [email protected]

Abstract. In the IoT (Internet of Things), devices have to be prevented from maliciously accessed. The CapBAC (Capability-Based Access Control) model is proposed to make IoT devices secure. In the CapBAC model, an owner of a device issues a capability token, i.e. a set of access rights to a subject. Here, the subject is allowed to manipulate the device according to the access rights authorized in the capability token. Suppose a subject sbi is allowed to get data from a device d2 but not allowed to get data from a device d1 . The subject sbi can get the data of the device d1 in the device d2 after another subject sbj brings the data from the device d1 to the device d2 . Here, the data in the device d1 illegally flow to the subject sbi . In order to prevent illegal information flow, an OI (Operation Interruption) protocol is proposed in our previous studies, where illegal get operations are interrupted. However, in the OI protocol, a subject sbi can get data dt1 of a device d1 generated at time τ even if the subject sbi is not allowed to get the data dt1 at time τ . In this case, the data dt1 come to the subject sbi later than expected by the subject sbi to get the data dt1 , i.e. the data dt1 flow late to the subject sbi . In this paper, we newly propose a TBOI (Time-Based OI) protocol to prevent late information flow in addition to illegal information flow from occurring. Keywords: IoT (Internet of Things) · Device security · CapBAC (Capability-Based Access Control) model · Illegal information flow · Information flow control · Late information flow · TBOI (Time-Based Operation Interruption) protocol

1

Introduction

Information systems are required to be secure in presence of malicious accesses. Thus, various types of models and methods are proposed, such as cryptography [14,15] and access control models [1]. Cryptography is used to prevent every information from being forged, stolen, or disclosed by a subject which is granted c Springer Nature Switzerland AG 2020  L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 93–104, 2020. https://doi.org/10.1007/978-3-030-33506-9_9

94

S. Nakamura et al.

no permission for the information. In the access control models, only an authorized subject is allowed to manipulate an object in an authorized operation. However, even if a subject is not allowed to get data in an object oi , the subject can get the data by accessing another object oj [1]. Here, illegal information flow from the object oi via the object oj to the subject occurs. We have to prevent illegal information flow among subjects and objects in the access control models. In the LBAC (Lattice-Based Access Control) model [17] proposed to prevent illegal information flow among subjects and objects, each entity is assigned a security class. Illegal information flow is defined based on the relations among classes and every operation implying the illegal information flow is prohibited. In our previous studies, various types of protocols to prevent illegal information flow are proposed. In the papers [6–8], types of protocols to prevent illegal information flow occurring in distributed database systems are proposed based on the RBAC (Role-Based Access Control) model [18]. In the papers [10,11], protocols to prevent illegal information flow occurring in P2PPSO (Peer-to-Peer Publish/Subscribe with Object concept) systems are proposed based on the TBAC (Topic-Based Access Control) model [13]. The IoT (Internet of Things) is composed of various types and millions of nodes including not only computers but also devices like sensors and actuators [4,16]. Here, the traditional access control models such as the RBAC [18] and ABAC (Attribute-Based Access Control) [19] models are not adopted for the IoT due to the scalability of the IoT. Hence, the CapBAC (Capability-Based Access Control) model is proposed [3]. Here, an owner of a device issues a capability token to a subject like users and applications. The capability token is defined to be a set of access rights, d, op for a device d and an operation op. The subject is allowed to manipulate the device d in an operation op only if the capability token including the access right d, op is issued to the subject. In addition, capability token includes the validity period which shows the capability token is valid from when to when. Each subject can manipulate devices according to the capability token during the validity period of the capability token. Suppose a subject sbi is issued a capability token including a pair of access rights d1 , get and d2 , put of a pair of devices d1 and d2 by owners of the devices. Suppose the device d1 is a sensor and the device d2 is equipped with a pair of sensor and actuator. A sensor just gives sensor data to a subject. On the other hand, an actuator supports an action to store data to the device. A subject sbj is issued a capability token including an access right d2 , get by an owner of the device d2 . First, the subject sbi gets the sensor data dt1 obtained by the sensor d1 and then gives the data dt1 to the device d2 . Next, the subject sbj gets the data dt1 from the device d2 . Here, the subject sbj can obtain the data dt1 of the device d1 via the device d2 although the subject sbj is not issued a capability token including the access right d1 , get. Here, the device d1 is a source one of information flow. This is illegal information flow from the device d1 to the subject sbj . In our previous studies, an OI (Operation Interruption) protocol is proposed to prevent illegal information flow in the IoT based on the CapBAC model [9,12]. First, the legal and illegal information flow relations

A TBOI (Time-Based Operation Interruption) Protocol

95

among subjects and devices are defined based on the CapBAC model. Then, it is checked whether or not the illegal information flow occurs based on the information flow relations. If the illegal information flow occurs, the operation is interrupted at the device. Hence, every illegal information flow is prevented from occurring. In the OI protocol, every subject sbi can not get data of a device d1 in another device d2 even if the subject sbi is allowed to get data of the device d2 . However, every subject sbi can get data dt1 of a device d1 generated at time τ even if the subject sbi is not allowed to get the data dt1 at time τ . This means, every illegal information flow is prevented but every information flow of data generated in the device d1 before the validity period of the access right d1 , get to a subject is ignored in the OI protocol because information flow relations are based on only devices. Here, the older information comes to a subject than what the subject expects to get. Since the information comes to a subject late, the information flow is referred to as late information flow. In this paper, first, we define information flow relations based on not only devices but also time. In addition, we newly propose a TBOI (Time-Based OI) protocol to prevent late information flow in addition to illegal information flow. In Sect. 2, we discuss the system model. In Sect. 3, we define types of information flow relations based on the CapBAC model. In Sect. 4, we newly propose the TBOI protocol to prevent late information flow in addition to illegal information flow in the IoT.

2 2.1

System Model CapBAC (Capability-Based Access Control) Model

In order to make information systems secure, types of access control models [2,18,19] are widely used. Here, a system is composed of two types of entities, subjects and objects. A subject s issues an operation op to an object o. Then, the operation op is performed on the object o. Here, only an authorized subject s is allowed to manipulate an object o in an authorized operation op. Most of the access control models are based on ACLs (access control lists) such as RBAC (Role-Based Access Control) [18] and ABAC (Attribute-Based Access Control) [19] models. An ACL is a list of access rules specified by an authorizer. Each access rule s, o, op is composed of a subject s, an object o, and an operation op. This means, a subject s is granted an access right of an object o and an operation op. In the ACL system, if a subject s tries to access the data of an object o in an operation op, a service provider has to check whether or not the subject is authorized to manipulate the object o in the operation op by using the ACL, i.e. s, o, op in the ACL. In scalable systems like the IoT, the ACL gets also scalable and it is difficult to maintain and check the ACL. Hence, the RBAC and ABAC models are not suitable for the scalable distributed systems where there is no centralized coordinator and each node is an autonomous process which makes a decision by itself.

96

S. Nakamura et al.

The CapBAC (Capability-Based Access Control) model is proposed as an access control model for the IoT [3]. There is an owner of each device. In the CapBAC model, an owner of a device first issues a capability token CAPi to a subject sbi . The capability token CAPi issued to the subject sbi is defined to be a set of access rights. An access right is a pair d, op of a device d and an operation op. A subject sbi is allowed to manipulate a device d in an operation op only if a capability token CAPi including an access right d, op is issued to the subject sbi , i.e. d, op ∈ CAPi . Otherwise, the access request is rejected. In the paper [5], the distributed CapBAC model is proposed. Here, there is no intermediate entity between each pair of a subject and a device to implement the access control. Each capability token includes the validity period. Here, an access request to manipulate a device d in an operation op of a subject sbi whose capability token CAPi expires is rejected even if the capability token CAPi includes the access right d, op. Let gtsi .st be the first time when a subject sbi is allowed to issue a get operation to a device s. On the other hand, gtsi .et is the end time when a subject sbi is allowed to issue a get operation to a device s. That is, the subject sbi is allowed to issue a get operation to a device s from gtsi .st to gtsi .et. 2.2

Devices

In the IoT, a set D of devices d1 , . . . , ddn (dn ≥ 1) are interconnected in networks. In this paper, we consider three types of devices, sensor, actuator, and hybrid device. A sensor device just obtains data collected by sensing events which occur in physical environment. An actuator device acts according to the action request from a subject. Let SB be a set of subjects sb1 , . . . , sbsbn (sbn ≥ 1). A subject sbi gets and puts data from a sensor s and to an actuator a, respectively. In addition, a subject sbi issues both get and put operations to a hybrid device h like robots and cars. Let dts be data collected by a sensing device s at time τ . dts .t shows the time τ when the data dts are collected by a device s. Through manipulating devices, data are exchanged among devices and subjects. If data in a device d1 flow into another device d2 , the device d1 is referred to as a source device of the device d2 . Let sbi .DT and d.DT be sets of data which flow into the subject sbi and device d, respectively. For each device d, the set d.DT of source devices is manipulated each time a subject manipulates the device d, which is initially φ. If the device d gets data dt by sensing events, the data dt are added to the set d.DT of the device d. On the other hand, let sbi .D and d.D be sets of source devices whose data flow into the subject sbi and the device d, respectively. Here, sbi .D = ∪dts ∈sbi .DT s and d.D = ∪dts ∈d.DT s. Let minTes be the oldest generation time of the data which are generated in the device s and flow into an entity e.

3

Information Flow Relations

In this section, we define types of information flow relations on devices and subjects based on the CapBAC model. A subject sbi can get data from a device

A TBOI (Time-Based Operation Interruption) Protocol

97

d (∈ D) only if an access right d, get is granted to the subject sbi . Let IN (sbi ) be a set of devices whose data a subject sbi is allowed to get. That is, a subject sbi is issued a capability token CAPi including an access right d, get, i.e. IN (sbi ) = {d | d, get ∈ CAPi } (⊆ D). Definition 1. A device d flows to a subject sbi (d → sbi ) iff (if and only if) d.D = φ and d ∈ IN (sbi ). If d → sbi holds, data brought to the device d may be brought to the subject sbi . Otherwise, no data flow from the device d into the subject sbi . Definition 2. A device d legally flows to a subject sbi (d ⇒ sbi ) iff d → sbi and d.D ⊆ IN (sbi ). The condition “d.D ⊆ IN (sbi )” means that the data in the device d are allowed to be brought into the subject sbi . Definition 3. A device d illegally flows to a subject sbi (d → sbi ) iff d → sbi and d.D ⊆ IN (sbi ). The condition “d.D ⊆ IN (sbi )” means that the data in some device of d.D are not allowed to be brought into the subject sbi . Definition 4. A device d timely flows to a subject sbi (d ⇒t sbi ) iff d ⇒ sbi and ∀s ∈ d.D (gtsi .st ≤ minTds ≤ gtsi .et). In the legal information flow relation (⇒), only devices are checked. On the other hand, in the timely flow relation (⇒t ), time is considered. The condition “∀s ∈ d.D(gtsi .st ≤ minTds ≤ gtsi .et)” means that only the data dts generated while the subject sbi is allowed to issue a get operation to the device s flow into the subject sbi . Definition 5. A device d flows late to a subject sbi (d →l sbi ) iff d ⇒ sbi and ∃s ∈ d.D (¬(gtsi .st ≤ minTds ≤ gtsi .et)). The condition “∃s ∈ d.D (¬(gtsi .st ≤ minTds ≤ gtsi .et))” means that the data dts generated while the subject sbi is not allowed to issue a get operation to the device s flow into the subject sbi . This means, the subject sbi can get the data dts even if the data dts are not generated while the subject sbi is allowed to get the data dts . Example 1. Let us consider a sensor device s and a hybrid device h as shown in Fig. 1. Suppose a capability token including a pair of access rights s, get and h, put is issued to a subjects sbi . On the other hand, a capability token including the access rights s, get and h, get is issued to another subjects sbj . We also suppose gtsi .st = t1 , gtsi .et = t2 , gtsj .st = t3 , gtsj .et = t4 , gthj .st = t3 , and gthj .et = t4 . Here, IN (sbi ) = {s} and IN (sbj ) = {s, h}. First, the sensor s collects data by sensing events occurring around itself and obtains the data dts at time τ . Here, minTss = τ . Next, the subject sbi gets the data dts from

98

S. Nakamura et al.

the sensor s. Here, s ⇒ sbi holds because s → sbi holds and s.D (= {s}) ⊆ IN (sbi ) (= {s}). In addition, s ⇒t sbi holds because s ⇒ sbi holds, s.D = {s}, and gtsi .st (= t1 ) ≤ minTss (= τ ) ≤ gtsi .et(= t2 ). Here, minTis = τ . Then, the hybrid device h collects data by sensing events occurring around itself and obtains the data dth . Here, minThh = τ  . After that, the subject sbi puts the data dts to the hybrid device h. Here, minThs = τ . Finally, the subject sbj gets data from the hybrid device h. Here, h ⇒ sbj holds because h → sbj holds and h.D (= {s, h}) ⊆ IN (sbj ) (= {s, h}). However, h →l sbj holds because h ⇒ sbj holds, s ∈ h.D, and minThs (= τ ) ≤ gtsj .st (= t3 ). If the capability token CAPj does not include the access right s, get, h → sbj holds because h → sbj holds and h.D (= {s, h}) ⊆ IN (sbj ) (= {h}). In this case, the information of the sensor s illegally flows into the subject sbj .

Fig. 1. Information flow in the IoT system.

4

Protocols

In this paper, we consider three types of devices, sensor, actuator, and hybrid devices. Sensor and hybrid devices collect data by sensing events occurring around them. Once a device d gets data by sensing events, the device d is added to the set d.D of the device d. Actuators and hybrid devices get data which are collected by other devices like sensors and hybrid devices. Subjects issue get

A TBOI (Time-Based Operation Interruption) Protocol

99

operations to sensors and hybrid devices to get data collected by the sensors and hybrid devices. On the other hand, subjects issue put operations to actuators and hybrid devices to store data which the subjects get from sensors and hybrid devices. 4.1

An OI (Operation Interruption) Protocol

In the IoT with the CapBAC model, a capability token CAPi which is a set of access rights is issued to a subject sbi . If the capability token CAPi obtained by the subject sbi includes an access right d, op, the subject sbi is allowed to manipulate the device d in an operation op. If the data dt1 in a device d1 brought into a device d2 are brought into a subject sbi which is not allowed to get the data dt1 in the device d1 , the data dt1 in the device d1 illegally flow into the subject sbi . In Sect. 3, the illegal information flow is defined based on the CapBAC model. In this section, we discuss the OI (Operation Interruption) protocol [9,12] to prevent illegal information flow based on the information flow relations among subjects and devices. IN (sbi ) is a set of devices whose data are allowed to be got by a subject sbi , i.e. IN (sbi ) = {d | d, get ∈ CAPi }. sbi .D is a set of devices whose data are brought to the subject sbi . d.D is a set of source devices whose data are brought to the device d. For each subject sbi and device d, the sets sbi .D and d.D are manipulated, which are initially φ. If a subject sbi issues a get operation to a device d, it is checked whether or not the subject sbi is allowed to get all the data which may be brought into the subject sbi . If at least one device whose data are not allowed to be brought into the subject sbi exists, i.e. d.D ⊆ IN (sbi ), the get operation is interrupted to prevent illegal information flow. The OI protocol is shown as follows: [OI (Operation Interruption) Protocol] 1. A device d gets data by sensing events occurring around the device d. a. d.D = d.D ∪ {d}; 2. A subject sbi gets data from a device d, i.e. d → sbi holds. a. If d ⇒ sbi , the subject sbi gets the data from the device d and sbi .D = sbi .D ∪ d.D; b. Otherwise, the get operation is interrupted at the device d; 3. A subject sbi puts data to a device d. a. Data obtained by the subject sbi are brought to the device d and d.D = d.D ∪ sbi .D; When a subject sbi tries to get data from a device d, the subject sbi sends an access request to the device d with the capability token CAPi . On receipt of the access request from the subject sbi , the device d checks whether or not the legal information flow condition is satisfied according to the capability token CAPi attached in the access request from the subject sbi . If the legal condition is satisfied, the device d sends a reply of the access request to the subject sbi .

100

S. Nakamura et al.

Otherwise, the access request of the subject sbi is interrupted at the device d. Here, the device d informs the subject sbi that the access request is denied. Figure 2 shows an example of the OI protocol. Here, an access right s, get is not granted to the subject sbj differently from the Example 1. First, the sensor s collects data by sensing events occurring around itself, i.e. s.D = {s}. Next, the subject sbi gets data from the sensor s because s ⇒ sbi holds. Hence, the subject sbi can get the data from the sensor s and sbi .D = sbi .D (= φ) ∪ s.D (= {s}) = {s}. Then, the hybrid device h collects data by sensing events occurring around itself, i.e. h.D = {h}. After that, the subject sbi puts the data to the hybrid device h. Here, h.D = h.D (= {h}) ∪ sbi .D (= {s}) = {h, s}. Finally, the subject sbj tries to get the data from the hybrid device h. Here, h → sbj holds because h → sbj and h.D (= {h, s}) ⊆ IN (sbj ) (= {h}). Hence, the get operation of the subject sbj is interrupted at the hybrid device h to prevent illegal information flow as shown in Fig. 2.

Fig. 2. OI protocol.

4.2

A TBOI (Time-Based OI) Protocol

In Example 1, since no illegal information flow occurs, no operation is interrupted in the OI protocol. However, the subject sbj can get the data dts generated at time τ via the hybrid device h although the subject sbj is not allowed to get the data dts at time τ . Here, the data dts generated at time τ come to the subject sbj although the subject sbj expects to get data generated at time between t3

A TBOI (Time-Based Operation Interruption) Protocol

101

and t4 which is later than τ . Hence, late information flow from the sensor s to the subject sbj occurs (s →l sbj ). In this paper, we newly propose an TBOI (Time-Based OI) protocol to prevent late information flow in addition to illegal information flow based on the information flow relations among subjects and devices. Here, time is considered differently from the OI protocol. In the OI protocol, each device makes an authorization decision based on sets of devices whose data flow into each entity. On the other hand, in the TBOI protocol, the time when data are generated and validity period of an access right to get the data are used in addition to sets of devices. dts .t is the time when the data dts are collected by a device s. Every subject sbi is allowed to issue an get operation to a device s at time between gtsi .st and gtsi .et. sbi .DT is a set of data which flow into a subject sbi . d.DT is also a set of data which flow into a device d. Here, sbi .D = ∪dts ∈sbi .DT s and d.D = ∪dts ∈d.DT s. minTes is the oldest generation time of the data which are generated in the device s and flow into an entity e. For each subject sbi and device d, the variable minTes is manipulated. If a subject sbi issues a get operation to a device d, it is checked whether or not every data dts in d.DT which may be brought into the subject sbi are generated during the validity period of the subject sbi by using the variables minTds , gtsi .st, and gtsi .et. If the data dts generated at time when the subject sbi is not allowed to get the data dts exist, i.e. ∃s ∈ d.D (¬(gtsi .st ≤ minTds ≤ gtsi .et)), the get operation is interrupted to prevent late information flow. The TBOI protocol is shown as follows: [TBOI (Time-Based OI) Protocol] 1. A device d gets data by sensing events occurring around the device d at time τ. a. d.D = d.D ∪ {d}; b. If minTdd = N U LL, minTdd = τ ; 2. A subject sbi gets data from a device d, i.e. d → sbi holds. a. If d ⇒t sbi , i. The subject sbi gets data from the device d and sbi .D = sbi .D ∪ d.D; ii. For each device s such that s ∈ (sbi .D ∩ d.D), minTis = min(minTis , minTds ); iii. For each device s such that s ∈sbi .D and s ∈ d.D, minTis = minTds ; b. Otherwise, the get operation is interrupted at the device d; 3. A subject sbi puts data to a device d. a. Data obtained by the subject sbi are brought to the device d and d.D = d.D ∪ sbi .D; b. For each device s such that s ∈ (d.D ∩ sbi .D), minTds = min(minTds , minTis ); c. For each device s such that s ∈d.D and s ∈ sbi .D, minTds = minTis ; Figure 3 shows an example of the TBOI protocol. Here, the scenario is same as the Example 1. First, the sensor s collects data by sensing events occurring around itself at time τ , i.e. s.D = {s} and minTss = τ . Next, the subject sbi

102

S. Nakamura et al.

gets data from the sensor s because s ⇒t sbi holds. Hence, sbi .D = sbi .D (= φ) ∪ s.D (= {s}) = {s} and minTis = minTss = τ . Then, the hybrid device h collects data by sensing events occurring around itself at time τ  , i.e. h.D = {h} and minThh = τ  . After that, the subject sbi puts the data to the hybrid device h. Here, h.D = h.D (= {h}) ∪ sbi .D (= {s}) = {h, s} and minThs = minTis = τ . Finally, the subject sbj tries to get the data from the hybrid device h. Here, h →l sbj holds because minThs (= τ ) ≤ gtsj .st (= t3 ). Hence, the get operation of the subject sbj is interrupted at the hybrid device h to prevent late information flow as shown in Fig. 3.

Fig. 3. TBOI protocol.

5

Concluding Remarks

In order to make IoT secure, we take the CapBAC (Capability-Based Access Control) model. In this paper, we consider three types of devices, sensor, actuator, and hybrid devices. In the IoT, a subject sbi can get the data of the device d1 from another device d2 to which the data of the device d1 are brought although the subject sbi is not allowed to get the data from the device d1 . Here, illegal information flow occurs. In our previous studies, the OI (Operation Interruption) protocol where operations implying illegal information flow are interrupted is proposed to prevent illegal information flow in the IoT based on the CapBAC model. However, in the OI protocol, late information flow such that a subject sbi

A TBOI (Time-Based Operation Interruption) Protocol

103

can get data dt1 of a device d1 generated at time τ even if the subject sbi is not allowed to get the data dt1 at time τ occurs. In this paper, we newly proposed the TBOI (Time-Based OI) protocol to prevent not only illegal information flow but also late information flow from occurring. Acknowledgments. This work was supported by Japan Society for the Promotion of Science (JSPS) KAKENHI Grant Number JP17J00106.

References 1. Denning, D.E.R.: Cryptography and Data Security. Addison Wesley, Boston (1982) 2. Fernandez, E.B., Summers, R.C., Wood, C.: Database Security and Integrity. Adison Wesley, Boston (1980) 3. Gusmeroli, S., Piccione, S., Rotondi, D.: A capability-based security approach to manage access control in the Internet of Things. Math. Comput. Model. 58(5–6), 1189–1205 (2013) 4. Hanes, D., Salgueiro, G., Grossetete, P., Barton, R., Henry, J.: IoT Fundamentals: Networking Technologies, Protocols, and Use Cases for the Internet of Things. Cisco Press, Indianapolis (2018) 5. Hern´ andez-Ramos, J.L., Jara, A.J., Mar´ın, L., Skarmeta, A.F.: Distributed capability-based access control for the internet of things. J. Internet Serv. Inf. Secur. 3(3/4), 1–16 (2013) 6. Nakamura, S., Duolikun, D., Enokido, T., Takizawa, M.: A flexible read-write abortion protocol to prevent illegal information flow among objects. J. Mob. Multimed. 11(3&4), 263–280 (2015) 7. Nakamura, S., Duolikun, D., Enokido, T., Takizawa, M.: A write abortion-based protocol in role-based access control systems. Int. J. Adapt. Innov. Syst. 2(2), 142–160 (2015) 8. Nakamura, S., Duolikun, D., Enokido, T., Takizawa, M.: A read-write abortion protocol to prevent illegal information flow in role-based access control systems. Int. J. Space-Based Situated Comput. 6(1), 43–53 (2016) 9. Nakamura, S., Enokido, T., Barolli, L., Takizawa, M.: Capability-based information flow control model in the IoT. In: Proceedings of the 13th International Conference on Innovative Mobile and Internet Services in Ubiquitous Computing, pp. 63–71 (2019) 10. Nakamura, S., Enokido, T., Takizawa, M.: Information flow control in object-based peer-to-peer publish/subscribe systems. Concurr. Comput. Pract. Exp. (accepted) 11. Nakamura, S., Enokido, T., Takizawa, M.: Causally ordering delivery of event messages in P2PPSO systems. Cogn. Syst. Res. 56, 167–178 (2019) 12. Nakamura, S., Enokido, T., Takizawa, M.: Evaluation of an OI (operation interruption) protocol to prevent illegal information flow in the IoT. In: Proceedings of the 22nd International Conference on Network-Based Information Systems, pp. 3–14 (2019) 13. Nakamura, S., Ogiela, L., Enokido, T., Takizawa, M.: An information flow control model in a topic-based publish/subscribe system. J. High Speed Netw. 24(3), 243– 257 (2018) 14. Ogiela, L.: Intelligent techniques for secure financial management in cloud computing. Electron. Commer. Res. Appl. 14(6), 456–464 (2015)

104

S. Nakamura et al.

15. Ogiela, M.R., Ogiela, L.: On using cognitive models in cryptography. In: Proceedings of IEEE the 30th International Conference on Advanced Information Networking and Applications (AINA-2016), pp. 1055–1058 (2016) 16. Oma, R., Nakamura, S., Duolikun, D., Enokido, T., Takizawa, M.: An energyefficient model for fog computing in the internet of things (IoT). Internet Things Eng. Cyber Phys. Hum. Syst. 1–2, 14–26 (2018) 17. Sandhu, R.S.: Lattice-based access control models. IEEE Comput. 26(11), 9–19 (1993) 18. Sandhu, R.S., Coyne, E.J., Feinstein, H.L., Youman, C.E.: Role-based access control models. IEEE Comput. 29(2), 38–47 (1996) 19. Yuan, E., Tong, J.: Attributed based access control (ABAC) for web services. In: Proceedings of the IEEE International Conference on Web Services (ICWS 2005) (2005)

Enhancement of Binary Spray and Wait Routing Protocol for Improving Delivery Probability and Latency in a Delay Tolerant Network Evjola Spaho1(B) , Klodian Dhoska2 , Leonard Barolli3 , Vladi Kolici1 , and Makoto Takizawa4 1 Department of Electronics and Telecommunication, Faculty of Information Technology, Polytechnic University of Tirana, Mother Teresa Square, No. 4, Tirana, Albania [email protected], [email protected] 2 Department of Production-Management, Faculty of Mechanical Engineering, Polytechnic University of Tirana, Mother Teresa Square, No. 4, Tirana, Albania [email protected] 3 Department of Information and Communication Engineering, Fukuoka Institute of Technology (FIT), 3-30-1 Wajiro-Higashi, Higashi-Ku, Fukuoka 811-0295, Japan [email protected] 4 Department of Advanced Sciences, Hosei University, 3-7-2, Kajino-cho, Koganei-shi, Tokyo 184-8584, Japan [email protected]

Abstract. In this paper, we enhanced the Binary Spray and Wait (BS&W) routing protocol and create two versions of Spray and Wait (S&WV1 and S&W-V2) and evaluate and compare their performance in a Delay Tolerant Network (DTN). The network is created from pedestrians, cars and buses of public transport, equipped with smart devices that move and exchange information in an urban area in Tirana city, Albania. Different simulations are conducted to evaluate the performance of the enhanced protocols. Simulations are done using the Opportunistic Network Environment (ONE) simulator. We use the delivery probability and average latency as evaluation metrics. Based on simulation results, we found that our proposed versions S&W-V1 and S&W-V2 improve the delivery probability and average latency.

1

Introduction

Delay Tolerant Networks (DTNs) enable communication where connectivity issues like sparse connectivity, long delay, high error rates, asymmetric data rate, and no end-to-end connectivity exists. In DTNs, mobile nodes can send and receive data, carry data as relays and forward data in opportunistic way upon contacts. In order to handle disconnections and long delays, DTNs use store-carry-and-forward approach. c Springer Nature Switzerland AG 2020  L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 105–113, 2020. https://doi.org/10.1007/978-3-030-33506-9_10

106

E. Spaho et al.

Smart devices equipped with different communication interfaces like Bluetooth and WiFi are the main computing and communication platform nowdays. These smart devices can be used to carry and forward messages in DTNs. DTNs are occasionally connected networks, characterized by the absence of a continuous path between the source and destination [1,2]. DTN is the “challenged computer network” approach that is originally designed from the Interplanetary Internet, and the data transmission is based upon the store-carryand-forward protocol for the sake of carrying data packets under a poor network environment such as space [1]. Different copies of the same bundle can be routed independently to increase security and robustness, thus improving the delivery probability and reducing the delivery delay. However, such approach increases the contention for network resources (e.g., bandwidth and storage), potentially leading to poor overall network performance. DTNs get around the lack of end-to-end connectivity with an architecture that is based on message switching. It is also intended to tolerate links with low reliability and large delays. The architecture is specified in RFC 4838 [3]. Bundle protocol has been designed as an implementation of the DTN architecture. A bundle is a basic data unit of the DTN bundle protocol. Each bundle comprises a sequence of two or more blocks of protocol data, which serve for various purposes. In poor conditions, bundle protocol works on the application layer of some number of constituent Internet, forming a store-and-forward overlay network to provide its services. In order to handle disconnections and long delays in sparse opportunistic network scenarios, DTN uses store-carry-and-forward approach. A network node stores a bundle and waits for a future opportunistic connection. When the connection is established, the bundle is forwarded to an intermediate node, according to a hop-by-hop forwarding/routing scheme. This process is repeated and the bundle will be relayed hop-by-hop until reaching the destination node. There are different research works with focus on routing in DTNs. In [4–21] authors deal with routing in DTNs. In this paper, we enhanced Binary Spray and Wait (B-S&W) protocol. For the simulations we use the Opportunistic Network Environment (ONE) [22] simulator. This simulation environment can generate different movement models and offers various DTN routing algorithms. Its graphical user interface visualize both mobility and message passing in real time. The simulation results show that for the proposed versions of protocol, the delivery probability and average latency is improved. The remainder of this paper is as follows. Spray and Wait protocol and its enhanced versions are presented in Sect. 2. The simulation system design and simulation scenarios are described in Sect. 3. In Sect. 4 are shown the simulation results. Finally, the conclusions and future work are presented in Sect. 5.

Enhancement of Binary Spray and Wait Routing Protocol

2 2.1

107

Spray and Wait Routing Protocol and Its Enhanced Version Spray and Wait Routing Protocol

Spray and Wait (S&W) [17] is a routing protocol that attempts to gain the delivery ratio benefits of replication-based routing as well as the low resource utilization benefits of forwarding-based routing. The S&W protocol is composed of two phases: the spray phase and the wait phase. When a new message is created in the system, a number L is attached to that message indicating the maximum allowable copies of the message in the network. During the spray phase, the source of the message is responsible for “spraying”, or delivery, one copy to L distinct “relays”. When a relay receives the copy, it enters the wait phase, where the relay simply holds that particular message until the destination is encountered directly. 2.2

Binary Spray and Wait Routing Protocol

In Binary Spray and Wait (B-S&W), the source of a message initially starts with L copies. Any node A that has n > 1 message copies (source or relay), and encounters another node B (with no copies), hands over to B n/2 and keeps n/2 for itself. When it is left with only one copy, it switches to direct transmission. 2.3

Enhanced Versions of Spray and Wait Routing Protocol

In the enhanced versions of Spray and Wait the changes are done only in the spray phase. Different from B-S&W where the sending node sprays to the encountered node n/2 and keeps n/2 for itself, in order to improve delivery probability we changed the value for S&W-V1 to 3n/5 and keeps 3n/5 and for S&W-V2 to 7n/10 and keeps 7n/10. When it is left with only one copy, it switches to direct transmission. The algorithm of S&W-V1 and S&W-V2 is shown in Fig. 1.

3

Simulation System and Scenarios

The simulation is realized for a part of Tirana city in Albania by importing the map from Open Street Map (OSM) [23] as presented in Fig. 2. Simulations are carried out using the ONE simulator. We create a DTN with 200 nodes. In our simulated scenario, there are 75 cars, 75 pedestrians and 50 buses, all equipped with a smart device that move according to map-based movement model and exchange information among them. The simulation time is 4 h. In Fig. 3 is shown the initial position of all nodes. All network nodes use a transmission range of 10 m. The buffer size varies from 1 MB to 6 MB. The event generator is responsible for generating bundles with sizes uniformly distributed in the ranges [500 kB, 1 MB]. The data bundles ttl is set 300 min. The simulation parameters are shown in Table 1.

108

E. Spaho et al.

Fig. 1. Algorithm of S&W-V1 and S&W-V2.

For the considered parameters we evaluate the performance of 3 different versions of S&W protocols: B-S&W, S&W-V1 and S&W-V2. We use the following metrics to measure the performance of different routing protocols: delivery probability and average latency. • Delivery probability is the ratio of number of delivered messages to that of created messages. • Average latency is the average time elapsed from the creation of the messages at source to their successful delivery to the destination.

4

Simulation Results

In Fig. 4 are shown the simulation results of delivery probability vs. message creation interval. Increasing the message creation interval will increase also the

Enhancement of Binary Spray and Wait Routing Protocol

109

Fig. 2. Tirana city map imported from OSM.

Fig. 3. Nodes initial positions.

delivery probability of all protocols. The enhanced version of S&W perform better than B-S&W. In Fig. 5 are shown the simulation results of the average latency vs. message creation interval. From the results we can notice that the enhanced versions perform better than B-S&W because more copies of the messages are created in the network and they can reach faster the destination. In Fig. 6 are shown the simulation results of delivery probability vs. buffer size. The increase of buffer size have a positive effect on the delivery probability

110

E. Spaho et al. Table 1. Simulation parameters and their values. Parameters

Values

Number of mobile nodes

200

Simulation time

14400 s

Map size

2.5 km × 2.5 km

Buffer size

1, 2, 3, 4, 5, 6 MB

Interface type

Bluetooth/WiFi

Interface Transmission Speed 2 MBps Interface Transmission range 10 m Message TTL

300 min

Pedestrians speed

1.8–5.4 km/h

Cars speed

5.4–27 km/h

Buses speed

25.2–36 km/h

Message size

500k, 1M

Warm up time

100 s

Message creation interval

[25–35], [55–65], [85–95], [115–125] s

1

Binary SprayAndWait Modified SprayAndWait-V1 Modified SprayAndWait-V2

Delivery Probability

0.8

0.6

0.4

0.2

0

25-35

55-65

85-95

115-125

Message Creation Interval (s)

Fig. 4. Results of delivery probability vs. message creation interval.

for all versions of S&W. The enhanced versions S&W-V1 and S&W-V2 perform better in terms of delivery probability. The simulation results of the average latency vs. buffer size are presented in Fig. 7. For small buffer size 1 MB and 2 MB, B-S&W performs better than two other versions. For bigger buffer size, best performance is achieved from enhanced versions.

Enhancement of Binary Spray and Wait Routing Protocol 3000

Binary SprayAndWait Modified SprayAndWait-V1 Modified SprayAndWait-V2

Average Latency (s)

2500

2000

1500

1000

500

0

25-35

55-65

85-95

115-125

Message Creation Interval (s)

Fig. 5. Results of average latency vs. message creation interval. 1

Binary SprayAndWait Modified SprayAndWait-V1 Modified SprayAndWait-V2

Delivery Probability

0.8

0.6

0.4

0.2

0

1

2

3

4

5

6

Buffer size (MB)

Fig. 6. Results of delivery probability vs. buffer size. 3000

Binary SprayAndWait Modified SprayAndWait-V1 Modified SprayAndWait-V2

Average Latency (s)

2500

2000

1500

1000

500

0

1

2

3

4

5

6

Buffer size (MB)

Fig. 7. Results of average latency vs. buffer size.

111

112

5

E. Spaho et al.

Conclusions

In this work, we enhanced the B-S&W routing protocol in the spray phase by increasing the number of copies the sending node sprays to the encountered node and the number of copies it keeps. Then, we evaluated and compared the performance of B-S&W routing protocol with its enhanced version S&W-V1 and S&W-V2 in a DTN based on Tirana city. For evaluation we considered delivery probability and average delay. Simulation results showed that the proposed versions of protocol have better delivery probability and lower average latency compared with B-S&W. In the future, we would like to improve the performance of the S&W routing protocol in terms of overhead ratio. We also would like to create an energyaware S&W routing protocol for DTNs evaluate its performance and compare with different routing protocols considering different scenarios and parameters.

References 1. Fall, K.: A delay-tolerant network architecture for challenged internets. In: Proceedings of the International Conference on Applications, Technologies, Architectures, and Protocols for Computer Communications, SIGCOMM 2003, pp. 27–34 (2003) 2. Delay-and disruption-tolerant networks (DTNs) tutorial. NASA/JPL’s Interplanetary Internet (IPN) Project (2012). http://www.warthman.com/images/DTN Tutorial v2.0.pdf 3. Cerf, V., Burleigh, S., Hooke, A., Torgerson, L., Durst, R., Scott, K., Fall, K., Weiss, H.: Delay-tolerant networking architecture. IETF RFC 4838 (Informational), April 2007 4. Massri, K., Vernata, A., Vitaletti, A.: Routing protocols for delay tolerant networks: a quantitative evaluation. In: Proceedings of ACM Workshop PM2HW2N 2012, pp. 107–114 (2012) 5. Massri, K., Vitaletti, A., Vernata, A., Chatzigiannakis, I.: Routing protocols for delay tolerant networks: a reference architecture and a thorough quantitative evaluation. J. Sens. Actuator Netw. 1–28 (2016). https://doi.org/10.3390/jsan5020006 6. Demmer, M., Fall, K.: DTLSR: delay tolerant routing for developing regions. In: Proceedings of the 2007 ACM Workshop on Networked Systems for Developing Regions, p. 6 (2007) 7. Ilham, A.A., Niswar, M.: Agussalim: evaluated and optimized of routing model on delay tolerant network (DTN) for data transmission to remote area. In: Proceedings of FORTEI, pp. 24–28. Indonesia University, Jakarta (2012) 8. Uchida, N., Ishida, T., Shibata, Y.: Delay tolerant networks-based vehicle-tovehicle wireless networks for road surveillance systems in local areas. Int. J. SpaceBased Situated Comput. 6(1), 12–20 9. Bylykbashi, K., Spaho, E., Barolli, L., Xhafa, F.: Routing in a many-to-one communication scenario in a realistic VDTN. J. High Speed Netw. 24(2), 107–118 (2018) 10. Bylykbashi, K., Spaho, E., Barolli, L., Xhafa, F.: Impact of node density and TTL in vehicular delay tolerant networks: performance comparison of different routing protocols. Int. J. Grid Util. Comput. 7(3), 136–144 (2017)

Enhancement of Binary Spray and Wait Routing Protocol

113

11. Jain, S., Fall, K., Patra, R.: Routing in a delay tolerant network. In: Proceedings of ACM SIGCOMM 2004 Conference on Applications, Technologies, Architectures, and Protocols for Computer Communication, Portland, Oregon, USA, 30 August–3 September 2004, pp. 145–158 (2004) 12. Zhang, Z.: Routing in intermittently connected mobile ad hoc networks and delay. IEEE Commun. Surv. Tutor. 8(1), 24–37 (2006) 13. Soares, V.N.G.J., Rodrigues, J.J.P.C., Farahmand, F.: GeoSpray: a geographic routing protocol for vehicular delay-tolerant networks. Inf. Fusion 15(1), 102–113 (2014) 14. Burgess, J., Gallagher, B., Jensen, D., Levine, B.N.: MaxProp: routing for vehiclebased disruption-tolerant networks. In: Proceedings of the IEEE Infocom, April 2006 15. Lindgren, A., Doria, A., Davies, E., Grasic, S.: Probabilistic routing protocol for intermittently connected networks. draft-irtf-dtnrg-prophet-09. http://tools.ietf. org/html/draft-irtf-dtnrg-prophet-09 16. Vahdat, A., Becker, D.: Epidemic routing for partially connected ad hoc networks. Technical report CS-200006, Duke University, April 2000 17. Spyropoulos, T., Psounis, K., Raghavendra, C.S.: Spray and wait: an efficient routing scheme for intermittently connected mobile networks. In: Proceedings of ACM SIGCOMM 2005 – Workshop on Delay Tolerant Networking and Related Networks (WDTN 2005), Philadelphia, PA, USA, pp. 252–259 (2005) 18. Bylykbashi, K., Spaho, E., Barolli, L., Takizawa, M.: Comparison of spray and wait and epidemic protocols in different DTN scenarios. In: Proceedings of the 12th International Conference on Broad-Band Wireless Computing, Communication and Applications (BWCCA-2017), pp. 218–229 (2017) 19. Spaho, E., Bylykbashi, K., Barolli, L., Takizawa, M.: Routing in a DTN: performance evaluation for random waypoint and steady state random waypoint using NS3 simulator. In: Proceedings of the 12th International Conference on P2P, Parallel, Grid, Cloud and Internet Computing (3PGCIC-2017), pp. 133–141 (2017) 20. Spaho, E., Dhoska, K., Bylykbashi, K., Barolli, L., Kolici, V., Takizawa, M.: Performance evaluation of energy consumption for different DTN routing protocols. In: Proceedings of the 21st International Conference on Network-Based Information Systems (NBiS-2018), pp. 122–131 (2018) 21. Spaho, E., Dhoska, K., Bylykbashi, K., Barolli, L., Kolici, V., Takizawa, M.: Performance evaluation of routing protocols in DTNs considering different mobility models. In: Proceedings of the 15th International Symposium on Frontiers of Information Systems and Network Applications (FINA-2019), pp. 205–214 (2019) 22. Keranen, A., Ott, J., Karkkainen, T.: The ONE simulator for DTN protocol evaluation. In: Proceedings of the 2nd International Conference on Simulation Tools and Techniques (SIMUTools-2009) (2009). http://www.netlab.tkk.fi/tutkimus/ dtn/theone/pub/theonesimutools.pdf 23. Open street map. http://www.openstreetmap.org/

Data Exchange Algorithm at Aggregate Level in the TWTBFC Model Yinzhe Guo1(B) , Ryuji Oma1 , Shigenari Nakamura1 , Tomoya Enokido2 , and Makoto Takizawa1 1 Hosei University, Tokyo, Japan {yinzhe.guo.3e,ryuji.oma.6r}@stu.hosei.ac.jp, [email protected] 2 Rissho University, Tokyo, Japan [email protected]

Abstract. In the TBFC (Tree-Based Fog Computing) and TWTBFC (Two-Way TBFC) models the electric energy consumed by fog nodes and servers can be reduced in the fog computing (FC) model. Here, fog nodes are hierarchically structured in a height-balanced tree, where a root node is a cloud of servers, leaf nodes are edge nodes which communicate with devices, and each node receives data from child nodes and sends the processed data to a parent node. In the TWTBFC model, nodes send processed data to not only a parent node but also each child node. In order to reduce the network traffic in the TWTBFC model, only aggregate nodes at some level collect the output data of every other aggregate node, i.e. aggregate data. Since only target actuators are to be activated, the aggregate data has to be only delivered to target actuators. Nodes whose descendant actuators are target ones are relay nodes. On receipt of aggregate data, only relay nodes forward the aggregate data to the child nodes. We evaluate the new TWTBFC model in terms of energy consumption of nodes and number of messages transmitted to deliver aggregate data to edge nodes. Keywords: Energy-efficient fog computing · IoT (Internet of Things) · Two-way TBFC (TWTBFC) model · Aggregate node

1

Introduction

The IoT (Internet of Things) [5,7] is composed of not only computers like servers and clients but also millions of devices, i.e. sensors and actuators installed in various things like glasses and cars [11,14]. Compared with traditional information systems like the cloud computing (CC) model [4], the IoT is more scalable and huge amount of data from sensors are transmitted in networks and are processed by application processes on servers. The fog computing (FC) model [16] is proposed to reduce the network and server traffic of the IoT (Internet of Things). On the other hand, huge amount of electric energy is consumed by nodes. In order to not only increase the performance but also reduce the electric energy c Springer Nature Switzerland AG 2020  L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 114–124, 2020. https://doi.org/10.1007/978-3-030-33506-9_11

Data Exchange Algorithm at Aggregate Level in the TWTBFC Model

115

consumption of the IoT, the TBFC (Tree-based Fog Computing) model is proposed in our previous studies [3,11,12,15]. Here, fog nodes are hierarchically structured in a height-balanced tree. A root node is a cloud of servers and leaf nodes are edge nodes which receive sensor data from sensors and send actions to actuators. Each fog node has one parent node and child nodes. Each node receives input data from the child nodes. Then, the fog node processes the input data and sends the output data, obtained by processing the input data to a parent node. A server in a cloud finally receives data processed by fog nodes. Then, the servers delivers actions to actuators through networks of fog nodes. While the traffic of servers and networks can be reduced, it takes time to deliver actions to actuators. The TWTBFC (Two-Way TBFC) model [9,10] is also proposed to reduce delay time to deliver actions to actuators. Here, a node not only sends output data to a parent node in a same way as the TBFC model but also forwards the output data to the child nodes. In addition, some level is taken as aggregate level. Nodes at aggregate level are aggregate nodes [8]. Each aggregate node collects the output data from every other aggregate nodes. Then, each aggregate node obtains aggregate data which is a collection of output data of all the aggregate nodes. Then, the aggregate data is transmitted from each aggregate node down to the descendant edge nodes. Then, edge nodes make a decision on actions and activate child actuators by sending the actions. Since the aggregate data is transmitted to every edge node, more number of messages are transmitted in networks. On the other hand, only some edge node is required to activate its child actuators. Actuators to be activated for the aggregate data are target ones. Nodes whose descendant actuators are target ones are referred to as relay ones. In order to reduce the network traffic, we propose a new model where only relay nodes forward the aggregate data to the child nodes. In the evaluation, we show the number of messages and energy consumption of nodes to obtain the aggregate data and deliver the aggregate data to target edge nodes. In Sect. 2, we present the TWTBFC model of the IoT. In Sect. 3, we present the power consumption and computation module of a fog node. In Sect. 4, we evaluate the TWTBFC model.

2 2.1

Two-Way Tree-Based Fog Computing (TWTBFC) Model TBFC Model

The fog computing (FC) model [16] to efficiently realize the IoT [11] is composed of sensor and actuator devices, fog nodes, and clouds. Clouds are composed of servers like the cloud computing (CC) model [4]. In the TBFC (Tree-Based Fog Computing) model [12,15], fog nodes are hierarchically structured in a heightbalanced tree as shown in Fig. 1. Here, the root node f denotes a cloud of servers. Fog nodes at the bottom level are edge nodes which communicate with sensors and actuators.

116

Y. Guo et al.

Fig. 1. TBFC model.

Each node fR has cR (≥ 0) child nodes fR , . . . , fR,cR . Here, fRi shows the ith child node of the fog node fR and in turn fR is a parent node of the node fRi . ch(fR ) is a set {fR1 , . . . , fR,cR } of child nodes of a node fR . pt(fRi ) is a parent node fR of a node fRi . For example, the second child node of a root node f is f2 , and the first child node of the node f2 is f21 . Thus, the label R of a fog node fR is a sequence of numbers and shows a path from the root node f to the fog node fR . Let an(fR ) be a set of ancestor nodes of a node fR . dn(fR ) shows a set of descendant nodes of a node fR and sn(fR ) is a set of nodes which are at the same level as a node fR . In the cloud computing (CC) model, an application process p is performed on servers to process sensor data sent by sensors in networks. In this paper, an application process p is assumed to be linear, i.e. a sequence of subprocesses p0 , p1 , . . . , ph−1 . The edge subprocess ph−1 takes input data from sensors. The root subprocess p0 is performed on a root node f , i.e. servers. Each subprocess pi takes data from a subprocess pi−1 and gives the processed data to a subprocess pi+1 . In the TBFC tree of height h, each subprocess pi is performed on nodes of level i. Let p(fR ) show a subprocess to be performed in a node fR . A node fR takes input data dRi from each child node fRi (i = 1, . . . , lR ). DR shows a collection of the input data dR1 , . . . , dR,lR from child nodes fR1 , . . . , fR,lR , respectively. The node fR obtains output data dR by doing the computation f (pR ) on the input data DR . Then, the node fR sends the output data dR to a parent node pt(fR ).

Data Exchange Algorithm at Aggregate Level in the TWTBFC Model

117

A notation |d| shows the size [Byte] of data d. Let iR and oR be the size of |DR | and |dR | of input data DR and output data dR , respectively. The ratio |oR |/|iR | is the output ratio ρR of a node fR . Here, oR = ρR · iR . For example, if a fog node fR obtains an average value of the input data dR1 , . . . , dR,lR , the output ratio ρR is 1/lR . 2.2

Aggregate Nodes

Each fog node fR receives input data DR = {dR1 , . . . , dR,lR } from the child nodes fR1 , . . . , fR,lR and obtains output data dR by processing the output data DR . At some level l of the tree, each fog node sends the output data dR to every other node fv and receives output data dv from every other fog node fv in addition to sending the output data dR to the parent node pt(fR ). Then, each fog nodes fR obtains a collection ADl of output data from every node at the same level, i.e. here, nodes of level l are referred to as aggregate nodes and the level l is aggregate level in the tree. Let ANl be a set of aggregate nodes of level l in the tree. The data ADl is an aggregate data which is a set of output data obtained by all the aggregate nodes of the level l, i.e. ADl = {ds  fs ∈ ANl }. By using the aggregate data ADl , actuators to be activated are decided. An actuator to be activated for the aggregate data ADl is referred to as target actuator. Here, a node which is an ancestor of a target actuator is referred to as relay node. Each aggregate node fR sends the aggregate data ADl to each relay child node. Let RNR (∈ ch(fR )) be a subset of relay child nodes of a node fR . Even if a non-relay child node fRi receives the aggregate data ADl , the node fRi does not forward ADl to any child node fRij . Thus, only a relay node fRi forwards the aggregate data ADl to child nodes fRij . Eventually, a relay edge node fR receives the aggregate data ADl . The relay edge node fR makes a decision on actions to be performed on child target actuators and issues the actions to the target actuators. Target actuators are localized in some area for the aggregate data ADl as shown in Fig. 2. A node fR is referred to as broadcast node if every descendant edge node is a relay one. This means, actuators in an area covered by a broadcast node are activated. On receipt of the aggregate data ADl , a broadcast node fR forwards the aggregate data ADl to every child node. Every descendant node of a broadcast node fR is a broadcast node. An aggregate node forwards the aggregate data ADl to relay nodes. A relay node forwards the aggregate data ADl to relay nodes. Eventually, a relay node fR forwards the aggregate data ADl to every child node. Here, fR is a broadcast node. A level at which the node fR exists is a broadcast (b) level. At higher level than the b level, a relay node send the aggregate data ADl to only relay nodes, i.e. unicasts ADl to each relay nodes Let us consider a node fR which has child fog nodes fR1 , . . . , fR,cR . Let xR stand for the size |dR | of the output data dR . The size xR of the output data dR cR of a node fR is given as xR = ρR · (Σi=1 xRi ). Here, ρR is the output ratio of the node fR . If fR is an edge node, each size xRi shows the size of the sensor data dRi from a child sensor sRi . Thus, the size xRi of the output data dRi of each child

118

Y. Guo et al.

Fig. 2. Relay nodes and target actuators.

node fRi of level l can be obtained, and then the size xR of an aggregate node fR is calculated. Each aggregate node fR of aggregate level l obtains the aggregate cs xsi ). The aggregate data data ADl whose size asl (= |ADl |) is Σfs ∈ANl (ρs · Σi=1 ADl of size asl is forwarded to target edge nodes. On receipt of the aggregate data ADl , a relay edge node fR decides on actions and sends the actions to the target actuators aR1 , . . . , aR,alR .

3 3.1

Power Consumption and Computation Models of a Fog Nodes Upward Transmission

A fog node fR is assumed to be implemented to be a sequence of input (IR ), computation (CR ), and output (OR ) modules. The input module IR receives input data dRi from a child node fRi and the output module OR sends output data dR to a parent node pt(fR ). The computation module CR is a subprocess p(fR ) which generates the output data dR by processing input data DR = dR1 , . . . , dR,CR . In this paper, we assume the IR , CR , and OR modules are sequentially performed in a fog node fR on receipt of the input data DR . It takes time to perform the IR , CR , and OR modules of a node fR . Let T IR (x), T CR (x), and T OR (x) show the execution time [sec] of the input IR ,

Data Exchange Algorithm at Aggregate Level in the TWTBFC Model

119

computation CR , and output OR modules of a node fR for data of size x, respectively. The execution time T CR (x) depends on the computation complexity of a subprocess p(fR ). In this paper, the computation complexity of the subprocess p(fR ) is assumed to be O(x) or O(x2 ). That is, the execution time T CR (x) of the computation module (CR ) is ctR · CR (x) where CR (x) = x or CR (x) = x2 and ctR is a constant. A pair of execution time T IR (x) and T OR (x) to receive and send data of size x, respectively, are proportional to the data size x, i.e. T IR (x) = rtR · x and T OR (x) = stR · x, where stR and rtR are constants. Thus, the execution time T CR (x), T IR (x), and T OR (x) are given as follows: T CR (x) = ctR · CR (x). T IR (x) = rtR · x.

(1) (2)

T OR (x) = stR · x.

(3)

It takes time T FR (x) [sec] for each node fR to receive and process input data DR of size x and send the output data dR to a parent node pt(fR ): T FR (x) = T IR (x) + T CR (x) + δR · T OR (ρR · x).

(4)

Here, if fR is a root node, δR = 0, else δR = 1. The execution time T IR (x) of the IR module realized in a Raspberry Pi 3 model B [2] node is five times longer than the execution time T OR (x) of the OR module, i.e. rtR = 5 · stR and ctR = rtR /2 [13]. That is, ctR : stR : rtR = 1 : 2.5 : 0.5. EIR (x), ECR (x), and EOR (x) show the electric energy [J] consumed by the input IR , computation CR , and output OR modules [11] of a node fR for data of size x, respectively. In this paper, we assume each node fR follows the SPC (Simple Power Consumption) model [5–7]. The power consumption of a node fR to perform the computation module CR (= p(fR )) is maxER [W]. In a Raspberry Pi Model B, node fi , maxEi = 3.7 [W]. The energy consumption ECR (x) [J] of the computation module CR of a node fR to process data of size x (> 0) is ECR (x) = maxER [W] · T CR (x) [sec]. A pair of the electric power P IR and P OR [W] are consumed to perform the input IR and output OR modules, respectively [5–7]. P IR and P OR are reR · maxER and seR · maxER , respectively, where 0 < seR ≤ reR ≤ 1. For example, seR = 0.676 and reR = 0.729 in the Raspberry Pi 3 model B node fR [13]. The energy consumption EIR (x) and EOR (x) [J] to receive and send data of size x (> 0) are EIR (x) = P IR [w] ·T IR (x)[sec] and EOR (x) = P OR [w] · T OR (x)[sec], respectively. Each node fR consumes the energy EFR (x) to reduce and process the input data DR of size x and send the processed data dR of size ρR · x: EFR (x) = EIR (x) + ECR (x) + δR · EOR (ρR · x) = (reR · T IR (x) + T CR (x) + δR · seR · T OR (ρR · x)) · maxER = (reR · rtR · x + ctR · CR (x) + δR · seR · stR · ρR · x) · maxER . (5)

120

3.2

Y. Guo et al.

Downward Transmission

Each aggregate node fR consumes electric energy and takes time to collect the aggregate data ADl from other aggregate nodes of aggregate level l as shown in Fig. 2. ANl is a set of aggregate nodes at level l. Each aggregate node fR of level l sends the output data dR to and receives the output data ds from every other aggregate node fs . Let os be the size |ds | of the output data ds . Then, the aggregate node fR forwards the aggregate data  ADl to the child nodes fR1 , . . . , fR,cR . The aggregate data ADl is a set {ds  fs ∈ ANl } of output data of every aggregate node. The size asl (= |ADl |) of the aggregate data ADl is: cs asl = Σfs ∈ANl |os | = Σfs ∈ANl (ρs · Σi=1 osi ).

(6)

It takes time AEXR of an aggregate node fR to send the output data dR and to receive the output data ds from every other aggregate node fs : AEXR = T OR (oR ) · |ANl | + Σfs ∈ANR T IR (os ).

(7)

Then, a relay aggregate node fR of aggregate level l sends the aggregate data ADl to relay child nodes. Let RNR (⊆ ch(fR )) be a set of relay child nodes of a node fR . The total time AT OR [sec] of a relay aggregate node fR is given as follows: AT OR = T OR (oR ) · |ANl | + Σfs ∈ANl T IR (os ) + T OR (asl ) · |RNR |.

(8)

The relay aggregate node fR consumes the energy AEOR [J] as follows: AEOR = (seR · T OR (oR ) · |ANl | + seR · T OR (|asl |) + reR · Σfs ∈ANl T IR (os )) · maxER .

(9)

A descendant relay node fR of the aggregate nodes receives the aggregate data ADl . If fR is a relay node, the node fR forwards the aggregate data ADl to the relay child nodes. The execution time AT OR of a relay node fR is as follows: AT OR = T IR (asl ) + T OR (asl ) · |RNR | = rtR · asl + stR · asl · |RNR |.

(10)

Each node fR of level k (< l) consumes energy AEOR to forwards the aggregate data ADl to the descendant edge nodes. AEOR = (reR · T IR (asl ) + seR · T OR (asl )) · maxER = (reR · rtR · asl + seR · stR · asl · |RNR |) · maxR .

(11)

The higher the aggregate level l is, the smaller size of the aggregate data ADl and the fewer number of messages are exchanged among the aggregate nodes. However, the more number of messages are transmitted to deliver the aggregate data ADl to edge nodes.

Data Exchange Algorithm at Aggregate Level in the TWTBFC Model

4

121

Evaluation

We evaluate the TWTBFC model of the IoT in terms of electric energy consumption of fog nodes and number of messages transmitted by fog nodes. The TWTBFC model is composed of fog nodes structured in a tree. In this paper, we consider a height-balanced k-ary tree of fog nodes, whose height is h. The output ratio ρR of each fog node fR is assumed to be the same ρ, i.e. ρR = ρ. We assume a root node is a server f with a pair of Inter Xeon E5-2667 CPUs [1], where the minimum electric power consumption minE0 is 126.1 [W] and the maximum electric power consumption maxE0 is 301.3 [W]. Each fog node fR is realized by a Raspberry Pi 3 Model B [2]. Here, the minimum power minER is 2.1 [W] and the maximum power maxER is 3.7 [W] [3]. The computation ratio CRR of each fog node fR is 0.879/4.75 = 0.185, where the computation rate of the root node f is 1. This means, the computation speed of the node fR is 18.5 [%] of the root node f . ANl is a set of aggregate nodes at aggregate level l. There are k l (= |ANl |) aggregate nodes at aggregate level l in the tree. As presented in the preceding section, each aggregate node fR exchanges the output data  dR with every other aggregate node and obtains the aggregate data ADl (= fs ∈ANl ds ). Each aggregate node fR sends the output data dR to (k l − 1) aggregate nodes and receives output data from the other (k l − 1) aggregate nodes. Hence, totally, k l · (k l − 1) messages are transmitted. Here, the size of sensor data which each edge node receives from the child sensors is assumed to be one. A node of level h − 1 receives data of total size k from k edge nodes and sends the output data dR of size ρ·k. Thus, each aggregate node fR receives the output data DR of size (ρk)h−1−l · k and generates the output data of size (ρk)h−l−2 . Hence, the total size k l · (k l − 1) · (ρk)h−l−1 of data is exchanged among the aggregate nodes. For the aggregate data ADl , the target actuators are in some area. In this paper, we assume there is one broadcast node at broadcast level b and every descendant edge node of the broadcast node is a target one. At level q (l ≤ q < b), one relay node sends the output data ADl to one child relay node. Then, a broadcast node sends the output data ADl to k child nodes. Thus, (b − l) messages are transmitted to deliver the aggregate data ADl to the broadcast node. The node fR sends the aggregate data ADl to k child nodes and each child node forwards the message ADl to k child nodes. Thus, totally, k + k 2 + . . . + k h−1−b = k ·(1−k h−1−b )/(1−k) messages are transmitted. For a broadcast node fR of broadcast level b, there are k h−1−b descendant edge nodes. Here, totally (b − l) + k · (1 − k h−1−b )/(1 − k) messages are transmitted. The total size of data transmitted is [(b − l) + k · (1 − k h−1−b )/(1 − k)] · k l · (k l − 1) · (ρk)h−1−l . We assume k = 2, the height h of tree is 10 (h = 10) and the output ratio ρ = 0.5 in the evaluation.

122

Y. Guo et al.

Fig. 3. Total size of data transmitted (b = 5).

Fig. 4. Total size of data transmitted (l = 5).

Data Exchange Algorithm at Aggregate Level in the TWTBFC Model

123

Figure 3 shows the ratio of the total size of data transmitted at the aggregate level l(1 ≤ l ≤ 8) to the level 8. Here, a broadcast level b is five (b = 5). The higher the aggregate level l is, the smaller the total size of data transmitted. Especially, if the aggregate level l is larger than 5, the total size of data transmitted exponentially increases. Figure 4 shows the ratio of the total size of data transmitted for the broadcast level b (b ≥ l) where l = 5. The higher the broadcast level b is, the more volume of data is transmitted.

5

Concluding Remarks

In this paper, we proposed the modified model to efficiently realize the TWTBFC model. Here, one aggregate level l is selected and aggregate nodes at the aggregate level l collect output data of every aggregate node as the aggregate data ADl . A target edge node is one whose actuators are activated for the aggregate data ADl . A fog node whose descendant edge nodes are target ones is a relay node. Only target actuators have to be activated. The aggregate data ADl has to be delivered to only edge nodes of target actuators. In this paper, only a relay node forwards the aggregate data ADl to its relay nodes. In the evaluation, we showed the number of messages and energy consumption of nodes to exchange output data and forward aggregate data to descendant nodes can be reduced.

References 1. Dl360p gen8. www8.hp.com/h20195/v2/getpdf.aspx/c04128242.pdf?ver=2 2. Raspberry pi 3 model b. https://www.raspberrypi.org/products/raspberry-pi-3model-b/ 3. Chida, R., Guo, Y., Oma, R., Nakamura, S., Duolikun, D., Enokido, T., Takizawa, M.: Implementation of fog nodes in the tree-based fog computing (TBFC) model of the IOT. In: Proceedings of the 7th International Conference on Emerging Internet, Data and Web Technologies (EIDWT-2019), pp. 92–102 (2019) 4. Creeger, M.: Cloud computing: an overview. Queue 7(5), 3–4 (2009) 5. Enokido, T., Ailixier, A., Takizawa, M.: A model for reducing power consumption in peer-to-peer systems. IEEE Syst. J. 4, 221–229 (2010) 6. Enokido, T., Ailixier, A., Takizawa, M.: Process allocation algorithms for saving power consumption in peer-to-peer systems. IEEE Trans. Ind. Electron. 58(6), 2097–2105 (2011) 7. Enokido, T., Ailixier, A., Takizawa, M.: An extended simple power consumption model for selecting a server to perform computation type processes in digital ecosystems. IEEE Trans. Ind. Inf. 10, 1627–1636 (2014) 8. Guo, Y., Oma, R., Nakamura, S., Duolikun, D., Enokido, T., Takizawa, M.: Data and subprocess transmission on the edge node of TWTBFC model. In: The 11th International Conference on Intelligent Networking and Collaborative Systems (INCoS-2019) (2019, Accepted) 9. Guo, Y., Oma, R., Nakamura, S., Duolikun, D., Enokido, T., Takizawa, M.: Evaluation of a two-way tree-based fog computing (TWTBFC) model. In: Proceedings of the 13th International Conference on Innovative Mobile and Internet Services in Ubiquitous Computing (IMIS-2019), pp. 72–81 (2019)

124

Y. Guo et al.

10. Guo, Y., Oma, R., Nakamura, S., Duolikun, D., Enokido, T., Takizawa, M.: A twoway flow model for fog computing. In: Proceedings of the Workshops of the 33rd International Conference on Advanced Information Networking and Applications (WAINA-2019), pp. 612–620 (2019) 11. Oma, R., Nakamura, S., Duolikun, D., Enokido, T., Takizawa, M.: An energyefficient model for fog computing in the internet of things (IoT). Internet Things 1–2, 14–26 (2018) 12. Oma, R., Nakamura, S., Duolikun, D., Enokido, T., Takizawa, M.: Evaluation of an energy-efficient tree-based model of fog computing. In: Proceedings of the 21st International Conference on Network-Based Information Systems (NBiS-2018), pp. 99–109 (2018) 13. Oma, R., Nakamura, S., Duolikun, D., Enokido, T., Takizawa, M.: Evaluation of data and subprocess transmission strategies in the tree-based fog computing (TBFC) model. In: Proceedings of the 22nd International Conference on NetworkBased Information Systems (NBiS-2019) (2019, Accepted) 14. Oma, R., Nakamura, S., Duolikun, D., Enokido, T., Takizawa, M.: A fault-tolerant tree-based fog computing model. Int. J. Web Grid Serv. (IJWGS) (2019, Accepted) 15. Oma, R., Nakamura, S., Enokido, T., Takizawa, M.: A tree-based model of energyefficient fog computing systems in IoT. In: Proceedings of the 12th International Conference on Complex, Intelligent, and Software Intensive Systems (CISIS-2018), pp. 991–1001 (2018) 16. Rahmani, A.M., Liljeberg, P., Preden, J.S., Jantsch, A.: Fog Computing in the Internet of Things. Springer, Heidelberg (2018)

Trust-Based Game-Theoretical Decision Making for Food-Energy-Water Management Suleyman Uslu1 , Davinder Kaur1 , Samuel J. Rivera2 , Arjan Durresi1(B) , and Meghna Babbar-Sebens2 1

Indiana University-Purdue University Indianapolis, Indianapolis, IN, USA {suslu,davikaur}@iu.edu, [email protected] 2 Oregon State University, Corvallis, OR, USA {sammy.rivera,meghna}@oregonstate.edu

Abstract. Decision making has been an essential aspect of the life of both individuals or organizations. We all have to face situations where we need to decide whether a daily one such as having coffee or tea in the breakfast or selecting a graduate school for Ph.D. The importance of a decision increases as the duration of the impact of its results and the number of people that are affected increases. Food-Energy-Water (FEW) is one of the fields where the impacts can stay for a long time and affect many people and areas. In this paper, we proposed a game theory-based approach for decision making among FEW actors sharing a finite amount of continuous resource where actors have different weights on their trust and the amount of share that they receive in their payoff functions. Then, we run simulations on scenarios utilizing a more realistic discrete solution set for actors. Results have shown that when actors place more weight on trust in their payoff function, they tend to propose fairer solutions that are closer to the consensus point. Also, they move towards that point faster compared to actors with low trust weight.

1

Introduction

One of the considerably essential problems that people deal with is decision making. Although the decision itself could be challenging to make, it becomes more burdensome when multiple stakeholders are included. Food-Energy-Water (FEW) is one of the fields where the results of the decisions could be vital for a significant part of the community. Also, it satisfies the first condition, which is involving multiple stakeholders from farmers to government agencies and administrative people. A sophisticated Decision Support System (DSS) can help actors in FEW fields by (i) organizing the development and evaluation of feedback, (ii) highlighting the supported solutions, and (iii) demonstrating the results of proposed and approved solutions. In this study, we first have actors to propose solutions for a split of a finite continuous resource, and then we use a more realistic and sophisticated solution c Springer Nature Switzerland AG 2020  L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 125–136, 2020. https://doi.org/10.1007/978-3-030-33506-9_12

126

S. Uslu et al.

set and have actors to propose solutions from this set at each round. Solutions proposed by actors are rated by other actors, which are used as trust measurements by our measurement theory-based trust framework. First, we propose a game theory-based approach for finite resource sharing problem. Actors try to maximize their payoffs which is a function of both the share of the resource that they receive based on the proposed solution and the trust that they receive from the community as ratings to the proposed solution. Then, we present our DSS considering modern and primitive approaches [5,9,13–15] combined with our trust management framework. We run simulations on the realistic solution set and investigate the effect of the weight of the trust in the payoff function. Finally, we present that as actors put more weight on trust, they tend to propose solutions closer to the consensus, which decreases the number of rounds to reach an agreement. The paper is organized as follows. In Sect. 2, background and related work regarding decision making and trust are presented. In Sect. 3, some details of the trust management framework is explained, and the formulas that are used for this study is presented. In Sect. 4, a game theory-based approach that utilizes trust is proposed, and its results are presented. In Sect. 5, considering the game theory approach, simulation of the trust management framework for realistic data and its results are presented. Then, we conclude the paper in Sect. 6.

2

Background and Related Work

In this section, we present the background and related work for decision making, trust, and its applications. 2.1

Decision Making

Economics, society, and environment are such fields that require decisions having significant impacts on people’s lives; however, a sophisticated decision-making mechanism can help people to overcome the difficulty of making decisions in those fields. One example of such outcomes is the global financial crisis of 2008, where more than half of the population of the world is affected. As the world is becoming more complex, the decisions cannot be expected to stay simpler. Moreover, engaging multiple stakeholders from different fields with a different point of views, and maybe with competing goals, could make decisions even more complicated [9]. Whereas the stakeholders might have conflicting or competing objectives, they can also be experts in the field and expressing their ideas could be their primary reason to attend the decision making. Their goal, consequently, becomes reaching a consensus by idea discussion and expression, including personal development in the field. This could require multiple rounds with expression, discussion, and alteration of the ideas. In [5], minimization of modifications approach has been proposed for expert solutions.

Trust-Based Game-Theoretical Decision Making for Food-Energy-Water Management

127

Besides decision making, there can be other ways of conversation and communication among participants such as surveys, focus groups, interviews, workshops, and scenario analysis [6]. Also, an algorithm is designed and proposed to utilize the feedback of the actors for a sophisticated optimization [4]. 2.2

Trust and Its Applications

Trust is a context-dependent concept and can be highly effective directly or indirectly in the decision making process [22–24]. Using a trust model which is capable of historic measurements could require the process to be computerized [8, 12,16]. To measure and anticipate trust between entities of a network or members of a community, a measurement theory-based trust management framework is proposed [21]. The list of applications of this framework include stock market prediction using Twitter data and trust management of Internet of Things, in cloud computing, and for fake user and news detection [10,17–20,28,29]. Several uses of trust in decision making are the transactions over the internet [7], trust-based consumer decision making models for e-commerce [11], multi-stakeholder decision-making model for water allocation problem [1], and a generic framework for consensus reaching [2]. We proposed a DSS using our measurement theory-based trust management framework for the natural resource sharing problem in FEW [26]. An advanced version of this DSS with the capability of utilizing ratings of ratings is also proposed in [25].

3

Trust Framework

A framework which is based on measurement theory is proposed to measure the trust among parties in a community [21]. In Food-Energy-Water (FEW) sectors, one candidate of the interactions that could be used to generate a trust network between actors could be the ratings that actors assign to the solutions of a specific problem proposed by other actors. In other words, when actors gather and discuss a problem regarding FEW fields and propose solutions, their ratings to the proposed solutions can be used to measure the trust between the actors in that specific field. In [21], the two main property of the trust is described as the impression, denoted by m, and the confidence, denoted by c. Higher ratings lead to a higher impression value, whereas consistent ratings are required to achieve a higher confidence value. As a part of the trust modeling, we calculate the impression as the average of the ratings as shown in Eq. 1. Corresponding confidence can be calculated as shown in Eq. 2 where rA:B is a rating from actor A to actor B as one measurement. N A:B r A:B m (1) = i=1 i N  N A:B − r A:B )2 i i=1 (m cA:B = 1 − 2e where e = (2) N (N − 1)

128

S. Uslu et al.

In addition to the capability of measuring trust using ratings and calculating the impression and confidence values, our framework can anticipate the trust between two entities, which are not interacted yet, through inference rules. The two inference rules we utilize in our framework are the transitivity and the aggregation. Although there are several transitivity and aggregation formulas proposed in [21], we decided to use the TP1 and AP1 which are based on the multiplication of the impression for the transitivity and the average for the aggregation as shown in Eqs. 3 and 5. Corresponding error values can also be calculated using Eqs. 4 and 6. = mST ⊗ mT D = mST mT D mSD T  = eST ⊗ eT D = (eST )2 (mT D )2 + (eT D )2 (mST )2 eSD T SD mSD T1 ⊕ mT2 = SD eSD T1 ⊕ eT2 =

4

mSD T1 

+ 2

mSD T2

1 2 ((eSD )2 + (eSD T2 ) ) 22 T1

(3) (4) (5) (6)

Decision Making Using Game Theory and Trust

Specifically, for perfect information games where the strategy of all players are known by everyone, players can maximize their payoff using backward induction. Considering the game tree, which shows the possible actions of the players at each turn, the last player selects the leaf, which gives him the best return on his last turn. Since this information is also known by the other players, second from the last player would select the node in his turn where he maximizes his payoff considering the strategy of the last player. This strategy escalates from the leaves to the root of the tree where the first player takes an action [3,27]. If an actor in the decision making can decide the values of all the parameters which affect his payoff, he can easily propose a solution that maximizes his payoff. However, in real life, there can be parameters which are decided by other actors. Moreover, a trade-off can exist between the parameters that an actor decide their value and the parameters whose values are decided by other actors. To overcome this trade-off situation and propose a solution, which is assigning values to the parameters, actors apply weights on these parameters. If a parameter has more weight, actors tend to increase the value that they assign to that parameter. Another question arises from the situation where some of the parameters in the payoff equation cannot be assigned by the actor himself but the other actors. In perfect information games, since the strategy of others are known to the actor, he can take an action considering this information and still maximize his payoff. However, this is not the case in real-life scenarios. Every actor has their strategy, and those strategies are not necessarily known to other actors. Moreover, an actor can change his strategy even during decision making. We show our game theory approach starting with a basic scenario and then present more realistic ones.

Trust-Based Game-Theoretical Decision Making for Food-Energy-Water Management

129

In our decision-making scenarios in FEW fields, actors propose solutions to a split of a finite resource, give ratings for other actors’ solutions, and receive a rating for their solutions. Those ratings are used to build trust between the actors. Shares in the split that is proposed by the actor are the parameter that the actor can decide directly whereas the ratings. Therefore the trust is a parameter that is decided by the other actors, maybe by the whole community collectively. The interesting point is that an actor might affect the decision of the other actors on the parameters of the second type through his decisions on the parameters of the first type. If an actor proposes a fairer solution where he might lose some of his shares, he can receive better ratings from the community, which helps him increase his trust. Payoff function of an actor can be defined as the weighted average of his share in the split and the trust of the community he gains during the decision making as shown in Eq. 7 where Fa is the payoff of actor a, ws is the weight of the share, wt is the weight of the trust, sa is the share in the split, and ta is the trust of the actor. (7) Fa = ws sa + wt ta In the first scenario, the sum of the shares each actor requests for themselves in their proposals is less than or equal to the available amount of the resource as shown in Eq. 8 where ai is an actor, A is the set of actors, sai :aj is the amount of resource ai gives to aj , and M is the total amount of the resource. In this basic scenario, since the resource is enough to satisfy every actor, there is no need to negotiate.  sai :ai ≤ M (8) ai ∈A

In the second scenario, which leads to a negotiation, the sum of the requested amounts of the resource is more than the total amount as shown in Eq. 9. Although we can assume that each actor maximizes their payoff and propose it as a solution, they cannot receive the amount that they requested due to exceeding the total resource amount.  sai :ai > M (9) ai ∈A

Considering Eq. 7, we first assume that the weight on the share is much higher than the weight on the trust, as shown in Eq. 10. In other words, actors first maximize their share until they are satisfied, then maximize their trust. ws  wt

(10)

Since trust is calculated historically and the previous measurements cannot be changed, to maximize the trust, actors need to maximize the ratings for the current round. If the impression is calculated as the average of the ratings, actors maximize the sum of the ratings. However, it is more realistic that they try to maximize the ratings in a fairer way such as maximizing the multiplication of the ratings as shown in Eq. 11 which leads to an effort to satisfy other actors equally.

130

S. Uslu et al. ∗

In Eq. 11, fra refers to the function of ratings that the actor a∗ is maximizing, ∗ A is the set of actors, and rai :a is the rating from actor ai to a∗ .  ∗ ∗ rai :a (11) fra = a∗ ,ai ∈A, ai =a∗

Since the actors try to receive equal ratings from the other actors, after they request the amount of resource that would satisfy them, they split the remaining resource among the other actors weighted by their initial requests. In Eq. 12, actor a∗ receives equal rating from all other actors where ai and aj also belong to the actor set A and they are different then a∗ . ∗

∗

rai :a = raj :a | ∀a∗ , ai , aj ∈ A, a∗ = ai = aj

(12)

If the rating function for actors is defined as the ratio of the received amount of resource from an actor to the requested amount, as shown in Eq. 13, and if there are 3 actors, a, b, and c, the amount of resource that actor a should give to b and c to maximize the Eq. 11 is given in Eqs. 14 and 15. This can also be generalized to n actors by changing the denominator as the sum of the requested amounts from all other actors. saj :ai sai :ai 1 − sa:a = b:b ∗ sb:b s + sc:c 1 − sa:a = b:b ∗ sc:c s + sc:c

rai :aj =

(13)

sa:b

(14)

sa:c

(15)

However, in real life, the weights on the parameters are usually comparable. Actors maximizing their share first without considering the feedback from the community may prevent actors from reaching a consensus in the decision making. Therefore, we neglect the condition given in Eq. 10 and have rounds where actors propose a solution maximizing their payoff with the prediction of the ratings received from the other actors. When the consensus condition, given in Eq. 8, is not met, actors also update their rating functions, adding a multiplier that is less than 0 to the Eq. 13, to push other actors to be fairer in their proposals. We run two different tests to see the effect of the weights on the share and the trust received from the community. In the first test, actors weight 0.3 on the trust and 0.7 on the share of the resource that they receive. As shown in Fig. 1, actors start proposing solutions as they maximize their payoff. Then, move towards to the consensus point since they cannot receive their estimated ratings from the other actors and also simply if there is no consensus, the solutions are not accepted. When actors increase the weight of the trust, from 0.3 to 0.6, in their payoff function, they tend to start with a solution that is closer to the consensus point compared to the weight of 0.3. Also, they move towards the consensus point faster, as shown in Fig. 2.

Trust-Based Game-Theoretical Decision Making for Food-Energy-Water Management

131

Fig. 1. When actors have 0.3 weight on the trust, they start proposing solutions away from the consensus point and also move towards it gradually.

Fig. 2. When actors increase the weight of trust to 0.6, they start proposing solutions closer to the consensus point and also move towards it faster.

In Figs. 1 and 2, the curves show the payoff functions at each round. The actors propose the solution, that is the point on the curve, where the payoff is maximized.

5

Simulation and Results

In Sect. 4, we proposed game theory-based solutions for consensus reaching in a decision making for the split of a finite resource. However, real-life problems are more complex than just distributing a finite amount of resource. In FEW fields, the actors, including administrative people and farmers, need to decide the amount of water used as well as types of fertilizer and crop for a specific land. Although the water is a finite and continuous resource that they need to split, they also need to decide the other parameters. We precomputed solutions for actors to propose during decision making. One sample solutions is given in Table 1. A solution includes values for all parameters for all actors. In our simulations, we have 5 actors, and the parameters are ground water, surface water, crop choice, and fertilizer choice.

132

S. Uslu et al. Table 1. Sample solution for 5 actors with 4 parameters Parameters

Actor 1

Actor 2

Actor 3

Actor 4

Actor 5

Ground water

11.4798125 35.5564315 12.8204529 21.5142238 10.1051691

Surface water

42.8265841 12.7084777 13.3468875 16.9873113 4.7743236

Crop choice

2

3

2

2

1

Fertilizer choice 2

1

3

1

1

We generated a solution set with 500 solutions that an actor can choose from. Then, we normalize the profits to [0–1] scale. Each actor sorts the solutions mostly based on their profit but using a function where the profits of other actors are also significant, as shown in Table 2, and starts proposing solutions from the first one in the sorted set. Table 2. Normalized profits of solutions sorted for actor 1 P roA1 P roA2 P roA3 P roA4 P roA5

Wavg Ravg

1.000

0.110

0.249

0.592

0.019

0.697 0.243

0.736

0.501

0.426

0.416

0.701

0.646 0.511

0.754

0.451

0.010

0.942

0.516

0.644 0.480

0.782

0.681

0.022

0.694

0.044

0.613 0.360

We investigated the effect of the weight of the trust in the payoff function. We generated different combinations of actors with different trust weights. In the first group of decision makings, the trust weights of actors are the same if they are in the same decision making. In the first decision making, they all weight one on the trust, which means that they want to maximize their trust first. In the other decision makings, we decrease the trust weight of all actors to 0.3, 0.1, and 0.01 to see the influence of the trust weight to the number of rounds to reach a consensus. Although we do not define the consensus point for the simulations, we provide the metric for how close the solutions are to each other. We define the average distance of solutions to the center of mass of solutions as the metric for solutions closeness where the center of mass is defined as the weighted average of the solutions where the weights are the trust of each proposer. We run the test for 20 rounds and present the results for each group. As shown in Fig. 3, when actors have a high weight on the trust that they receive from the community, the solutions proposed by the actors converge more quickly. After they propose their first solution, they quickly realize that they can increase their payoff by increasing their trust. They propose the next solution, which gives them less resource but more to other actors, from their sorted list of solutions for the next round. Sometimes, they even skip the next one or several

Trust-Based Game-Theoretical Decision Making for Food-Energy-Water Management

133

solutions and propose a solution that would increase their trust much higher than the next solution. As the weight of trust decreases to 0.3, it starts taking more rounds to converge and reach some degree of consensus with a less average distance to the center of mass of solutions. When we decrease it to 0.1 or 0.01, the actors do not propose a new solution in the next 20 rounds, which jams the decision making.

Fig. 3. When the actors have high trust weight in their payoff functions, they tend to propose a new and fairer solution and receive higher ratings which results in higher trust. Also, it leads to a faster convergence regarding the average distance of solutions to the center of mass of the all proposed solutions.

Fig. 4. When the actors have different trust weight in the same decision making, if the majority of the people has high trust weight, they can still reach an agreement level comparable to the best scenario where they all have high trust weight in a comparable number of rounds.

In addition to the first scenario, where actors in the same decision making have the same weight of trust on their payoff functions, we also investigate the

134

S. Uslu et al.

effect of having a distribution of high and low weights on the trust parameter of the payoff functions as shown in Fig. 4. Starting with the case where all actors weight one on trust, we replace one actor with someone having 0.1 weight on trust. When the number of actors with low weight on trust is 1 and 2, the average distance to the center of mass of the solutions increases, which means that the result is less acceptable. Also, it starts to require more rounds to reach that agreement level. However, both parameter, consensus degree, and number of rounds is still close to the best case where the actors have all high weight on trust. When the actors with low trust weight become the majority, the difference becomes more evident that they can be distinguished from the groups with actors where high trust weight is the majority.

6

Conclusion

In this paper, we proposed a game theory-based approach for a finite resource sharing problem considering the trust between the actors in the decision making. Actors can propose a solution maximizing their payoff regarding their share and trust by predicting the ratings they receive from other actors. When their estimations do not match the real ratings, they update their predictions and propose a new solution in the next round until a consensus is reached. When the problem becomes more complicated than a finite resource sharing, actors can utilize a precomputed solution set to propose solutions at each round. We prepared two different scenarios. In the first one, actors have the same trust weight in their payoff functions, and there are four decision makings where actors have 1, 0.3, 0.1, and 0.01 weights on trust. In the other scenario, the actors have either high or low trust weight. We start with the group where all actors have high weight and replace a high trust weight actor with a low trust weight actor at each trial until we have all actors having low weight on trust. Results have shown that when the actors have a high weight on trust, they tend to propose a solution that is closer to the consensus point. Also, they move towards the consensus point faster compared to actors with low trust weight. Simulations also gave us similar results. Groups of actors with high trust weight can reach a better agreement level with a smaller distance of proposed solutions. Also, they can reach a better level of agreement in a fewer number of rounds. Also, when the actors with low trust weight are the minority, the results are more comparable with the best case. However, when they become the majority, the distinction becomes more evident. Acknowledgments. This work was partially supported by the National Science Foundation under Grant No. 1547411 and by the U.S. Department of Agriculture (USDA) National Institute of Food and Agriculture (NIFA) (Award Number 2017- 67003-26057) via an interagency partnership between USDA-NIFA and the National Science Foundation (NSF) on the research program Innovations at the Nexus of Food, Energy and Water Systems.

Trust-Based Game-Theoretical Decision Making for Food-Energy-Water Management

135

References 1. Alfantoukh, L., Ruan, Y., Durresi, A.: Trust-based multi-stakeholder decision making in water allocation system. In: International Conference on Broadband and Wireless Computing, Communication and Applications, pp. 314–327. Springer (2017) 2. Alfantoukh, L., Ruan, Y., Durresi, A.: Multi-stakeholder consensus decisionmaking framework based on trust: a generic framework. In: 2018 IEEE 4th International Conference on Collaboration and Internet Computing (CIC), pp. 472–479. IEEE (2018) 3. Aumann, R.J.: Backward induction and common knowledge of rationality. Games Econ. Behav. 8(1), 6–19 (1995) 4. Babbar-Sebens, M., Minsker, B.S.: Interactive genetic algorithm with mixed initiative interaction for multi-criteria ground water monitoring design. Appl. Soft Comput. 12(1), 182–195 (2012) 5. Dong, Y., Xu, J.: Consensus Building in Group Decision Making. Springer, Heidelberg (2016) 6. Hamilton, S.H., ElSawah, S., Guillaume, J.H., Jakeman, A.J., Pierce, S.A.: Integrated assessment and modelling: overview and synthesis of salient dimensions. Environ. Model Softw. 64, 215–229 (2015) 7. Josang, A.: Trust-based decision making for electronic transactions. In: Proceedings of the Fourth Nordic Workshop on Secure Computer Systems (NORDSEC 1999), pp. 496–502 (1999) 8. Jøsang, A., Aˇzderska, T., Marsh, S.: Trust transitivity and conditional belief reasoning. In: IFIP International Conference on Trust Management, pp. 68–83. Springer (2012) 9. Kambiz, M.: Multi-Stakeholder Decision Making for Complex Problems: A Systems Thinking Approach with Cases. World Scientific (2016) 10. Kaur, D., Uslu, S., Durresi, A.: Trust-based security mechanism for detecting clusters of fake users in social networks. In: Workshops of the International Conference on Advanced Information Networking and Applications, pp. 641–650. Springer (2019) 11. Kim, D.J., Ferrin, D.L., Rao, H.R.: A trust-based consumer decision-making model in electronic commerce: the role of trust, perceived risk, and their antecedents. Decis. Support Syst. 44(2), 544–564 (2008) 12. Kim, Y.A., Song, H.S.: Strategies for predicting local trust based on trust propagation in social networks. Knowl.-Based Syst. 24(8), 1360–1371 (2011) 13. Morecroft, J.D.: System dynamics: portraying bounded rationality. Omega 11(2), 131–142 (1983) 14. Morecroft, J.D.: A systems perspective on material requirements planning. Decis. Sci. 14(1), 1–18 (1983) 15. Morecroft, J.D., Larsen, E.R., Lomi, A., Ginsberg, A.: The dynamics of resource sharing: a metaphorical model. Syst. Dyn. Rev. 11(4), 289–309 (1995) 16. Ruan, Y., Durresi, A.: A survey of trust management systems for online social communities-trust modeling, trust inference and attacks. Knowl.-Based Syst. 106, 150–163 (2016) 17. Ruan, Y., Durresi, A.: A trust management framework for cloud computing platforms. In: 2017 IEEE 31st International Conference on Advanced Information Networking and Applications (AINA), pp. 1146–1153. IEEE (2017)

136

S. Uslu et al.

18. Ruan, Y., Durresi, A., Alfantoukh, L.: Trust management framework for internet of things. In: 2016 IEEE 30th International Conference on Advanced Information Networking and Applications (AINA), pp. 1013–1019. IEEE (2016) 19. Ruan, Y., Durresi, A., Alfantoukh, L.: Using Twitter trust network for stock market analysis. Knowl.-Based Syst. 145, 207–218 (2018) 20. Ruan, Y., Durresi, A., Uslu, S.: Trust assessment for internet of things in multiaccess edge computing. In: 2018 IEEE 32nd International Conference on Advanced Information Networking and Applications (AINA), pp. 1155–1161. IEEE (2018) 21. Ruan, Y., Zhang, P., Alfantoukh, L., Durresi, A.: Measurement theory-based trust management framework for online social communities. ACM Trans. Internet Technol. (TOIT) 17(2), 16 (2017) 22. Shambour, Q., Lu, J.: A hybrid trust-enhanced collaborative filtering recommendation approach for personalized government-to-business e-services. Int. J. Intell. Syst. 26(9), 814–843 (2011) 23. Shambour, Q., Lu, J.: A trust-semantic fusion-based recommendation approach for e-business applications. Decis. Support Syst. 54(1), 768–780 (2012) 24. Sutcliffe, A.G., Wang, D., Dunbar, R.I.: Modelling the role of trust in social relationships. ACM Trans. Internet Technol. (TOIT) 15(4), 16 (2015) 25. Uslu, S., Kaur, D., Rivera, S.J., Durresi, A., Babbar-Sebens, M.: Decision support system using trust planning among food-energy-water actors. In: International Conference on Advanced Information Networking and Applications, pp. 1169–1180. Springer (2019) 26. Uslu, S., Ruan, Y., Durresi, A.: Trust-based decision support system for planning among food-energy-water actors. In: Conference on Complex, Intelligent, and Software Intensive Systems, pp. 440–451. Springer (2018) ¨ 27. Zermelo, E.: Uber eine anwendung der mengenlehre auf die theorie des schachspiels. In: Proceedings of the Fifth International Congress of Mathematicians, vol. 2, pp. 501–504. Cambridge University Press, Cambridge (1913) 28. Zhang, P., Durresi, A., Barolli, L.: Survey of trust management on various networks. In: 2011 International Conference on Complex, Intelligent and Software Intensive Systems (CISIS), pp. 219–226. IEEE (2011) 29. Zhang, P., Durresi, A., Ruan, Y., Durresi, M.: Trust based security mechanisms for social networks. In: 2012 Seventh International Conference on Broadband, Wireless Computing, Communication and Applications (BWCCA), pp. 264–270. IEEE (2012)

Energy-Efficient Purpose Ordering Scheduler Tomoya Enokido1(B) and Makoto Takizawa2 1 2

Faculty of Business Administration, Rissho University, Tokyo, Japan [email protected] Department of Advanced Sciences, Faculty of Science and Engineering, Hosei University, Tokyo, Japan [email protected]

Abstract. Distributed applications are composed of multiple objects. An object is an unit of computation resource. Conflicting transactions have to be serialized to keep objects mutually consistent. In this paper, the energy-efficient purpose ordering (EEPO) scheduler is proposed to not only serialize multiple conflicting transactions in the significant order of purposes assigned to the transactions but also reduce the total electric energy consumption of servers by omitting meaningless methods. Keywords: Energy-aware information systems · Transactions · Purposes · The EEPO scheduler · The EERO scheduler · The RO scheduler

1

Introduction

A subject doing a job function plays a role [1,2] in an enterprise. In the rolebased access control (RBAC) model [1–3], a role is a set of access rights. An access right is given in a pair o, op of an abject o [4] and a method op. A subject granted a role including an access right o, op can manipulate the object o through the method op by issuing a transaction. A transaction [5,6] is an atomic sequence of methods issued by a subject to manipulate objects. Conflicting methods [6] issued by multiple transactions have to be serialized on an object to keep the object mutually consistent. There are various ways to serialize multiple conflicting methods like timestamp ordering (TO) [5] and FIFO [5,6]. In our previous studies, the role ordering (RO) scheduler [3] is proposed to serialize multiple conflicting transactions in the significant order of roles granted to subjects and authorization relation [1,2,7] of roles. The RO scheduler does not consider to reduce the total electric energy consumption of servers to perform methods on objects. The energy-efficient role ordering (EERO) scheduler [8] is proposed to not only serialize multiple conflicting transactions in the significant order of roles granted to subjects but also reduce the total electric energy consumption of servers by omitting meaningless methods. In the RO and EERO schedulers, a subject granted more significant roles is more significant than other c Springer Nature Switzerland AG 2020  L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 137–149, 2020. https://doi.org/10.1007/978-3-030-33506-9_13

138

T. Enokido and M. Takizawa

subjects granted less significant roles. Then, a method issued by a more significant subject is performed prior to other methods issued by less significant subjects. However, this assumption is not true in some types of applications. For example, suppose a president would like to access to a bank account just for checking but a manager would like to access to the bank account for transferring money to make payment. The purpose payment is more significant than the purpose check in a purpose point of view. Hence, a method issued by the manager should be performed prior to a method issued by the president. In this paper, a subject assigns a transaction with a purpose which is a subset of roles granted to the subject. We first define the purpose-oriented dominant relation among subjects. Then, the energy-efficient purpose ordering (EEPO) scheduler is proposed to not only serialize multiple conflicting transactions in the significant order of purposes but also reduce the total electric energy consumption of servers by omitting meaningless methods. We evaluate the EEPO scheduler in terms of the total electric energy consumption of servers and the execution time of each transaction compared with the RO and EERO schedulers. In Sect. 2, we discuss the significancy of transactions, meaningless methods, and power consumption model of a server. In Sect. 3, we propose the EEPO scheduler. In Sect. 4, we evaluate the EEPO scheduler compared with the RO and EERO schedulers.

2 2.1

System Model Object-Based Systems with RBAC Model

A server cluster S is composed of multiple servers s1 , ..., sn (n ≥ 1) and multiple clients cl1 , ..., cll (l ≥ 1) interconnected in reliable networks. Let O be a set of objects o1 , ..., om (m ≥ 1) [4]. Each object oh is a unit of computation resource like a file and is an encapsulation of data and methods to manipulate the data. Objects are distributed on multiple servers. A pair of methods op1 and op2 conf lict if and only if (iff) the result obtained by performing the methods depends on the computation order. Otherwise, op1 and op2 are compatible. A transaction is an atomic sequence of methods [5]. A transaction Ti is initiated in a client cls and issues methods to servers to manipulate objects. In this paper, we assume each transaction Ti serially issues methods. Each transaction Ti initiated in a client cls is given an identifier tid(Ti ) = V (Ti ), id(cls ) where V (Ti ) is a logical time of the client cls when Ti is initiated and id(cls ) is an identifier of the client cls . For every pair of transaction identifiers tid(Ti ) (= V (Ti ), id(cl1 )) and tid(Tj ) (= V (Tj ), id(cl2 )), tid(Ti ) < tid(Tj ) iff 1) V (Ti ) < V (Tj ) or 2) id(cl1 ) < id(cl2 ) and V (Ti )  V (Tj ). A role R is a collection of access rights in the role-based access control (RBAC) model [1,2]. An access right is specified in a pair o, op of an object o and a method op. If a subject Sub is granted a role R including o, op, the subject Sub is allowed to invoke a method op on an object o. Let Srole be a family {R1 , ..., Rq } of roles granted to a subject Sub. Let Subi denote a subject which initiates a transaction Ti .

Energy-Efficient Purpose Ordering (PO) Scheduler

2.2

139

Significancy of Methods

Class methods are ones for creating and dropping an object. Object methods are ones for manipulating data in an object. Object methods are furthermore classified into change and output types. In an output type method, data is derived from an object. Change type methods are furthermore classified into full and partial types. In a full type method, whole data in an object is fully changed. In a partial type method, a part of data in an object is changed. A method op1 semantically dominates op2 on an object o (op1   op2 ) iff an application considers op1 to be more significant than op2 . op1 is semantically equivalent with op2 (op1 ∼ = op2 ) if op1  op2 and op2  op1 . op1 is more semantically signif icant than op2 (op1  op2 ) if op1  op2 and op1 ∼ = op2 . op1 and op2 are semantically uncomparable (op1  op2 ) iff neither op1  op2 nor op2  op1 . Definition. A method op1 is more signif icant than another method op2 on an object o (op1  op2 ) iff (1) op1 is a class type and op2 is an object type, (2) op1 is a change type and op2 is an output one, (3) op1 is a f ull change type and op2 is an partial one, or (4) op1 and op2 are a same object type and op1  op2 . A method op1 is signif icantly equivalent with op2 (op1 ≡ op2 ) iff op1 and op2 are a same type and op1 ∼ = op2 . op1 signif icantly dominates op2 (op1  op2 ) iff op1  op2 or op1 ≡ op2 . op1 and op2 are signif icantly uncomparable (op1  op2 ) iff neither op1  op2 nor op2  op1 . Suppose a file object F supports six methods create, drop, modify, insert, delete, and read as shown in Fig. 1. modify  insert since modify is a full change type method and insert is a partial change type method. class type

change type

create

insert

output type

modify drop

full type

read delete partial type

Fig. 1. Significancy of methods.

2.3

Significancy of Roles

In object-based systems, subjects and objects are referred to as entities. Each entity ei is given one security class sc(ei ) [9]. A security class sc1 can f low into sc2 (sc1 → sc2 ) iff the information in an entity e1 of a security class sc1 can flow into another entity e2 of a security class sc2 . sc1 and sc2 are equivalent (sc1 ≡ sc2 ) iff sc1 → sc2 and sc2 → sc1 . sc1 precedes sc2 (sc1 ≺ sc2 ) iff sc1 → sc2 but sc2 → sc1 . sc2 dominates sc1 (sc1  sc2 ) iff sc1 ≺ sc2 or sc1 ≡ sc2 .

140

T. Enokido and M. Takizawa

Definition. An object o1 is more signif icant than o2 (o1  o2 ) iff sc(o1 )  sc(o2 ). A pair of objects o1 and o2 are signif icantly equivalent (o1 ≡ o2 ) iff sc(o1 ) ≡ sc(o2 ). o1 signif icantly dominates o2 (o1  o2 ) iff o1  o2 or o1 ≡ o2 . o1 and o2 are signif icantly uncomparable (o1  o2 ) iff neither sc(o1 )  sc(o2 ) nor sc(o2 )  sc(o1 ). Definition. Let α1 and α2 be a pair of access rights o1 , op1  and o2 , op2 . An access right α1 is more signif icant than α2 (α1  α2 ) iff (1) o1  o2 , (2) op1  op2 and o1 ≡ o2 , or (3) α1  α3 and α3  α2 for some access right α3 . A pair of access rights α1 and α2 are signif icantly equivalent (α1 ≡ α2 ) iff (1) op1 ≡ op2 and o1 = o2 , or (2) o1 ≡ o2 and o1 = o2 . α1 signif icantly dominates α2 (α1  α2 ) iff α1  α2 or α1 ≡ α2 . α1 and α2 are signif icantly uncomparable (α1  α2 ) iff neither α1  α2 nor α2  α1 . Let A be a set of access rights. An access right β is maximally reachable from another access right α (β  α) iff β  α and there is no access right γ such that γ  β in A. Definition. A role R1 signif icantly dominates R2 (R1  R2 ) iff (1) for some access right α in R2 , there is an access right β ∈ R1 - R2 such that β  α in R1 ∪ R2 and (2) for every access right β ∈ R1 , there is no access right α ∈ R2 such that α  β in R1 ∪ R2 . A role R1 is signif icantly equivalent with R2 (R1 ≡ R2 ) iff R1  R2 and R2  R1 . R1 and R2 are signif icantly uncomparable (R1  R2 ) iff neither R1  R2 nor R2  R1 . A least upper bound R1  R2 is a role R3 such that R3  R1 and R3  R2 and there is no role R4 such that R3  R4  R1 and R3  R4  R2 . A greatest lower bound R1  R2 is similarly defined. Here, R1  · · ·  Rm  Ri  R1  · · ·  Rm holds but R1 ∩ · · · ∩ Rm  Ri  R1 ∪ · · · ∪ Rm may not hold. Definition. Let R1 and R2 be families of roles. R1 signif icantly dominates R2 (R1  R2 ) iff R∈R1 R  R∈R2 R. R1 and R2 are signif icantly equivalent (R1 ≡ R2 ) iff R1  R2 and R2  R1 . R1 and R2 are significantly uncomparable (R1  R2 ) iff neither R1  R2 nor R2  R1 . 2.4

Significancy of Transactions

We first define the dominant relation of subjects with respect to the significancy of roles and authorized relation: Definition. A subject Subi precedes Subj on a role R (Subi ⇒R Subj ) iff Subi grants R to Subj or Subi ⇒R Subk ⇒R Subj for some subject Subk . A pair of subjects Subi and Subj are equivalent on R (Subi ≡R Subj ) iff Subi ⇒R Subj and Subj ⇒R Subi . A pair of subjects Subi and Subj are independent with respect to R (Subi R Subj ) iff neither Subi ⇒R Subj nor Subj ⇒R Subi .

Energy-Efficient Purpose Ordering (PO) Scheduler

141

Subject-oriented (SO) Dominant Relation. A subject Subi subjectoriented (SO) dominates Subj (Subi SO Subj ) iff (1) Srolei  Srolej , (2) Subi ⇒R Subj for some role R ∈ Sroleij and Subj ⇒R Subi for every R ∈ Sroleij if Srolei  Srolej , or (3) Subi SO Subk SO Subj for some subject Subk . Suppose each subject Subi issues a transaction Tt with purpose P rolet (⊆ Srolei ). We define the dominant relation of subjects with respect to the significancy of purposes of transactions. Purpose-oriented (PO) Dominant Relation. For a pair of transactions Tt and Tu issued by subjects Subi and Subj , respectively, the subject Subi purposeO oriented (P O) dominates Subj (Subi P tu Subj ) with respect to the purposes of transactions Tt and Tu iff (1) P rolet  P roleu or (2) Subi ⇒R Subj for some role R ∈ P roletu and Subj ⇒R Subi for every R ∈ P roletu if P rolet  P roleu . The SO-dominant relation SO is transitive. However, the PO-dominant relaO PO tion P tu is not transitive since the PO-dominant relation tu is only defined for a pair of transactions Tt and Tu . We define the SO- and PO-dominant relations of transactions based on the SO- and PO-dominant relations of subjects issuing the transactions, respectively. Definition. For a pair of conflicting transactions Ti and Tj , – Ti SO-dominates Tj (Ti SO Tj ) iff Subi SO Subj . O – Ti P O-dominates Tj (Ti P O Tj ) iff Subi P tu Subj . Let D show a dominant relation of transactions for a dominant type D ∈ {SO, PO}. A pair of transactions Ti and Tj are equivalent (Ti ≡D Tj ) iff Ti D Tj and Tj D Ti . A pair of transactions Ti and Tj are independent (Ti D Tj ) iff neither Ti D Tj nor Tj D Ti . 2.5

Meaningless Methods

Let T be a set {T1 , ..., Tk } (k ≥ 1) of transactions. Let SH be a schedule of the transactions in a set T where every transaction in the schedule SH is serially performed in the following serial precedent relation: Definition. A transaction Ti serially precedes Tj in a schedule SH (Ti →SH Tj ) iff (1) Ti D Tj , or (2) tid(Ti ) < tid(Tj ) if Ti D Tj or Ti ≡D Tj . A schedule SH is a totally ordered set T, →SH . A schedule SH is serializable iff the serial precedent relation →SH is acyclic. A schedule SH = T, →SH  is legal iff T1 →SH T2 if T1 D T2 , or tid(T1 ) < tid(T2 ) if T1 D T2 or T1 ≡D T2 for every pair of T1 and T2 in T. In order to make a schedule legal, methods from transactions are required to be buffered until all the transactions are initiated. Definition. A schedule SH = T, →SH  is RS-partitioned into the subschedules SHf = Tf , →SHf  (f = 1, ..., d): 1. Tf ∩ Tg = φ for every pair of subschedules Hf and Hg and T1 ∪ · · · ∪ Td = T.

142

T. Enokido and M. Takizawa

2. T1 →SHf T2 if T1 D T2 , or tid(T1 ) < tid(T2 ) if T1 D T2 or T1 ≡D T2 for every pair of transactions T1 and T2 in each SHf . 3. T1 →SH T2 if T1 →SHf T2 for every pair of transactions T1 and T2 in each SHf . 4. For every pair of subschedules SHf and SHg , if Tf 1 →SH Tg1 for some pair of transactions Tf 1 in SHf and Tg1 in SHg , there is no pair of transactions Tf 2 in SHf and Tg2 in SHg such that Tg2 →SH Tf 2 . Definition. A schedule SH of T is RS-serializable with respect to subschedules SH1 , ..., SHd iff SH is RS-partitioned into the subschedules SH1 , ..., SHd . It is straightforward for the following theorem to hold. Theorem. A history SH is serializable if SH is RS-serializable with respect to some RS-partition SH1 , ..., SHd of SH. Suppose a schedule SH is RS-partition into the subschedules SH1 , ..., SHd . Definition. A method op1 serially precedes op2 in a subschedule SHf (op1 →SHf op2 ) iff (1) the methods op1 and op2 are issued by a same transaction Ti and op1 is issued before op2 , (2) the methods op1 and op2 are issued by a pair of transactions Ti and Tj , respectively, and Ti →SHf Tj , or (3) op1 →SHf op3 →SHf op2 for some method op3 . Let SHfoh be a local subschedule of methods which are performed on an object oh in a subschedule SHf . Definition. A method op1 serially precedes another method op2 in a local subh op2 ) iff op1 →SHf op2 . schedule SHfoh (op1 →oSH f Suppose an object oh supports six methods as shown in Fig. 1 and a method insert serially precedes another method modify in a local subschedule SHfoh h (insert →oSH modify) on the object oh . Here, the insert method is not required f to be performed on the object oh if the modify method is surely performed on the object oh , i.e. the modify method can absorb the insert method. Definition – A full change method op1 absorbs another partial change method op2 in a local h op1 , and there is no class or subschedule SHfoh on an object oh if op2 →oSH f oh   h output method op such that op2 →SHf op →oSH op1 , or op1 absorbs op f   and op absorbs op2 for some method op . – An output method op1 absorbs another output method op2 in a local subh op1 , and there is no class or schedule SHfoh on an object oh if op2 →oSH f o   h h change method op such that op2 →SHf op →oSH op1 , or op1 absorbs op f   and op absorbs op2 for some method op . – A class method op1 for dropping an object oh absorbs another change method h op1 , and there op2 in a local subschedule SHfoh on an object oh if op2 →oSH f oh   h is no class or output method op such that op2 →SHf op →oSH op1 , or op1 f    absorbs op and op absorbs op2 for some method op .

Energy-Efficient Purpose Ordering (PO) Scheduler

143

A method op is not required to be performed on an object oh if the method op is absorbed by another method op in a local subschedule SHfoh . Definition. A method op is meaningless iff the method op is absorbed by another method op in the local subschedule shofh of an object oh . 2.6

Power Consumption Model of a Server

In class methods and change type methods, data is written into an object. On the other hand, in output type methods, data is read from an object. In this paper, methods are classified into read (r) and write (w) types of methods. Methods which are being performed and already terminate are current and previous at time τ , respectively. Let RPt (τ ) and W Pt (τ ) be sets of current r and w methods on a server st at time τ , respectively. Here, Pt (τ ) = RPt (τ ) ∪ W Pt (τ ). Let rti (oh ) and wti (oh ) be methods issued by a transaction Ti to read and write data in an object oh on a server st , respectively. By each method rti (oh ) in a set RPt (τ ), data is read in an object oh at rate RRti (τ ) [B/sec] at time τ . By each method wti (oh ) in a set W Pt (τ ), data is written in an object oh at rate W Rti (τ ) [B/sec] at time τ . Let maxRRt and maxW Rt be the maximum read and write rates [B/sec] of r and w methods on a server st , respectively. The read rate RRti (τ ) (≤ maxRRt ) and write rate W Rti (τ ) (≤ maxW Rt ) are given as f rt (τ ) · maxRRt and f wt (τ ) · maxW Rt , respectively. Here, f rt (τ ) and f wt (τ ) are degradation ratios for read and write methods, respectively. 0 ≤ f rt (τ ) ≤ 1 and 0 ≤ f wt (τ ) ≤ 1. The degradation ratios f rt (τ ) and f wt (τ ) 1 1 and wrt ·|RPt (τ )|+|W are given as |RPt (τ )|+rw Pt (τ )| , respectively. Here, 0 t ·|W Pt (τ )| ≤ rwt ≤ 1 and 0 ≤ wrt ≤ 1. The read laxity lrti (τ ) [B] and write laxity lwti (τ ) [B] of methods rti (oh ) and wti (oh ) show how much amount of data are read and written in an object oh by the methods rti (oh ) and wti (oh ) at time τ , respectively. Suppose that a pair of methods rti (oh ) and wti (oh ) start on a server st at time stti , respectively. At time stti , the read laxity lrti (τ ) is rbh [B] where rbh is the size of data in an object oh . The write laxity lwti (τ ) is wbh [B] where wbh is the size of data to be written in an object oh . Here, lrti (τ ) = rbh - Σττ=stti RRti (τ ) and lwti (τ ) = wbh - Σττ=stti W Rti (τ ). Let Et (τ ) be the electric power consumption [W] of a server st at time τ . maxEt and minEt show the maximum and minimum electric power consumption [W] of the server st , respectively. The power consumption model for a storage server (P CS model) [10] is proposed. According to the PCS model, the electric power Et (τ ) [W] of a server st to perform multiple r and w methods at time τ is given as follows: ⎧ W Et if |W Pt (τ )| ≥ 1 and |RPt (τ )| = 0. ⎪ ⎪ ⎪ ⎨ W REt (α) if |W Pt (τ )| ≥ 1 and |RPt (τ )| ≥ 1. (1) Et (τ ) = ⎪ if |W Pt (τ )| = 0 and |RPt (τ )| ≥ 1. REt ⎪ ⎪ ⎩ if |W Pt (τ )| = |RPt (τ )| = 0. minEt

144

T. Enokido and M. Takizawa

The server st consumes the electric power REt [W] if |W Pt (τ )| = 0 and |RPt (τ )| ≥ 1. The server st consumes the electric power W Et [W] if |W Pt (τ )| ≥ 1 and |RPt (τ )| = 0. The server st consumes the electric power W REt (α) [W] = α · REt + (1 - α) · W Et [W] where α = |RPt (τ )| / (|RPt (τ )| + |W Pt (τ )|) if |W Pt (τ )| ≥ 1 and |RPt (τ )| ≥ 1. Otherwise, a server st consumes the minimum electric power minEt . Here, minEt ≤ REt ≤ W REt (α) ≤ W Et ≤ maxEt . The processing power consumption P Et (τ ) [W] of a server st at time τ is Et (τ ) - minEt . The total processing energy consumption T P Et (τ1 , τ2 ) of a server st from time τ1 to τ2 is given as T P Et (τ1 , τ2 ) = Σττ2=τ 1 P Et (τ ).

3

Energy-Efficient Purpose Ordering (EEPO) Scheduler

We discuss energy efficient purpose ordering (EEP O) scheduler to not only make transactions RS-serializable with PO-dominant relation but also reduce the total energy consumption of a server cluster S. A transaction Ti first sends a begin request bi to every target object. Then, the transaction Ti issues methods and lastly issues either a commit (cmi ) or abort (abi ) request to the objects. Each client cls manipulates a variable cfs where initially cfs = 1. Each client cls periodically sends a f ence message k to make an RS-partition, which carries k.f (= cfs ). Each time a client cls sends a fence message k, cfs = cfs + 1 in the client cls . Each object oh has a variable fh where initially fh = 1. Each time an object oh receives a fence message k where k.f = fh from every client, fh = fh + 1 in the object oh . Transactions whose begin requests are received before a fence message k compose an RS-partition and are sorted in the serial precedence relation →SHf . There are a set RQh of local receipt queues RQh1 , ..., RQhl , a global receipt queue GRQh , and an auxiliary global receipt queue AGRQh for each object oh . On receipt of a method opi from a transaction Ti initiated on a client cls , the method opi is enqueued into a local receipt queue RQhs for the client cls (s = 1, ..., l) on an object oh . Begin requests and fence messages are moved to AGRQh to make an RS-partition. Transactions in an RS-partition are serialized in the serial precedence relation →SHf . Methods are moved to GRQh and are performed in the serial precedence relation →SHf . The following conditions have to be satisfied to realize the RS-serializability: Role-Based Serializability (RS) Conditions 1. Methods in every global receipt queue GRQh are sorted in the serial precedence relation →SHf . 2. For a method opi from a transaction Ti , if opi precedes a method opj conflicting with opi from another transaction Tj in some GRQh , opi from Ti precedes a method opj conflicting with opi from Tj in every GRQh . The EEPO scheduler for an object oh handles methods to realize the RSserializability by the RS procedure as shown in Algorithm 1. Suppose an RS-partition SHf is composed of begin requests preceding a fence k where k.f is the minimum in AGRQh of an object oh . Each begin request bi

Energy-Efficient Purpose Ordering (PO) Scheduler

145

Algorithm 1. RS procedure Input: GRQh , AGRQh , {RQh1 , ..., RQhl }. Output: RS-partitioned GRQh . /* The following procedures are used to manipulate a queue Q for a method op. */ – top(Q): a top element in Q. – enqueue(op, Q): op is enqueued into Q. – tail(Q): a tail element in Q. – dequeue(Q): a top element in Q is dequeued. – RSsort(op, Q, e1 , e2 ): op is inserted between elements e1 and e2 in Q, and requests between elements e1 and e2 in Q are sorted in the serial precedence relation →SHf . procedure RS(GRQh , AGRQh , {RQh1 , ..., RQhl }) if there is a fence k where k.f = fh in every RQhs then for every local receipt queue RQhs do while top(RQhs ) = k do opi ← dequeue(RQhs ); if opi = bi then enqueue(opi , AGRQh ); else /* opi is not a begin request bi . */ if bi is between the top of AGRQh and a fence k then RSsort(opi , GRQh , top(GRQh ), k ); else if bi is between a pair of fences k and k then RSsort(opi , GRQh , k , k ); else if bi is between a fence k and the tail of AGRQh ) then RSsort(opi , GRQh , k , tail(GRQh )); end if end if end while opi ← dequeue(RQhs ); opi is removed; /* fence k is removed. */ end for enqueue(k, AGRQh ); enqueue(k, GRQh ); fh = fh + 1; end if end procedure

of a transaction Ti holds a transaction identifier tid(Ti ) and list Li of methods issued by the transaction Ti . Here, begin requests in the RS-partition SHf can be totally ordered in the serial precedent relation →SHf of transactions. Hence, a local subschedule SHfoh of methods can be created on an object oh by sorting lists of methods held in begin requests according to the serial precedent relation →SHf . Methods in GRQh are performed on an object oh by the Delivery procedure as shown in Algorithm 2.

4

Evaluation

We evaluate the EEP O scheduler in terms of the total electric energy consumption [J] of a server cluster S and the average execution time [msec] of each transaction compared with the RO [3] and EERO [8] schedulers. We consider a homogeneous server cluster S composed of five servers s1 , ..., s5 . Every server st (t = 1, ..., 5) follows the same data access model and power consumption

146

T. Enokido and M. Takizawa

Algorithm 2. Delivery procedure Input: GRQh . Eh and TEh are sets of current methods and transactions on oh . Output: Performing methods on an object oh . /* Procedures to check methods and transactions being performed on oh . */ – Mcompatible(op, Eh ): true if Eh = φ or a method op does not conflict with every method in Eh , otherwise false. – Tcompatible(T (op), TEh ): true if TEh = φ or a transaction T (op) issuing a method op does not conflict with every transaction in TEh , otherwise false. – Meaningless(op): true if a method op is meaningless in the local subschedule o SHf h and there is a method op in a global receipt queue GRQh where the method  op absorbs the method op, otherwise false. procedure Delivery(GRQh ) op ← top(GRQh ); if op = fence then if Mcompatible(op, Eh ) and Tcompatible(T (op), TEh then op ← dequeue(GRQh ); Eh ← Eh ∪ {op}; if T (op) ∈ TEh then TEh = TEh ∪ {T (op)}; end if if Meaningless(op) then Eh = Eh - {op}; if op = cmi or op = abi then TEh = TEh - {T (op)}; end if else perform(op); end if end if else if Eh = φ and TEh = φ then every begin request bi preceding the fence op in AGRQh is removed; op is removed from GRQh and AGRQh ; /* the fence op is removed. */ end if end if end procedure

model as shown in Table 1. Parameters of each server st are given based on the experimentations [10]. There are five objects o1 , ..., o5 in a system. Each server st holds one object oh (t = h). The size of data in each object oh is randomly selected between 50 and 80 [MB]. Each object oh supports six types of methods as shown in Fig. 1. There are five subjects Sub1 , ..., Sub5 . There are three roles R1 , R2 , and R3 owned by Sub1 , where R1  R2  R3 . Here, Sub1 Ri Sub2 , Sub1 Ri Sub3 , Sub1 Ri Sub4 , Sub1 Ri Sub5 for every role Ri (i = 1, ..., 3). Sub2 R3 Sub4 and Sub3 R3 Sub5 . Srole1 = {R1 , R2 , R3 }, Srole2 = Srole3 = {R2 , R3 }, and Srole4 = Srole5 = {R3 }. Here, Srole1 SO Srole2 = Srole3 SO Srole4 = Srole5 . The subject Sub1 issues transactions with a purpose P role1 = {R3 } (⊆ Srole1 ). Other transactions are assigned with purposes as P role2 = P role3 = {R2 , R3 } and P role4 = P role5 = {R3 }. Here, P role2 = P role3 P O P role1 = P role4 = P role5 . Each subject Subi (i = 1, ..., 5) initiates a same number l (0 ≤ l ≤ 1,200) of transactions on each of five clients cl1 , ..., cl5 . The total number tn (= l · 5) (0 ≤ tn ≤ 6,000)

Energy-Efficient Purpose Ordering (PO) Scheduler

147

Table 1. Homogeneous cluster S. Server st maxRRt

rwt wrt minEt W Et

80 [MB/sec] 45 [MB/sec] 0.5

45 RO EERO EEPO

35

RO

30 25

EERO EEPO

20 15 10 5 0

REt

0.5 39 [W] 53 [W] 43 [W]

12

40

Average execution time [msec]

Average total electric energy consumption [KJ]

st

maxW Rt

0

1000

2000 3000 4000 Number tn of transactions

5000

6000

Fig. 2. Average total energy consumption [KJ] of a server cluster.

AET1 AET2 AET3 AET4 AET5

10 8

AET4 AET5

6

AET3 AET2

4

AET1

2 0

0

1000

2000 3000 4000 Number tn of transactions

5000

6000

Fig. 3. Average execution time AETi in the RO scheduler.

of transactions are issued to manipulate objects. We assume each transaction issues full change, partial change, and output methods. The total amount of data of an object oh are fully written and read by each full change and output methods, respectively. On the other hand, a half size of data of an object oh are written into the object oh by partial change methods. Each transaction issues three methods randomly selected from twenty methods on the five objects. The starting time of each transaction Ti is randomly selected in a unit of one second between 1 and 3600 [sec]. Figure 2 shows the average total electric energy consumption [KJ] of the server cluster S to perform the number tn of transactions in the RO, EERO, and EEPO schedulers. The average total electric energy consumption of the server cluster S in the EEPO algorithm is almost the same as the EERO scheduler. In the EERO and EEPO schedulers, meaningless methods which are not required to be performed on each object are omitted. As a results, the average total electric energy consumption of the server cluster S can be more reduced in the EERO and EEPO schedulers than the RO scheduler. Suppose a transaction Ti starts at time sti and commits at time eti . Here, the execution time ETi of the transaction Ti is eti - sti [msec]. Figures 3, 4, and 5 show the average execution time AETi of each transaction issued by the same subject Subi in the server cluster S to perform the total number tn of transactions in the RO, EERO, and EEPO schedulers, respectively. In the RO and EERO schedulers, transactions are ordered based on the SO-dominant relations. As a result, the average execution time AET1 of transactions issued by the subject Sub1 is the minimum in the RO and EERO schedulers since the subject Sub1 is more significant than the other subjects. Following Figs. 3 and 4, the more significant subject is, the shorter average execution time a transaction

148

T. Enokido and M. Takizawa 12 AET1 AET2 AET3 AET4 AET5

10 8

AET5 AET4

6

AET3 AET2

4 AET1

2 0

0

1000

2000 3000 4000 Number tn of transactions

5000

6000

Fig. 4. Average execution time AETi in the EERO scheduler.

Average execution time [msec]

Average execution time [msec]

12

AET1 AET2 AET3 AET4 AET5

10 8

AET1

AET4

AET5

6 4

AET2

AET3

2 0

0

1000

2000 3000 4000 Number tn of transactions

5000

6000

Fig. 5. Average execution time AETi in the EEPO scheduler.

issued by the subject implies in the RO and EERO schedulers. On the other hand, transactions are ordered based on the PO-dominant relations in the EEPO scheduler. As a result, the average execution time of AET1 of transactions issued by the subject Sub1 is the same as the average execution times AET4 and AET5 issued by subjects Sub4 and Sub5 , respectively, since P role1 = P role4 = P role5 . Following Fig. 5, the more significant transaction with respect to purpose is, the shorter average execution time a transaction implies in the EEPO scheduler. Following the evaluation, the more significant transactions with respect to purposes, the earlier performed in the EEPO scheduler. The average total electric energy consumption of a server cluster can be more reduced in the EEPO and EERO schedulers than the RO scheduler. The average total electric energy consumption of a server cluster in the EEPO scheduler is the same as the EERO scheduler.

5

Concluding Remarks

In this paper, we newly proposed the EEPO scheduler to not only serialize multiple conflicting transactions in the significant order of purposes assigned to transactions but also reduce the total electric energy consumption of a server cluster by omitting meaningless methods. We evaluated the EEPO scheduler compared with the RO and EERO schedulers. The evaluation results show the total electric energy consumption of a server cluster in the EEPO scheduler is the same as the EERO scheduler. The total electric energy consumption of a server cluster can be more reduced in the EEPO and EERO scheduler than the RO scheduler. In addition, the more significant transactions with respect to purposes, the earlier performed in the EEPO scheduler.

References 1. Ferraiolo, D.F., Kuhn, D.R., Chandramouli, R.: Role Based Access Control. Artech House, Norwood (2005)

Energy-Efficient Purpose Ordering (PO) Scheduler

149

2. Sandhu, R.S., Coyne, E.J., Feinstein, H.L., Youman, C.E.: Role-based access control models. IEEE Comput. 29(2), 38–47 (1996) 3. Enokido, T.: Role-based serializability using role ordering schedulers. J. Interconnect. Netw. 7(4), 437–450 (2006) 4. Object Management Group Inc.: Common object request broker architecture (CORBA) specification, version 3.3, part 1 – interfaces (2012). http://www.omg. org/spec/CORBA/3.3/Interfaces/PDF 5. Bernstein, P.A., Hadzilacos, V., Goodman, N.: Concurrency Control and Recovery in Database Systems. Addison-Wesley, Boston (1987) 6. Gray, J.N.: Notes on database operating systems. Lect. Notes Comput. Sci. 60, 393–481 (1978) 7. Sandhu, R.S.: Lattice-based access control models. IEEE Comput. 26(11), 9–19 (1993) 8. Enokido, T., Duolikun, D., Takizawa, M.: Energy-efficient role ordering scheduler. In: Proceedings of the 20th International Conference on Network-Based Information Systems (NBiS-2017), pp. 78–90 (2017) 9. Denning, D.E., Denning, P.J.: Cryptography and Data Security. Addison-Wesley Publishing Company, Boston (1982) 10. Sawada, A., Kataoka, H., Duolikun, D., Enokido, T., Takizawa, M.: Energy-aware clusters of servers for storage and computation applications. In: Proceedings of the 30th IEEE International Conference on Advanced Information Networking and Applications (AINA-2016), pp. 400–407 (2016)

NFC-Based Commissioning of Adaptive Sensing Applications for the 5G IIoT Hadil Abukwaik(B) , Christian Groß, and Markus Aleksy ABB Corporate Research Center, Ladenburg, Germany {hadil.abukwaik,christian.d.gross,markus.aleksy}@de.abb.com

Abstract. Improving commissioning mechanisms of field devices is a necessity under the massively increasing number of these devices today and in the future 5G-enabled industrial automation systems. Hence, we propose a simplified commissioning for an edge computing-based plug & use functionality using NFC tags with information models. This approach is expected to increase the efficiency in commissioning the field devices as it reduces the required manual user input. We also present two application scenarios for 5G-enabled field devices, which benefit from our proposed approach.

1

Introduction

The increasing need for customized products and flexible production lines participated in triggering the forth industrial revolution. However, the current systems usually rely on wire-line technologies to connect sensors and actuators, which is not the best fit for the desired flexibility. Thus, the 5G networks come into the picture as the enabling technology for interconnecting entities of production system in a flexible wireless manner, while still fulfilling the quality of service (QoS) requirements. That is, the 5G mobile communication standard offers new features and characteristics that make mobile communication for the first time suitable in an industrial automation context thereby forming the Industrial Internet of Things (IIoT). As 5G networks are considered for a wide range of IoT applications with different QoS requirements, they offer a key feature which is dividing the network into slices each having its own QoS characteristics and allocated its network traffic. Having such a feature in industrial automation would enable operating high-bandwidth video surveillance application in parallel with a control application requesting latencies in the sub millisecond scale. Another relevant key feature of 5G for industrial automation is the support for NarrowBand-IoT (feNB-IoT), which strives for a massive machine type communication (mMTC) [4]. feNB-IOT classify under Low Power Wide Area (LPWA) networks and is expected to allow increased communication coverage and optimized power consumption, thereby, enabling an increased battery life time of field devices with a reduced functional complexity. Furthermore, in comparison to previous 4G mobile technologies, feNB-IoT makes use of repetitions to reach also devices with difficult to reach physical locations. c Springer Nature Switzerland AG 2020  L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 150–161, 2020. https://doi.org/10.1007/978-3-030-33506-9_14

Simplified Commissioning for 5G IIoT Applications

151

To sum up, 5G is rendered a suitable communication technology for industrial automation supporting ultra-low latency QoS requirements, massive machine to machine communication, and deployment of private 5G cells called campus networks. However, the constant massive increase in the number of sensors and actuators with their individual QoS requirements increases the complexity. Their commissioning, deployment, and integration tasks become expensive and errorprone. This problem then directly contradicts the vision of Industry 4.0 of introducing flexible and re-arrangeable production lines in plug and produce manner. Thus, the need arises for a plug and use commissioning procedure to reduce the deployment, engineering, and maintenance effort of future manufacturing setups [4,15]. This paper introduces a novel commissioning approach for field devices in a 5G industrial setup that reduces the amount of manual user input by using Near Field Communication (NFC) tags as well as a central information model for deriving device configurations. The approach is highlighted on two industrial automation scenarios with specific requirements that drive solution design decisions.

2

Overview of Proposed Approach and Relevant Scenarios

In this section, we detail our solution for realizing the plug and use functionality in an 5G-enabled industrial automation context that utilizes NFC tags. Then, we shed the light on two scenarios from the industrial automation and their functional and non-functional requirements, which would benefit from our approach. The core idea is to attach NFC tags to each field and edge device, containing the information about (i) the device type and (ii) a unique serial number as well as (iii) the International Mobile Subscriber Identity (IMSI) of the Subscriber Identity Module (SIM) chip card that is used for the 5G communication. For each device type we assume that there exists a device type definition, which describes all the properties and functions a device has. The device type definition is stored in a central repository during the production of the device such that it can be access later during the commissioning procedure. The unique serial number allows for the unique identification of a device and its digital twin in the cloud. The IMSI is used for the unique identification of a field device in the network layer and to grant access to the 5G mobile network. During the design of our commissioning approach, we divided the commissioning of a given field device into the shown steps in Fig. 1. The connection setup establishes the communication link between field devices such that data can be exchanged between the two endpoints. Then, the device identification and registration of the field device happens such that each entity is known to the system and can be identified uniquely. Necessarily, the device authorization takes place to set the access control of the field device

152

H. Abukwaik et al.

Fig. 1. Steps of the field device comissioning approach

ensuring that only legitimate communication partners are involved in the communication. Optionally, the device localization determines the physical location of a field device in case that the location is relevant for the later configuration process. The device configuration ensures that its all parameters are set correctly. Finally, the device integration takes place to allow the device into the overall system and to put it into an operational state. Below, we detail the two scenarios that would advantage from our proposed approach. In both scenarios, we assume that all sensors and actuators are interconnected via a 5G campus network and that the IoT edge device is installed on-site with a local instance of a monitoring app and a closed-loop controller app. Massive Additive Sensing Scenario. Sensing importance is growing for variant operational goals including monitoring of process quality, timing, energy, and material use. This applies for autonomous plants where all sensors and actuators are connected to one or more private 5G base stations residing on a large industrial area. To collect the respective metrics and increase the process resolution, it is necessary to have an efficient way of deploying the rapidly increasing number of sensors in such plants. In this scenario, we assume that the communication entities are connected to a local IIoT edge device that oversees relaying communication between the sensors and the cloud. Furthermore, we assume that the sensor allows for an Internet Protocol (IP)-based communication either by being directly connected via 5G to the edge device or by communicating via another dedicated IP-capable gateway. Control Loop Scenario. In a low network latency control loop of an industrial application, a 5G base station connects sensors, actuators, and a local IIoT edge device that runs the controller software. In such a loop, the local controller sends a control command to the actuators, while the sensors measure the resulting state change of the system and return it to the controller. The controller then processes the feedback, plans the next command, and send it out to the actuators. The status of the control loop is constantly reported via an edge device to the cloud, where an operator monitors and if necessary adjusts the control loop process. Looking at both scenarios, we find the following common non-functional requirements: (1) scalability to fit the increasing the number of field devices, applications, and network traffic, (2) flexibility in order to easily add and remove devices, (3) efficiency of operations with respect to cost and time needed for adapting the system, (4) reliability such that changes to the system always

Simplified Commissioning for 5G IIoT Applications

153

result in a desired and stable system state, and (5) security to prevent unwanted changes of the system. On the other hand, both scenarios have functional requirements to be fulfilled by the our solution. From a device perspective the approach shall support device identification, device authentication, device connectivity, device localization, and device configuration. From a network perspective the approach shall support network slicing to support various different QoS profiles and device integration.

3

Demonstrator Architecture

In this section, we present our design decisions and two architectural views for the proposed simplified commissioning approach on the Additive Sensing Scenario that we described in the previous section. Our designed architecture is based on standard communication technologies and is implemented using Commercial-offthe-shelf (COTS) libraries and toolkits. We describe later in the implementation in Sect. 4. 3.1

Design Decisions and Patterns for IIoT Edge Computing

We took a number of design decisions on the cloud, device, and engineering levels. On the cloud level, we chose to create proprietary information models of our endpoint devices (e.g., the smart sensor and the edge) for the sake of higher preciseness and expressiveness. The models can be reused in other deployment projects with little modification. Furthermore, information models are provided manually during the engineering process to increase the efficiency and maintainability. Information models are deployed on the cloud with the goal in mind of increasing interoperability, improving remote monitoring of the system and supporting security. On the field/edge device level, we decided to make use of network scan techniques to automatically discover devices. Furthermore, the edge communication has been realized using a message-based protocol to achieve a better scalability and performance compared to transaction-oriented communication. On the engineering tool level, we chose to rely on a mobile application to increase the efficiency of complex commissioning processes. Moreover, it should improve user experience and usability. Further, the built-in Global Positioning System (GPS) of state-of-the-art smartphones allows for the precise localization of field devices. The mobile application communicates via Representational State Transfer (REST) application programming interfaces (APIs) using Hypertext Transfer Protocol Secure (HTTPS) to support interface-level interoperability and technology independence. This approach also provides flexibility regarding programming languages used for the implementation. A variety of design patterns can be applied when developing an edge computing solution for industrial automation. On one hand, on the application layer of the system, the functionality of the edge should make use of a modular design, where the functionality is composed of a set of independent modules.

154

H. Abukwaik et al.

The modules could be started and stopped depending on the required functionality to limit the resources footprint for realizing a certain functionality. Below we describe the design patterns [17] that we adopted into our proposed solution. The publisher/subscriber design pattern can be used to organize the communication among the modules. By using publish-subscribe-communication among the modules, we can decouple the communication among the modules relationwise (i.e., they do not need to know each other), time-wise (i.e., they do not need to run at the same time), and synchronization-wise (i.e., they do not need to be interrupted during their activities during publishing or receiving) easing the deployment and removal of modules during runtime. The singleton design pattern restricts the instantiation of a class to one single instance, i.e., one exact object, with a global pointer to it. It centralizes the coordination of actions among the system elements. For example, this is useful for managing the logging functionality by having one instance of the logger with global access for the entire system allowing a centralized log management. The proxy design pattern provides a placeholder for another object to control accessing it. It creates an extra level of indirection for control reasons and protects the object from undesired complexity. This pattern enables safe and secure communication by relaying the entire communication to the outside through a single point. Four IoT design patterns are also proposed by Qanbari et al. [14] to design, build, and engineer edge applications. These patterns include the edge provisioning pattern that offers a container-based solution that helps in ensuring that edge devices are started with a reliable baseline environment and can be reconfigured in an efficient way. This is facilitated through docker images utilizing a layered and versioned file system. Thus, devices pull needed layers only rather than the whole image. Also, devices can roll back to the image latest or any working version in case of failure. In addition, Kurschl and Beer [13] propose a cloud computing model that is mainly based on pipes and filters design pattern. This model combines the concept of wireless sensor networks with the cloud computing paradigm. This combination allows increasing the processing power and lifetime along with scalability of the data storage infrastructure and the ability to share the results more easily. By using the digital twin concept for field and edge devices as well as the networking elements that exhibit a well-defined service interface, we can offer an easy and understandable user interface that allows for the handling of the increasing complexity in todays and future industrial automation. In addition, it enables a simplified traceability of entities of an industrial installation over time. 3.2

Static View

The demonstrator components include a Smart Sensor (i.e., a field device that we can multiply as needed for the concrete scenario), an IoT Edge Gateway (i.e., real-time controller for field devices), the Cassia Gateway (i.e., that allows

Simplified Commissioning for 5G IIoT Applications

155

Fig. 2. Class diagram of the demo.

the wireless communication between field devices and their controller), the ABB cloud-based services platform, and a Mobile App (i.e., offers the additive sensing functions). The class diagram in Fig. 2 gives a close look at the structure of the on-site components of our demo. This includes the smart sensor, the IoT Edge Gateway components, and the Mobile App. For the smart sensor, it is a device that has a type model and a configuration model, which get predefined by the local network manager. It also has a device manger that takes care of its interactions starting

156

H. Abukwaik et al.

from registration and connection and ending up with sending the sensed data. With regards to the Edge Gateway component, it consists of subcomponents for: Device Registry and Discoverer (Yellow Classes): We use here the previously described publisher/subscriber design pattern. The device discovery event triggers the device registry handler that adds the device to the network using the Local Registry. The later posts a device registration event that get delivered to interested subscribers like the Ability Data Processor that invokes the Ability Device Registration. Here we use the previously described factory design pattern, in which we create a new device through a common interface without exposing the details about the device type. As noted in the model (green classes), we have two levels of registration the first is locally on the edge level and the second is on the cloud-based level, each with a different purpose. The device discovery event triggers the device registry handler that adds the device to the network using the Local Registry. The later posts a device registration event that get delivered to interested subscribers like the Ability Data Processor that consequently invokes the Ability Device Registration. We use the factory design pattern, in which we create a new device through a common interface without exposing the details about the device type. Data Collector (Blue Classes): For wireless communication between the smart sensor and the edge, we use a bluetooth-based solution (i.e., Cassia Gateway APIs). Event Handler and Logger (Red Classes): As our designed demo has many interacting elements, we decided to use the Publish/Subscribe design pattern in order to keep the components loosely coupled with separated concerns. That is, we decoupled the components that send an event from the components receiving it through an event channel or a broker. Accordingly, subscriber components subscribe to topics of interest (e.g., temperature data received, device registered, device discovered, etc.) with the broker that delivers topic-related event once received from publisher components. Thus, subscribers and publishers are decoupled on the relation-wise (i.e., they do not need to know each other), timewise (i.e., they do not need to run at the same time), and synchronization-wise (i.e., they do not need to be interrupted during their activities during publishing or receiving). In addition to the decoupling benefits, the Publish/Subscribe pattern allows controlling and message-filtering through the broker and in our use case it is topic-based filtering. Furthermore, this pattern is more scalable than the client-server pattern. We also had a design decision to have a Logger that is dedicated for keeping a log of all clients’ actions for debugging purposes. Accordingly, we see that using the previously explained singleton design pattern serves our needs. Hence, the logger is used to log publisher posts including device discovery, device registration, publishing measurement data, subscription to a topic, device disconnection, etc. Beside its use in the development life cycle, a rich log is seen as an auditing tool that gathers details about the system workflow and maintain it in a centralized location.

Simplified Commissioning for 5G IIoT Applications

157

Fig. 3. Sequence diagram of the IoT edge commissioning stage.

Supporting Security (Green Class): We opted at reusing ABB cloud-based services platform that has already integrated security services (e.g., authentication and authorization of users, edges, and endpoint devices). It also implements the proxy design pattern and plays an interface role between the edge gateway and the cloud, to which the collected data will be sent. Although we do not have a security-driven design, as it is out of the focus of this demo, we are aware of potential security improvements. For example, the Bluetooth-based communication between the smart sensor and the edge gateway may not be considered as the most secure communication channel. 3.3

Dynamic View

The commissioning procedure described in this paper consists of two steps: the commissioning of the edge acting as local proxy to the cloud followed by the commissioning of the field device itself (see Fig. 3). For commissioning the edge device, we assume the following: The edge is already connected to the local network and is equipped with a 5G modem to connect the local 5G campus network. The campus network in turn can be managed via a local management and orchestration component (5G-M-and-O-Module), which oversees the admission control. The cloud-based platform hosts the type registry and the information model offsite and can be accessed over the Internet. During the production of the edge device an NFC tag is attached to the edge device. The engineers loaded this tag with information about the edge type, the serial number as well as the IMSI of the 5G SIM card. Furthermore, the device type definition is registered

158

H. Abukwaik et al.

in a central type registry containing the information about the properties and functions of the edge device. Thus, the installer starts the commissioning app on the mobile device, which establishes a connection to the cloud and authenticates itself against the cloud. Next, the installer scans the NFC tag of the edge device to obtain the serial number, edge type and IMSI. Based on the device type, the commissioning app requests the device type definition from the type registry, extracts the configuration parameters and optionally requests missing parameters from the installer. Subsequently, the app sends the gathered information to the information model on the cloud to create an object instance of the device. Finally, the information model reports the device creation status back to the mobile app, which is displayed to the installer. Afterwards, the edge device contacts the 5G management and orchestration module and registers its IMSI to get access to the local 5G network. Once the access has been granted, the edge device establishes a 5G connection with the base station and sends a status update to the information model which in turn notifies the installer via the mobile app. The second stage is the commissioning for the child field devices. For commissioning the field device, the installer scans the attached NFC tags, assigns already commissioned edge device as parent device, and if necessary modifies the parameter settings for the field device. Once the parameters have been set, the mobile application determines its location and associates it with the location of the field device. The information model creates the object instance and based on the parent device information informs the already running IoT edge device about the new field device by publishing a new device message. The IoT edge device in turn contacts the 5G management and orchestration module and registers the IMSI so that the field device gets access to the local 5G network. Once the information model has created the corresponding object instance for the field device, an update message is sent to the mobile app to inform the installer about the successfully registered field device. Once the registration process finishes, the installer starts the field device. Firstly, the field device establishes a 5G connection to the local private base station and in case of a field devicedriven setup, discovers the IoT edge device via means of either IP range scan or another rendezvous mechanism (e.g., mDNS or Bonjour). The IoT edge device in turn, responds with the Universally Unique Identifier (UUID) and the device configuration obtained from the information model during the registration step. Once the field device configuration step finishes, a status update message is sent to the information model by the edge device. The information model stores the updated information and pushes a state update message to the mobile app to inform the installer about the field device being in operational state. As soon as the field device has reached its operational state, it starts sending status update messages to the IoT edge device that automatically forwards them to the information model.

Simplified Commissioning for 5G IIoT Applications

4

159

Implementation

In this section, we explain the technologies we used to implement the App demo. We used Java and Python along with open source libraries as described below. To implement the publisher/subscriber pattern, we used the Google EventBus [5]. It allows publishers to post events and dispatch them to subscribers based on their type and allows the subscribers to register by themselves. In specific, we use the AsynchronousEventBus subclass for asynchronous event dispatching. The Message Queuing Telemetry Transport (MQTT) messaging protocol [18] is our choice for implementing the publisher/subscriber messaging. The MQTT Broker works on top of the TCP/IP protocol and embodies all the decoupling aspects that we mentioned earlier in Sect. 3. That is, it decouples publisher and subscribers and provides them with its hostname/IP and port. It also allows message storing for offline clients and it works asynchronously. We use this protocol for the communication between the Edge Gateway and the cloud-based platform components. The Javalin [9] is a lightweight Java and Kotlin web framework that offers REST APIs. This we mainly need for implementing the communication between two components (i.e., the Mobile App and the Edge Gatway) by creating and starting the REST server of th ethird component (i.e., the IOT Edge Device) and creating the RegisteredDeviceHandler. The OkSse [8] is an extension library for OkHttp [16] (i.e., open-source API designed for creating efficient HTTP-client) to create Server-Sent Event (SSE) client. We reuse this library to create the HTTP-clients who then send and receive synchronous calls. These calls are for sending device discovered messages, which is created by the RawDataCollector class and data collected, which is created by the DeviceDiscoverer class. The Cassia SDK [3] allows us to connect our Bluetooth Low Energy (BLE) sensors to the Cassia Bluetooth router. Once connected, the Cassia Bluetooth router acts an Internet gateway and syncs the BLE sensor data to the cloud. The GSON [6] is a Java serialization/deserialization library that we use to convert Java Objects into JSON and back. Specifically, we use it in converting the event message string received by a subscriber into the device that created the event. The Log4j [2] is a Java-based logging utility that allows simple logging using built-in logging levels (e.g., DEBUG and ERROR). It also implements our desired singleton design pattern. Further, it supports user-defined message objects, filtering, and asynchronous loggers. So, we use this library in implementing our Logger class.

5

Related Work

Our simplified IoT commissioning approach is related to Plug and Play (PnP) technology that was originally developed by Microsoft and Intel to compute

160

H. Abukwaik et al.

devices on a local computer bus. This was extended with the universal Plug and Play (uPnP) [10] that allows devices to discover each other on a computer networks. Researchers have been working on PnP approaches for more than a decade and here we present some of the relevant ones to our proposed approach. Koziolek et al. [12] proposed OpenPnP architecture, which uses standardized network discovery techniques for industrial field devices. It is like our approach in equipping field devices with information models formerly designed, automatically transferring configuration parameters to devices, and automatically connecting devices with controllers. However, it does not support automatic positioning and requires determining it manually. Hammerstingl and Reinhard [7] proposed a unified PnP architecture, which contains similar concepts as our approach, but it is based on classical fieldbuses and analog connections and accordingly allow rich information modeling or vendor interoperability. Kainz et al. [11] proposed AutoPnP approach for discovery and connection of production modules. Though, it requires substantial modeling perquisite for each different setting. Alkhabbas et al. [1] proposed inferring emergent configurations for engineering IoT systems by processing contextual information and following the MAPE-K loop from autonomic computing. However, due to safety reasons in our domain, fixed configuration parameters are still dominant.

6

Conclusion

In this paper, we presented edge-computing-based commissioning approach using NFC tags in combination with information modelling. We discussed its requirements and design decisions under two scenarios. Then, we presented our solution approach, which relies on the edge computing paradigm for managing the access control of devices in local 5G networks. The use of NFC tags that store the device type and serial number information together with the use of a central information model enables fast and fault tolerant IIOT device commissioning. In future, we plan to deploy our solution in a larger automation setup to gain further performance insights for real-world setup. We will also include cross network domain QoS negotiation mechanism into our commissioning procedure and to automate the generation of the field device configuration by exploiting data in the information model. Acknowledegements. This research was supported by the German Federal Ministry of Education and Research (BMBF) under grant number 16KIS0721. The responsibility for this publication lies with the authors.

References 1. Alkhabbas, F., Spalazzese, R., Davidsson, P.: Architecting emergent configurations in the internet of things. In: 2017 IEEE International Conference on Software Architecture (ICSA), pp. 221–224. IEEE (2017) 2. Apache: Apache log4j 2 (2019). https://logging.apache.org/log4j/2.x/. Accessed 07 Mar 2019

Simplified Commissioning for 5G IIoT Applications

161

3. Cassia Networks: Cassia sdk (2009). https://www.cassianetworks.com/products/ bluetooth-sdk/ 4. Feltrin, L., Tsoukaneri, G., Condoluci, M., Buratti, C., Mahmoodi, T., Dohler, M., Verdone, R.: NarrowBand-IoT : a survey on downlink and uplink perspectives (2018) 5. Google Inc.: Google eventbus (2019). https://google.github.io/guava/releases/22. 0/api/docs/com/google/common/eventbus/EventBus.html. Accessed 07 Mar 2019 6. Google Inc.: Gson 2.6.2 api (2019). https://google.github.io/gson/apidocs/com/ google/gson/Gson.html (2019). Accessed 07 Mar 2019 7. Hammerstingl, V., Reinhart, G.: Unified plug & produce architecture for automatic integration of field devices in industrial environments. In: 2015 IEEE International Conference on Industrial Technology (ICIT), pp. 1956–1963. IEEE (2015) 8. HERE Technologies: Oksse (2019). https://github.com/heremaps/oksse. Accessed 07 Mar 2019 9. Javalin: Javalin - a lightweight java and kotlin web framework (2019). https:// javalin.io/. Accessed 07 Mar 2019 10. Jeronimo, M., Weast, J.: UPnP Design: A Software Developers Guide to Universal Plug and Play by Example (2003) 11. Kainz, G., Keddis, N., Pensky, D., Buckl, C., Zoitl, A., Pittschellis, R., K¨archer, B.: Autopnp-plug-and-produce in der automation. atp Mag. 55(04), 42–49 (2013) 12. Koziolek, H., Burger, A., Doppelhamer, J.: Self-commissioning industrial IoTsystems in process automation: a reference architecture. In: 2018 IEEE International Conference on Software Architecture (ICSA), pp. 196–19609. IEEE (2018) 13. Kurschl, W., Beer, W.: Combining cloud computing and wireless sensor networks. In: Proceedings of the 11th International Conference on Information Integration and Web-based Applications & Services, pp. 512–518. ACM (2009) 14. Qanbari, S., Pezeshki, S., Raisi, R., Mahdizadeh, S., Rahimzadeh, R., Behinaein, N., Mahmoudi, F., Ayoubzadeh, S., Fazlali, P., Roshani, K., et al.: IoT design patterns: computational constructs to design, build and engineer edge applications. In: 2016 IEEE First International Conference on Internet-of-Things Design and Implementation (IoTDI), pp. 277–282 (2016) 15. Schulz, D.: Intent-based automation networks: toward a common reference model for the self-orchestration of industrial intranets. In: IECON Proceedings (Industrial Electronics Conference) (2016) 16. Square, Inc.: OkHttp (2019). https://square.github.io/okhttp/. Accessed 07 Mar 2019 17. Stal, M., Buschmann, F., Meunier, R.: Pattern-Oriented Software Architecture: A System of Patterns (1996) 18. Standard Oasis: Version 3.1.1 (2014). https://docs.oasis-open.org/mqtt/mqtt/v3. 1.1/os/mqtt-v3.1.1-os.html

SCHC-Based Solution for Roaming in LoRaWAN Wael Ayoub1(B) , Mohamad Mroue2 , Abed Ellatif Samhat2 , Fabienne Nouvel1 , and Jean-Christophe Pr´evotet1 1

Institut National des Sciences Appliqu´ees de Rennes—IETR-INSA, Rennes, France {wael.ayoub,fabienne.nouvel,jean-christophe.prevotet}@insa-rennes.fr 2 Faculty of Engineering - CRSI, Lebanese University, Hadath Campus, Hadath, Lebanon {mohamad.mroue,samhat}@ul.edu.lb

Abstract. To take advantage of IPv6 stack in IoT technologies, an efficient header compression scheme is required. Since 2004, many IPv6 header compression schemes have been proposed and some of them have been standardized by the IETF. In [9], Static Context Header Compression (SCHC) mechanism has been designed for Low Power Wide Area Networks (LPWAN). SCHC compression is based on a common static context stored in both the IoT device and the network side. This static context defines the compression and decompression rules of the headers. The SCHC framework is compatible with LoRaWAN v1.0 [2] but not with LoRaWAN v1.1 that supports roaming of devices during mobility between different LoRaWAN operators. During roaming, the header values of the protocol stack change and are no longer static. In this paper, we propose a solution based on SCHC to support roaming of devices during mobility between different LoRaWAN operators. We define a server to manage the context between operators. In addition, the LoRaWAN frame route and the communication scheme are updated. A testbed has been setup to show the time differences between current LoRaWAN network and our proposal. The results shows that our proposal improves the communication process and decreases the time delay to handle the transmitted messages “uplink” before the registration. Keywords: IoT communication · LPWAN Long-range · Mobility · Roaming · SCHC

1

· LoRaWAN · IPv6 ·

Introduction

LoRaWAN is an open standard architecture developed by LoRa Alliance [2] to provide a medium access control mechanism and enable End-Devices (ED) to communicate with one or more gateways (GW). A GW is seamless to an ED, where an ED sends data when available without considering location change, movement, and speed of motion. In addition, any GW that receives the message c Springer Nature Switzerland AG 2020  L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 162–172, 2020. https://doi.org/10.1007/978-3-030-33506-9_15

SCHC-Based Solution for Roaming in LoRaWAN

163

will forward it to the NS. In the latest version of LoRaWANv1.1 [1], specifications address the mobility of devices. They define the coverage of the home operator, as a home network server (H-NS), and the coverage of the foreign operator by a Visited Network Server (V-NS) as shown in Fig. 1. LoRaWAN specifications define two roaming scenarios: the passive and the handover. We discriminate between these two roaming types based on the LoRaWAN communication stack. This stack consists of three layers: the anchor, the LoRaWAN and the Gateway controller. The anchor is responsible for the communication between the NS and the Application Server (AS). It also communicates with the Join Server (JS) to manage the state of the device and the registration parameters to associate the device with the NS. The second layer of this stack is the L2 controller, which is the LoRaWAN protocol. This layer specifies the functionality of the LoRaWAN link layer found in the specification, such as Adaptive Data Rate (ADR) management, device location, communication, etc. The third layer is the gateway controller. This layer defines the communication between the NS and the GWs that cover different areas. This layer is responsible for the functions of the PHY layer, such as the radio access network, the power transmission, etc. As can be seen in Fig. 1, in case of no roaming, the device is associated with the H-NS which contains all three layers. After roaming, if the collaboration between the two network operators is passive, only layer three i.e. Gateway controller will be assigned to the V-NS. In this case, the V-NS will be considered as a GW to extend the H-NS coverage. In the handover collaboration, the second and third layers will be assigned to the V-NS. The device will be associated with the V-NS and a registration process is required to access the network to send/receive data. However, the anchor layer is maintained with the H-NS. Therefore, any packet received from the ED by the V-NS will be forwarded to the H-NS and then sent to the corresponding AS. In addition to the three layers specified in LoRaWAN v1.1, the SCHC protocol [9] is a protocol for compressing/decompressing headers in the communications stack. This mechanism is placed as a layer between the Anchor and LoRaWAN layers in case of no roaming. The SCHC standard is defined to install this layer on the ED and the NS or on a separate server within the home network. Since the H-NS reserves the Anchor layer to communicate with the AS, the logic is to keep the SCHC layer with the anchor layer. Therefore, the communication procedure will be as shown in Fig. 2. In both roaming cases, the SCHC layer and the anchor were reserved for H-NS. Then, the communication procedure will be as follows. In passive communication, the ED sends the uplink to the GW. Uplink is the message transmitted by ED towards AS. Then, the GW forwards the LoRaWAN frame to the V-NS. Because the device is not assigned to the V-NS, there is no network session key to decrypt the payload of the frame. Thus, the LoRaWAN frame is forwarded to the H-NS. In the H-NS, the frame is decrypted using the network session key and the frame payload is executed. This payload will be decompressed using the SCHC layer. Then, it will be forwarded using the anchor to the corresponding AS. In the AS, the received payload will be decrypted using the application session key and the data will be available to the user. In handover communication, the device is associated with

164

W. Ayoub et al.

Fig. 1. Communication schemes support on LoRaWANv1.1

the V-NS, the received LoRaWAN frame is decrypted in the V-NS and the compressed payload is sent back to H-NS. Then, the payload is decompressed using the SCHC layer in the H-NS and the final payload is sent to the AS to decrypt the user data. In this case, it is not necessary to share the context of the device as it moves between different foreign network providers. But if we consider that the V-NS and the AS are in the same network topology, this form of implementation is not ecient in terms of bandwidth usage, routing, power consumption, and communication latency. During downlink i.e. transmitted message from AS to ED, latency is an important and considerable parameter, especially if the downlink is an acknowledgment (ACK). For example in LoRaWAN class A devices, two windows open after an uplink, with one second time for each. Therefore, within two seconds the device have to receive the ACK. Otherwise, the uplink is repeated. This consumes more energy, and causes collisions and interference with other devices as the number of devices in the area increases. Also, increasing listening time is not efficient in terms of energy consumption. From our observation, there is still no work related to SCHC mobility or LoRaWAN roaming. Next, we illustrate the proposal of the communication mechanism in Sect. 2. Then, this mechanism will be implemented and tested as shown in Sect. 3. Finally, Sect. 4 concludes and summarizes our contributions.

2

Roaming Enhancement

In this section, we propose a mechanism that allows the use of the two layers: anchor and SCHC, by the V-NS in order to optimize the communication routing. For this purpose, an Administration Management Server (AMS) is added to the LoRaWAN network. In parallel to the Join Server (JS), the AMS is responsible for the devices belonging to the H-NS. The AMS will manage the registration

SCHC-Based Solution for Roaming in LoRaWAN

165

Fig. 2. Protocol stack at each component

and exchange the SCHC context, the geographic location of the devices and perform a pre-registration with the V-NS. It will include a routing table for each device that belongs to the network. It can also save the data message rate and manage the power of the device. AMS is responsible for the network topology formed by the GWs and network operators to which the device can access. In addition, when AMS detects the geographical location [4,7] of the device, it verifies whether roaming is necessary or not. Then, AMS performs a preregistration process for a device and sends the new parameters through a downlink message ACK before the connection to the current GW is lost. This functionality can also be added to the H-NS, but it adds a load to the H-NS as the number of devices increases, and especially when it comes to SCHC. For this, it is preferable to add the AMS as an independent server in the LoRaWAN network that is formed by H-NS and its V-NS partners, as shown in Fig. 3. In loRaWAN, geolocation management is quite easy. Each GW of LoRaWAN has a wide coverage. In addition, the GW is a physical repeater of the uplinks/downlinks to the NS. An NS can cover a part of a city and more. Therefore, in LoRaWAN when it comes to geolocation, it is not necessary to wait for a point location for the car or wait exactly for GW coverage, since the device is associated with NS and not with GW. It is easy to deal with it as a car moving from a city covered by more than one GW that belongs to H-NS to another city covered by the GWs of V-NS(X). In this article, we deal with a wide geolocation specified by an operator’s coverage, while knowing the coverage of each GW will not make a difference Therefore, the management of the location of the devices is an important feature in the AMS. As shown in Fig. 3, while the car moves from position one (P1) to position two (P2), the AMS detects the degradation in the RSS of the packets received by the GWs of the H-NS. The AMS knows the covered area of each NS. The movement of the automobile can be predicted from the road maps. Therefore, as in [7], AMS can expect from the geographical location of the car and the known NS coverage the next network operator the car will cross. Then AMS contacts this operator to inform it about the car. When an uplink message is received from ED, the NS can accurately detect the closest GW to the car and can specify a more accurate location for the AMS.

166

W. Ayoub et al.

Fig. 3. Proposed LoRaWAN network

During the movement of a car between GWs, AMS has to face two cases. First, if the operator coverage is H-NS or V-NS with passive roaming collaboration, AMS does not step in the communication process. In this case AMS only used in the compressing/decompressing process of the SCHC package [3] and manage the SCHC context change. Secondly, if the car moves to a V-NS operator coverage with the handover collaboration, AMS enrolls the device in this NS to avoid the drop of the uplink message. This V-NS will then accept the uplink message received from the device and send it to H-NS since the message is encrypted with the previous network session key. After decrypting the packet with the H-NS network session key, it is sent to AMS to decompress the SCHC header and then sent to AS. In parallel, AMS will send all the required information about the device to V-NS(Y) in order to continue the registration process. Then, while sending the ACK on the downlink, V-NS(Y) informs the device to extend the RX2 window to continue with the registration process. The advantages of this mechanism is that the uplink message of the device is received and not discarded, the time of the registration process decreases and there is no delay in delivering the data to AS, especially if this data is important. Once the registration success, a network session key of V-NS(Y) will be assigned to the device, so that no more packets will be sent to H-NS and V-NS(Y) will deliver the data to AS decompressed using the AMS. In addition to managing the geographic location, AMS has to track the SCHC context change while the ED is moving between the networks. In fact, the ED is allowed to access either H-NS or its partners. Thus, AMS accesses and monitors the status of the device when it moves between these operators. AMS contains the

SCHC-Based Solution for Roaming in LoRaWAN

167

context used to compress and decompress the packet headers of the device. As the device network operator changes, the new operator will send the received packet from the ED to AMS to compress/decompress it. Then, the SCHC contexts will be managed by the AMS and no exchange between operators will be necessary during device mobility. Therefore, maintaining this responsibility in a singular part of the network is efficient, easy and manageable. As the location changes, IPv6 (layer L3) of the ED also changes. To avoid adding an L3 function to each NS, AMS can assume this responsibility and manage the IPv6 address of the ED. The management of the IPv6 address during mobility allows the continuity of the session between the device and the AS after mobility.

Fig. 4. Time difference at each position

168

W. Ayoub et al. Table 1. Measurements of the latency Time in ms Without AMS With AMS

3

P1 Uplink (Car to A) Forward (A to H-NS) Process and Forward (H-NS to AS) ACK (AS to car) Total

1509 20 420 150 2099

1509 20 460 185 2156

P2 Uplink (Car to D) Forward (D to X) Check and Forward (X to H-NS) Link to AMS (Send/Receive) Decompress SCHC Process and Forward (H-NS to AS) ACK (AS to Car) Include SCHC Compression Total

1509 50 80 X 326 120 230 2315

1509 49 80 40 300 115 265 2358

P3 Uplink (Car to F) Wait ACK Register to X Repeat Uplink (Car to F) Forward (F to Y) Check and Forward (Y to H-NS) Process and Forward (H-NS to AS) Include SCHC decompression Forward (Y to AS) Include SCHC decompression ACK (AS to Car) Total

1509 2000 5500 1509 20 50 442

1509 X X X 20 X X

X

415

228 11258

270 2214

Testing and Measurements

To test the proposed framework, a testbed was built to measure the communication delay. The testbed consists of three network operators, as shown in Fig. 3. Using three Raspberry Pi and LoRaWAN shields, with the help of the LoRaServer project [6], we created the Lora network and implemented the SCHC protocol on the ED and on the AMS. Three network servers are involved: HNS, V-NS(X) and V-NS(Y). GWs A and B belong to H-NS, C and D belong to V-NS(X) and E and F belong to V-NS(Y). The distance between each two GWs that belongs to the same operator is 500 m, i.e., A and B whereas it is 800 m between GWs of different operators, i.e., B and C. We consider that the transfer between H-NS and V-NS(X) is a passive collaboration, while it is a

SCHC-Based Solution for Roaming in LoRaWAN

169

handover collaboration with V-NS(Y). GWs A and B belong to H-NS, C and D belong to V-NS(X) and E and F belong to V-NS(Y). The mobile device is an Arduino UNO R3 with a LoRaWAN shield [8], programmed as Over-TheAir-Activation (OTAA). During roaming, the LoRaWAN parameters and the location of the devices changes, so Activation-By-Personalisation (ABP) will not support roaming. In addition, the Wireshark program is used to measure the time delay between every two components of the network. To make the implementation closer to reality, we used the Pfsense project [5] to manage the bandwidth, add delays and manage the routes between the different components. Otherwise, the delay between the components is less than 1 ms. The JS and the AMS are executed on the PC that is connected to the same network. The mobile device associated with the H-NS at the startup. Next, we will measure the time required for a complete communication procedure that begins when a device sends an uplink LoRaWAN frame to AS followed by a confirmation link sent from AS to device. This procedure is repeated in the three positions of the car, as shown in Fig. 3. In each position, we compared the time required for a complete procedure using the LoRaWANv1.1 network without AMS and the same procedure when AMS installed in the network. During the measurements, we configure the device to use a payload size of 155 bytes, the Spreading Factor (SF) is set to 10 and the bandwidth is set to 125 kHz. The payload of 155 bytes consists of 40 bytes representing uncompressed IPv6, 8 bytes for UDP, 4 bytes for CoAP, 1 byte for RuleID, 1 to 5 bytes for compression headers, and remain bytes are data. 3.1

At Position P1

In the P1 position, as shown in Fig. 3, the car is under the coverage of GW (A). As shown in Table 1 and Fig. 4, the car sends an uplink message to GW(A). Then, the uplink message is forwarded to the H-NS. In a network not compatible with AMS, the H-NS unpacks the package, then decompresses it and sends it to the AS. While, if AMS is supported, H-NS sends the packet to AMS to decompress it. Then, H-NS forwards this packet to AS. The additional time delay between the two implementations is the time needed to send and receive the packet between H-NS and AMS. As shown in Table 1 and Fig. 4 in P1, without AMS, in the downlink when the AS returns an acknowledgment, HNS compresses the downlink with a duration of 150 ms. But a communication time delay of 185 ms is added to compress the headers when AMS is used. Finally, the total time difference between AMS network support and the nonAMS network is 57 ms. This additional cost represents the link time between H-NS and AMS generated during the uplink and the downlink to achieve a complete communication procedure. Our proposal will add a time delay of 57 ms in the uplink during decompression and in the downlink during compression of the packet headers over the current implementation of LoRaWANv1.1. But with AMS, the context of SCHC is centralized within a server and avoids the dispersion of the ED context between the servers. In addition, technologies such

170

W. Ayoub et al.

as LoRaWAN are not developed for the use of applications in real time, so that 57 ms of latency will not be recognized. 3.2

At Position P2

In the P2 position, as shown in Fig. 3, the car is under the coverage of GW(D). As shown in Table 1 and Fig. 4, the car sends an uplink message to GW(D). This uplink message costs 1.50938 s of time. Then, GW(D) forwards the received LoRaWAN frame to V-NS(X). The link between GW(D) and V-NS(X) costs 50 ms to forward the received LoRaWAN frame. Since the collaboration with the H-NS is passive, the V-NS(X) verifies only the address of the device at the L2 layer and sends the LoRaWAN frame back to the H-NS. This procedure costs 80 ms which this depends on the process of the Raspberry Pi (RPi) and the link between the two RPis. In a LoRaWAN network without AMS, H-NS has the SCHC context of the device to decompress the headers of received packets. This process costs 326 ms. While in our proposal the network is compatible with AMS, it maintains the SCHC context of the device and decompresses the packet header. The cost of the link between H-NS and AMS is 40 ms for sending/receiving and 300 ms to decompress the package. Finally, the H-NS receives the decompressed header packet and sends it to AS with a time delay of 120 ms. After a successful uplink message is received, the AS will respond with an ACK in the downlink message. In the LoRaWANv1.1 network, the total latency for a complete procedure was 2,315 s, whereas it is 2,358 s for the network with AMS. The results show that in the case of passive collaboration, our proposal is not so much time consuming and improves the organization of the compression/decompression of SCHC in the network in a central server. 3.3

At Position P3

In the P3 position, as shown in Fig. 3, the car is under the coverage of GW(F). As shown in Table 1 and Fig. 4, the car sends an uplink message to GW(F). In the network without the AMS, the GW forwards the received frame to V-NS(Y). Since collaboration is handover, the LoRaWAN frame is dropped and the device has to register with the V-NS(Y). As shown in Fig. 4, the car waits two seconds to receive an ACK, but the frame was dropped. Without receiving confirmation, the device detects a loss of connection. Therefore, as shown in Table 1 and Fig. 4 for the position P3 without AMS, the total time delay is 1,509 for the uplink and 2 s are waiting for an ACK. Then, the device begins with the registration procedure with the V-NS(Y) [10]. The registration procedure OTAA costs 5.5 s in our case. Then, the device repeats the uplink message with a time delay of 1.509 s. After that, the V-NS(Y) forwards the packet to H-NS. Then, the packet is sent to AS after decompressing the headers. While for the network that supports AMS, a preregistration process is performed on the V-NS(Y) by the AMS. This mechanism prevents dropping the uplink message as shown in Fig. 4 at P3 with AMS. Moreover, when using the AMS, the following packets received from the device can be decompressed directly by sending to AMS. This

SCHC-Based Solution for Roaming in LoRaWAN

171

mechanism adds a time delay between the V-NS(Y) and the AMS but reduces the time delay between V-NS(Y) and H-NS. Then, the packet is sent directly to AS by V-NS(Y). This proposal also saves the cost of routing, especially if AS and V-NS(Y) are in the same network topology. In addition, it saves the use of bandwidth between V-NS(Y) and H-NS, and between H-NS and AS, as the number of devices belonging to H-NS increases. Finally, as shown in Table 1 and Fig. 4, the total time cost of the network without AMS is 11.258 s to deliver the first uplink from the device to AS, while for a network with AMS, the uplink message is delivered to AS with only 2,214 s.

4

Conclusions

In this paper, the LoRaWAN architecture is modified by adding an AMS to improve the mobility of the device; uses an IPv6 header compressed by SCHC. We implemented the AMS mechanism in a testbed to verify the improvement in term of time. The contributions of AMS were: decreases the time to handle the uplink message before the registration process during the LoRaWAN handover collaboration. Secondly, it saves the bandwidth and message rate of the device and avoids repeating the transmission and avoids transmitting the SCHC context of the device to the new NS. Third, reserve the power of the device by avoiding repeating the uplink messages. On the other hand, improve the transfer and continuity of the session in layer L3 as shown in [3]. Fourth, it improves the use of the SCHC protocol and saves the bandwidth from context exchange and the updates. Finally, it avoids the dispersion of the SCHC context saved in different NS during each movement of the device.

References 1. Lora alliance. https://www.lora-alliance.org/. Accessed 06 Oct 2018 2. Ayoub, W., Samhat, A.E., Nouvel, F., Mroue, M., Pr´evotet, J.: Internet of mobile things: overview of lorawan, dash7, and nb-iot in lpwans standards and supported mobility. IEEE Commun. Surv. Tutor. 21, 1561–1581 (2018). https://doi.org/10. 1109/COMST.2018.2877382 3. Ayoub, W., Nouvel, F., Hmede, S., Samhat, A.E., Mroue, M., Pr´evotet, J.C.: Implementation of SCHC in NS-3 simulator and comparison with 6LoWPAN. In: 26th International Conference on Telecommunications (ICT), HANOI, Vietnam, April 2019. https://hal.archives-ouvertes.fr/hal-02051757 4. Baharudin, A.M., Yan, W.: Long-range wireless sensor networks for geo-location tracking: design and evaluation. In: 2016 International Electronics Symposium (IES). pp. 76–80, September 2016. https://doi.org/10.1109/ELECSYM.2016. 7860979 5. Botelho, R., Souza, L.O.O., Pingle, J., Dillard, J., Beaver, S., Smith, M.: Open source security firewall from netgate. https://www.pfsense.org/. Accessed 5 Feb 2019 6. Cablelabs, Sidnfonds, Acklio: Loraserver project. https://www.loraserver.io/. 5 Feb 2019

172

W. Ayoub et al.

7. Fargas, B.C., Petersen, M.N.: GPS-free geolocation using LoRa in low-power WANs. In: 2017 Global Internet of Things Summit (GIoTS). pp. 1–6, June 2017. https://doi.org/10.1109/GIOTS.2017.8016251 8. Kooijman, M.: Arduino-lmic library. https://github.com/matthijskooijman/ arduino-lmic. Accessed 5 Dec 2017 9. LPWAN-IETF-WG: LPWAN static context header compression (SCHC) and fragmentation for IPv6 and UDP. https://tools.ietf.org/html/draft-ietf-lpwan-ipv6static-context-hc-09. Accessed 22 Dec 2017 10. Toussaint, J., Rachkidy, N.E., Guitton, A.: Performance analysis of the on-theair activation in LoRaWAN. In: 2016 IEEE 7th Annual Information Technology, Electronics and Mobile Communication Conference (IEMCON). pp. 1–7, October 2016. https://doi.org/10.1109/IEMCON.2016.7746082

Reputation System for IoT Data Monetization Using Blockchain Atia Javaid, Maheen Zahid, Ishtiaq Ali, Raja Jalees Ul Hussen Khan, Zainib Noshad, and Nadeem Javaid(B) COMSATS University, Islamabad 44000, Pakistan [email protected], http://www.njavaid.com/

Abstract. Internet of Things (IoT) is growing exponentially and bringing revolution in today’s modern society. IoT based smart devices are source of convenience to human life and producing huge amount of data on daily basis. This data is useful for consumers like industries, marketplaces, and researchers to extract valuable and functional data from raw data generated by these devices. This data is used by industries and developers to provide more efficient devices and services to users. Owner of the IoT device can generate revenue by selling IoT device data to interested consumers. However, on the other hand consumers do not trust the owner of IoT device for data trading and are not confident about the quality of data. Traditional systems for data trading have many limitations, such as they are centralized, lack reputation system, security and involve third party. Therefore in this paper, we have leveraged the IoT with blockchain technology to provide a trustful trading through automatic review system for monetizing IoT data. We have developed blockchain based review system for IoT data monetization using Ethereum smart contracts. All transactions are secure and payments are automated without any human intervention. Data quality is ensured to consumer through reviews and ratings about the data. Additionally, Ethereum blockchain system requires gas for every transaction. We have used 2 parameters: gas consumption, string input length and in terms of time and cost, and examined our model. Keywords: Message querring telemetry transport Review system · Metadata

1

· Blockchain · IoT ·

Introduction

The Internet of Things (IoT) is a world wide network where humans, devices and objects are connected with unique addresses. These devices have the capability to communicate, share and transfer data through centralized servers. IoT is growing exponentially year by year and gaining more attraction for all academic fields and industries. More than 8.5 billion devices are connected to the IoT and will increase up to 20.4 billion connected devices in 2020. The applications of c Springer Nature Switzerland AG 2020  L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 173–184, 2020. https://doi.org/10.1007/978-3-030-33506-9_16

174

A. Javaid et al.

IoT in daily life makes its importance more evident. These devices produce data of different types depending on the type of network. This data is used by public users, industries and many other technologies like smart home automation, health care, data trading market places, etc. This data is used to make human life more convenient and advance. Using IoT devices for the purpose of sensing as a service, huge amount of revenue could be generated. As number of connecting devices are increasing, there is huge threat to network security and data integrity. Traditional centralized systems because of security, automation, scalability and third party issues are not feasible for IoT devices to share or trade data and assets over the network. Centralized systems are costly, involve hacking and trust issues along with the threat of single point of failures [1]. Improving security and privacy regarding IoT is challenging because IoT devices are energy, memory and resource constrained [2]. The available amount of energy is utilized to perform needed application functions, because of which security and other important features are compromised. Many researchers contributed to handle such security issues like transmission field security [3], cloud storage field [4], digital signature and permission identification [5,6]. In the field of IoT and healthcare, researchers have worked on the IoT to collect data ubiquitously [7], to provide processed data in time for efficient healthcare purposes. The resource based IoT access methods are used to collect heterogenous data ubiquitously from cloud and mobile computing platforms. On the other hand, these devices are generating huge amount of data. Scalable data storage solution is necessary to handle large amount of data efficiently. Researchers have worked on such data storage issues to produce efficient solutions for the structured and unstructured data [8]. Blockchain is new emerging technology with a lot of advantages of security, trust and immutability. It is a decentralized ledger, previously only used for money transactions however, it is now widely applied to diverse fields like smart grids, IoT [9], smart cities [10], vehicle management [11] and many more. Once the transactions between sender and receiver are recorded in the open and distributed ledger, then this data cannot be tampered. In the mining process of blockchain, miner records the transaction in the block and broadcast this block in the network, so that all nodes have the same copy of transaction.

2

Related Work

Many studies have been done to leverage the advantages of blockchain in IoT. Blockchain and IoT integration have achieved automation in industries and in daily human life. Secure, scalable and adaptive industrial IoT platforms are needed by the industries to achieve their goals. Existing industrial IoT platforms are centralized systems having problems of single point failure. These systems also face the problems like accessibility, confidentiality and integrity. Semantic rules engine [12] and health based IoT systems [13] are facing the issue of single point failure. In [14] authors have presented the industrial IoT system for the smart factories to address the above mentioned problems. Directed Acyclic

Reputation System for IoT Data Monetization Using Blockchain

175

Graph (DAG) structured blockchain is used with efficient POW mechanism, making DAG structured blockchains more efficient for industrial IoT systems. Privacy and confidentiality are ensured through data authority management system. However, they do not tackle the storage and data quality issues related to sensor data. There is a major threat to traditional storage systems because of centralized storage. Traditional storage systems are vulnerable to single point of failures, and are costly due to involvement of third party. In [15] a blockchain based storage system named Sapphire is presented for analyses of data. For this storage system smart contracts are created based on object storage device approach for interaction of IoT devices with blockchain. Application specific operations are executed only and results are provided to users instead of all the data which decreases the storage and computational overhead of data analyses in IoT. However, this system is not matured enough to deal with scalability and security issues. IoT devices are used along with other electronic devices and protocols to perform different tasks and operations without the human involvement. These electronic and IoT devices and other network participants have to communicate with each other to complete desired tasks. These devices must be authenticated before entering the network and sharing data and resources with the other authenticated entities. Authentication is essential, otherwise network will become target for malicious users. Conventional centralized authentication systems for IoT are not efficient. In [16] authors have presented a decentralized system for authentication of devices. Blockchain system provides security features and virtual zones provides trustful environment where devices can authenticate each other. However, virtual zones introduce the computational overhead and delay for the authentication purpose. Different authors have used blockchain to tackle the several problems such as: data trading, energy trading, node recovery, efficient energy routing, edge servers participation, data rights management, healthcare issues, securing data, fair sharing of data and under water routing problems. In [17–27] authors, have provided solutions for the above mentioned problems using blockchain. 2.1

Motivation

In [28], authors used review system for ensuring data quality and integrity of data being traded through data marketplaces. Reviews are maintained by the blockchain based system for the new users to analyze the reviews before using the data. 2.2

Problem Statement

Earning revenue through the data produced by the IoT devices considering security, cost efficiency, automatic monetization and centralized governance are the challenging problems handled by the authors in [29].

176

A. Javaid et al.

• Security: Previously, generating revenue from IoT does not consider the rules and regulations provided from the owner of the IoT devices. Usage terms about the data are not well defined. • Cost Efficiency: Automatic monetization of IoT data is costly due to the involvement of third parties to exchange data and money. • Centralized Governance: Number of devices connected to the centralized platform will create a bottleneck, and if centralized system is attacked or failed, whole system will suffer the consequences. Authors have implemented a blockchain using smart contracts to provide secure, cost effective and decentralized solution for IoT data trading. However, they did not consider the trustworthiness of data owner and quality of data. Data consumers will purchase the data only if they have trust on the data owner. Previous IoT and blockchain systems do not provide a review system for IoT data to be traded. Review system is needed which holds the reviews from users who have used the data, so that other data users can trust the data they are using.

3

Proposed System Model

In this section, we present an Ethereum blockchain based solution to provide an automated review system for monetization of IoT data using smart contracts by taking motivation from [28] and [29]. Ethereum blockchain platform is used because it is open source, and public users can access it easily. It is a distributed virtual machine with crypto-currency payments and developers are free to execute their smart contracts. Smart contracts eliminate the third party risk associated with exchange of payments in IoT monetization. Proposed system comprises of seven main entities interacting with each other. Main entities of the system include: device owner having an IoT device, MQTT broker [28], review system, smart contracts, arbitrator, users and blockchain database. The process starts with creating a smart contract using solidity language where owner of the IoT device defines set of rules for the sale of IoT device data. Previous customers who had accessed the data, had provided reviews about the data used, which are stored in the blockchain. New customers who want to use this data can interact with the smart contract to get the reviews and ratings along with exchange of payments. Review system is available for the users to analyze the quality of data they want to use. Metadata, reviews and ratings are stored in the blockchain however, because of the security and storage purposes data is stored in MQTT broker. Owner sends data to MQTT broker and making use of review system [29], when any user request the data, reviews and ratings, smart contract is triggered and unique token is provided to the user. Smart contract also provide the customer unique token to the MQTT broker for authentication.

Reputation System for IoT Data Monetization Using Blockchain

3.1

177

System Model Components

The main components of the system model are discussed below. All main components contain Ethereum addresses for identification of their accounts in Ethereum blockchain. The interaction between the smart contracts and main components is as follows: • Owner: At first, IoT device contract is created by the owner of the device. Owner also specifies the rate for each topic in the contract depending on the topic information and length. Owner of device data encrypts the data which is to be stored in MQTT broker and sends private key to the smart contracts. Customers must have to deposit money to smart contracts before making transaction. Rates for the topics are decided in agreement with MQTT broker, because broker has to store them. Successful data access triggers the deposit function and payment is transferred to the ether wallet of the owner. • IoT Device Contract: All the logic and code of the system is written in the device contract. • MQTT Broker: MQTT broker holds the data and provides this data to user on demand. Smart contract provides the customer token to broker so that broker can authenticate the user. • Review System: The javascript code is connected with the web interface using the HTML code. Application binary interface (ABI) and bytecode values of the smart contract are ascribed by the javascript code. Web3 application programming interface is the communication medium between smart contracts and mining server. Functions defined in the javascript makes call to smart contract functions to execute the transactions. When user interact with the web interface, values are passed to the javascript functions as transaction input. Output is displayed when block is mined successfully. When user on the web page calls the function, transaction occurs and added to the block through mining. We have used geth interface to run the mining server. Remote procedure call (RPC) protocol is used by the geth interface. Web page and web browser are connected using Node.js and web page is created using HTML and CSS. User will open the web browser and then connected web page will also open. When a contract is deployed, user will choose the account address and write the review. User who wants to check the existing review, register review or modify review, will trigger the javascript and smart contract functions. • User: Users will interact with the smart contracts, broker and review system and will pay for the data they want to use. User is allowed to subscribe, review, access data, add review, modify review. User is provided with the decryption key to decrypt and use data. • Arbitrator: Arbitrator is a trusted entity selected by the owner and system administration. Arbitrator will get incentive, to act honestly and perform the two specified tasks from the owner. Owner will provide some privileges to arbitrator to download the data and provide reviews to handle dispute and fake reviews (Fig. 1).

178

4

A. Javaid et al.

Smart Contract: Review System for IoT Data Monetization

The proposed blockchain system focusses on keeping and managing reviews, ratings, and metadata about the data stored in MQTT broker. These reviews and ratings are stored in blockchain as a result of the user interactions. After using data, user writes review and rating before leaving the system depending upon his experience. In this Section, discussion about the functions and smart contract is provided. Smart contract code is written, using solidity language. • Add Topic: Firstly, IoTContract is created by the device owner using the constructor of the contract. Owner then identifies the default topics and MQTT broker’s address. Smart contract registers the MQTT broker and topic is added using the Add Topic function. • Deposit: Interested customers who want to access this data will deposit ethers using the Deposit function to the contract. • Subscribe: Before, accessing the topic customer has to subscribe the topic which they want to access. • Is Review Exist: User triggers this function to check whether reviews exist for the data he/she subscribed.

Review System for IoT Data Monezaon MQTT Broker

MQTT Connecon

Heading

MQTT Connecon

Owner Side

JS Services

Request Data

User Owner defines rules

Device Owner

Review System IoT Device

Request and Submit Reviews

Search Register Modify

Ethereum Blockchain

Fig. 1. Blockchain based patient driven interoperability.

Reputation System for IoT Data Monetization Using Blockchain

179

• Get Reviews: Customer will make request to smart contract for the reviews of the subscribed topic. Then this function returns the existing review for subscribed data. Data structure which saves the reviews consists “dt − contents” which is string value. • Get Ratings: Customer will make request to smart contract for the rating of the subscribed topic. Then this function returns the existing rating for subscribed data. Data structure which saves the reviews consists “dt − ratings” which is a rating value for the string. • Access: After, analyzing the review and ratings about data, if user is satisfied he/she can access the topic by calling access function. Smart contract then authenticates, if user has subscription and deposited money. Smart contract will grant access usage time, and unique token to user. The message is broadcasted and broker also have user unique token for user authentication. Every user will have a unique token because contract hashes the concatenation of several variables which include: owner and customer address, topic id, and total number of accesses. User access time for the data depends on the amount of money in his account. When account ethers are utilized, user is not able to access data further. • Get Data Content: If user want to get a copy of data to be downloaded on his system. User will invoke this function and get the data offchain. Smart contract will send decryption key to user, after user has invoked this function. User using this decryption key will decrypt the data. To download data, user must have enough balance in his account. • Update Subscription: When user disconnects or account balance is used, broker will call this function and customer balance is updated according to the usage time or downloaded data. Smart contract then sends the money to the owner account. • Refund : In case, if there are ethers left in the user’s account and user disconnects, then these ethers will be refunded to the user. This function will also be triggered, when arbitrator finds problem in the user downloaded file. • Set Reviews: User and arbitrator will provide reviews after using and analyzing data and is provided with incentive through incentive mechanism.

5

Results and Discussion

In Ethereum blockchain gas is of fundamental importance because it reflects the computational complexity of the transactions having number of different operations. As a result of any request when transaction occurs, smart contracts are executed. In the network at every node instructions are executed and there is a cost for every operation which is expressed as number of gas units. Transaction is the operation which adds something to the blockchain or modifies its state. Transaction cost or gas cost for a transaction depends on the size and complexity of the smart contract. Transaction cost is the cost of sending data to blockchain and is concerned with the transaction’s base cost and contract deployment cost

180

A. Javaid et al. 70000

Execut ion Gas Transact ion Gas

Gas Consum pt ion

60000 50000 40000 30000 20000 10000 0 Access

Get Dat a Set Reviews

Refund

Fig. 2. Gas consumption for smart contract functions Execut ion Gas Transact ion Gas

Gas Consum pt ion

20000

15000

10000

5000

0 Get Rat ings

Updat e

Get Deposit ReviewExist

Fig. 3. Gas consumption for smart contract functions

which are 21000 and 32000 in our case. We can calculate transaction cost of any transaction, if we know the gas cost using Eq. 1 [29]. T otalCostgwei = GasU sed × GasCost

(1)

In Eq. 1 gas used, is the amount of gas used by smart contract or any single operation of smart contract. Gas cost represents the unit price of gas with cryptocurrency unit of gwei. In smart contract cost test we have set the default gas price as 10gwei. Different operations of the smart contract uses different gas amount. Some smart contract function costs are given in Table 1. The data trading contract and data review contract are created once and their costs are $989099 and $90898 respectively. At first, the owner of the IoT device adds the topic to MQTT broker and cost of this transaction is $0.15103. Then customer will perform initial ether deposit and subscribe the topic. The cost for these two operations is $0.03880732 and $0.21558128 respectively. Then customer can access to data and make request for reviews about the data. These two operations have the

Reputation System for IoT Data Monetization Using Blockchain

181

140,000

Gas Amount

120,000

100,000

80,000

60,000

43

8

20

37

String Length as Trx Input Value

Fig. 4. Gas consumption for string input length

Fig. 5. Mining time for string input length Table 1. Cost of functions in Ether and USD. (Functions)

Gas used Cost (Ethers) Cost (USD)

Topic added

92091

0.00092091

Deposit

23663

0.00023663

0.03880732

Subscribe

131452

0.00131452

0.21558128

0.15103

Access

70,000

0.0007

0.1148

Get reviews

24476

0.00024476

0.04014064

Get data content 23115

0.00023115

0.0379086

Refund

18352

0.00018352

0.03009728

Set reviews

59362

0.00059362

0.0973537

cost $0.1148 and $0.04014064 respectively. If user is satisfied about the data he/she will invoke the get data content to download the data and this operation have the cost of $0.0379086. If user did not want to download data money will be refunded to that user when he invokes the refund function. This operation has the cost of $0.03009728. At the end user will set reviews and this operation has total consumption of 59362 gas and actual cost in ethers is $0.00059362 ether, and USD (0.09735368).

182

A. Javaid et al.

Figure 2 shows the result for access, get data content, refund and set review functions. Access function has gained the highest execution and transaction costs as compared to other smart contract functions. The user when invokes the access function smart contracts check whether user has the subscription and provides a unique token to the user. When this user makes other transactions this token is used for authentication of that specific user. Because of these authentications involved with these functions, they have more cost and gas consumption. Figure 3 shows the result for get deposit, get ratings, is review exist and update subscription time functions. Ethereum yellow paper is referred [30] for more information regarding to transaction and execution gas. Figure 4 shows the relationship between review string length and gas used to submit that review. Amount of gas used is increased when string contains more characters. 3200 gas is consumed when review string length is 32 because gas consumption and review length are directly proportional to each other. Figure 5 shows the mining time for string input length. Graph shows that there is no relationship between review length and mining time. At smaller review string length mining time is greater and for larger review string length mining time is smaller. We can conclude from the results that mining time is not related to review string length, it depends on the network conditions and miner’s choice. Customer who has given a detailed review about the data, cost of the transaction gas will be increased, however mining time depends upon the network conditions.

6

Conclusion

IoT devices are becoming everyday part of human life and data produced by these devices is used to improve the modern civilization. In this paper, we have presented blockchain based review system for trading data of IoT devices. This system provides confidence to users, that quality of the data is satisfactory. Trustful automated payments are done using smart contracts by eliminating third party risk. In order to consider data integrity and security at first only metadata is provided and reviews cannot be modified because it is a blockchain based system. All transactions are done through Ethereum smart contracts and there is log of every transaction in the blockchain. The security and immutability of the system is ensured using blockchain based system. The smart contracts are implemented in solidity language and the system is designed using Visual studio code. User interface is developed using HTML and CSS.

References 1. Perera, C., Zaslavsky, A., Christen, P., Georgakopoulos, D.: Sensing as a service model for smart cities supported by internet of things. Trans. Emerg. Telecommun. Technol. 25(1), 81–93 (2014) 2. Xie, R., He, C., Xie, D., Gao, C., Zhang, X.: A secure ciphertext retrieval scheme against insider kgas for mobile devices in cloud storage. Secur. Commun. Netw. (2018)

Reputation System for IoT Data Monetization Using Blockchain

183

3. Fan, L., Lei, X., Yang, N., Duong, T.Q., Karagiannidis, G.K.: Secure multiple amplify-and-forward relaying with cochannel interference. IEEE J. Sel. Top. Signal Process. 10(8), 1494–1505 (2016) 4. Shen, J., Gui, Z., Ji, S., Shen, J., Tan, H., Tang, Y.: Cloud-aided lightweight certificateless authentication protocol with anonymity for wireless body area networks. J. Netw. Comput. Appl. 106, 117–123 (2018) 5. Tao, M., Ota, K., Dong, M., Qian, Z.: AccessAuth: capacity-aware security access authentication in federated-IoT-enabled V2G networks. J. Parallel Distrib. Comput. 118, 107–117 (2018) 6. Chen, J., He, K., Yuan, Q., Xue, G., Du, R., Wang, L.: Batch identification game model for invalid signatures in wireless mobile networks. IEEE Trans. Mob. Comput. 16(6), 1530–1543 (2016) 7. Xu, B., Da Xu, L., Cai, H., Xie, C., Hu, J., Bu, F.: Ubiquitous data accessing method in IoT-based information system for emergency medical services. IEEE Trans. Ind. Inform. 10(2), 1578–1586 (2014) 8. Jiang, L., Da Xu, L., Cai, H., Jiang, Z., Bu, F., Xu, B.: An IoT-oriented data storage framework in cloud computing platform. IEEE Trans. Ind. Inform. 10(2), 1443–1451 (2014) 9. Yu, B., Wright, J., Nepal, S., Zhu, L., Liu, J., Ranjan, R.: IoTChain: establishing trust in the internet of things ecosystem using blockchain. IEEE Cloud Comput. 5(4), 12–23 (2018) 10. Sharma, P.K., Park, J.H.: Blockchain based hybrid network architecture for the smart city. Future Gener. Comput. Syst. 86, 650–655 (2018) 11. Jiang, T., Fang, H., Wang, H.: Blockchain-based Internet of vehicles: distributed network architecture and performance analysis. IEEE Internet Things J. 6, 4640– 4649 (2018) 12. El Kaed, C., Khan, I., Van Den Berg, A., Hossayni, H., Saint-Marcel, C.: SRE: semantic rules engine for the industrial Internet-of-Things gateways. IEEE Trans. Ind. Inform. 14(2), 715–724 (2017) 13. Hossain, M.S., Muhammad, G.: Cloud-assisted industrial internet of things (IIoT)enabled framework for health monitoring. Comput. Netw. 101, 192–202 (2016) 14. Huang, J., Kong, L., Chen, G., Wu, M.Y., Liu, X., Zeng, P.: Towards secure industrial IoT: blockchain system with credit-based consensus mechanism. IEEE Trans. Ind. Inform. 15, 3680–3689 (2019) 15. Xu, Q., Aung, K.M.M., Zhu, Y., Yong, K.L.: A blockchain-based storage system for data analytics in the internet of things. In: New Advances in the Internet of Things, pp. 119–138. Springer, Cham (2018) 16. Hammi, M.T., Hammi, B., Bellot, P., Serhrouchni, A.: Bubbles of trust: a decentralized blockchain-based authentication system for IoT. Comput. Secur. 78, 126–142 (2018) 17. Mateen, A., Javaid, N., Iqbal, S.: Towards energy efficient routing in blockchain based underwater WSNs via recovering the void holes, MS thesis, COMSATS University Islamabad (CUI), Islamabad, Pakistan, July 2019 18. Naz, N., Javaid,N., Iqbal, S.: Research based data rights management using blockchain over ethereum network, MS thesis, COMSATS University Islamabad (CUI), Islamabad, Pakistan, July 2019 19. Javaid, A., Javaid, N., Imran, M.: Ensuring analyzing and monetization of data using data science and blockchain in loT Devices, MS thesis, COMSATS University Islamabad (CUI), Islamabad, Pakistan, July 2019

184

A. Javaid et al.

20. Zainab Kazmi, H.S., Javaid, N., Imran, M.: Towards energy efficiency and trustfulness in complex networks using data science techniques and blockchain, MS thesis, COMSATS University Islamabad (CUI), Islamabad, Pakistan, July 2019 21. Zahid, M., Javaid, N., Babar Rasheed, M.: Balancing electricity demand and supply in smart grids using blockchain, MS Thesis, COMSATS University Islamabad (CUI), Islamabad, Pakistan, July 2019 22. Noshad, Z., Javaid, N., Imran, M.: Analyzing and securing data using data science and blockchain in smart networks, MS thesis, COMSATS University Islamabad (CUI), Islamabad, Pakistan, July 2019 23. Ali, I., Javaid, J., Iqbal, S.: An incentive mechanism for secure service provisioning for lightweight clients based on blockchain, MS thesis, COMSATS University Islamabad (CUI), Islamabad, Pakistan, July 2019 24. Khan, J.H., Javaid, N., Iqbal, S.: Blockchain based node recovery scheme for wireless sensor networks. MS thesis, COMSATS University Islamabad (CUI), Islamabad, Pakistan, July 2019 25. Samuel, O., Javaid, N., Awais, M., Ahmed, Z., Imran, M., Guizani, M.: A blockchain model for fair data sharing in deregulated smart grids. In: IEEE Global Communications Conference (GLOBCOM), July 2019 26. Rehman, M., Javaid, N., Awais, M., Imran, M., Naseer, N.: Cloud based Secure Service Providing for IoTs using Blockchain. In: IEEE Global Communications Conference (GLOBCOM) (2019) 27. Awais, M., Javaid, M., Imran, M.: Energy efficient routing with void hole alleviation in underwater wireless sensor networks, MS thesis, COMSATS University Islamabad (CUI), Islamabad, Pakistan, July 2019 28. Suliman, A., Husain, Z., Abououf, M., Alblooshi, M., Salah, K.: Monetization of IoT data using smart contracts. IET Netw. 8, 32–37 (2018) 29. Park, J.S., Youn, T.Y., Kim, H.B., Rhee, K.H., Shin, S.U.: Smart contract-based review system for an IoT data marketplace. Sensors 18(10), 3577 (2018) 30. Wood, G.: Ethereum: a secure decentralised generalised transaction ledger. Ethereum Proj. Yellow Pap. 151, 1–32 (2014)

Blockchain Based Balancing of Electricity Demand and Supply Maheen Zahid, Ishtiaq Ali, Raja Jalees Ul Hussen Khan, Zainib Noshad, Atia Javaid, and Nadeem Javaid(B) COMSATS University Islamabad, Islamabad, Pakistan [email protected], [email protected], [email protected], [email protected], [email protected], [email protected]

Abstract. The growing number of Renewable Energy Sources (RES) in the energy system provides new market approaches according to price and decentralized generation of electricity. Local market, in which consumers and prosumers can trade locally by generation of electricity through RES directly within their community. This approach creates a balance between generation and consumption in a decentralized manner. In this paper, a distributed technology of Blockchain is used, which highlights the decentralized nature of local market. It provides a decentralized market platform for trading locally without the need of central intermediary through Periodic Double Auction (PDA) mechanism. With the introduction of Smart Grid (SG) systems, there have been improvements in how utility companies interact with customers with regards of electricity usage. However, there is a tendency for the data of users to be compromised in SG. In this proposed system, users are able to do trading through PDA and get access of their own previous history. The blockchain provides transparency, traceability and is utilized to mitigate the above mentioned problems. Smart contacts, are used to exclude the third party to provide a transparent system between users in the network. Keywords: Smart Grid Double Auction

1

· Blockchain · Electricity trading · Periodic

Introduction

Electricity plays a vital role in the development of the latest technologies. It is the basic necessity for the economic development of any country. It is helpful in various areas: transportation, education, banking and computing etc and it enables several revolutions. The integration of Information and Communication Technology (ICT) in the traditional grid it becomes Smart Grid (SG). SG is an emerging technology of this modern era [1]. It provides the facility of bi-directional communication between the utility and its consumers. It introduces the recently advanced energy generation ways such as solar panels and charging/discharging of Electric Vehicles (EV) [2]. Smart Home (SH) is an important entity to make c Springer Nature Switzerland AG 2020  L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 185–198, 2020. https://doi.org/10.1007/978-3-030-33506-9_17

186

M. Zahid et al.

any city smart. The SH directly communicates with SG through Smart Meter (SM) to get electricity according to its demand. SG enables consumers to properly manage their electricity consumption through the Home Energy Management System (HEMS) [3]. The data recorded from the SM is transmitted directly to SG in real-time through communication medium. SM’s data is stored in the SG database and used for billing and also available for research purpose [4,5]. However, SG is a centralized system and it does not provide data access to the users. Furthermore, the data stored in the SG systems may easily be tempered by malicious attacks, which tend to increase the amount of electricity bill without knowing the consumer and utility companies. Users do not know why they are paying huge amount of bills. There are also many issues facing by consumer and SG about users data such that: traceability, authorization, immutability, security of data and single point of failure. To solve the above aforementioned issues, a new emerging technology of blockchain is introduced. A scientist named Satoshi Nakamoto gives the idea of “Bitcoin” digital currency, a form of electronic cash in 2008. A concept of Peer-to-Peer (P2P) online trading of digital currency is introduced without the need of an intermediary party [6]. For electronic payments on the Internet, a trusted third party is required as a financial institution. Moreover, for P2P trading digital signatures are used as an initiator to verify the existence of both parties. The cryptographic proof is necessary to build trust that shows the willingness and trustworthiness of two parties. In this article, Satoshi proposed the system of P2P trading. However, this concept is beneficial for online trading, when any company or enterprise wants to use it practically. So, to implement this concept in a real-time environment, an idea proposed about the innovative technology [7] named “Blockchain” in 2015. According to this, Blockchain technology changed the vision of the business model all over the world. It is an innovative disruption in the world of computers, social networks. It is defined as structural database for decentralized storage of data [8]. A block is a collection of data containing related information known as transactions [9]. A timestamp shows the creation time of a block, when many transactions are done and after that, a block is generated. A block comprises of header and body. A header contains hash of the previous block, through which the current block is combined with the previous one and current block knows about its previous block through its hash. The body contains all the transactions and their hashes. Once a block is a part of the chain then, it is very difficult to be tempered. When block is created and validated then, it becomes the part of the chain due to its hash. Blockchain is classified into three types: public, private and consortium blockchain. Public is open, anyone can be a part of this. Private has some restrictions so, only authorized nodes are the part of the network. Consortium blockchain contains half features of public and half features of private. Its data is publicly available to users for read-only, while the write access is available to just authorized nodes [10]. Different characteristics of these three types of Blockchain are shown in Table 1.

Blockchain Based Balancing of Electricity Demand and Supply

187

Different consensus mechanisms are also used for the validation of blocks and after consensus the block becomes the part of the chain. There are different kinds of consensus mechanisms which are as following: Proof of Work (POW), Proof of Stake (POS), Proof of Identity (POI) and Proof of Authority (POA) [10]. The base of all other types of consensus mechanisms are POW. Table 1. Characteristics of blockchain types Type

Openness

Decentralization Write

Read

Public

Anyone

Complete

Everyone

Private

Specific individuals Partially

Consortium Specific groups

Partially

Any one

Specific nodes Specific Specific nodes Anyone

Now-a-days Blockchain is used in different paradigms: in health monitoring, data sharing, getting feedback from users, decentralized trading. To minimize the chances of single point failure, unauthorized persons are not the part of any network to prevent malicious attacks. In this paper, the solution of blockchain based different smart contracts are proposed that creates a system of P2P trading through Periodic Double Auction (PDA). Authorized users are just able to get access to their own data history. Due to immutability, the users can easily see their original record file and this data is not altered by anyone. Different smart contracts are used to eliminated the third party concept. It is beneficial for users to prevent them to pay the huge amount of taxes in their bills. Due to the deployment of smart contracts, the users pay a small amount of ethers for the transaction. Section 2 described and discussed the related work. In Sect. 3, provided the implementation details of the proposed model. Section 4 summarized the main finding of this paper.

2

Related Work

In [11], Blockchain based P2P energy trading platform is proposed for efficient transaction of power between prosumers and consumers. The suggested platform is used two different types of P2P trading scenarios: one is Pure P2P and the other is Hybrid P2P. In Pure P2P trade, energy is used as a transaction item and in Hybrid P2P, energy tag is used as a transaction. A Tag is assigned to block for validation and transaction. The production of electrical energy from different domains are divided into ten different categories. Some of them are consumer-oriented and some of them are supply-oriented. In [12], the authors presented the concept of sovereign Blockchain technology, which prevents data tempering from any malicious source. It also maintained the records of tempered data in side Blockchain. Consumers send their requests of electricity according to their demand to SG through SM. The users monitored

188

M. Zahid et al.

their usage and also know about how much electricity they demand. If the tempering happened then, the record of this tempered data is also kept in side chain of the sovereign Blockchain. Authors in [13], discussed the idea of distributed trading of electricity in which nodes are used as: consumers and prosumers for P2P trading. They proposed a model consists of two layers: one layer is Multi-Agent System (MAS) used for the sustenance of prosumer side information of generation of electricity. MAS enabled users to negotiate the prices and form coalition. The second layer showed the secure and trustworthy trading through transactions between MAS and Social Coordination Agent (SCA). The trading is secure due to Blockchain based settlement system in which the double-chain and high-frequency verification mechanism worked parallelly, which helped to make trading transparent. In every negotiation system, data is chained one by one and stored in the first chain. The high-frequency mechanism are used to detect any malicious activity occurred between contract and ledger. In this paper [14], the authors proposed the concept of Federated Power Plants (FPP) and Virtual Power Plants (VPP). VPP made P2P transactions through self-organizing users. They also discussed the incentive mechanism during trade between prosumers and consumers. Two different types of strategies are proposed: Coordination between Distributed Energy Resources (DER) into VPP and P2P energy trading platform. Noor et al. [15] worked on the Demand Side Management (DSM) to improve the reliability of the whole system. A game theoretic approach is used in simple DSM model, which does not only minimize the peak-to-average ratio in SG. It also reduced the dips in a load profile of users. Blockchain technology implemented a DSM system to make P2P trading system secure. In [16], the maintenance and skillful usage of natural resources are essentially related and beneficial for utilization of electricity. The use of the enormous powers, i.e., coal, water, daylight, geothermal resources, wind and gravitational forces for electricity generation. In [17], the authors proposed a framework, which contains seven components. These components are used to implement private Blockchain in Brooklyn case study. In this paper, the authors introduced the merged concept of MG and Blockchain to introduced MG energy market. Homes are acting as prosumers and they generate electricity through PV. The physical layer is showing the infrastructure of MG. Virtual layer displays the price and trading mechanism. Regulation layer tells about the government policies, taxes and rules. In [18], the authors addressed the issue of Demand Response (DR) with a decentralized system and encouraged the consumers use less energy consumption in on peak hours. The authors proposed the framework of Distributed Load Balancing Trade Framework (LBTF) with two different schemes: Utility-Grid contract and MG contract. Consortium Blockchain is used and POW consensus algorithm is public key encryption, digital signatures and hash functions are used to maintain security of users’ information.

Blockchain Based Balancing of Electricity Demand and Supply

189

Pop et al. [19] used a Blockchain based model for storing, managing and validation of DR in low or medium voltages of SGs. Blockchain stored all that data in P2P distributed ledger network which is collected from IOT smart metering devices in a secure manner which is called as temper proof mannered. Authors implemented a new smart contract named as a self-enforcing smart contract which is used for checked out and tracked each Distributed Energy Prosumer (DEP) profiles. All DEPs are enrolled in DR program. Their penalties or rewards are calculated here and it also detects the unbalanced grid energy for DR events. In [20], the authors described the P2P energy trading system using coalition formation method. Surplus energy and depicted energy MGs are also able to communicate directly with each other. They know about each other state and if they want to share electricity, they trade easily with each other. A Blockchain-based system provides this facility to maintain their state record, whether they want to sell or buy electricity to fulfill the demand of their consumers. Blockchain provided secure transactions and a consensus mechanism allows every valid transaction to add in a block. The utilities realized their role in the power systems. When the consumers’ need a power supply to fulfill their electricity demands. They directly contact utilities. The authors in [21] described the responsibilities of utilities. They collect domestic communities information because there is a rapid increase in the usage of electricity. These modern systems fully spread the power markets and also increased the accessibility of decentralized renewable power production. However, the utilities are able to modernize their business models and support SG markets by proficient knowledge. Wang et al. proposed the system of transactions between MGs through Blockchain and they are using the continuous double auction mechanism for trading [22]. The authors proposed the concept of Unspent Transaction Out (UTXO) model. In this system, the authors used Continuous Double Auction (CDA) with Blockchain technology parallelly to achieved low-cost transactions and transparent data of MG. Satoshi is used as a digital currency which is the sixteenth part of the bitcoin. It is also used as a token. The mechanism of trading of electricity and transfer of tokens are very helpful. MGs can sell/ buy their energy with each other. MGs are able to fulfill the requirements of their consumers. In this system, the unique data structure of Blockchain confirmed the security of data. However, in this system, the authors ignored the power fluctuation in the main SG. In [23], the authors discussed the issues of management and control of sustainable energy forms. To solved these problems, blockchain implemented with energy through the internet and gives the new concept of energy internet. It consists of renewable energy generation, Energy Storage Devices (ESD) and internet is used for connectivity between them. The energy internet involved various energy forms and different participants. The main contributions are to introduce the compatibility of energy internet and blockchain technology. Blockchain is implemented in many companies and is helpful to provide decentralized applications, however, the excessive power usage is considered. It is just suitable for

190

M. Zahid et al.

some small communities and the practical implementation on a huge commercial level needs more resources. Smart Community (SC) is a necessary part of the Internet of Energy (IoE) which connected all the RES, SG and EV. Permissioned blockchain is used on the basis of smart contracts for secure and private communication. The authors proposed reputation based Delegated Byzantine Fault Tolerance (DBFT) consensus algorithm for energy trading. Users also take electricity from traditional SG or electricity generated through RES, it depends on EV user. Furthermore, this system is designed for SGs, however, prosumers are not considered to participate in it. In [24], authors proposed the privacy preserving mechanism through private blockchain. Authors in [25,26] proposed incentive mechanism and repudiation Table 2. Summary of related work of blockchain Proposed models

Achievements

Limitations

[10] Energy Internet with blockchain

Reduced costs

Reliability and excessive power consumption

[11] LEM

Short term electricity trading, transparent

High Energy Consumption

[12] Grid Monitoring System

Customer utility control

Storage Capacity is small

[13] MAS system

Efficient negotiation System

Interruption of third party

[14] FPP platform

Confidential coordination between VPPs

Cost consumption is very high

[15] Game theoretic model Reduce peak-to-average ratio and reduced dips

Storage capacity is less

[16] Brooklyn MG

Scalability and robustness Cost is high

[17] LBTF

Privacy and integration of Computation time stored data between nodes is very high

[18] Decentralized DR system

DSM system

[20] P2P using distributed Trust and robustness coalition formation method

Multi stakeholder market Specific for only two MGs

[22] UTXO

Low cost transactions and Ignored the power transparency fluctuation in the main SG

[23] DBFT

Energy trading

[36] Pure P2P and hybrid P2P

Invalid transaction, energy Privacy and no access loss, cost efficient control of users

[37] POC

Saved the labor cost, minimized the human interaction

Third party is involved

Fixed prices for users

Blockchain Based Balancing of Electricity Demand and Supply

191

system through blockchain for data storage of IoT devices on cloud edge network. In [27–29] authors used blockchain for data trading and store health records of patients on IPFS. Authors in [30–32] performed node recovery in wireless sensor networks and provide secure communication in crowd sensor networks through blockchain (Table 2). 2.1

Problem Statement

Home Energy Management Controller (HEMC) is a centralized system which monitors and controls the home appliances. A consumer requests its electricity demand through SM to SG which can reveal the consumption behavior in front of third party [33]. In such a way, all the data of users can be easily hacked and the malicious person can extract the behavior of users about the routine of the SH dwellers or inhabitants. Authors in [34] proposed cloud based systems to save the large amount of data. SG uses cloud to store vast amount of the information of users. They send their requests of electricity demand and get early responses of their requests. However, the communication between users and cloud is not secured. The record maintained about users detail cannot be seen by users. Users are not able to get access of their own profile history, i.e., demand history of electricity. So, any malicious person can hack the cloud server and can change the users data. In result, the users received huge amount of bills. SG is a single entity to fulfill the electricity demand of consumers. In [36], the authors provide the facility of P2P trading through DSO between users to divide the load of SG. However, authorization of users is not considered. Anonymous users can also become the part of the trading. Moreover, the users are not able to know about the available amount of electricity from DSO. In [37], authors used Blockchain for decentralized trading. However, the users purchase electricity on the defined prices of utility which is considered as one sided market. Users pay taxes to third party which acts as a communication link between prosumers and consumers. The data of users is maintained by DSO. However, there is also an issue of data integrity and confidentiality. In this paper, the Blockchain is implemented for decentralized and P2P trading between different consumers and prosumers. Authorized users can just participate through PDA [22] and do negotiation on prices. They also get access their own previous history of electricity (selling and buying). If an unauthorized user wants to change the data in Blockchain, or wants to add any malicious block in the chain the hash of the block will be changed and a new block is generated. Due to distributed ledger, P2P decentralized trading and immutable nature, the existing block of data remains same.

3

Proposed System Model

This section describes the methodology adopted for the study. The description of the system model represents in Fig. 1 is given below. This proposed model is designed by taking motivation from figs in [37].

192

M. Zahid et al.

Maintain History

Monitors User Data

SmartGrid Save Data

Access History

User

Market Mechanism 3.

2.

Market Layer

Information Layer

Authentication

Registration

User

1.

Profile Login

User

User Layer

Smart Meter

PV System

Consumer

Prosumer

Two Way Communication

One Way Communication

Fig. 1. System model

Access History

Smart Grid

Blockchain Based Balancing of Electricity Demand and Supply

3.1

193

User Layer

The user layer contains all the entities who purchase electricity from the market for the beginning and running of processes that are necessary for routine works. Prosumers and consumers can directly communicate with each other without any central party. They can share their information through blockchain either they want to sell electricity or they need to buy energy from any prosumer. There is a small community in which the registered users are able to do trading. All the information about available electricity sell and buy can be seen by authorized users. A home which has PV panel for electricity generation it acts as a prosumer. When a home has surplus energy after its own usage, then it is able to sell this surplus energy to those homes which are energy deficient. All those homes which purchase electricity to fulfil their electricity demand called consumers. This layer connects with information layer and users provide some information to get registered themselves for become the part of the network. 3.2

Information Layer

It is a second layer of the system model. It comprises of two parts: registration and authentication. When a user wants to be a part of this system then, he must register himself through interface by providing some necessary information his name, password and email address. In addition to this, the user gets its specific unique ID and every user has its own profile. All the information of user is saved on Blockchain and the data stored in the form of hash in block. In this paper, users are firstly registered themselves and after the authentication they get access to see the history of their previous usage. When user enters its unique ID and password, the hash will be calculated and matched with the copy of the existing hash which is already saved in Blockchain. If the user can enter wrong ID or password, then the hashes are different and it does not matched with the already calculated hashes. The system shows a message or give pop up notification, your ID or password is incorrect. The users demand and generation of electricity measured from SM details either this user have surplus amount of energy or energy deficient. On the basis of this information, the deficient and surplus energy details saved in a blockchain and it is a distributed ledger. Every user in the blockchain network have the record of all the transactions happened. The users demand and generation of electricity measured from SM details either this user has excess amount of energy or needs energy to buy. On the basis of this information, the deficient energy and surplus energy details saved in a blockchain and it is a distributed ledger. Every user in the network have the record of all the transactions done in the blockchain. 3.3

Market Layer

First of all, the consumer gets login from his unique ID and after that, the consumers can place their order in an open market according to their requirement

194

M. Zahid et al.

and also their suggested prices for buying and selling. Prosumers give a response to that consumer who submit its required demand of electricity to buy. The PDA helds for price negotiation between consumer and prosumer. The price auction occurs within a specific time slot for one bid. When bidding is done, then trading occurs. A message is broadcasted in blockchain to SG provides this amount of electricity to that consumer. Every user is known by its special account address in blockchain. It provides the required quantity of electricity to that specific consumer. The transaction cost is deducted for double auction and for the transfer of electricity from the consumer account. All the data of sell/purchase is maintained in blockchain through hashes. It is a distributed ledger which maintains all the records of every transaction in a block. A block creates when all the transactions are added in it for single bidding. A single block can store more than one transaction and it depends upon the consensus mechanism is using in the system. After the generation and validation of a block, it becomes the part of the chain. The users can also be able to see the history of their own usage of energy. The history contains all the information of user details i.e., meter id, amount of electricity demand, electricity sell, electricity buy and timestamp. SG maintains all the history record of every user. Consumer and prosumer can access their own history of sell and purchase through profile. They can also see that how much extra quantity purchased from prosumer, how much electricity consumed. Blockchain provides transparency and immutability. In this layer, blockchain used to make the system secure and users access their own history. SG acts as a prosumer here. SG is connected physically with all the consumers. So, when the bidding is going to be happen between prosumers and consumers, for transfer of energy to that specific consumer, then Prosumer sends message to SG to transfer electricity to a specific consumer. Every prosumer provides its surplus energy to SG. After the authentication, then user can also be able to do trading and also get access their own data: how much electricity sells at which time and at which price. It checks all the buying and selling transactions. When all the transactions are done then a block is generated and after the validation of a block, it becomes the part of the chain. Hashes of transactions are calculated. These hashes are used to verify about the ownership of sender through its address. Every node in a network has its specific address and all the nodes in the network are anonymous. So, no one can see the name or other details of any user, the address of the account is just visible. For energy trading within a small community, electricity is generated through PV panels and the PDA mechanism is used. PDA used closed order book system for trading. A particular clearing price is obtained for every single time slot t. In this specific time slot, one bidder could be entertained. Consumer can send their demand of particular amount of electricity and price into the market for bidding. When prosumers received the order, then they reply their own prices and quantity for bidding. After that, the prices are matched between both of them. Then, prosumer sends encrypted message through its SM to SG, provide this specific amount of electricity to this specific consumer. SG decrypts this message through the public key of user. The price of per unit of sell/ purchase of electricity is followed by the tariff price given by utility market. Prosumers cannot send their

Blockchain Based Balancing of Electricity Demand and Supply

195

surplus energy to consumers below the minimum price limit of tariff. Consumers cannot purchase the electricity per unit more than the maximum price limit already defined by the utility. Electricity cannot be sell or buy from SG in this local market. SG is also acts as a prosumer here because it is physically connected to every SM to provide electricity.

4

Simulations and Results

In this section, the implementation details is based upon an open source platform of Ethereum blockchain, which inherits some technical features and block creation after validation. Market mechanism and authentication of users by RBACS are implemented through smart contracts which are written in solidity language. The specifications for implementation setup are: intel(R) core (TM) m3-7Y30 [email protected] GHz, 8 GB RAM, 64 bit Operating system and X64-based processor. The programming language is solidity to write smart contract. Javascript using for front user interactive form. The tools used to develop this system are following: Visual Studio Code: It is an open source editor designed by Microsoft for different windows and operating systems. It is used for code compilation and also supports java script for user interface. Ganache: It is a virtual emulator which has ten addresses of Ethereum accounts and each account contains 100 ethers. It is used as a wallet to test the smart contracts, to run test and to execute different commands. Meta mask: It is the extension which is added in a browser to create connectivity between ganache and smart contract for transactions. For simulations, python language is used and Spyder is used as a tool to perform all of these results are showing here. Different graphs are showing about the transaction cost and execution cost of various smart contracts. Transaction cost is the total gas consumption of sending data to the blockchain [6]. Transaction gas depends on four things: Base cost of a transaction is 21000 gas. Base fee is the cost of an operation which retrieve the sender address from the signature. The minimum deployment cost of the contract is 32000 gas, zero byte data or cost for a code of transaction and the cost of nonzero byte of data or the cost for a code of transaction. Execution cost is also included in a transaction cost. Execution gas is the cost which depends on the cost of computation operation of every line of code in a function. Basically, the execution cost is the everything cost which is used as a runtime cost used for the calling of single method or function. Figure 2 shows the execution and transaction costs of the contract to deploy and calling of every function. The contract is used to do double auction at which the global variables call. This graph represents the difference in transaction cost and execution of every function. The total transaction and execution gas consumed for the deployment of these contract is $0.8541 and $0.4310, respectively. The different bars show the different amount for execution and transaction gas. Transaction cost of every function is high because it contains execution cost of that function and also it contains the deployment of every function in smart contact (Fig. 3).

196

M. Zahid et al.

Fig. 2. Double sided auction

Fig. 3. Data access

5

Conclusion and Future Work

Blockchain is used to do decentralized trading. Through PDA mechanism, users are able to purchase electricity according to their own suggested prices. In this proposed work, authorized users are just able to do trading and also achieved immutability, decentralization and data security. In this paper, users are also able to see their own history (buy and sell electricity). Authentic users are just able to do trading and access their history through interface. Smart contracts are used to remove third party. PDA is used to do negotiation in prices and purchase electricity according to demand. Gas consumption prices are also analyzed through calculation of transaction cost and execution cost of smart contracts. The challenge of this work is limited amount of storage. In future, decentralized storage is used to store large amount of data on it.

Blockchain Based Balancing of Electricity Demand and Supply

197

References 1. Newbery, D., Strbac, G., Viehoff, I.: The benefits of integrating European electricity markets. Energy Policy 94, 253–263 (2016) 2. Manshadi, S.D., Khodayar, M.E., Abdelghany, K., Uster, H.: Wireless charging of electric vehicles in electricity and transportation networks. IEEE Trans. Smart Grid, to be published. http://ieeexplore.ieee.org/document/7837718/ 3. Andraeand, A.S.G., Edler, T.: On global electricity usage of communication technology: trends to 2030. Challenges 6(1), 117–157 (2015) 4. Kabalci, Y.: A survey on smart metering and smart grid communication. Renew. Sustain. Energy Rev. 57, 302–318 (2016) 5. Salahuddin, M., Alam, K.: Information and communication technology, electricity consumption and economic growth in OECD countries: a panel data analysis. Int. J. Elect. Power Energy Syst. 76, 185–193 (2016) 6. Nakamoto, S.: Bitcoin: a peer-to-peer electronic cash system (2008) 7. The promise of the blockchain: the trust machine. http://mt.sohu.com/20161021/ n470943606.shtml. Accessed 10 May 2019 8. Yan, Y., Zhao, J.H., Wen, F.S., Chen, X.Y.: Blockchain in energy systems: concept, application and prospect. Electr. Power Constr. 38(2), 12–20 (2017) 9. Szczerbowski, J.J.: Transaction costs of blockchain smart contracts. Law Forensic Sci. 16, 2 (2018) 10. Wu, J., Tran, N.: Application of blockchain technology in sustainable energy systems: an overview. Sustainability 10(9), 3067 (2018) 11. Sikorski, J.J., Haughton, J., Kraft, M.: Blockchain technology in the chemical industry: machine-to-machine electricity market. Appl. Energy 195, 234–246 (2017) 12. Gao, J., Asamoah, K.O., Sifah, E.B., Smahi, A., Qi Xia, H., Xia, X.Z., Dong, G.: Gridmonitoring: secured sovereign blockchain based monitoring on smart grid. IEEE Access 6, 9917–9925 (2018) 13. Luo, F., Dong, Z.Y., Liang, G., Murata, J., Xu, Z.: A distributed electricity trading system in active distribution networks based on multi-agent coalition and blockchain. IEEE Trans. Power Syst. 34, 4097–4108 (2018) 14. Morstyn, T., Farrell, N., Darby, S.J., McCulloch, M.D.: Using peer-to-peer energytrading platforms to incentivize prosumers to form federated power plants. Nature Energy 3(2), 94 (2018) 15. Noor, S., Yang, W., Guo, M., van Dam, K.H., Wang, X.: Energy demand side management within micro-grid networks enhanced by blockchain. Appl. Energy 228, 1385–1398 (2018) 16. Khaqqi, K.N., Sikorski, J.J., Hadinoto, K., Kraft, M.: Incorporating seller/buyer reputation-based system in blockchain-enabled emission trading application. Appl. Energy 209, 8–19 (2018) 17. Mengelkamp, E., G¨ arttner, J., Rock, K., Kessler, S., Orsini, L., Weinhardt, C.: Designing microgrid energy markets: a case study: the Brooklyn microgrid. Appl. Energy 210, 870–880 (2018) 18. Inayat, K., Hwang, S.O.: Load balancing in decentralized smart grid trade system using blockchain. J. Intell. Fuzzy Syst. 1-11 19. Pop, C., Cioara, T., Antal, M., Anghel, I., Salomie, I., Bertoncini, M.: Blockchain based decentralized management of demand response programs in smart energy grids. Sensors 18(1), 162 (2018)

198

M. Zahid et al.

20. Thakur, S., Breslin, J.G.: Peer to peer energy trade among microgrids using blockchain based distributed coalition formation method. Technol. Econ. Smart Grids Sustain. Energy 3(1), 5 (2018) 21. Green, J., Newman, P.: Citizen utilities: the emerging power paradigm. Energy Policy 105, 283–293 (2017) 22. Wang, J., Wang, Q., Zhou, N., Chi, Y.: A novel electricity transaction mode of microgrids based on blockchain and continuous double auction. Energies 10(12), 1971 (2017) 23. Di Silvestre, M.L., Gallo, P., Ippolito, M.G., Sanseverino, E.R., Zizzo, G.: A technical approach to the energy blockchain in microgrids. IEEE Trans. Ind. Inf. 14(11), 4792–4803 (2018) 24. Samuel, O., Javaid, N., Awais, M., Zeeshan, A., Imran, M., Guizani, M.: A blockchain model for fair data sharing in deregulated smart grids. In: IEEE Global Communications Conference (GLOBCOM 2019) (2019) 25. Rehman, M., Javaid, N., Awais, M., Imran, M., Naseer, N.: Cloud based secure service providing for IoTs using blockchain. In: IEEE Global Communications Conference (GLOBCOM 2019) (2019) 26. Ali, I., Javaid, N., Iqbal, S.: An incentive mechanism for secure service provisioning for lightweight clients based on blockchain, MS thesis, COMSATS University Islamabad (CUI), Islamabad, Pakistan, July 2019 27. Javaid, A., Javaid, N., Imran, M.: Ensuring analyzing and monetization of data using data science and blockchain in loT devices, MS thesis, COMSATS University Islamabad (CUI), Islamabad, Pakistan, July 2019 28. Naz, M., Javaid, N., Iqbal, S.: Research based data rights management using blockchain over ethereum network, MS thesis, COMSATS University Islamabad (CUI), Islamabad, Pakistan, July 2019 29. Kazmi, H., Zainab, S., Javaid, N., Imran, M.: Towards energy efficiency and trustfulness in complex networks using data science techniques and blockchain, MS thesis, COMSATS University Islamabad (CUI), Islamabad, Pakistan, July 2019 30. Khan, R.J.H., Javaid, N., Iqbal, S.: Blockchain based node recovery scheme for wireless sensor networks, MS thesis, COMSATS University Islamabad (CUI), Islamabad, Pakistan, July 2019 31. Mateen, A., Javaid, N., Iqbal, S.: Towards energy efficient routing in blockchain based underwater WSNs via recovering the void holes, MS thesis, COMSATS University Islamabad (CUI), Islamabad, Pakistan, July 2019 32. Noshad, Z., Javaid, N., Imran, M.: Analyzing and securing data using data science and blockchain in smart networks, MS thesis, COMSATS University Islamabad (CUI), Islamabad, Pakistan, July 2019 33. Shakeri, M., Mohsen Shayestegan, S.M., Reza, S., Yahya, I., Badariah Bais, M., Akhtaruzzaman, K.S., Amin, N.: Implementation of a novel home energy management system (HEMS) architecture with solar photovoltaic system as supplementary source. Renew. Energy 125, 108–120 (2018) 34. Bukhsh, R., Javaid, N., Khan, Z.A., Ishmanov, F., Afzal, M., Wadud, Z.: Towards fast response, reduced processing and balanced load in fog-based data-driven smart grid. Energies 11(12), 3345 (2018) 35. Liu, Y., Wu, L., Li, J.: Peer-to-peer (P2P) electricity trading in distribution systems of the future. Electr. J. 32, 2–6 (2019) 36. Park, L., Lee, S., Chang, H.: A sustainable home energy prosumer-chain methodology with energy tags over the blockchain. Sustainability 10(3), 658 (2018) 37. Mengelkamp, E., Notheisen, B., Beer, C., Dauer, D., Weinhardt, C.: A blockchainbased smart grid: towards sustainable local energy markets. Comput. Sci.-Res. Dev. 33(1–2), 207–214 (2018)

Data Replication Based on Cuckoo Search in Mobile Ad-Hoc Networks Takeru Kurokawa and Naohiro Hayashibara(B) Kyoto Sangyo University, Kyoto, Japan {i1986061,naohaya}@cc.kyoto-su.ac.jp

Abstract. Mobile ad-hoc network consists of a collection of mobile devices that are interconnected each other. It has attracted attention in various areas such as information sharing and intelligent transportation. In particular, information sharing on MANET in an emergency is crucial for victims. In this situation, information on relief supplies and safety confirmation should be replicated to keep it on the network as long as possible. On the other hand, the cost of replicated information should be taken into account since the storage of each mobile device is limited. In this paper, we proposed a data replication protocol based on the cuckoo search that is a meta-heuristic algorithm inspired by the egg-laying habits of cuckoos to improve the availability of low demand data in MANET. It realizes the space-efficient data replication without a significant impact on data availability.

1

Introduction

Mobile ad-hoc network (MANET) is widely used by many applications such as the information sharing system in an emergency [10], indoor location tracking [1] and intelligent transportation in a smart city. Now, we suppose the information sharing in the case of emergency (e.g., disaster). In this case, there is no network infrastructure available. So, MANET that consists of mobile phones is an alternative to a network infrastructure (e.g., wifi access points) for sharing information among victims. Victims in a shelter upload information on disaster relief supplies, medical support, livelihood support, and confirmation of individual safety. The number of replication of data is often in proportion to the number of requests in most of replication approaches, for example, Path replication [3] and Owner replication [8]. Thus, data that are frequently required by users create many replicas. So, data on disaster relief supplies and medical, livelihood support are required by many people and replicated actively. Contrary to this, data with low demand creates the small number of replicas. This type of data could be disappeared easily in MANETs by node’s leaving or churn. On the other hand, those data could be important for specific individuals, for example, the safety information on family and relatives. In general, the number of replicas is correlated with the availability of the data. The availability of the data increases as the number of its replicas increases. c Springer Nature Switzerland AG 2020  L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 199–209, 2020. https://doi.org/10.1007/978-3-030-33506-9_18

200

T. Kurokawa and N. Hayashibara

Meanwhile, the cost of the storage of each node becomes high in this case. Thus, there exists a tradeoff between data availability and storage cost. Now, we assume a mobile ad-hoc network that consists of mobile devices and information is shared on it in the case of emergency (e.g., disaster). In this case, we need to take into account the storage cost in each device because it is strictly limited. We also make low demand (but still important) information available on the network as long as possible at the same time. For this purpose, we propose a novel approach for data replication on MANETs based on Cuckoo search. Cuckoo search [13] is a meta-heuristic algorithm inspired by the egglaying habits of cuckoos. It is known as an efficient solution for nonlinear global optimization problems, and it is proven that it guarantees the global convergence by using L´evy flight [5]. In this paper, we use Cuckoo search for allocating the replicated data to nodes in MANET. Our approach is based on the protocol proposed in [7], which is a replication protocol to improve the availability of low demand data. We extend the protocol using Cuckoo search for properly allocating replicated data. It improves the efficiency of the storage cost without a significant reduction of data availability. We also measure the availability of data, the storage cost, the hit ratio of search queries, and the average distance to data while searching on the proposed protocol, and compare with other protocols.

2

Related Work

First, we introduce some of the replication protocols in MANETs and P2P systems. Then we explain Cuckoo search for solving optimization problems. 2.1

Replication in MANETs and P2P Systems

Many data replication protocols in MANETs and P2P systems have been developed so far. Owner replication [8] is a simple replication protocol which replicates data to the node that issues a search request for it. It means that one replicated data corresponds to the search query. Thus, the cost of replicating data is low. However, this protocol may require a lot of messages for searching for data. Path replication [3] creates replicas of the data and allocates them to all nodes along the path from the requesting node to the providing node in a search query for it. This protocol reduces the search traffic (i.e., the number of messages) by a factor of three compared to Owner replication. On the other hand, it imposes higher storage cost than Owner replication because of the larger number of replicated data. Kageyama and Shibusawa proposed the replication protocol for improving the availability of low demand files in super-node based P2P systems [7]. It replicates the high demand files using Owner replication. In addition, it computes a demand forecast for low demand files and decides the number of replicas to be created.

Data Replication Based on Cuckoo Search in Mobile Ad-Hoc Networks

2.2

201

Cuckoo Search

Cuckoo search proposed by Yang and Deb [13] is a metaheuristic algorithm inspired by the brood parasitic behavior of the common cuckoo species. They lay their eggs in the nest of other species of birds. A cuckoo egg closely mimics the eggs of the host ones. The host birds may incubate and hatch the cuckoo egg, or they may recognize the intruding egg and abandon the nest. We now explain Cuckoo search algorithm shown in Algorithm 1.

Algorithm 1. Cuckoo search via L´evy flight 1: Initialize: Objective function : f (x), x = (x1 , ..., xd )T ; Initial population of host nests : xi (i = 1, 2, ..., n); t←0  Counter of the current generation tmax ← Maximum number of generations 2: while t < tmax do randomly. 3: Choose a nest xt−1 j 4: Search a new nest xti by L´evy walk. )), then 5: if (F (xti ) > F (xt−1 j 6: Replace xt−1 by xti . j 7: end if 8: Some of the worse nests are abandoned in probability pa and new ones are built. 9: Sort the list of nests and find the best nest. 10: t←t+1 11: end while

First, each cuckoo chooses a nest xt−1 randomly, and a nest xti by using L´evy j flight. Then they are evaluated by the function F . The cuckoo replaces xt−1 by j t−1 t t t xi if F (xi ) > F (xj ) and dumps an egg in the nest xi . Then, it finds the best nest at the generation t by sorting the list of nests. Meanwhile, some of the worse nests are abandoned with a probability pa ∈ [0, 1] and new ones are built. The line 6 in Algorithm 1 is the brood parasitic behavior of cuckoos. A new is computed by L´evy flight in Eq. 1. nest (or solution) xt+1 i xt+1 = xti + α × step length i

(1)

step length is a step length of L´evy flight, which obeys L´evy distribution, and α is a weight parameter for the step length. We set α = 1 since this configuration is used in most cases [13].

3

System Model

We assume that a MANET consists of a set of nodes N = {N1 , N2 , ..., N|N | }, which are mobile devices (e.g., smartphones, laptop computers, wearable

202

T. Kurokawa and N. Hayashibara

devices). Each node has a unique identifier. Each pair of nodes can communicate with each other. It means that a node in the network has a multi-hop path to any other node. We can obtain the remaining amount of battery RBit and storage RSit of a node Ni at time t. We also define the capacity of battery CBi and storage CSi of a node Ni . Moreover, Ati is defined as a duration (minutes) that Ni joins the network at time t. We conducted simulations by a cycle-based simulator. Thus, Ati is represented as the number of cycles. The MANET we assume is a dynamic distributed system. Nodes leave from the network by going out of the communication range of any other node or turning off. In addition, nodes can join the network as new nodes. Once a node leaves from the network, it joins it again with another identifier without any data and logs in the previous execution. It means that we assume fail-stop failure. The topology of a MANET is assumed to be a random geometric graph.

4

Data Replication Based on Cuckoo Search

Our approach is an extension of the protocol proposed by Kageyama and Shibusawa [7] for improving the availability of low demand data in MANETs. It allocates the replicated data based on the evaluation of nodes regarding the residual battery and storage. Although it assumes a super-node based P2P system, we assume a fully decentralized P2P system. We also use Cuckoo search [13] to find an eligible node for allocating replicated data. First of all, we introduce the replication protocol proposed in [7]. It replicates the data based on the prediction of requests to each data. It assumes that the distribution of the data requests obeys Poison distribution as follows. P (X = t) =

λt −λ e t!

(2)

λ is the average number of requests on certain data in the Eq. 2. So, the probability of requests to the data at time t is represented as the equation. Every node measures the number of requests to data that are stored in the node. The prediction of data requests is computed based on the exponential smoothing. Let M Rt and P Rt be the number of requests which is measured, and the predicted number of requests. So, P Rt+1 will be represented as follows. P Rt+1 = δ · M Rt + (1 − δ) · P Rt

(3)

Data is distinguished as low demand if P Rt+1 for the data becomes lower than the bottom five percent of the probability distribution P (t + 1). 4.1

Allocation of Replicated Data

As we mentioned in Sect. 3, we assume that data are distinguished and labeled in the frequency of requests So, data on disaster relief supplies, medical support and livelihood support are expected to be labeled as high demand because many

Data Replication Based on Cuckoo Search in Mobile Ad-Hoc Networks

203

people could search and require this type of data. We assume that this type of data is replicated by Owner replication in this paper. On the other hand, data that are labeled as low demand based on the prediction of requests (Eq. 3) are replicated and allocated by the proposed data allocation protocol based on Cuckoo search to reduce the storage cost of each node. The main purpose of the proposed protocol to improve the availability of this type of data without the additional cost on storage. We use Cuckoo search for the allocation of replicated data in MANET. Each node Ni has two parameters, the capacity of battery CBi and storage CSi . Moreover, it has three variables, the remaining amount of battery RBit and storage RSit , and the duration Ati at time t. The proposed protocol evaluates the nodes in the network based on the following equation and computes the evaluation value Ei of each node Ni .    RBit RSit α× +β× Ei = ln × Ati CBi CSi α and β are weights for the evaluation where α, β = [0, 1]. To simplify the capacity of battery and storage, CBi and CSi are defined as 100.0. Thus, RBit and RSit are variables from 0.0 to 100.0. We assume that data are disappeared from the network by nodes’ leaving. A node that has a large enough amount of battery are not likely to leave the network because of the battery shortage. The proposed protocol tends to allocate replicated data into such a node. On the other hand, it takes into account the residual space of storage. It computes Ei for each node Ni and searches the node that has the maximum value of Ei by Cuckoo search. It uses L´evy walk that is known as an efficient algorithm for a wide-area search. Since the topology of a MANET is usually represented as a graph, we use the L´evy walk algorithm for unit disk graphs [12]. The original Cuckoo search [13] assumes the optimization problem on continuous values. On the other hand, our target is the optimization of discrete values to find the optimal node for allocating replicated low demand data. So, we have to modify the Cuckoo search into the discrete one. 4.1.1 Discretization of Cuckoo Search We assume that the topology of MANET is represented as a random geometric graph. Cuckoo search finds a node Ni with a high Ei value. For this purpose, we modify the original Cuckoo search into the discrete version as follows: 1. 2. 3. 4.

5

We assume the optimal node Ni regarding Ei . Find a new node Nj by L´evy walk starting from Ni . Compare the evaluation values Ei and Ej of Ni and Nj . Replace the optimal one by Nj if Ei < Ej .

Performance Evaluation

The purpose of the evaluation is to clarify the following things.

204

T. Kurokawa and N. Hayashibara

• Data availability of replicated data in MANETs. • Storage cost in replicated data. We evaluate the things listed above by comparing with other replication protocols, Owner replication [8], Path replication [3], the protocol proposed by Kageyama [7]. We implemented these protocols and the proposed one on PeerSim simulator [6]. PeerSim is a cycle-based simulator for Peer-to-Peer systems implemented in Java. 5.1

Environment and Scenarios

We assume that the number of nodes |N | = 2000 and they are located uniformly at random in the 1000×1000 field. Each node is interconnected with the neighbor nodes that are within a certain communication range of it, hence the topology of a MANET is a random geometric graph. The storage residual of each node is set 100, and the battery residual of it is decided randomly from 50 to 100. We execute 500 cycles for a simulation run. We assume that 50 unique data (files) uploaded to the network. We also assume two scenarios regarding the ratio of high demand and low demand data as follows: • Scenario 1: 50% of data is low demand and the rest of data are high demand. • Scenario 2: 80% of data is low demand and the rest of data are high demand. 5.2

Data Availability

One of the purposes of data replication is improvement of data availability. In MANET, data may be disappeared by nodes’ leave. Thus, data availability is an essential criterion in the evaluation. We now define the data availability DA in each cycle of a simulation run as follows. Dt (4) DAt = Doriginal + Dreplicated Doriginal and Dreplicated are the total number of the original data and that of the replicated data, respectively. The total number of data available at time t t t + Dreplicated . Thus, the data availability is denoted as Dt where Dt = Doriginal t DA at time t is represented in Eq. 4. Figures 1 and 2 show the availability of high demand data and that of low demand data in Scenario 1, respectively. There is no difference between the approaches we compared. On the other hand, Kageyama’s protocol and the proposed protocol improves the availability of low demand data compared to Owner replication and Path replication. In particular, Kageyama’s protocol is better than the proposed one from 280 cycles though they are almost the same until 250 cycles (see Fig. 2). Figures 3 and 4 show the availability of high demand data and that of low demand data in Scenario 2, respectively. There is also no difference between the

Data Replication Based on Cuckoo Search in Mobile Ad-Hoc Networks 1

1 Owner Path Kageyama Cuckoo

Owner Path Kageyama Cuckoo

0.8

0.8

0.6

0.6

Data Availability

Data Availability

205

0.4

0.2

0.4

0.2

0

0 0

100

200

300

400

500

0

100

200

Num of cycle

300

400

500

Num of cycle

Fig. 1. The availability of high demand data in Scenario 1.

Fig. 2. The availability of low demand data in Scenario 1.

approaches, just like Fig. 1. Figure 4 shows that Kageyama’s protocol is slightly better than the proposed protocol from 150 to 250 cycles, and the difference becomes bigger from 300 cycles. Owner replication and Path replication drop the availability of low demand data at 100 cycles, and the gap between others is getting bigger. 1

1 Owner Path Kageyama Cuckoo

0.8

0.8

0.6

0.6

Data Availability

Data Availability

Owner Path Kageyama Cuckoo

0.4

0.2

0.4

0.2

0

0 0

100

200

300

400

Num of cycle

Fig. 3. The availability of high demand data in Scenario 2.

5.3

500

0

100

200

300

400

500

Num of cycle

Fig. 4. The availability of low demand data in Scenario 2.

Storage Cost

The storage cost is crucial for replication protocols. Improvement of availability of data usually imposes the cost of storage. In other words, there exists a tradeoff between them. We measure the average residual of storage of each node and the cumulative storage usage. Figures 5 and 6 show the average residual of storage of each node and the cumulative storage usage in Scenario 1. Figure 5 indicates that the proposed protocol is most efficient regarding the storage cost of each node. It improves

206

T. Kurokawa and N. Hayashibara

the average residual of storage more than 20% compared to Kageyama’s protocol at 180 cycles or later. Figure 6 shows that the difference between the proposed protocol and Kageyama’s one regarding the cumulative storage usage becomes bigger from 300 cycles. 100

6×10

6

5×10

6

Owner Path Kageyama Cuckoo

Cumulative Occupancy

Avarage Storage Remaining

80

60

40

20

4×106

3×106

2×106

1×106 Owner Path Kageyama Cuckoo

0

0 0

100

200

300

400

500

0

100

Num of cycle

200

300

400

500

Num of cycle

Fig. 5. The average storage residual of each node in Scenario1.

Fig. 6. The cumulative storage usage in Scenario 1.

Figures 7 and 8 show the average residual of storage of each node and the cumulative storage usage in Scenario 2. Figure 7 indicates that the proposed protocol and Owner replication show a similar tendency on the average residual of storage. On the other hand, the difference between the proposed protocol and Kageyama’s protocol becomes more significant from an early time (50 cycles). Figure 8 also shows the efficiency of the proposed protocol on the cumulative storage usage compared to Kageyama’s protocol that increases the storage cost linearly. In the case that has a large proportion of low demand data, the storage efficiency of the proposed protocol is emphasized compared to Kageyama’s protocol. 3×106

100

Owner Path Kageyama Cuckoo

2.5×106

6

Cumulative Occupancy

Avarage Storage Remaining

80

60

40

2×10

1.5×106

1×106

20 500000 Owner Path Kageyama Cuckoo 0

0 0

100

200

300

400

Num of cycle

Fig. 7. The average storage residual of each node in Scenario2.

500

0

100

200

300

400

Num of cycle

Fig. 8. The cumulative storage usage in Scenario 2.

500

Data Replication Based on Cuckoo Search in Mobile Ad-Hoc Networks

5.4

207

Hit Ratio and Average Distance to Data

We measured the hit ratio of search queries and the average distance to the required data when it was found in the network. The average distance between the searcher and the node that stores the data is represented by the hop count in the network. We compare four protocols, Owner replication [8], Path replication [3], the protocol proposed by Kageyama et al. [7], and our proposed protocol. Tables 1 and 2 show the results on the hit ratio and the average distance to the required data in Scenario 1 and 2, respectively. Our proposed protocol and Kageyama’s protocol show the same result as Owner replication regarding high demand data because they use Owner replication for replicating high demand data. The proposed protocol is slightly worse than Kageyama’s protocol on low demand data. This is because the number of replication of each data is reduced in the proposed one for saving the storage usage of each node. On the other hand, it is still better than Owner replication and Path replication. Table 1. The comparison on the hit ratio and the distance to the data in Scenario 1. Protocol

Demand Hit ratio Avg. distance

Proposed

High Low

81% 88%

8.09 12.88

Kageyama et al. [7] High Low

81% 97%

8.09 9.28

Owner replication

High Low

81% 55%

8.09 17.57

Path replication

High Low

88% 63%

8.47 14.30

Table 2. The comparison on the hit ratio and the distance to the data in Scenario 2. Protocol

Demand Hit ratio Avg. distance

Proposed

High Low

75% 92%

8.20 10.93

Kageyama et al. [7] High Low

75% 98%

8.20 8.42

Owner replication

High Low

75% 56%

8.20 17.00

Path replication

High Low

78% 70%

7.05 14.44

208

6

T. Kurokawa and N. Hayashibara

Conclusion

We proposed the allocation mechanism of replication data based on Cuckoo search for improving the availability of low demand data and implemented it in the modified version of the replication protocol proposed in [7]. The primary purpose of this paper is to clarify the impact of Cuckoo search on the availability of low demand data and the storage cost of replicated data. There exists a tradeoff between the availability and the storage cost. Our simulation result shows that the proposed protocol improves the efficiency of both the average storage cost of each node and the cumulative storage usage compared to Kageyama’s protocol. On the other hand, the proposed protocol is getting worse than Kageyama’s one regarding the availability of replicated data as time passes though both of them are almost the same on it at an early period. Our proposed protocol is slightly worse than Kageyama’s one regarding the hit ratio of search queries and the average distance to the required data. However, It is still useful when considering the storage cost of each device. When we suppose the information sharing in the case of a disaster, the availability of data is essential as well as the storage cost of mobile devices, the hit ratio and distance to the required data. Our proposed protocol shows the efficiency of the storage of each device without a significant impact on the availability, especially on low demand data. Thus, our protocol is suitable for such a situation. In future work, we will further study to improve the availability take into account several factors of each node (mobile devices).

References 1. Ali, A., Latiff, L.A., Fisal, N.: GPS-free indoor location tracking in mobile ad hoc network (MANET) using RSSI. In: 2004 RF and Microwave Conference (IEEE Cat. No. 04EX924), pp. 251–255 (2004). https://doi.org/10.1109/RFM.2004.1411119 2. Chandrakala, C.B., Prema, K.V., Hareesha, K.S.: Improved data availability and fault tolerance in MANET by replication. In: 2013 3rd IEEE International Advance Computing Conference (IACC), pp. 324–329 (2013). https://doi.org/10. 1109/IAdCC.2013.6514244 3. Cohen, E., Shenker, S.: Replication strategies in unstructured peer-to-peer networks. SIGCOMM Comput. Commun. Rev. 32(4), 177–190 (2002). https://doi. org/10.1145/964725.633043 4. Hara, T.: Data management issues in mobile ad hoc networks. Proc. Jpn. Acad. Ser. B 93(5), 270–296 (2017). https://doi.org/10.2183/pjab.93.018 5. He, X., Wang, F., Wang, Y., X.S.Y.: Nature-inspired algorithms and applied optimization. In: Studies in Computational Intelligence, vol. 744. Springer (2018) 6. Jelasity, M., Montresor, A., Jesi, G.P., Voulgaris, S., Arteconi, S., Hales, D., Marcozzi, A., Picconi, F.: Peersim (2016). http://peersim.sourceforge.net/ 7. Kageyama, J., Shibusawa, S.: Replication that prevents low-demand files from disappearing in P2P network. In: Proceedings of The First Forum on Data Engineering and Information Management (2009). (in Japanese)

Data Replication Based on Cuckoo Search in Mobile Ad-Hoc Networks

209

8. Lv, Q., Cao, P., Cohen, E., Li, K., Shenker, S.: Search and replication in unstructured peer-to-peer networks. In: Proceedings of the 16th International Conference on Supercomputing, ICS 2002, pp. 84–95. ACM, New York (2002). https://doi. org/10.1145/514191.514206 9. Mukilan, P., Wahi, A.: CDRA: consistency based data replication algorithm for manet. Int. J. Comput. Appl. 51(14), 1–8 (2012) 10. Sakano, T., Kotabe, S., Komukai, T., Kumagai, T., Shimizu, Y., Takahara, A., Ngo, T., Fadlullah, Z.M., Nishiyama, H., Kato, N.: Bringing movable and deployable networks to disaster areas: development and field test of mdru. IEEE Netw. 30(1), 86–91 (2016). https://doi.org/10.1109/MNET.2016.7389836 11. Shen, H.: An efficient and adaptive decentralized file replication algorithm in P2P file sharing systems. IEEE Trans. Parallel Distrib. Syst. 21(6), 827–840 (2010). https://doi.org/10.1109/TPDS.2009.127 12. Shinki, K., Nishida, M., Hayashibara, N.: Message dissemination using l´evy flight on unit disk graphs. In: Proceedings of the 31st IEEE International Conference on Advanced Information Networking and Applications (AINA 2017), Taipei, Taiwan ROC (2017) 13. Yang, X.: Suash deb: cuckoo search via l´evy flights. In: 2009 World Congress on Nature Biologically Inspired Computing (NaBIC), pp. 210–214 (2009). https:// doi.org/10.1109/NABIC.2009.5393690

Trusted, Decentralized and Blockchain-Based M2M Application Service Provision Besfort Shala1,2(&), Ulrich Trick1, Armin Lehmann1, Bogdan Ghita2, and Stavros Shiaeles2 1

Research Group for Telecommunication Networks, Frankfurt University of Applied Sciences, Frankfurt/M., Germany [email protected] 2 Centre for Security, Communications and Network Research, University of Plymouth, Plymouth, UK

Abstract. Decentralized M2M service platforms enable the integration of enduser-based M2M applications and end-user-located M2M resources without the use of central entities or components in the system architecture. Sharing enduser-based M2M applications with other users’ part of an M2M community allows the creation of new and complex M2M applications. However, a fully decentralized system often leads to several trust issues regarding the behavior of end-users and M2M applications. A powerful measure to overcome possible limitations of decentralized M2M service platforms and to replace the missing control authority are trust relationships among the nodes. Therefore, this publication proposes a novel concept for trusted M2M application service provision. Moreover, it introduces the integration of blockchain elements and trust evaluation techniques to optimize the M2M application service provision. A trust consensus protocol is integrated in order to secure the decision-making process among the stakeholders which optimizes several aspects, such as peer joining, service registration and application configuration. Keywords: Trust application


P2P
Blockchain
M2M
Security
Service and

1 Introduction Being part of the end-user environment, intelligent M2M devices have a great potential for supporting smart environments when they are used in the creation of new complex M2M applications which are accessible for other users. To realize the integration of such devices, the participation of the end-user in the M2M application service provision process is required. In this context, it is important to provide a fully decentralized architecture for M2M application service provision in order to avoid end-user environment limitations and problems, such as the need for large resources for service development and maintenance, high costs for operating service platforms and the lack of reliability of the platform as result of single point of failures. The literature review provides a considerable number of different M2M service platforms [1]. However, analyses regarding decentralized architectures and end-user © Springer Nature Switzerland AG 2020 L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 210–221, 2020. https://doi.org/10.1007/978-3-030-33506-9_19

Trusted, Decentralized and Blockchain-Based M2M

211

integration have shown that most of them do not completely fulfill these requirements [2]. The authors in [3] introduce an enhanced Dynamic Service Overlay Network (eDSON) platform with a focus on distributed service provision which supports the provision of user individual services and operates on distributed servers in the Internet. However, the platform is maintained by a centralized operator and does not integrate the end-user in the service composition process. Another approach for M2M application service provision is presented in [2, 4] where every end-user part of the network can provide or consume M2M services and can act as a decentralized M2M service provider. The authors in [9] highlight the necessity of trust because of the increasing “risks, threats and vulnerabilities at component, device, system, service and human levels” in the world of Information and Communication Technology (ICT). However, existing limitations of end-user-based and decentralized M2M services are the lack of trust in the network and the controllability or access control of joining and leaving peers. Moreover, a trustworthy mechanism, which considers only highly trusted and fully accepted services by all the members in the service composition process, is nonexistent. To overcome trust issues in decentralized M2M communities, the authors in [5–7] introduce a fully decentralized trust evaluation system which covers the trustworthiness of new and existing entities. Besides those, the presented trust evaluation system covers several aspects of a peer and a service in the trust evaluation process. Additionally, it optimizes the data storage system of the trust results by integrating the blockchain technology. For using the blockchain technology and its elements in the M2M environment, several integration possibilities are proposed in [7]. However, the consensus protocols used in different blockchain applications have several limitations, such as high computational effort, high energy consumption and lack of trustworthiness. In order to integrate blockchain with its features to M2M service platforms, a fully suitable and trust-based consensus protocol is proposed in [5]. The aim of this publication is to optimize several levels of M2M application service provision starting with the M2M community admission, the M2M service registration and the M2M application configuration. Therefore, blockchain elements and trust provision principles are combined to introduce a novel trusted and blockchain based M2M application service provision system. The remainder of this paper is organized as follows. Section 2 describes the M2M application service provision according to [2, 4]. The decentralized trust evaluation system and the integrated blockchain elements are introduced in Sect. 3. Section 4 presents the optimization approach by integrating a trust-based consensus protocol to the M2M service provision life cycle. Finally, Sect. 5 gives a conclusion of the presented methodologies and approaches.

2 P2P-Based M2M Application Provision The authors in [2] propose an M2M service platform where the end-user is integrated into the application creation process. To ensure independency in the application creation process, central entities, such as central platform or network element provider, are

212

B. Shala et al.

removed from the architecture. To reduce the cost of operating an M2M service platform and to increase the acceptance of an M2M solution, existing resources in the end-user environment are reused. Moreover, the proposed M2M service platform integrates different M2M device technologies and allows them to be combined with each other.

Fig. 1. Cooperative M2M application service provision [2]

Using a GUI, even with less technical knowledge, the end-users have the possibility to design individual M2M application services and make them available for other endusers or central service providers. They additionally have the possibility to cooperate with each other in order to provide complex or so called cooperative M2M application services. After this kind of M2M application service is modelled, it will be configured automatically and autonomously by connecting the specific instances of services that are involved in the cooperative M2M application service. Figure 1 shows that a complex service consists of the combination of several distributed services that are networked together. The combination may consider same services with same functionality (service aggregation) or services with different functionality (service composition) [4]. In order to ensure a decentralized system architecture, the author in [2] proposes to use a Peer-to-Peer (P2P) network for communication and information storage between the peers. To create a social network and interest groups among the participating nodes, the authors in [2, 4] introduce an M2M community. M2M application services are described by machine-readable State Chart XML (SCXML). The application requires an application interface (described by an Interface Description (IDS)) with which it forms an application service in order to be consumable for other entities.

Trusted, Decentralized and Blockchain-Based M2M

213

Some additional information regarding the M2M service platform presented in [2, 4] are provided in Sect. 4. Moreover, Sect. 4 lists several limitations and introduces a novel, trust- and block-based optimization approach for M2M application provision.

3 Blockchain-Based Trust Model In order to ensure a secure environment in an M2M community, trust relationships between the participating peers are required. Therefore, the trustworthiness of peers and the services they provide should be evaluated. Several trust definitions depending on the application domain and the context exists. Regarding ICT environments, the authors in [8] state that the preference of an entity for decision-making with other entities and service consumption is affected by trust. Specifically, they claim that “trust evaluation is especially significant in ICT environments where a huge number of entities mutually interact with each other to provide and consume information or resources”. The literature provides several trust evaluation and management approaches which are evaluated in [6]. Most of them do not provide a solution for bootstrapping peers or new services which are provided to the community. Moreover, they do not provide a secure mechanism to store the trust scores computed through the trust evaluation process. Therefore, the authors in [5] propose a trust model which optimizes the storage system and includes other trust aspects for evaluation. Moreover, the trust model includes blockchain elements combined with a newly introduced Trust Consensus Protocol. The trust model (from now on called trust evaluation system) and the blockchain are going to be explained in the following. 3.1

Trust Evaluation System

Several security and trust limitations on existing M2M service platforms [1] can be mitigated if there is an overview about the trust scores of the participating peers and services. Therefore, the authors in [5] introduce a completely decentralized and community-based trust evaluation system. An overview of the architecture of the trust evaluation system is shown in Fig. 2. For the trust evaluation, several aspects in the M2M community are considered, such as the service functionality, service quality, service acceptance, peers’ behavior and participation willingness in several community tasks. To increase the reliability of the data integrity and to support integrity check-ups for data stored in the P2P overlay, the authors in [7] proposes to include blockchain elements in the trust evaluation system. To ensure a decentralized environment without centralized entities, the authors in [5] suggest to distribute all community tasks among the participating nodes. More specifically, the Trust Evaluation System consists of three main parts, the Service Trust Evaluation, Behavior Trust Evaluation and Task Trust Evaluation. The Service Trust Evaluation includes Service Testing, which covers functional and performance testing and aims to identify the initial trust score of a new service. Moreover, it includes Service Monitoring and Service Rating where the behavior and the performance of a service is monitored by considering several parameters. Besides them, services are rated by other users based on their individual experience on using the

214

B. Shala et al.

services. Service Monitoring and Service Rating are used to evaluate the trust score of an existing service. The results of Service Testing, Service Monitoring and Service Rating are computed using a Service Trust Evaluation Function which concludes with a Partial Trust Score of the Service. Another part, the Behavior Trust Evaluation, is used to check the integrity of a service by comparing the data stored in the blockchain with the data stored in the P2P overlay. The results of the integrity check-up are used to increase (if integrity remains) or decrease (if integrity fails) the trust score of the peer. The third part, the Task Trust Evaluation, evaluates the participation of a peer in community tasks, such as acting as a Test Agent or Blockchain node.

Fig. 2. Trust evaluation system [5]

3.2

Blockchain and Trust Consensus Protocol

The blockchain technology is a subbranch of the so-called distributed ledger which can be defined as an asset database that can be shared across a network of multiple sites, geographies or institutions [9]. The ledger is maintained through cryptographical principles where changes are made available for all network members. One of the key elements of a blockchain is the consensus protocol which is used to agree for the same copy of the ledger among the participating nodes. Specifically, the term consensus protocol is defined in [10] as “a series of procedures from approving a transaction as an official one and mutually confirming said results”. Several publications in scientific libraries and in the industry have proposed different consensus protocols with specific characteristics where the most relevant ones are reviewed and evaluated in [5]. The review has shown that most of them require computational effort for achieving consensus and validating new transactions. Moreover, they do not provide a fair way to select a node to become the leader who proposes a new block. Another drawback of existing approaches is that they do not consider the trustworthiness of blockchain inputs and nodes who are storing something in the blockchain.

Trusted, Decentralized and Blockchain-Based M2M

215

Fig. 3. Trust consensus protocol [5]

To overcome these limitations, the authors in [5] introduce a novel Trust Consensus Protocol (Trust-CP). The protocol consists of five main phases (see Fig. 3): Trust peer Filtering, Random Selection, Block Creation, Trust Weighted Voting, Trust Reward/Punishment. The key aspect of the proposed consensus protocol can be described as follows: All nodes part of the M2M community also participate in the blockchain network. The trust score of every node is continuously evaluated to ensure trustworthiness in the network. During the lifetime, transactions are sent from one node to other nodes. All transactions are assigned with the trust score of the transaction initiator (trust score if the sender node). These transactions are unconfirmed and are waiting to be approved by the blockchain network. Before the approval process starts, the transactions have to be included in a block. This is done by so-called block creators (leaders) which are nodes selected from the blockchain network to perform these tasks. The authors in [5] propose that for every round of block generation, an algorithm is going through the nodes to select randomly one of them as block creator based on its trust score (see Fig. 3, Phase I and II). After that, the block creator will collect pending transactions to a block (see Fig. 3, Phase III), where it should consider only transactions with a good trust score. The generated block will be broadcasted to other nodes for validation and confirmation (see Fig. 3, Phase IV). Other nodes will receive and

216

B. Shala et al.

verify the block by checking the trust score of the block creator node, the trust score of the transactions part of the block and the hash values of the block. If the block contains the right information and also fulfills the criteria of the system, it will then be positively voted by the validating node and the block is forwarded to other nodes. The criteria are fulfilled if the block is created by a trusted block creator, the block contains the right hashes and the transactions part of the block are also trusted. If the block does not meet the conditions, it will receive a no-vote. The votes are weighted based on the trust score of the validators. The different actions performed by the nodes part of the blockchain are rewarded or punished accordingly by increasing or decreasing the trust score of the performing node (see Fig. 3, Phase V).

4 Integration of Trust Consensus Protocol for M2M Application Service Provision 4.1

Joining P2P Network and M2M Community

The peers acting as service providers and service consumers are connected P2P in [2, 4]. To fully decentralize the whole M2M application service provision, the authors in [2, 4] integrate a P2P layer into their layer model of decentralized networking. The P2P layer consists of the P2P communication layer which enables the information exchange between the peers using M2M communication protocols and the P2P overlay layer which realizes the distributed data storage using protocols, such as Chord or Gnutella. To join the P2P network, the authors in [4] propose to use a webpage/server for registration and for finding the contact information about the bootstrapping nodes of the P2P network. Afterwards, the new node will receive the necessary information to join the network. Additionally, the authors in [2, 4] introduce an M2M community to enable social networking in the P2P network. This M2M community is organized through the interface descriptions of the services which are provided by the peers. However, the authors in [4] do not consider the security aspects regarding joining a network and do not provide a concept how this process can be achieved in a secure and trustful manner. To optimize the entry to the P2P network and the associated M2M community, this publication proposes to integrate blockchain elements which have several benefits. Therefore, it is proposed to use the introduced trust evaluation system and the trust consensus protocol to manage the joining/leaving process in the M2M community. Therefore, this publication proposes a novel joining mechanism (an overview is shown in Fig. 4) for peers interested to enter the P2P network/M2M community. First, a new peer who wants to enter the M2M community needs to contact a bootstrapping node. The subdomains for these nodes can be resolved by using DynDNS. After a new peer has contacted a bootstrapping node (1), the bootstrapping node will test the peer and its services to issue an entry trust score for the new peer (2). The testing consists of evaluating the functional behavior and the performance of the new service. The

Trusted, Decentralized and Blockchain-Based M2M

217

computed entry trust score is sent (3) to all the other peers’ part of the P2P network. Continuously, all participating peers will collect and store all joining proposals sent from different bootstrapping nodes in their local storage (4). Afterwards, the Trust Consensus Protocol (5) is going to be applied by all peers in the network. The leader is first selected based on a trust-based and random selection algorithm (only peers with high trust scores are considered). Based on the list of proposals, the leader will create a transaction consisting of a list of interesting nodes which should be elected for joining the M2M community. This list will be sent afterwards to other nodes which all act as validating nodes by checking the transaction and vote on it. The voting consists of analyzing the list of joining nodes and their trust score. Furthermore, the trust score of the leader is checked. If the predefined criteria are fulfilled (trust score of joining proposals and leader should be high), the transaction will get a positive vote which are then weighted with their own trust scores. If the required trust threshold for the voting is achieved, the list of the selected joining nodes (mentioned in the transaction) will be admitted to the M2M community. Further information regarding the Trust-CP can be found in Sect. 3.

Fig. 4. Trust-CP-based joining

218

4.2

B. Shala et al.

M2M Service Registration

The authors in [2, 4] design a Service/Application Registry (SAR) which operates in the overlay and is used to manage all services in the M2M community. A new service is registered in the SAR by a peer which is acting as a service provider. The service provider then stores the combination of service ID and contact information and makes the Interface Description (IFD) of the service available to other end-users. It could be that several end-users offer the same services in terms of identical IFDs but with different contact information. The same services acting as individual instances of their services are stored with the same service ID in the SAR. End-users acting as service consumers can look up using keywords for specific services and have the possibility to select an instance of a service for the application configuration. A disadvantage of the existing approach is that every end-user acting as service provider can register any service without considering the functionality or the security of it. Other end-users (service consumers) will not have the possibility to check if the instance of a service is trustworthy or not. The authors in [6] propose to test the functionality and performance of the new services and to obtain the initial trust score which is made available for the whole community. However, even though the approach presented in [6] gives a better trust overview for all M2M community members, it does not mitigate that untrustworthy services are registered in the SAR. To optimize the service registration process, this publication proposes to use the proposed Trust-CP for decision making among the nodes in order to agree on the same trust score for a new service. Figure 5 shows the registration of a new M2M service and the creation of a new and extended Interface Description (IFD) (containing trust information about the service) based on the tests performed by the community members. First, every service provider has to register the individual instance of a new service with the SAR (1). The IFD of the new service will be stored in the SAR. Other peers’ part of the M2M community can make a request to the SAR for new services (2). The SAR will then provide the service IFDs of the new registered services to the requesting peers (3). The different community peers will test and evaluate the new services and every testing peer will create based on the test results a new extended IFD of the new service (4). The extended IFDs will be shared among the M2M community and a consensus for the same extended IFD is required (5). The Trust-CP ensures that all participating nodes agree on the same copy of the IFD. On basis of the Trust-CP, the selected leader and the validating nodes decide whether or not to accept the registration of that service. The new extended IFD will be registered (6) in the SAR and can be used by the community to decide whether or not they want to subscribe to the service. Same can also be done for deregistration of a service, where the community members test an existing service and agree using the Trust-CP for deleting the IFD from the SAR. Moreover, the service registration approach presented in this section helps to figure out if the joining process of a peer (described in Sect. 4.1) is done correctly or not by comparing the trust score computed by testing the service during the joining process with the trust score computed in the service registration step.

Trusted, Decentralized and Blockchain-Based M2M

219

Fig. 5. M2M service registration and extended M2M service IFD

4.3

Cooperative M2M Application Design

Every end-user in the M2M community has the possibility to design new cooperative M2M applications by configuring and selecting different available services based on the own interest. However, as mentioned before, multiple end-users can offer different instances of the same M2M service and the decentralized M2M service architecture does not support a centralized coordination regarding the direction of the peers to which specific instances of a service they should connect to realize the cooperative M2M application. The authors in [2, 4] introduce a combination of random and manual selection of service instances which is not a fair and secure solution for M2M application configuration. As result, an unstable or malfunctioning cooperative M2M application is created and provided in the M2M community. To overcome this problem, the authors in [6] propose to consider only service instances with a good trust score for being part in the random selection and application composition process. However, the approach in [6] does not provide a fair way to select the service instances and also relies only on the subjective decision of every single service consumer. In order to optimize the trust-based selecting approach, this publication proposes to include all participating peers in a fair voting system regarding the designed cooperative M2M application. This also enables load balancing among the peers regarding the participation of their services in a cooperative M2M application. The novel application configuration for a cooperative and distributed M2M application service is shown in Fig. 6. Specifically, this means that after one end-user has designed a cooperative M2M application, the corresponding SCXML application description is stored in the P2P overlay where other end-users can retrieve the description and can analyze the service chain defined in the application description. If a service consumer (SC) wants to use the application defined in the SCXML application description (1), it will send a request (2) to the M2M community where other participating nodes will propose (based on the application description) an instance of a service for every position in the service chain (3). These proposals will be sent to all community members (4). Using the Trust-CP

220

B. Shala et al.

(5), a leader is selected which takes one proposal (based on predefined trust criteria) from the pool of service chain proposals and sends them to all nodes for voting and validation. Other nodes receive the proposal and check if the leader and the instances selected for being part of the service chain are trustworthy. The voting will conclude with the decision about the definitive list of instances upon which all community members agreed. This list of service chain instances will be sent to the service consumer (6) which then will start contacting the specific service instances in order to allow the application configuration. Finally, after the application configuration, the application execution will start, considering only trustworthy services part of the service chain.

Fig. 6. M2M application configuration

5 Conclusion The benefits of using the blockchain technology are ensuring data integrity and nonrepudiation. Moreover, blockchain-based consensus protocols enable the agreement of the same state of art in a fully decentralized network and motivate the participating nodes to be actively involved in decision-making processes. Trust evaluation systems provide the possibility to measure the trust score of entities in a community. Trust relationships between participating nodes acting in a network without a central manager are very important to overcome potential security risks. The features of blockchain and trust evaluation systems are used to improve existing end-user based M2M service platforms and communities. Therefore, this research publication proposes to integrate trust and the trust evaluation processes in different parts of the M2M application service provision lifecycle. Considering decision making and data integrity, the blockchain and a novel Trust Consensus Protocol is also integrated. The improvements of the M2M application service provision process includes a secure and trustworthy joining mechanism to the P2P overlay and the M2M community. Moreover, the service registration and the cooperative M2M application design is improved through the use of trust and consensus protocol.

Trusted, Decentralized and Blockchain-Based M2M

221

Acknowledgment. The research project P2P4M2M providing the basis for this publication is partially funded by the Federal Ministry of Education and Research (BMBF) of the Federal Republic of Germany under grant number 03FH022IX5. The authors of this publication are in charge of its content.

References 1. Kim, J., Lee, J., Kim, J., Yun, J.: M2M service platforms: survey, issues, and enabling technologies. IEEE Commun. Surv. Tutor. 16(1), 61–76 (2014) 2. Steinheimer, M., Trick, U., Fuhrmann, W., Ghita, B.: Autonomous decentralised M2M application service provision. In: Proceedings of the Seventh International Conference on Internet Technologies and Applications (ITA 17), Wrexham, UK, pp. 18–23. IEEE (2018) 3. Kim, Y.J., Kim, E.K., Nam, B.W., Chong, I.: Service composition using new DSON platform architecture for M2M service. In: International Conference on Information Networking (lCOIN), pp. 114–119 (2012) 4. Steinheimer, M., Trick, U., Fuhrmann, W., Ghita, B., Frick, G.: M2M application service provision: an autonomous and decentralised approach. J. Commun. 12(9), 489–498 (2017) 5. Shala, B., Trick, U., Lehmann, A., Ghita, B., Shiaeles, S.: Novel trust consensus protocol and blockchain-based trust evaluation system for M2M application services. IoT – Eng. Cyber Phys. Hum. Syst. 7, 100058 (2019). https://doi.org/10.1016/j.iot.2019.100058. ISSN: 2542-6605 6. Shala, B., Trick, U., Lehmann, A., Ghita, B., Shiaeles, S.: Trust-based composition of M2M application services. In: 10th IEEE International Conference on Ubiquitous and Future Networks (ICUFN 2018), Prague, Czech Republic (2018) 7. Shala, B., Trick, U., Lehmann, A., Ghita, B., Shiaeles, S.: Blockchain-based trust communities for decentralized M2M application services. In: Xhafa, F., Leu, F.Y., Ficco, M., Yang, C.T. (eds.) Advances on P2P, Parallel, Grid, Cloud and Internet Computing, 3PGCIC 2018. Lecture Notes on Data Engineering and Communications Technologies, vol. 24. Springer, Cham (2018) 8. ITU-T, Recommendation Y.3052: Overview of trust provisioning in information and communication technology infrastructures and services (2017) 9. Distributed Ledger Technology: Beyond Blockchain. Government Office for Science (2016). https://assets.publishing.service.gov.uk/government/uploads/system/uploads/attachment_ data/file/492972/gs-16-1-distributed-ledger-technology.pdf. Accessed 10 July 2019 10. Survey on Blockchain Technologies and Related Services FY 2015 Report, Japan (2016). https://www.meti.go.jp/english/press/2016/pdf/0531_01f.pdf. Accessed 22 July 2019

A New Mobile Agent System for Sharing Disaster Information Under Unstable Network Conditions Natsuki Matsumoto and Tetsuya Shigeyasu(B) Prefectural University of Hiroshima, Hiroshima, Japan [email protected]

Abstract. Current network systems destined for collecting disaster information can be categorized into two types: (1) centralized systems collecting all information to a center server, and (2) distributed systems collecting information to several local servers at geographically distributed area. By the system (1), we can understand the disaster condition from a higher perspective. However, the system requires reliable connections. On the other hand, the system (2) can easier to obtain disaster information although it is limited by geographically. Both systems of (1) and (2) are useful for disaster relief activities even though these systems contribute to different perspective. In this paper, we discuss a new disaster information sharing system based on mobile agent. The proposed system implemented on Raspberry pi are evaluated on our test bed.

1

Introduction

In a time of disaster such as earthquake, It is needed to fast and adequate relief activities for reducing damages. For the activities, it is also important to collect and analyze disaster information at real-time. Nowadays, several systems have been proposed for offering the disaster information. Disaster On-line board service [1], 00000JAPAN [2], Twitter [3], Facebook [4] are the some of those information services. These information services offer useful information for reducing and avoiding disaster damages if sufferers can reach the service. Those services, however, most likely to be unusable state under the disaster-stricken due to crash the public network. In fact, Major top 3 mobile phone operators: NTT docomo, AU, SoftBank did not work at Kure area in Japan when the “Heavy rain in July 2019”. Hence, it is expected to develop a new system offering disaster information realtime even in the such severe condition for public network. By the way, current network systems destined for collecting disaster information can be categorized into two types: (1) centralized systems collecting all information to a center server, and (2) distributed systems collecting information to several local servers at geographically distributed area. By the system (1), we can understand the disaster condition from a higher perspective. However, the c Springer Nature Switzerland AG 2020  L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 222–230, 2020. https://doi.org/10.1007/978-3-030-33506-9_20

A New Mobile Agent System for Sharing Disaster Information

223

system requires reliable connections. On the other hand, the system (2) can easier to obtain disaster information although it is limited by geographically. Both systems of (1) and (2) are useful for disaster relief activities even though these systems contribute to different perspective. In this paper, we discuss a new disaster information sharing system based on mobile agent. The proposed system implemented on Raspberry pi are evaluated on our test bed.

2

Disaster Information Collection at both of Local/wide Area Based on Mobile Agent Technique

In a time of disaster, for reducing damages of disaster we need many kinds of information: location of shelters, confirmation of evacuees’ safely, supporting information, traffic information, meteorological information. For these kinds of information, we have to prepare to collect from either local area and wide area, according to these features. For examples, confirmation of evacuees’ safely, information for aid delivery, traffic information, meteorological information are better for collecting from wide area. On the other hand, although location of shelters and escape routes are the useful for all evacuees, it is better for collecting from local area due to these kinds of information are varied by regions. Therefore, we propose new information collection system integrating information from both of local/wide area based on mobile agent (MA) technique. The reasons why we develop system based on mobile agent technique are as follows: (1) MA can be distributed into any area, and it can work autonomously, and (2) MA can migrate its location on its own will, among evacuees’ smartphones and other network devices. For the above (1), our system can collect information fast at the specific area, and we can achieve load dispersion by allocating a MA to each area. For the above (2), MA can migrate device to operate and collect disaster information more efficiently at the adequate physical location. This migration ability of MA can useful for offering collected information to the other evacuees. Migrating its device during its operation, MA can easy to collaborate with the other MAs. This enables distributed computing on disaster information collection even in the central server is disconnected from the other areas affected by disaster. Our proposal introduces two kinds of MA: wide area MA and local area MA. In a time of disaster, any smartphone employing our proposal, starts at least one local are MA. Local area MAs a categorized as master MA or slave MA. 2.1

Roles of MAs

Wide area and local area MAs collecting disaster information Fig. 1 are described as follows:

224

N. Matsumoto and T. Shigeyasu

Master MA Master MA organizes a unit of information collection group (ICG). ICG consists of at least one master MA and slave MAs. master MA collects and offers disaster information from slave MAs of its ICG. Slave MA Slave MA belongs a ICG. Slave MA collects and processes disaster information, and uploads it to its master MA. Our proposal divides disaster affected area into some ICGs. Each ICG collects disaster information distributedly. Wireless connectivities are only required among ICG nodes. So, the proposal achieves higher availability than the conventional disaster information services based on public communication infrastructures. This mechanism makes for robust for collecting local disaster information. Wide area MA For collecting disaster information such as location of shelters, aid delivery information and information for relief activities, our proposal dispatches wide area MA from center server (e.g. disaster countermeasure office) to ICGs. The Wide area MA goes round among the ICGs except for missing network reachability. Wide area MA collects disaster information from master MA of each visited ICG. The information collected by wide area MA are integrated at the center server, then, those information are analyzed to utilize for the relief activities. In addition, for improving load dispersion and simplifying data processing, our system makes master MAs for each kind of disaster information one by one. 2.2

System Overview

This section describes overview of the proposal. 2.2.1

Building a ICG

System Startup We assume that all smart phones in a area, install platform of our proposal (MA platform) before disaster. First, all smart phones in a area receive Emergency alert from cellphone operator, it starts MA platform as smartphone application. The platform application generates MA as follows: MA Generation Phase As described in the previously, master MA constructs ICG, and collects disaster information around the ICG. The initial state of the MA platform generate one local MA but the MA does not decide its state to either master or local. Then, the generated local MA changes its state to master of local according the master selection process. Figure 2 shows the process. 1. After starting MA platforms, a node wait for a random time, and broadcasts its Hello packet. the Hello packet contains node location obtained by GPS (Global Positioning System). The Hello packet will be delivered to all node within the transmission range of the node.

A New Mobile Agent System for Sharing Disaster Information

225

Fig. 1. System overview of proposal

2. A node received the Hello packet change its mode to slave MA. Then, returns Accept packet. The Accept packet contains node location of the node, also. After returning the Accept packet, the node behaves as slave MA of the node which transmitted the Hello packet. 3. If a node which transmitted the Hello packet received Accept packet correctly, the node sets its mode as master MA, and it behaves master of the slave MAs which returned Accept packet. A node acting as master MA records the ID of slave MA, and collaborates these node as a member of its ICG, hereafter. On the other hand, if the two nodes acting as master MA of same kind of disaster information, go close to each other, two ICGs organized by the two nodes will be integrated into one ICG. By repeating the above step 1–3, ICGs are constructed at the disaster areas according the geographical conditions. After deciding the mode of MA on nodes, slave MAs periodically report disaster information obtained after previous report period, to master MA of its ICG. The slave MA also reports its current location obtained by GPS with the report packet. The master MA updates location tables for slave MA of its ICG. According to the updated location tables, master MA maintains geographical distribution of the ICG. If the current location of the master MA is far from the

226

N. Matsumoto and T. Shigeyasu

center point of the ICG, the master MA migrates its location to the other device close to the center point than the current node. By the migration, master MA can control its ICG while reducing transmission overhead. At the migration, the master MA selects next device to home considering two conditions: (1) device location is close to the center point of ICG, and (2) whether the device has enough remaining computational resource for disaster information processing. If the current master MA is lost due to some reasons, after one of slave MA failing to report to master MA, transmits the new Hello packet with reconstruction message and re-constructs the ICG, again.

Fig. 2. Master selection process.

A New Mobile Agent System for Sharing Disaster Information

227

Disaster Information Collection for Local Area After all MAs on the node of a ICG set the mode, all nodes start to collect the disaster information. The owners of the smartphone input the information such as, information of impassable road, scene of a fire. the collected information will be reported to the master MA of their ICG. Disaster Information Sharing Master MA integrates the disaster information reported by master MAs, and shares useful information for local distressed area with master MA of its ICG. 2.2.2 Disaster Information Collection for Wide Area This section describes how to collect the disaster information from ICGs widely distributed over the disaster distressed area. The overview of the collection of the disaster information on our proposal is shown in Fig. 3. As described as previously, all disaster information will be collected at the master MA of the ICG. However, personal information (e.g. evacuee’s name, age) is also collected at the master MA. Although these information is very useful for operating shelters and scheduling the aid delivery, it is strictly required to restrict information browsing by un authorized persons. Hence, in our proposal, to make these information delivery more secure, local information will be passed to wide area MA after encryption by public key. For the encryption, our system utilizes the public key encryption system. Public key are delivered to all nodes before disaster, and private key is stored only countermeasure office. Then, the all information will be carried by wide area MA can be browsed by the officer of countermeasure office.

3

Experimental Evaluation

In the previous section, we discussed a new disaster information sharing system based on mobile agent. This section reports the results of experimental evaluation of the prototype system of our proposal. Experiments has been done using 5 Raspberry Pis with our developed mobile agent system by JAVA language. Conditions used for the evaluations are listed in Table 1. Table 1. Experimental conditions Parameter

Values

Micro computer

Raspberry Pi Zero WH

OS

Raspbian GNU/Linux 9.4 istretch j

CPU

ARMv6 rev7 i700 MHz j

Main memory

512 MB

Wireless connectivity IEEE802.11n WirelessLAN Java version

1.8.0 65

228

N. Matsumoto and T. Shigeyasu

Fig. 3. Disaster information collection for wide area.

Our prototype system proceeds as following steps. 1. Select master MA and set all MA mode according to the previous section 2. Slave MAs collect disaster information from its Raspberry Pi 3. At each period for reporting, slave MAs reports collected disaster information to the master MA of its ICG 4. Master MA stores received disaster information, into DB on its Raspberry Pi. On the experimental evaluation, we checked (1) whether the master MA selection process works correctly, and (2) Whether all collected disaster information can be stored into the DB on master MA’s Raspberry Pi. For the evaluation for (1), in order to check the disaster information of slave MA, can be reflected to master MA or not, we modified the disaster information of slave MA several times during the evaluations. Figure 4 shows the snap shot of the experimental evaluations. Figures 5, 6 and 7 show the screen shot of the node where master/slave MA resident. As the figure shows, the MA exchanged Hello packet and Accept packet among the neighboring MAs. Figure 6 shows the two screen shots of slave MA and master MA. As the figure shows, the disaster information of slave MA correctly transmitted to master MA, and master MA collects all disaster information from all slave MAs, correctly.

A New Mobile Agent System for Sharing Disaster Information

Fig. 4. Experimental evaluation

Fig. 5. Screenshot of the node where a master MA residents

Fig. 6. Screenshot of the node where a slave MA residents (1)

229

230

N. Matsumoto and T. Shigeyasu

Fig. 7. Screenshot of the node where a slave MA residents (2)

These experimental results confirmed that our developed prototype of our mobile agent based disaster information system works correctly.

4

Discussion

In this paper, we have discussed the mobile agent system destined for collecting and sharing disaster information among wide area and local area for reducing effects of natural disaster. The results of experimental evaluation informed that the prototype system employing our proposed mobile agent system well works for exchanging information. In the future, we further discuss the how the disaster information system based on mobile agent technique utilize its mobility characteristics under disaster affected conditions, and, brush up our system for realizing.

References 1. Ministry of Public Management AHome Affairs APosts and Telecommunications: Disaster Information Board service (Online) http://www.soumu.go.jp/menu seisaku/ictseisaku/net anzen/hijyo/dengon.html. (In Japanese. Accessed 14 Dec 2018) 2. Ministry of Public Management AHome Affairs APosts and Telecommunications: Notice for using “00000JAPAN”. http://www.soumu.go.jp/menu kyotsuu/ important/kinkyu01 000125120.html. (In Japanese. Accessed 14 Dec 2018) 3. Twitter, Inc. F Twitter, Twitter, Inc. (online). https://twitter.com/. Accessed 14 Dec 2018 4. Facebook, Inc F Facebook, Facebook, Inc. (online). https://ja-jp.facebook.com/. Accessed 14 Dec 2018

A Deep Hybrid Collaborative Filtering Based on Multi-dimension Analysis Chunyan Zeng1, Songnan Lv1, Shangli Zhou1(&), and Zhifeng Wang2 1

Hubei Key Laboratory for High-Efficiency Utilization of Solar Energy and Operation Control of Energy Storage System, Hubei University of Technology, Wuhan, China [email protected] 2 Department of Digital Media Technology, Central China Normal University, Wuhan, China

Abstract. In order to solve the problem that the existing neural collaborative filtering methods are not comprehensive to mine the latent information of embedded vectors, a deep hybrid collaborative filtering based on multidimension analysis is proposed. The idea is to use different feature fusion methods for the embedded vectors of users and items to obtain multiple dimensional fusion features, so that the information explored by different methods can complement each other, and the model can better discover the interaction between users and items. Experimental results show that, compared with the single-method of dimension analysis, the multi-dimension analysis can effectively improve the model’s ability to mine the interaction between users and items, and improve the performance of the recommender system.

1 Introduction Collaborative filtering [1] technology is one of the most important technologies in recommender system, which is to understand the relationship between users and items according to the interaction information between users and items. Matrix factorization (MF) [2, 3] maps users and items into a latent space, then respectively uses a latent vector to represent feature of users and items, and finally uses the inner product of latent vectors to represent the interactive relationship between users and items. The traditional matrix factorization method uses the inner product [4] to explore the relationship between latent vectors, which makes matrix factorization method inevitably suffer from the limitation of inner product [3]. Inner product is an operation of reducing dimension, which undoubtedly leads to the loss of interactive information. For the simple and fixed inner product, it is difficult to mine the complex interaction between users and items in low-dimensional latent space. Therefore, matrix factorization is difficult to further improve the accuracy of recommendation. In recent years, a great deal of research has focused on extracting more interactive information from the implicit features of users and items. He et al. proposed a Neural Collaborative Filtering (NCF) [5], which used neural network to model embedded vectors (i.e., latent vectors) of users and items. They designed a generalized matrix factorization (GMF) model and a multi-layer perceptron (MLP) learning the model of © Springer Nature Switzerland AG 2020 L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 231–240, 2020. https://doi.org/10.1007/978-3-030-33506-9_21

232

C. Zeng et al.

interaction respectively. NCF makes matrix factorization more flexible and easy to extend to nonlinear space. However, the models designed in this framework are all based on the same assumption that each feature point in the embedded vector is independent of each other and provides the same contribution to the prediction. This assumption is unrealistic, because each feature point in the embedded vector can be considered as some attribute of the user or item, which is not necessarily independent of each other. Therefore, He et al. proposed an outer product-based neural collaborative filtering (ONCF) [6], which used the outer product to process embedded vectors to obtain a more expressive 2D mapping matrix, and used convolutional neural network (CNN) to learn the higher-order correlation between embedded vectors. In the above research, the embedded vectors of users and items are analyzed in different dimensions to obtain more information about the interaction between users and items. Multi-layer perceptron can theoretically approach any continuous function and retain the original data well, but cannot effectively capture the dimension correlations between embedded vectors. The outer product takes into account the correlation modeling of dimension correlations, which can discover more information than the traditional inner product method. In this work, we propose a deep hybrid collaborative filtering based on multidimension analysis (MDA), which uses multi-dimensions analysis methods to deal with the embedded features of users and items, so as to explore more interactive relations between users and items. The 1D vector is simply concatenate, which can retain the original information of embedded vector, and using a standard MLP model to learn the interactive relationship, which gives the model great flexibility and non-linear. 2D mapping matrix is obtained by outer product operation, not only contains the dimension correlations between embedded vectors, at the same time the diagonal elements corresponds to the inner product of the traditional matrix factorization method as a result. Outer product provides the higher-order correlation for the model, and 2D matrix format has been confirmed that interactive relationship can be learned by CNN, and has better generalization ability than DNN. The main works of this paper are as follows. 1. We propose the framework of deep hybrid collaborative filtering based on multidimension analysis, which uses multi-dimensions analysis methods to process the embedded vectors of users and items, so as to extract more interactive information from the embedded vectors. 2. We conduct extensive experiments on an implicit feedback data in the real world. Our method shows better performance and proves that multi-dimensions analysis can better explore interactive information. 3. This multi-dimensions analysis is a new attempt of matrix factorization method, which makes some explorations to explore the interactive relationship between users and projects.

A Deep Hybrid Collaborative Filtering Based on Multi-dimension Analysis

233

2 Proposed Methods We first introduce the framework of the deep hybrid collaborative filtering based on multi-dimension analysis. Then, we elaborates the model uses concatenate and outer product method to refactor implicit features of users and items. DNN and CNN are used to analyze the two refactored features respectively, and their analyze results will be fused.

yˆui

yui

concatenate DNN

CNN

concatenate User Embedded vector

outer product

User Embedded vector

0 1 0 0 0 0 0 User u

Item Embedded vector

Q

N ×K

Item Embedded vector

0 0 0 1 0 0 0 Item i

Fig. 1. The framework of deep hybrid collaborative filtering based on multi-dimension analysis

2.1

General Framework

Figure 1 shows the framework of deep hybrid collaborative filtering based on multidimension analysis. u and i respectively represent the user set and item set. Let the score data of the original data set convert into implicit data, and rui represent the implicit feedback [7] of users and items. It is defined that if the user has scored the movie, the interaction has been observed, otherwise it has not.
rui ¼

1; if interaction is observed; 0; otherwise:

ð1Þ

The input layer transforms the ID of user and item to one-hot encoding, which will I be used as the feature vectors vU u and vi of user u and item i. The feature vectors are input into the embedding layer, the embedding layer is essentially a full connection

234

C. Zeng et al.

layer, which can map sparse feature vectors into a dense vector. We can get the embedded vectors pu and qi from Eq. 2. pu ¼ P T v U u

qi ¼ QT vIi

ð2Þ

where P 2 RMK and Q 2 RNK are embedded matrix of user features and item features, respectively; K, M and N respectively donate the embedding dimension, user feature dimension and item feature dimension. P and Q is the target matrix that will be optimized when features embedded. In the refactor layer, two different methods are used to refactor the embedded vectors of users and items. Because different refactor methods may not have the same embedding dimension to achieve the best performance, if the output of the embedding layer is shared, the performance of the model will be limited. And to make the model more flexible, unique embedding vectors are provided for different refactor methods. The refactored features have different representation forms. In the hidden layer, appropriate networks are adopted to explore the interaction relations in these different representation forms. The last hidden layers of each network will be concatenated to get the predicted score ^yui from Sigmoid function, and it will be trained by minimizing the loss value between the predict score ^yui and the target score yui . 2.2

1D Analysis by Concatenation

Simple vector concatenation does not explain any interaction between the user and item embedded vectors, but it can preserve the original form of the embedded vector. Since neural networks have been shown to be able to approximate any continuous function, it is feasible to use DNN learning interaction between embedded vectors directly. The definition of the DNN model is as follows: 

p z1 ¼ /1 ðpu ; qi Þ ¼ u qi

 ð3Þ

/2 ðz1 Þ ¼ a2 ðW2T z1 þ b2 Þ

ð4Þ

/L ðzL1 Þ ¼ aL ðWLT zL1 þ bL Þ

ð5Þ

^yui ¼ rðhT /L ðzL1 ÞÞ

ð6Þ

where Wx , bx and ax donate the weight matrix, bias vector and activation function of the x-th neural network layer, respectively. ReLU function is selected as the activation function, because ReLU encourages sparse activation, which is very suitable for sparse data and can alleviate the over-fitting of the model. When setting up the network structure, the common tower structure is adopted, that is, the number of hidden layer neurons near the input layer is large, and the number of neurons closer to the output layer is smaller, so that the network can better learn the abstract features.

A Deep Hybrid Collaborative Filtering Based on Multi-dimension Analysis

2.3

235

2D Analysis by Outer Product

In order to obtain the correlation between embedded vector dimensions, we use outer product to deal with embedded features. The details are shown in Eq. 7. E ¼ pu  qi ¼ pu qTi

ð7Þ

where E is an interactive mapping matrix of size K  K. The first reason why the outer product is chosen as the feature fusion method is that the outer product is more meaningful than the simple concatenation [8]. The concatenation method can only retain the original information and cannot build any correlation modeling. Secondly, the diagonal elements of the interactive mapping matrix after the outer product are exactly the operation results of the inner product of the traditional matrix factorization method. This makes the outer product contain more information than the traditional matrix factorization method. Since the interactive mapping matrix is a 2D matrix, it has the properties of image. The interaction relations in the interaction mapping matrix can be regarded as local features in the image. As we all know, CNN has a great success in image field, so learning interactive mapping matrix using CNN can achieve better results than using DNN (Fig. 2).

Interactive map

Convolution Convolution Pooling Fully connected layer layer-1 Layer layer-2

64

64 × 64

64 × 64 × 32

64 × 64 × 32

32 × 32 × 32

Fig. 2. Using CNN to learn interactive mapping matrix

2.4

Learning Model

Considering that the model uses implicit feedback data, we consider the value of yui as a label, where 1 indicates that item i is related to user u, and 0 is not. The sigmoid function, which is common in binary tasks, is used as the activation function of the output layer, the model limits the output ^yui of the model to the range of [0, 1], and gives the model probability interpretation, that is, ^yui represents the degree of correlation between item i and user u. According to the above Settings, the likelihood function can be defined as shown in Eq. 8. Y Y ^yui pðg; g jP; Q; hf Þ ¼ ð1  ^yui Þ ð8Þ ðu;iÞ2g

ðu;iÞ2g

236

C. Zeng et al.

Taking the logarithm of the likelihood function, we can get: X X L¼ log ^yui þ logð1  ^yui Þ ðu;iÞ2g

ðu;iÞ2g

¼

X

yui log ^yui  ð1  yui Þ logð1  ^yui Þ

ð9Þ

ðu;iÞ2g [ g

where g is the positive example set, and g is the negative example set; P and Q respectively represents the embedded matrix of users and items when embedding, and hf represents the neural network parameters of learning interactive relationship. The model parameters are learned by maximizing the likelihood function, and the optimization can be realized by Adam.

3 Experiments 3.1

Dataset

MovieLens dataset have been widely used to evaluate collaborative filtering algorithms. We use the movielens-100 k dataset, which contains a total of 100,000 ratings of 1,682 movies by 943 users, each of whom has rated at least 20 movies. This experiment adopts the leave-one-out evaluation, that is, the most recent interaction behavior of each user is used as the test set, and the rest is used as the training set. Follow the general strategy when evaluating, a random sample of 100 items which have no interaction with users is used as a negative samples. 3.2

Evaluation Indicators

We used the Hit Rate (HR) and the Normalized Discount Cumulative Gain (NDCG) [9] as evaluation indicators for the top-K recommendation task. For the top-K recommendation task, HR directly evaluates whether the items in the test set appear in the top-K recommendation list. NDCG calculates the ranking of the hit items by assigning a higher score to the items in the recommendation list. The specific calculation formula is as follows. HR@K ¼

Number of Hits@K jGTj

8 K > < DCG@K¼ P i¼1

2rðiÞ 1 log2 ði þ 1Þ

> : NDCG@K¼ DCG@K iDCG@K

ð10Þ

ð11Þ

where jGTj represents the number of items in all test sets in Eq. 10. In Eq. 11, DCG represents the cumulative discount gain, calculates the ranking performance of the prediction list, and iDCG represents the cumulative discount gain of the optimal ranking list.

A Deep Hybrid Collaborative Filtering Based on Multi-dimension Analysis

3.3

237

Baselines

In this paper, the proposed deep hybrid collaborative filtering based on multidimension analysis is compared with the following methods in this field. Item-KNN [10]: This is a standard item-based collaborative filtering method. BPR [11]: Bayesian Personalized Ranking, which optimizes personalized ranking tasks, is often used as a baseline for item-based recommendations. MLP [5]: This is a model proposed in the NCF framework, using multilayer perceptron instead of inner product to learn the nonlinear relationship between users and items. Conv-NCF [6]: The outer product based neural collaborative filtering method is a state-of-the-art neural collaborative filtering method. CNN network is used to learn the outer product results of embedded vectors. 3.4

Parameter Setting

For each positive samples in the training set, we sample 4 items which have no interaction with users as negative samples. And we use the mini-batch Adam method to optimize the model with setting the learning rate to 0.001 and the batch size to 128. The dimension of embedded vector is a very important parameter. For the setting of embedded vector dimension, we test the model with embedded dimension of [8, 16, 32, 64]. It is worth noting that too large embedded dimension may lead to overfitting and performance degradation. Since the last hidden layer of each feature fusion method determines the performance of the method, the output of this layer is called Prediction Vector. If the size of predicted vector is 16, the number of layers for three-layer DNN network which processes concatenation results is allocated as [64, 32, 16]. Similarly, the number of nodes in the last full connection layer of the CNN network which processes the outer product result is also 16. Feature fusion methods use different size of predicted vectors will affect the performance of the model. 3.5

Performance Comparison

For the deep hybrid collaborative filtering based on multi-dimension analysis, the size of embedded vector dimension and the predicted vector dimension of each feature fusion method are two important hyper-parameters relatively. The size of embedded vector dimension determines the quality of feature extraction and the size of predictive vector dimension determines the quality of each feature fusion method. Therefore, when compare performance, we should consider the effects of these two hyperparameters on the model.

238

C. Zeng et al.

(a) HR@10

(b) NDCG @10

Fig. 3. Performance comparison of models in different embedded dimensions

Model Performance Comparison under Different Embedded Vector Dimensions Figure 3 shows the HR@10 and NDCG@10 of the model under different embedded dimensions. For ItemKNN, we tested the neighborhood of different sizes and take the best performance for comparison. For the deep hybrid collaborative filtering based on multi-dimension analysis, the multi-dimension is 1D and 2D which are generated by the concatenation method and the outer product method, and the predicted vector of both methods is 32, that is, the number of layers of DNN is allocated [128, 64, 32], and the number of nodes of the last full connection layer of CNN network is 32. It can be found that the deep hybrid collaborative filtering based on multidimension analysis achieves the best performance under several embedded vector dimensions. Among them, ConvNCF and MLP have better performance, while the accuracy of MDA which combined with these two methods increases by 2% and 5% respectively. This can intuitively explain that the multi-dimension analysis methods can complement each other, so as to better explore the preferences of users. It can be noticed that when the embedded vector dimension increases to 32, HR@10 starts to stabilize, and on the contrary, NDCG@10 decreases. Therefore, the setting of embedded vector dimension needs to be obtained through experiments, and too high set value will lead to overfitting. Model Performance Comparison Between Different Predicted Vector Settings Table 1 records the influence of the prediction vector settings in the deep hybrid collaborative filtering based on multi-dimension analysis on the model performance. According to previous experiments, the best setting for embedding vector dimensions is 32. Therefore, when comparing the influence of predicted vector on model performance, the embedded vector dimension is set as 32, and we test model with the predicted vector dimensions of [8, 16, 32, 64].

A Deep Hybrid Collaborative Filtering Based on Multi-dimension Analysis

239

Table 1. Performance comparison between different predicted vector settings. Embedded dimensions = 64

Predicted vector of outer product

8 16 32 64

Predicted vector of concatenation method 8 16 32 64 0.732 0.742 0.736 0.747 0.732 0.744 0.744 0.742 0.742 0.734 0.743 0.758 0.752 0.745 0.753 0.758

By comparing HR@10 of different prediction vector settings, it can be found that the model achieves the best performance when the dimension of prediction vectors of concatenation method and outer product method are 64. However, it’s not indicated that the larger the dimensions of the prediction vector, the better the performance of the model. Some combinations will make the model performance very poor, so the setting of prediction vector also needs to be verified by experiments, which is one of the inevitable problems in the framework of multi-model fusion.

4 Summary and Future Work In this paper, we explore matrix factorization method in recommender system based on collaborative filtering. A deep hybrid collaborative filtering based on multi-dimension analysis is proposed to complete the interaction between users and items by multidimension analysis methods, so as to improve the recommendation performance. At the same time, this paper also designs a model based on this framework, which fused concatenation method and outer product method to refactor feature vector into a variety of dimensions. Concatenation method can retain the original information, and the outer product method can explore the correlation between embedded vector dimensions, so as to obtain better performance than the two methods used alone. In the future work, we will continue to study the feature fusion method to refactor feature vectors into different dimensions and try to combine more dimension analysis methods. In addition, the work will be extended to accommodate more types of features, such as images, video, text, etc., so that the model can easily deal with more fields. Acknowledgments. This research was supported by National Natural Science Foundation of China (No. 61901165, No. 61501199), Excellent Young and Middle-aged Science and Technology Innovation Team Project in Higher Education Institutions of Hubei Province (No. T201805), Hubei Natural Science Foundation (No. 2017CFB683), and self-determined research funds of CCNU from the colleges’ basic research and operation of MOE (No. CCNU18QN021).

240

C. Zeng et al.

References 1. Zhang, H., et al.: Discrete collaborative filtering. In: The 39th International ACM SIGIR Conference. ACM (2016) 2. Koren, Y.: Factorization meets the neighborhood: a multifaceted collaborative filtering model. In: Proceedings of the 14th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, Las Vegas, Nevada, USA, 24–27 August. ACM (2008) 3. He, X., et al.: Fast matrix factorization for online recommendation with implicit feedback (2017) 4. Rendle, S.: Factorization machines. In: IEEE International Conference on Data Mining. IEEE (2011) 5. He, X., et al.: Neural collaborative filtering (2017) 6. He, X., et al.: Outer product-based neural collaborative filtering (2018) 7. Liang, D., et al.: Modeling user exposure in recommendation (2015) 8. He, X., Chua, T.-S.: Neural factorization machines for sparse predictive analytics. In: Proceedings of the 40th International ACM SIGIR Conference on Research and Development in Information Retrieval (2017) 9. He, X., et al.: TriRank: review-aware explainable recommendation by modeling aspects. In: The 24th ACM International. ACM (2015) 10. Sarwar, B., et al.: Item-based collaborative filtering recommendation algorithms. In: International Conference on World Wide Web (2001) 11. Rendle, S., et al.: BPR: Bayesian personalized ranking from implicit feedback (2012)

An Energy Efficient Mechanism for Downlink and Uplink Decoupling in 5G Networks Christos Bouras(B) , Georgios Diles, and Rafail Kalogeropoulos Computer Engineering and Informatics Department, University of Patras, Patras, Greece [email protected], {diles,kaloger}@ceid.upatras.gr Abstract. In current cellular networks, cell association is heavily based on the Downlink signal power and all devices are associated with the same Base Station in Downlink and Uplink. While as of now this technique has been proved adequate in homogeneous networks where all BSs have similar transmission levels, in increasingly dense heterogeneous networks rate is heavily dependent on the load, which can significantly vary from Base Station to Base Station. Due to increased demands for usage over several devices in heterogenous networks, large disparities in the Downlink pose a threat to the quality of services rendered by the network and this technique seems obsolete. Uplink and Downlink decoupling is the proposed solution, where the Downlink cell association is not necessarily based on the same criteria as Uplink. We propose using SINR and Path Loss with Range Extension as factors for choosing the appropriate Base Station for connection in Downlink and Uplink respectively, taking into consideration the Base Stations’ Resource Block availability, to avoid overloading Base Stations and we will use simulations to test our theory.

1

Introduction

Recent 4G cellular networks’ design foundations are based on Macro cells (MCells) that featured the same characteristics throughout the network. They were preferred because of their high transmit power, which ensured high Signal to Interference and Noise Ration (SINR). Their similarities extended to the number of users they can support [3]. Their focus was centered on improving peak rate and spectral efficiency to offer the best user experience. This architecture is commonly referred to as Homogeneous Networks. Although until now, this approach was adequate, in recent years network traffic demands keep rising making the deployment of heterogeneous networks (HetNets) necessary in order to adequately correspond to these needs. Heterogeneous networks consist of Macro cell Base Stations (MBSs) and Small cell Base stations (SBSs) that are scattered among a Macrocell Base Station’s vicinity [13]. HetNets have been implemented already successfully in 3G and 4G networks, but they were not designed as a part of them from the beginning, making it necessary to implement fundamental c Springer Nature Switzerland AG 2020  L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 241–252, 2020. https://doi.org/10.1007/978-3-030-33506-9_22

242

C. Bouras et al.

changes in the design of 5G networks to ensure their successful implementation. With the rise in user demands, 5G networks need to be user oriented in order to make them accessible from all users and flexible to work with. With the shift to hetnets, the connection between Uplink (UL) and Downlink (DL) is differentiated from homogeneous networks and not necessarily based on the same Base Station (BS) [9]. Traditionally the throughput required for DL was vastly higher than the one used in UL, resulting in a lack of symmetry between the resources needed to accommodate network traffic for each one [12]. Moving on to HetNets, the transmit power of all transmitters in the UL is almost identical since most of them are battery powered portable devices. The UL throughput needs keep rising, mostly thanks to applications that make equal use of both uploading and downloading such as social networks, real time streaming or server connection for video games and UL’s independency to distance between nodes and the amount of traffic in the network. Separating UL and DL into two different subnetworks that make up our network enables us to individually design a model for each network for the sake of eliminating interference with neighboring cells and increase cell association as well as throughput. In this design the User Equipment (UE) can decide on how to establish connection with the BSs, either by connecting with the same cell or with different cells (MBSs or SBSs) during Ul or Dl communication. HetNets are already applied, and became denser with more SBSs scattered among the network. Several papers have explored the notion of Downlink and Uplink Decoupling (DUD). In [2] it is suggested that it is part of a broader “device centric architectural vision”, since the set of network nodes used to connect a certain device to the grid as well as the functions of these nodes in a particular communication session, are tailored to that specific device and session [2]. A disruptive architectural design to study the gains of DuD in UL capacity was proposed in [4]. Additionally, previous attempts have studied the energy efficiency of this method since UL/DL decoupling allows for more flexibility in switching-off some BSs and also for saving energy at the terminal side [10]. Worth mentioning is Range Extension (RE) for UL, by adding a selection offset to the reference signals of Smallcells (Scells) to increase coverage and alleviate traffic from Macrocells (Mcells). Offsets greater that 3–6 dB cause interference on the DL, leading to the development of methods to try and combat these interferences [8]. On the pursuit of energy efficiency for non peak periods, [11] suggest two types of algorithms, for application to any network type, centralized and distributed. In this paper, we try to improve existing models on this subject, which propose that different BSs are set responsible for UL and DL connection with users, not limiting MBSs and SBSs to DL or UL connections in any way. We will provide an allocation algorithm that will efficiently match a UE to a BS on either UL or DL or both. The proposed mechanism suggests that unused BSs be shut down to minimize resource and energy usage, while still providing adequate QoS and allowing more users to access and connect to the network.

An Energy Efficient Mechanism for Downlink and Uplink Decoupling

243

In remaining sectors, we will complete our proposal. Specifically in Sect. 2 we will take a look on the system model of DUD. In Sect. 3 we propose a more efficient algorithm for matching users to the right BS. In Sect. 4 we will analize our simulations setup. In Sect. 5 we will present and analyze the results from our simulations and finally in the end, in Sect. 6 we draw our conclusions and propose our suggestions for future work.

2

System Model

Small Cells become smaller (nano, pico etc) so transmit power differences between MBSs and SBSs are constantly rising, raising the need to optimally match a UE with the appropriate BS for its UL/DL needs. With this technology a UE will now be able to effectively choose Small or Macro Cells to connect to, based on its requirements. That means that even if it is connected to a certain BS in one channel, it can still choose a different BS to connect to. Traditional cell association suggests that both UL and DL is based on the maximum DL Received Power (RP) as measured at the UE [11]. In our research we assume that DL association is based on SINR, while UL association is based on pathloss, where we also apply Range extension (PLRE). Each user calculates their desired rate and then computes the number of RBs necessary from the BSs to achieve such rate, and tries to connect with their desired BSs. We consider a multi-tier HetNet that consists of Macro cell Base Stations (MBs), Small cell Base Stations (SCs) and User Equipments (UEs). Suppose that we have a set of MBSs (M = 1, ...., | M |), a set of SMSs (S = 1, ...., | S |) and a set of UEs (U = 1, ...., | U |). All users want to transmit and receive to both directions (UL,DL) that can be considered as separate channels in the network. We consider that users are arranged in space following homogeneous Poisson point process (PPP) φ of intensity λS , λN . Finally each BS has a maximum number of users that it can serve simultaneously quoted as ni , i = 1, ...., | U | and also features a number of available Resource Blocks (RBs). The main focus of UL/DL Decoupling is to offload MBSs and distribute load among remaining SBSs in the MBSs vicinity, enabling better performance for users. In a traditional coupled example each user would link to a single BS, based solely on best DL performance regardless of UL performance. In decoupled models each user will connect to the best possible BS for DL and UL respectively. In our case users connect to a MBS for DL based on DL SINR and to a SBS based on Path loss with RE, seeking the BS that can provide their desired Data Rate (DR) with the least amount of RBs. In our research we will assume a typical metropolitan area network scenario. We assume that MBSs are mostly mounted above rooftop levels in order to provide continuous uninterrupted coverage in Large Cells. SBSs are based below rooftop levels city-wide to cover up for Non-Line-Of-Sight (NLOS) conditions, that stationary users, or users roaming the streets may experience. These SBSs are able to provide high throughput in areas with high user density especially for indoor terminals, by increasing spatial reuse and thus reducing the number of users per cell [10].

244

C. Bouras et al.

We assume that there is no interference between two users located within the same cell as they can be each assigned to non-interfering sets or resource blocks (RB). A RB a flexible resource structure, where the time-frequency spectrum is divided into orthogonal RBs [3]. First, we calculate Data Rate (DR) as: DR = BRB ∗ log2 (1 + SIN Rj,i ),

(1)

where Brb corresponds to the bandwidth of a specific RB and SIN Rj,i is the signal to interference and noise ratio between user j and Base Station i. The total DR for the whole system is equal to the sum of the DRM (Mcell DR) and DRS (Scell DR). This allows for more flexible resource allocation schemes enabling us to achieve higher spectral efficiency. To compute the number of RBs that a user (suppose user j) needs from a specific Base Station, in order to achieve the desired rate, we will use the following equation:   gj Rj,i = , (2) BR B ∗ log2 (1 + SIN Rj,i ) where gj denotes the UE throughput demands and DRi the desired Data Rate for the user j. Next we shall define the parameters suggested for DUD, PL and SINR. We will define PL for the distance dependent Path loss model, when a user is connected with MBSs (P LM ) as well as SBSs (P LS ). Pathloss is a metric to measure the signal loss in our wireless communication network. The equations that describe them are given below, both for UL and DL: P LM = 128.1 + 37.6 ∗ log10 d P LS = 140.7 + 36.7 ∗ log10 d

(3) (4)

And SINR for DL can be calculated as: DL SIN Ri,j =

3 3.1

Pj gi,j d-a i,j , ΣkB

(5)

Matching Algorithm Our Approach

Taking into consideration that each BS has an upper bound on its RB capacity we consider as W the number of Resource Blocks (RBs) available for each BS. In this research we consider that all users are equal, with no user having a priority for serving, even though they have different needs and requirements from their respective matched BS. Each user obviously wants to receive and send data and do so with a desired Rate. We assume that all MBSs share the same capacity and all SBSs share the same capacity as well. This issue greatly resembles the Knapsack Problem (KP), on which each Base Station resembles a knapsack,

An Energy Efficient Mechanism for Downlink and Uplink Decoupling

245

while users are different objects. We want to maximize Data Rate, so dr can be considered as the profit for selecting a user and the RBs that each user asks for, can be considered as the weight of each user. We will analyze the Knapsack problem for a complete overview later. Since we focus on UL, we will try to apply Range Extension only on Path loss (our metric for UL association), to satisfy as many users as possible. Range Extension means that while each BS has their own UL border and DL border, now all SBSs are provided with a positive offset, an advantage over MBSs to expand their UL border in order to offload MBSs from some users. Prioritizing SBSs over MBSs in Uplink is expected to yield positive results in the total number of users satisfied and lead to a more uniformally distributed user allocation over the network. 3.1.1 The Knapsack Problem In the KP we have a set of objects, each one with a specific value (vi ) and its respective weight (wi ). We also have a knapsack, with a limited space of W and the goal is to fit as many objects in the knapsack as we can in order to achieve the maximum profit. Each object can either be selected or not selected, so the problem is usually referred to as the 0–1 Knapsack Problem. Obviously since the carrying capacity of the knapsack is limited, subsequently the amount of objects we can carry is limited. Given a set of items (suppose n items) we want to maximize our profit [1]: n 

Ui Xi

(6)

i=1

Suppose that we can carry a set of m (m < n) objects. For all objects we define (Xi ) and Xi = 1 when an object belongs in the set of chosen objects or Xi = 0, when an object is not selected. Obviously for our set of selected objects: n 

Ui Xi ≤ W

(7)

i=1

3.2

The Associations

An association denotes a connection between a user and BSs. This connection can be either on UL or on DL and a user can be connected to different BS for their UL and DL connections, or the same BS for both. Each user i ⊇ N, can sign a “contract” which includes the identities of the associated with the user BSs for UL and DL connection. Think of a two BS example. Let possibilities for association be the following: user 1 prefers an association with BS 1 in the UL direction and BS 2 in the DL direction based on the utility function we use, so user 1 has a preferred association of {U L1, DL2}. For the two BS example, we define the preferences as following: {U L1, DL2} > {U L1, DL1} > {U L2, DL1} > {U L2, DL2}. For each BS j ⊇ B, we define two separate lists of preferred relations for UL and DL direction, over the set of possible associations.

246

C. Bouras et al.

3.3

The Algorithm

The proposed algorithm aims to provide a stable algorithm that optimizes produced results. The network starts with no users associated with any BS and we consider that there are no limitations to the assignments of users. Each user creates a list consisting of pairs, one for their UL connection and one for their DL connection where each user selects their preferred BS for UL and DL based on PLRE for UL and on SINR on DL. In the UL channel we apply Range Extension on PL, meaning SBSs are favored over MBSs with a positive offset. Range extension is applied if the user attempts a connection with a MBs only, to favor SBSs, in an attempt to offload MBSs. Regarding BSs, users eligible for matching are selected based on the number of RBs they (the BSs) have available. When a user’s (let’s assume user i) connection with the BS (assuming BS j) is accepted in the UL direction we use U Lj to suggest a connection, or DLj to suggest a connection with the user on DL direction. In other words we have a list for each BS stating a connection on UL or DL with each user. At an initial stage we can assume that neither users nor BSs have a reason to drop a possible connection. In the main phase, users will rank their preferences (with the available contracts) using the proposed scheme (UL: PLRE, DL: SINR). At this point each user has created a preference list. Then each user will submit a request at the BSs for acceptance in the UL or DL direction. It is obvious that each user wants to connect with its most preferred BS in each direction, namely the first match in their matching list. As some BSs may not accept some users’ connection, each rejected user attempts to connect with their next most preferred BS, based on the same parameters, meaning that their list of preferences is taken into account in descending order. Now each BS will have to prioritize its available connections and accept requests or decline them. We consider that each user must connect to some station both on UL and on DL direction. Which means that for user i there should be at least two connections with BSs (let’s assure BS j and BS k, where j and k are not necessarily different). After each user produces its preference list, they calculate the number of RBs needed to achieve their desired rate. Each UE transmits to its preferred BS along with their needed RBs. Each BS builds its preference list over the available set of connections based on the number of RBs that each user requests. Each BS has its own limit of users it can accept. All BSs accept users, until they can no longer accept any more users.

An Energy Efficient Mechanism for Downlink and Uplink Decoupling

247

Algorithm 1. Pseudocode for the Matching Algorithm 1: 2: 3: 4: 5: 6: 7: 8: 9: 10: 11: 12: 13: 14: 15: 16: 17: 18: 19: 20: 21: 22: 23: 24: 25: 26: 27: 28: 29: 30: 31: 32: 33:

U: Denoting all users MB: Denoting all Macrocell Base Stations (29) SB: Denoting all Smallcell Base Stations (45) for num of users= 100, 200, 500, 1000,2000 do for i = 1 to M B do ACCM(i): Empty; end for for i = 1 to SB do ACCS(i): Empty; end for for i = 1 to U do Create BS preference list for DL over SINR; Create BS preference list for UL over PL, applying RE in favor of SBs; Transmit request to most preferred BS for DL, UL; Calculate number of RBs to achieve wanted Rate; end for for J = M B and J = SB do for i = 1 to J do If total BS RBs < user wanted RBs => accept; If(best PL BS accepts user) then subtract user given RBs from total BS RBs If(no BS accepts user) then check next user end for end for for i = 1 to M B do If MBS(i) serves no users Then shutdown; end for for i = 1 to SB do If MBS(i) serves no users Then shutdown; end for end for

As users are rejected by a BS, then they submit a contract proposal to their next set of preferred BSs for connection. Again each BS shall create a new waiting list (consisted of all the users that can connect with them). The algorithm only concludes when there are no users left unassociated with a BS. Base Stations with no users matched to them, shut down to reduce energy waste.

4

Simulation Setup

For the evaluations performed in this paper, we will model a 5G System in a MATLAB based network simulator following the distance dependent pathloss model for macro cells and small cells. MATLAB provides standard functions

248

C. Bouras et al.

and an intuitive GUI for the design, simulation and verification of Advanced Communication Systems such as mobile networks. It is usually used for research and educational use. MATLAB simulations allow us to perform evaluations that are more difficult or impossible to perform with real systems and study the behavior of our mechanism in a highly controlled, reproducible environment. We executed simulations for as low as 100 users and as high as 2000 users for a network approach of 29 (fixed) macro cell BSs and 36 or 45 small cell BSs. Our network consists only of the colored hexagons, and all grey colored hexagons are not considered part of our simulation network. At the center of each hexagon we encounter the MBs displayed as a large triangle and two or three small triangles that represent SBs.

Fig. 1. Our MATLAB simulated network, consisting of 29 MBSs, 45 BSs and a total of 200 users

In our simulations we will consider an area that consists of Macrocell Base Stations (omni-directional with an inter-site distance of 375 m) and Smallcell Base Stations (omni-directional with a radius equal to 50 m). As for our simulation deployment scenario, we will simulate a network and model its performance for different numbers of users. That way we hope to create a model to study the network’s scalability potentials. At first, we consider 100 users that seek to use the resources of our network and later this number escalates to 200, 500, 1000 and finally 2000 users. All users have are randomly generated with a personalized chance of appearing inside our area of interest that is served from a cell and a small chance to appear beyond our network. In the downlink network as well as the uplink network, all users have their personalized demands for data rate that range from 2048 (Kbps) to 32768 (Kbps) for the downlink and for the uplink, their demands range from 2048 (Kbps) to 16384 (Kbps). All our simulation parameters will be later presented in the following array (Table 1):

An Energy Efficient Mechanism for Downlink and Uplink Decoupling

249

Table 1. Simulation parameters Parameter

Setting

Propagation model

Macro cell propagation model

DL Bandwidth

100 MHz

UL Bandwidth

100 MHz

User distribution

Poisson point process

Network deployment

29 Mcells and 45 Scells

Number of users

100/200/500/1000/2000

Carrier frequency

3.5 GHz DbM,

Modulation UL Scheme 64QAM

5

Simulation Results

Applying Range Extension on the UL results in an increased number of successful user associations. DRs are proven to be lower in PLRE, by a small margin in comparison to simply applying PL for DUD in the UL channel. The similar DRs with the increased number of associations, result in an increased total network throughput and leads to higher user satisfaction. Channel quality is quite important in the overall satisfaction of throughput demands and as the number of users increases, it is all the more difficult for a large percentage of users to connect to a BS, let alone their preferable BS, which results in setting an upper bound for the total produced DR. While Small Cells have a large coverage in the UL and they are in fact, able to satisfy a respectable amount of users, as the number of users increases, we notice that the DR for PLRE is following the same pattern as the DR based on simple PL association. It is important to note how DRs seem to increase and decrease as the number of users increases. This not only due to channel quality but it also indicates the need for Adaptive Range Extension (ARE) with different offsets dependent on the BS load or the number of users and possibly indicates that each network should feature different numbers of SBs dependent on the expected user load. Applying PLRE as the UL metric, we see that more users are connected with Small cells in comparison to simple DUD with PL for UL. Considering that increasing the number of network SBSs is a viable option, this ensures a more homogeneous distribution of UEs between the network nodes and less congestion on Mcells, which, in turn, leads to a better distribution of the network resources. Small cells are used increasingly and steadily by the proposed method, which combined with the increased total user associations, can prevent network congestion in extreme scenarios. Additional studies could examine the best positioning of Scells in order to maximize user allocation, DRs and/or total throughput. The fact that the UEs connect to the node to which they have the lowest PL results in reduced UL interference. As a result, users are provided with a higher UE SINR allowing the use of higher modulation schemes which results in higher

250

C. Bouras et al.

Fig. 2. Association percentage for PLRE over PL as UL metric

Fig. 3. Average DRs for PLRE over PL as UL metric

speeds for the UL channel. In the DUDe case, UEs are distributed more evenly between the nodes. As the number of UEs increases, all the available nodes are heavily congested in extreme scenarios, but for lower congestion scenarios, some BSs may be able to shut down, and since PLRE favors SBSs, MBSs can lower their transmit power for energy efficiency. The produced results indicate that Range Extension in PL, seems promising in DUD, which in turn, especially for future networks where the network load is expected to increase in the UL and Macro layer.

An Energy Efficient Mechanism for Downlink and Uplink Decoupling

251

Fig. 4. Number of small cells associations for PLRE vs PL as UL metric

6

Future Work and Conclusion

From the produced results, we see that as the number of network users rises, Base stations are increasingly having problems satisfying most of the users. In this regard, we applied Range Extension on Path Loss for association in the Uplink channel to valorise Small cell Base Stations. The results are promising, with a homogeneous user distribution across the network and higher association rates. The DRs of the proposed mechanisms show potential of improvement, which we expect to be possible by applying Adaptive Range Extension on Path Loss. User association on wireless mobile networks is a matter of intensive scientific research activities, which makes it essential to study this matter in conjunction with other study fields. As possible candidates, we expect machine learning and game theory to help us understand the mechanics behind overcomplicated network scenarios, to refine existing mechanisms and radically increase efficiency in these networks. Game theory can model existing and possible scenarios and enable us to provide different allocation mechanisms for different scenarios based on the users’ needs and strategies, while machine learning can provide us with the tools to alternate these mechanisms in real time or predict these scenarios.

References 1. Lee, C., Lee, Z., Su, S.: A new approach for solving 0/1 Knapsack Problem. In: 2006 IEEE International Conference on Systems, Man and Cybernetics, Taipei, pp. 3138–3143 (2006) 2. Boccardi, F., Andrews, J., Elshaer, H., Dohler, M., Parkvall, S., Popovski, P., Singh, S.: Why to decouple the uplink and downlink in cellular networks and how to do it. IEEE Commun. Mag. 54(3), 110–117 (2016)

252

C. Bouras et al.

3. Boostanimehr, H., Bhargava, V.K.: Unified and distributed qosdriven cell association algorithms in heterogeneous networks. IEEE Trans. Wireless Commun. 14(3), 1650–1662 (2015) 4. Elshaer, H., Boccardi, F., Dohler, M., Irmer, R.: Downlink and uplink decoupling: a disruptive architectural design for 5G networks. In: 2014 IEEE Global Communications Conference, pp. 1798–1803 (2014) 5. Cao, J., Zhu, D., Lei, M.: Uplink-downlink interference alignment in TDD-based cellular networks. In: 2013 IEEE 24th Annual International Symposium on Personal, Indoor, and Mobile Radio Communications (PIMRC), pp. 353–357 (2013) 6. Zhang, J., Soldati, P., Liang, Y., Zhang, L., Chen, K.: Pathloss determination of uplink power control for UL comp in heterogeneous network. In: 2012 IEEE Globecom Workshops, pp. 250–254 (2012) 7. Samdanis, K., Kutscher, D., Brunner, M.: Self-organized energy efficient cellular networks. In: 21st Annual IEEE International Symposium on Personal, Instanbul, Indoor and Mobile Radio Communications, pp. 1665–1670 (2010) 8. Liu, L., Chang, Y., Qin, R., Zhang, C., Yang, D.: On UL-DL imbalance mitigation for HSPA heterogeneous network. In: 2014 IEEE 80th Vehicular Technology Conference (VTC2014-Fall), pp. 1–7 (2014) 9. Shi, M., Yang, K., Xing, C., Fan, R.: Decoupled heterogeneous networks with millimeter wave small cells. IEEE Trans. Wireless Commun. 17(9), 5871–5884 (2018) 10. Rost, P., Maeder, A., Perez-Costa, X.: Asymmetric uplink-downlink assignment for energy-efficient mobile communication systems. In: 2012 IEEE 75th Vehicular Technology Conference (VTC Spring), pp. 1–5, May 2012 11. Singh, S., Zhang, X., Andrews, J.G.: Joint rate and SINR coverage analysis for decoupled uplink-downlink biased cell associations in HetNets. IEEE Trans. Wireless Commun. 14(10), 5360–5373 (2015). https://doi.org/10.1109/TWC.2015. 2437378 12. Sun, S., Adachi, K., Tan, P.H., Zhou, Y., Joung, J., Ho, C.K.: Heterogeneous network: an evolutionary path to 5G, pp. 174–178, October 2015 13. Feng, Z., Feng, Z., Li, W., Chen, W.: Downlink and uplink splitting user association in two-tier heterogeneous cellular networks, In: 2014 IEEE Global Communications Conference, pp. 4659–4664, December 2014

Efficient 5G Network Decoupling Using Dynamic Modulation and Coding Scheme Selection Christos Bouras1,2(&), Vasileios Kokkinos1, and Evangelos Michos2 1

2

Computer Technology Institute and Press “Diophantus”, Patras, Greece {bouras,kokkinos}@cti.gr Computer Engineering and Informatics Department, University of Patras, Patras, Greece [email protected]

Abstract. The transmission power limitations in telecommunication systems are tackled by using low code rates and high order modulation schemes, thus achieving high spectral efficiency with lower costs per bit. This paper evaluates the User-Centric model for Fifth Generation wireless telecommunication systems, attempting to efficiently improve user terminal-Base Station communication aspects. We propose a resource-aware method of improving network coverage across all layers by choosing to decouple the overall network into two separate uplink and downlink networks. The algorithm fully respects each user’s throughput demands and solves the User Equipment – Base Station association problem efficiently by choosing the appropriate Modulation and Coding Scheme that maximizes spectral efficiency inside each macro cell coverage area. Results are evaluated for both acceptable frequency ranges defined in 5G New Radio protocol interfaces (namely, Frequency Range 1 and Frequency Range 2) and show that the proposed Modulation and Coding Scheme-based mechanism offers perfect Quality of Service preservation and augmented data rates in favor of ultimate user coverage, in both scenarios. Additionally, due to the extended resources Frequency Range 2 provides, the equivalent simulation not only offers increased data rates compared to Frequency Range 1, but also a lower number of devices unsupported by the mechanism.

1 Introduction Nowadays, the demands for high throughputs and network coverage are forcing new telecommunication systems each decade. Starting with analog telecommunication standards under the name of First Generation Telecommunication Networks (1G Networks) which were introduced in 1980, we have now evolved into the Fifth Generation Wireless Communication systems (5G networks), which are projected to be implemented by the year 2020 [1]. 5G networks promise to offer an integrated heterogeneous network (HetNet), supporting high data rates, accompanied with satellite communication systems that cover large geographical areas. In order for the aforementioned targets

© Springer Nature Switzerland AG 2020 L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 253–265, 2020. https://doi.org/10.1007/978-3-030-33506-9_23

254

C. Bouras et al.

to be fulfilled, many aspects of wireless telecommunications must be taken under high consideration, such as energy consumption, bandwidth, interference and latency and overcoming those barriers can undoubtedly help 5G networks in achieving their goals. 5G HetNet architecture is envisioned to be shifted from the existing network-centric (NC) model and towards an innovative user-centric (UC) model [2]. A user-oriented network model can offer improved connectivity aspects between a network subscriber and his equivalent Base Station (BS) inside dense urban HetNet deployments. The main difference between traditional homogeneous networks and heterogeneous ones is that a HetNet decouples the original network into uplink (UL) and downlink (DL) networks and considers them as two separate networks under different underlying architectures and system models. This offers an option for a User Equipment (UE) to connect to different BSs in the UL and DL, giving the decoupled networks increased freedom over managing UE-BS associations. HetNets are envisioned to extend existing macrocell infrastructures by installing small cells in areas near the macro cell borders, offering improved coverage and throughputs for UEs near cell borders. But, in order for 5G networks to succeed, the appropriate Modulation and Coding Scheme (MCS) must be selected and applied, since it determines the actual throughput for a UE that connects to a BS. In [1], the paper discusses 5G requirements as far as MCS is concerned and acknowledges high SE as a key requirement for 5G, which in general can be ensured by adopting a high order modulation and a low code rate at a high Signal-to-Noise Ratio (SNR). The authors of [2] suggest a power-efficient UC approach for HetNets, which targets at overall energy consumption mitigation while respecting the user’s Quality of Service (QoS). In [3], four different approaches are presented on efficiently selecting the appropriate MCS in the physical layer in order to improve over-the-air Spectral Efficiency (SE), developing each scenario to be applicable in different real-life situations. The authors of [4] proposed a deployment scenario for an Long Term Evolution (LTE) dense HetNet infrastructure through UL and DL decoupling and suggested a DL association based on the receiving power and an UL association based on the pathloss. In [5], novel multiple access 5G technologies are tested in a 5G field trial, with the results showing that as long as the LTE protocol is followed, Orthogonal Frequency Division Multiple Access (OFDMA) and turbo coding can achieve augmented SE over 100%. The authors of [6] targeted at improving power and spectral efficiency of the network while preserving QoS at the user’s point of view, accompanied by a user association algorithm which ensured the UE-BS according to the lowest energy consumption per BS. Lastly, in [7], the paper authors took advantage of the recently accepted Third Generation Partnership Project (3GPP) technical specifications for the 5G New Radio (NR) wireless communication systems and investigated different channel scenarios based on adjustable bandwidths and scalable sub-carrier spacing. Simulations regarding achievable throughputs and Block Error Rate (BER) that follow the low-density parity-check (LDPC) approach instead of turbo code revealed that system performances tend to increase if the LDPC coding scheme is used.

Efficient 5G Network Decoupling

255

In this paper, we suggest a context-aware approach that improves network coverage by decoupling the dense HetNet into two different UL and DL networks. The augmented needs and requirement that 5G networks dictate are satisfied through the adoption and application of the 5G NR radio interface protocol at our network’s physical layer. The algorithm preserves QoS and targets at a maximum SE inside the macro cell coverage area, thus providing higher data rates but without ensuring coverage for all users. The mechanism takes into consideration the Resource Block (RB) demands for each UE and begin the iterations from the UEs with the lowest RB demands in order to satisfy as many users as possible. Practical user data rates derive from the selection of the appropriate MCS. Simulations are carried out in both applicable 5G physical layer scenarios, namely Frequency Range 1 (FR1) that holds frequencies between 450 MHz and 6 GHz and Frequency Range 2 (FR2), holding frequencies from 24.25 GHz up to 52.6 GHz [8]. Results show that our mechanism succeeds at respecting user QoS demands and offers higher throughputs than originally requested. Moreover, simulation parameters of FR2 scenario reveal that the proposed mechanism not only behaves better at offering augmented data rates compared to the FR1 deployment scenario, but also a larger number of devices may be served using the same mechanism. The remainder of this paper is organized as follows. Section 2 describes the system model for the aforementioned small cells deployment scenario. Section 3 provides a thorough analysis of the proposed scheme and Sect. 4 offers a summary of the simulation formulas and results that evaluate our system model. Finally, in Sect. 5 we present our summarized conclusions for this paper and provide insights over future works.

2 System Model Our geographical area of interest shall be an urban HetNet where each cell has a fixed size, equal to all neighboring cells and inside every cell, there exists only one BS, positioned in the cell center. According to the technical specifications for 5G networks that follow the NR radio interface protocol (see [9]), as far as the physical layer in concerned, the DL network follows the conventional Orthogonal Frequency Division Multiplexing (OFDM) using a normal cyclic prefix, whereas the UL network complies with conventional. For a network to follow the NR standard means a uniform distribution of the available system frequency into Resource Blocks (RBs) and supplying each RB with a pre-defined number of 12 sub-carriers. The UC model illustration appears in Fig. 1.

256

C. Bouras et al.

Fig. 1. A network instance of the UC model scenario.

2.1

Path Loss Models

In this section of the paper, we begin to formulate the equations for the distancedependent path loss models for both the macro cell tier and the small cell tier, for an urban deployment. We accept that the path loss models are identical between the UL and DL network models and thus, according to [10], the propagation model as far as the macro cell tier is concerned is as accurate as: PLmacro ¼ 128:1 þ 37:6  log10 ðd Þ:

ð1Þ

where d stands for the UE-macro cell norm-2 distance in kilometers. Moving over for the propagation model for the small cell deployment in our designated dense urban model, the propagation model is computed as [11]: PLsmall ¼ 140:7 þ 36:7  log10 ðd Þ:

ð2Þ

where d once more denotes the norm-2 distance between the small cell and the UE in its vicinity. The problem considers no further losses due to wall loses (especially for the case of the user-installed femtocells inside home residents). Both PLmacro and PLsmall are measured in dB. The channel gain is the same for both the UL and DL networks (denoted as GUL and GDL respectively) and equal to: GUL ¼ GDL ¼ 10PL=10 :

ð3Þ

Efficient 5G Network Decoupling

2.2

257

Data Rates Model

For a UE to link to its equivalent BS, the UE is served with a number of RB provided by the linked BS. According to [10], for a fixed number of user throughput demands by the time the link must occur, the RB demands of a user depend from his throughput demands, the available bandwidth and its Signal-to-interference-plus-noise ratio (SINR) towards the desired BS. In detail, the RB demands can be expressed as: rbj;i ¼

th
j ; BRB  log2 1 þ SINRj;i

ð4Þ

where the de operator denotes ceiling function to the nearest integer equal or greater to the element inside, thj stands for the UE data rates demands, BRB is the bandwidth of the equivalent RB and SINRj;i is the SINR as experienced from the UE’ point of view. It is worth mentioning that since we choose to round the RB demands towards the next integer, the vast majority of the users will be provided with more bandwidth than the one requested (its actual value would be a floating number). We will later see in our simulation that this will result in preserving the Quality of Service (QoS) for all the network users served from a BS. DL Network: Under the OFDM architectural consideration, let SINRDL i;j be the SINR measured from an ith BS to a jth UE be, according to [12], equal to: SINRDL i;j ¼

Prad G i P i;j ; N0 Df þ i0 Prad  Gi0 ;j i0

ð5Þ

where Prad stands for the power radiation output from the BS, Gi;j corresponds the i channel gain between the specific UE and the BS, N0 is the white noise power spectral P  Gi0 ;j is the sumdensity, Df is the sub-carrier spacing (SCS). Additionally, i0 Prad i0

mation of the radiated power of every other BSs except the ith BS multiplied with the channel gain between these BSs and the equivalent jth UE. All calculations are measured over RBs instead of sub-carriers. To further utilize our channel’s capacity, we have concluded that adaptive MCS should be followed. The data rate as experienced from the user’s point of view, when the jth UE establishes a link towards an ith BS, highly depends on the number of RBs it receives upon requests and on the SINR between them. More specifically, if RDL j;i refers to the user’s data rate in the DL model, then [3]: X RDL WRB  crSINR  ð1  BLERSINR Þ; ð6Þ j;i ¼ jr j  r2RB

where jr j corresponds to the cardinality of the RBs required to fulfil the user’s throughput demands, WRB is the specified bandwidth of the RB, crSINR refers to the code rate of the selected modulation scheme and BLERSINR the DL SINR-dependent block error rate.

258

C. Bouras et al.

UL Network: Following the OFDM architecture and according to [2], we consider th th SINRUL j;i as the SINR measured from an i BS to a j UE to be equal to: SINRUL j;i ¼

Prad  Gj;i j P N0 Df þ j0 Prad  Gj0 ;i i

ð7Þ

where Prad corresponds to the power radiated from the UE, Gj;i is the UE-BS channel, j P  Gj0 ;i denotes the summation N0 and Df stay the same with the DL model and j0 Prad i from each network UE except the jth one, of the multiplication between the radiated power the UE with the equivalent UE-BS channel gain. Let RUL j;i be the user’s data rate in the decoupled UL network. This results in:  UL     crSINR  ð1  BLERSINR Þ; RUL j;i ¼ Ws  Ns

ð8Þ

where as Ws we refer to the bandwidth of a specific sub-carrier, NsUL denotes the cardinality of the sub-carriers needed in the UL network so that the user meets the desired throughput speeds, crSINR again is the modulation’s code rate and BLERSINR the block error rate that depends on the UL SINR.

3 Proposed Mechanism The suggested algorithm makes it possible for a UE to link up with possibly different BSs in the UL and DL. The algorithm chooses to maximize average throughput inside the macro cell coverage area, but without ensuring service towards all users. The mechanism takes into consideration the RB demands from each user and begins iterating from the UEs with the lowest RB demands in order to satisfy as many users as possible. Each UE is required to be matched with the appropriate MCS and each MCS is mapped to its equivalent Channel Quality Indicator (CQI), providing a total of 15 MCS-CQI sets, as described in [13]. The mechanism assumes knowledge of the available RBs for each BS, SINR measurements from each BS to each UE in and viceversa in both networks and the RB demands for every user in each case. Repetitively, each UE selects the best available BS candidate up until its data rate transmission demands are met. Each UE-BS association is possible only if there exist remaining RBs, otherwise we decide to select the next best candidate. The UE-BS association algorithm is presented below:

Efficient 5G Network Decoupling

259

begin for each macro BS do for each CQI = 1:15 do for each user do calculate user throughput(CQI); end for find average throughput(CQI) for macro area; end for find optimal average throughput(CQI) for macro area; select MCS based on average CQI; for each user do ); choose user with min( select optimal BS by finding max( ); if BS RBs are enough then associate user with BS; calculate user throughput(CQI); update available DL RBs; else ); try next best BS by new min(

end

end if end for end for repeat above steps for UL network

4 Performance Evaluation In this section, we present and analyze the MATLAB simulation results that derive from our proposed mechanism inside an urban HetNet that follows the 5G NR specifications. We consider a two-level ring topology, where the main 7 macrocells are the ones where users are located and thus, we are interested at and a total of 12 additional macrocells are added as a perimeter, simulating a real-life scenario where interference is experienced from neighboring cells. All macrocell BSs are situated at the cell center, surrounded by 3 small cells which are located near the cell edge so as to efficiently server users near cell edge that receive poor coverage from the cell’s macro BS. All BSs operate at their maximum power to ensure capabilities of maximum data rates. Users have a very high probability of spawning inside a macrocell which is in our area of interest and user throughputs are randomly assigned to UL and DL network users. Simulations parameters include users varying from 100 up to 400 in order to study dense HetNet scenarios.

260

C. Bouras et al.

Aiming at studying different simulation scenarios, we consider maximum applicable transmission bandwidths for both FR1 and FR2, leading the configuration of 100 MHz as FR1 setting and the selection of 200 MHz for the FR2 scenario. The number of RBs per BS derive from the same Release 15 specifications stated in [8]. Our mechanism will be simulated for both FR1 and FR2 representative deployment settings that best simulate a 5G network. According to [7], using larger SCS shortens the slot duration and enables fast data transmissions and thus, we assume as possible SCSs the configurations that maximize the number of RBs for both networks. In detail, FR1 setting considers 30 kHz as applicable SCS, while FR2 configures 60 kHz as SCS. Available RBs per BS directly depend on the sub-carrier configuration and vary across different sub-carrier settings, always as stated in Release 15. As for the MATLAB simulations, we evaluate the performance of the UC architecture by implementing our mechanism inside a HetNet and checking whether it reach the throughput goals. Evaluation metrics include calculating UE-BS successful associations over both networks and data rates measurements, again in our decoupled UL and DL networks. Simulation parameters follow the 5G NR simulation parameters of [7] and the 3GPP technical specifications of [8, 10, 11] and are summarized below in Table 1.

Table 1. Simulation parameters Parameter Area Modulation scheme DL bandwidth UL bandwidth Resource blocks (per BS) Carrier frequency RB bandwidth Cycle prefix SCS White noise density Macro cell inter-site distance Macrocell radius Small Cell radius BS antenna type UE antenna type Macro BS Prad i;max Small BS Prad i;max UE

Prad j;max

FR1 Setting 19 Macrocells/21 Small Cells QPSK/16QAM/64QAM 100 MHz 100 MHz 273 3.5 GHz 360 kHz Normal 30 −174 dBm/Hz 750 m

FR2 Setting 19 Macro Cells/21 Small Cells QPSK/16QAM/64QAM 200 MHz 200 MHz 264 30 GHz 720 kHz Normal 60 −174 dBm/Hz 750 m

375 m 50 m Omni-directional Omni-directional 30 W

375 m 50 m Omni-directional Omni-directional 30 W

1W

1W

0.2 W

0.2 W

Efficient 5G Network Decoupling

261

Fig. 2. Macrocell-small cell DL connections for FR1 and FR2 scenarios.

Fig. 3. Macrocell-small cell UL connections for FR1 and FR2 scenarios.

Figures 2 and 3 reveal the number of macrocell and small cell connections that occur during the FR1 and FR2 simulation scenarios, for both the cases of 30 kHz and 60 kHz for the SCS, respectively. For a device to successfully connect to a BS inside the ultra-dense HetNet, the BS must be able to provide the available RBs to the device, demands that derive from Eq. (4) and highly depend on the user requested throughput demands. For the DL network, as the number of network users increases, it is getting harder and harder for devices to connect to the station to which they have the optimal channel quality and thus, they are forced to rely on a small cell to server their needs.

262

C. Bouras et al.

This can be seen in Fig. 2, where while users increase, the percentage of user connections to macrocells decreases, whereas the percentage of user connections to small cell increases. As for the different SCS selection, the selection of 60 kHz seems dominant for macrocell connections, but for the small cells, the 30 kHz configuration is preferable. For the DL network, the reason why the 30 kHz configuration is optimal for macrocells while the 60 kHz configuration is optimal for small cells lies directly to the number of RBs the SCS configuration offers. More specifically, for the 60 kHz SCS selection in FR1, Table 1 defines a total of 273 available RBs, each one having a bandwidth of 360 kHz, whereas for the 60 kHz selection in FR2, the available RBs drop to 264, but with increased bandwidth of 720 kHz per RB. Since our mechanism rounds the RB demands towards the next integer and users that associate with a BS can use all the RBs they demanded (no same RB sharing is considered in the simulation), it becomes evident that in the case where users have increased demands and need only macrocells to cover their needs, is more important to supply the BSs with more RBs than to reduce the number of RBs and increase their bandwidth. On the other hand, in cases where users do associate with small cells, the association occurs because the RB demands of those users can be met by small cells which have less available RBs compared to macrocells (thus, the 30 kHz configuration of FR2). Comparing the performances of DL and UL networks with each other, the UL network allows an increased portion of small cell associations than the DL network. The reason why this happens is because in UL users have far less throughput demands than the DL users, so less RB demands, so better chances of finding an optimal small cell that can server their low demands. Additionally, the UL seems to be producing identical performances in for both FR1 and FR2 scenarios, regardless of the SCS selection, revealing that an optimal SCS selection is efficiently applicable in scenarios where users have increased throughput demands.

Fig. 4. Average DL data throughputs for FR1 and FR2 scenarios.

Efficient 5G Network Decoupling

263

Fig. 5. Average UL data throughputs for FR1 and FR2 scenarios.

In the figures above (namely, Fig. 4 for the DL and Fig. 5 for the UL), the average data rates for the users that have successfully connected to a BS are presented, for both FR1 and FR2 deployment scenarios. Equations (6) and (8) dictate that the experienced data rates per user are directly linked to the number of RBs he receives from the BS, the bandwidth of each RB and the coding rate that derives from the average optimal MCS selected for the macrocell area from our proposed mechanism. Simulation in both FR1 and FR2 scenarios revealed that even though adjustable MCS was available, the dominant MCS choice would eventually be the 64QAM, since this CQI maximizes average user throughput per macrocell area. We can observe that the (maximum) selection of 60 kHz that FR2 provides as SCS is optimal for both DL and UL networks. The augmented frequency resources that FR2 can provide seem to play a far more significant role in the user data rates than they played in associating users and BSs (see Figs. 2 and 3). The maximization of user data rates in both separated DL and UL networks can be achieved by selecting the maximum applicable SCS, which is the 60 kHz choice for the FR2 scenario. Augmented FR2 data rates occur due to the fact that FR2 increases the available bandwidth of each RB to 720 kHz (FR1 offers 360 kHz), a metric that is inextricably linked with the user data rates from Eqs. (6) and (8). Simulation results for the different number of users are presented below in Table 2. The mechanism (as promised) managed to achieve perfect 100% QoS preservation for all users by providing the users with far more data rates than the ones they demanded Additionally, we can observe that not only the FR2 scenario offers augmented average user throughputs, but also manages to cover a larger number of user devices, thus the unsupported users of FR2 scenario are far less than the ones from the FR1 scenario.

264

C. Bouras et al. Table 2. Simulation results for all users Number of users FR1 scenario (SCS 30 kHz) DL macrocell connections DL small cell connections DL unsupported users DL preferable MCS DL average data rates (Mbps) DL user QoS preservation DL QoS preservation (%) UL macrocell connections UL small cell connections UL unsupported users UL preferable MCS UL average data rates (Mbps) UL user QoS preservation UL QoS preservation (%)

100 users 200 users 300 users 400 users 99 1 0 64 QAM 12.48 100 100 94 6 0 64 QAM 8.32 100 100

167 11 22 64 QAM 13.06 178 100 176 24 0 64 QAM 7.98 200 100

196 28 76 64 QAM 12.23 224 100 254 34 12 64 QAM 7.55 288 100

245 29 126 64 QAM 13.99 274 100 132 41 47 64 QAM 7.45 353 100

FR2 scenario (SCS 60 kHz) DL macrocell connections DL small cell connections DL unsupported users DL preferable MCS DL average data rates (Mbps) DL user QoS preservation DL QoS preservation (%) UL macrocell connections UL small cell connections UL unsupported users UL preferable MCS UL average data rates (Mbps) UL user QoS preservation UL QoS preservation (%)

99 1 0 64 QAM 13.82 100 100 94 6 0 64 QAM 9.68 100 100

191 8 1 64 QAM 14.33 199 100 176 24 0 64 QAM 9.68 200 100

243 22 35 64 QAM 13.53 265 100 264 34 2 64 QAM 9.21 298 100

300 29 71 64 QAM 14.52 329 100 326 41 33 64 QAM 9.24 367 100

5 Conclusion and Future Work In this paper, we studied the UC approach for 5G networks that improves communication quality between UEs and BSs. The considered architecture demanded the decoupling of the existing network into two separate and independent networks, the DL and the UL. The formulated mechanism included an algorithm that provided optimal throughputs for all served UEs inside a macrocell by selecting the maximum available modulation scheme. Simulation results were evaluated for both 5G frequency ranges FR1 and FR2 defined in 5G NR and showed that the proposed mechanism offers perfect QoS preservation and augmented data rates in favor of ultimate user coverage.

Efficient 5G Network Decoupling

265

Additionally, the FR2 simulation provided both increased data rates compared to FR1 and lower portions of unsupported devices by the association mechanism. Future work may include additional mechanisms for interference mitigation or scenarios where users move at high speeds and the handover procedure is of utmost importance.

References 1. Kim, H.: Coding and modulation techniques for high spectral efficiency transmission in 5G and Satcom. In: 23rd European Signal Processing Conference (EUSIPCO), Nice, pp. 2746– 2750 (2015) 2. Ghaleb, A.M., Mansoor, A.M., Ahmad, R.: An energy-efficient user-centric approach for high capacity 5G heterogeneous cellular networks. Int. J. Adv. Comput. Sci. Appl. (IJACSA) 9(1), 405–411 (2018) 3. Bouras, C., Alexiou, A., Papazois, A., Kokkinos, V., Tsichritzis, G.: Modulation and coding scheme selection in multimedia broadcast over a single frequency network-enabled longterm evolution networks. Int. J. Commun Syst 25, 1603–1619 (2012) 4. Elshaer, H., Boccardi, F., Dohler, M., Irmer, R.: Downlink and uplink decoupling: a disruptive architectural design for 5G networks. In: IEEE Global Communications Conference, pp. 1798–1083 (2014) 5. Wang, J., et al.: Spectral efficiency improvement with 5G technologies: results from field tests. IEEE J. Sel. Areas Commun. 35(8), 1867–1875 (2017) 6. Mesodiakaki, A., Adelantado, F., Antonopoulos, A., Alonso, L., Verikoukis, C.: Energy and spectrum efficient user association in 5G heterogeneous networks. In: IEEE 27th Annual International Symposium on Personal, Indoor, and Mobile Radio Communications (PIMRC), Valencia, pp. 1–6 (2016) 7. Peralta, E., Levanen, T., Ihalainen, T., Nielsen, S., Ng, M.H., Renfors, M., Valkama, M.: 5G new radio base-station sensitivity and performance. In: 15th International Symposium on Wireless Communication Systems (ISWCS), pp. 1–6 (2018) 8. 3GPP: 3GPP TS 38.104, v15.1.0, Base Station (BS) radio transmission and reception. Release 15 (2018) 9. 3GPP: 3GPP TS 38.300, v15.3.1, 5G; NR; Overall Description; Stage 2. Release 15 (2018) 10. 3GPP: 3GPP TS 36.931, v13.0.0, Radio Frequency (RF) requirements for LTE Pico Node B. Release 13 (2016) 11. 3GPP: 3GPP TR 36.872, v12.1.0, Small Cell Enhancements for E-UTRA and E-UTRANPhysical layer aspects. Release 12 (2013) 12. Goldsmith, A.: Wireless Communications. Cambridge University Press, Cambridge (2005) 13. 3GPP: 3GPP TS 36.213, v15.5.0, Evolved Universal Terrestrial Radio Access (E-UTRA)Physical layer procedures. Release 15 (2019)

A Probabilistic Offloading Approach in Mobile Edge Computing Bhed Bahadur Bista(B) , Jiahong Wang, and Toyoo Takata Iwate Prefectural University, Takizawa City, Iwate 020-0693, Japan {bbb,wjh,takata}@iwate-pu.ac.jp Abstract. The mobile edge computing (MEC) is a new paradigm for providing computing at the edge of networks to support wireless devices to offload computational intensive tasks to MEC server for execution. In mobile environment, different users have different sizes of computation tasks with different target latency for smooth running of applications. Moreover, tasks will arrive at the MEC server for execution at different rate depending upon the time of the day or users density. In such varying environment, it is necessary to consider probabilistic approach to offload tasks for successful mobile edge computing. In this paper, we derive successful computation probability, successful communication probability and successful edge computing probability. We then simulate how the successful probabilities change for different sizes of task, target latency and task arrival rate.

1

Introduction

Software and hardware for wireless communication are improving year by year bringing high popularity of mobile devices specially smartphones. Due to ubiquitous mobile devices, more and more mobile applications requiring intensive computations such as face recognition, natural language processing, interactive gaming, augmented reality (AR), virtual reality (VR) and so on are emerging and attracting great attention both in academic and industrial fields [8,14,16,18]. Though in recent years, the computation and battery power of mobile devices have been improved significantly enabling mobile users to process complex computation tasks which were not possible a few years ago, many applications such as VR, AR and Tactile Internet [5] need intensive computing for their smooth running. Mobile devices are not capable in executing such applications and need to offload some of the tasks to resource rich servers wirelessly. Mobile Could Computing (MCC) [3,13,19] is envisioned as a promising approach for mobile offloading of tasks for mobile devices. For some applications, MCC may be suitable for remote execution of tasks but as MCC is physically resides further away from mobile users, i.e. away from mobile users’ access point (AP), for latency sensitive applications such as VR and AR, it is not suitable as it may not satisfy the latency requirements of the applications. To address the latency requirements challenges, a novel offloading and computing paradigm called Mobile Edge Computing (MEC) has been proposed where the computation tasks are performed at the edge of networks, i.e. within the cell of the c Springer Nature Switzerland AG 2020  L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 266–278, 2020. https://doi.org/10.1007/978-3-030-33506-9_24

A Probabilistic Offloading Approach in Mobile Edge Computing

267

mobile users [1,6,10]. In MEC, a resource rich server is attached to the AP and offloaded tasks do not need to travel further down the network to reach MCC thereby reducing the total offload time. Many offloading techniques and communication models are proposed for MEC which are well surveyed in [11]. Some of them are directly derived from MCC and fog computing techniques. In previous MEC offloading techniques, they mainly consider optimization of energy, latency and optimization of computation resources of MEC servers for making offloading decision. In this paper, we propose probabilistic approach for offloading tasks to MEC server in dynamic environment. For different latency, data size and task arrival rate at MEC server, we find the successful offloading probabilities of the task so that a user can make a decision whether to offload the task. The paper is organized as follows. In Sect. 2, we present some related works. In Sect. 3, we outline the communication model. In Sect. 4, we present the outline of the probabilistic offloading approach. In Sect. 5, we perform simulation and analyze the results and finally in Sect. 6 we present conclusion and future direction of our research.

2

Related Work

Since energy is one of the main problems in mobile devices, energy minimization problem in MEC has been considered in [4,17,21]. Computation offloading is an important technique for mobile devices with limited resources. In [2,15], authors consider binary offloading in which a particular task should be offloaded or locally computed depending upon the optimization constraints such as energy consumption and computation resources. In [12,20], authors consider partial offloading, i.e. some subtasks are offloaded while others are locally executed providing more flexibility in offloading strategies and reducing latency by partially executing subtasks. Authors in [7,9], consider the optimization of latency based on computation resources of MEC server, offloading policy, or trade-off between latency and communication performance. Most of the above works focus on latency and energy consumption. However, successful task offloading and edge computing probabilities for different characteristics of the environment has not been well explored. Offloading environment may vary continuously such as number of task arriving and queuing at MEC server for execution, interference from other user and/or base stations affecting upload and download speed of the task and so on. In such environment which cannot be determined with certainty before hand, it is essential to consider probabilistic approach whether a task will be successfully offloaded or not.

268

3

B. B. Bista et al.

Communication Model

In our communication model, we consider a set of mobile device users served by a Base Station (evolve NodeB (eNB) in LTE system) in which a resource rich MEC server is attached. A user which has a computationally intensive task uploads it to the MEC server via the eNB and downloads the computed result from the same eNB. As shown in Figs. 1 and 2, while a user is uploading the task to the MEC server, the eNB suffers from interference from other uploading users. Similarly, the user suffers from interference from eNBs while downloading the computed results from the eNB. The interference affects upload and download time of the task.

Fig. 2. Downloading from MEC server

Fig. 1. Uploading to MEC server

3.1

Offloading Time

Each task that is offloaded to the MEC server must satisfy some target latency in order to run the application at the user device smoothly. For example, to prevent users from feeling dizzy and nauseous, a VR system must guarantee the latency before human start noticing the lag. The total time including uploading, downloading and execution of the task must be less than the pre-defined target latency for the application. For task i with data size si the uploading time, T u , to the MEC server is calculated as shown below. Tu =

si riu

(1)

A Probabilistic Offloading Approach in Mobile Edge Computing

269

The size of the data after computation may be different from the original data. Let si be the size of the resulting data after computation, then the download time, T d , of the data, si , from the MEC server is given by Td =

si rid

(2)

where riu and rid are upload and download data rate respectively and are expressed as shown below riu = Wu log2 (1 + SIN Ru ),

rid = Wd log2 (1 + SIN Rd )

(3)

where Wu and Wd are upload and download bandwidth respectively, and SIN Ru and SIN Rd are signal to interference noise ratio at the eNB and user equipment (UE) respectively. Time T c that the MEC server takes to compute the task i can be expressed as shown below. ci (4) Tc = f where ci is the CPU cycles required to accomplish the computation task i and f is the computational capability, i.e. CPU cycles per second of the server. Many tasks will arrive at the MEC server for execution and will be queued. Now, if T w is the waiting time at the server then total offloading time for the task i will be T u + T d + T c + T w . For smooth running of an application, the offloading time of the application task i must not exceed required latency Tl , i.e. T u + T d + T c + T w ≤ Tl .

4

Probabilistic Offloading Approach for MEC

In order to find the successful mobile edge computing probability for a task, we need to first find successful computing probability and successful communication probability of the task. For successful communication probability, we further need to find successful upload and successful download probabilities of the task to and from the MEC server. 4.1

Successful Computation Probability

Total time a task remains in the MEC server is the sum of the task waiting time in queue and its execution time, i.e. T w + T c . Let T t be the target time the computation at the server must be finished to satisfy the required latency, then we have T w + T c ≤ T t . For a given task, its T c will be fixed and in order to finish the computation within T t , we need to find that the probability of task being in queue is less than T t − T c which can be expressed as shown Eq. 5 and it is the successful computation probability, Scp .

270

B. B. Bista et al.

Scp = P (T

w

    λ λ t c < T − T ) = 1 − exp −μ 1 − T −T μ μ t

c

(5)

where λ and μ are task arrival rate at the MEC server and its task execution rate respectively. 4.2

Successful Communication Probability

When User Equipment, U Ei , as shown in Fig. 1, uploads data to the MEC server Mi , Signal to Interference Noise Ratio (SINR), γiu , at the eN Bi is as shown in Eq. 6. Pi D−α γiu = (6)  i N0 + i=j Pj Dj−α where Pi is the transmit power of U Ei . Di is the distance between U Ei and eN Bi . α is the path-loss exponent. N0 is the Gaussian white noise. Pj is the transmit power of interfering U Ej and Dj is the distance between U Ej and eN Bi . For successful uploading of the task to the MEC server, the γiu must be above the upload threshold εuth . The probability that there is successful upload is given by the following Eq. 7.   −N0 u Pi Di−α u u   exp εth P (γi > εth ) = (7) Pi Di−α Pi D−α + Pj D−α εu i

i=j

j

th

When U Ei downloads data from eN Bi as shown in Fig. 2, the SINR it experiences is given by Eq. 8. γid =

PeN B R−α  i i N0 + i=j PeN Bj Rj−α

(8)

where PeN Bi is the transmit power of eN Bi and Ri is the distance between U Ei and eN Bi . PeN Bj is the transmit power of interfering eN Bj and Rj is the distance between U Ei and eN Bj . Similar to uploading case, for successful download the γid must be above the download threshold εdth . The probability that there is successful download is given by Eq. 9.   −N0 PeN Bi Ri−α d d d   exp εth P (γi > εth ) = (9) −α PeN Bi Ri−α εdth PeN Bi Ri−α + i=j PeN Bj Rj The upload and download thresholds are given as shown below.

A Probabilistic Offloading Approach in Mobile Edge Computing

271

D u Up

εuth = 2 W u Ts − 1, D d Up

εdth = 2 W d Ts − 1

(10)

where Du is the computation requested data packets, Up is the unit packet size, W u is the upload Bandwidth, Dd is the computation resulting data packets, W d is the download Bandwidth and Ts is the time slot. Uploading and downloading of a given task may require different number of time slots. Here for simplicity, we consider only one time slot, Ts . The successful communication probability, Smp , is the product of the successful upload probability and the successful download probability given by Eq. 11. Smp = P (γ0u > εuth )P (γ0d > εdth ) 4.3

(11)

Successful Edge Computing Probability

A task is successfully edge computed, if the offloading time of the task is less than or equal to the required latency of the task. The uploading and downloading of the task to and from the MEC server do not affect the computation of the task at the MEC server, i.e. they are independent events. The successful edge computing probability, Sep , therefore is the product of successful computation probability and the successful communication probability given by Eq. 12 Sep = Scp Smp

5

(12)

Simulation Results and Analysis

In our simulation, we assume a UE, randomly located inside a cell, offloads a task to the MEC server connected to its eNB for computation. We also assume that cells are hexagonal and the UE and its eNB experience interference from at least other three UEs and eNBs. The CPU cycles is based on the face recognition application in [16], where 1000 Megacycles is required to compute 5000KB of data. For simplicity, we assume that the size of upload and download data is same. The rest of the default simulation parameters are as shown in Table 1. As mentioned earlier, the total offloading time is T w + T u + T d + T c . Since, we are considering one time slot, T u + T d = 2Ts . For target latency Tl , the offloading time must be T w + 2Ts + T c ≤ Tl . For data packet Dc , the successful computation probability given in Eq. 5 satisfying target latency becomes     Dc λ Dc λ exp −μ 1 − P (T w ≤ Tl −(2Ts +T c )) = 1− Tl − (2Ts + T c ) (13) μ μ

272

B. B. Bista et al. Table 1. Default simulation parameters Parameter

Value

Upload/Download time slot Ts

10−3 (sec)

Target latency Tl

3.5 × 10−3 (sec)

Computation data size Dc

3 (packets)

Task arrival rate λ

3 (packets/slot)

Task service rate (execution rate) μ

18 (packets/slot)

Transmit power of U E

23 (dBm)

Transmit power of eN B

45 (dBm)

Path-loss exponent α

4

Data unit size Up

512 (bits)

Upload bandwidth W

u

5 × 106 (Hz)

Download bandwidth W

d

10 × 106 (Hz)

Gaussian white noise N0

−104 (dBm)

Computing rate of MEC server

10 × 109 (cycle/sec)

CPU cycle required to process unit data (512 bits) 12,800 CPU cycles

1

Successfull Computation Probability

0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 2

2.2

2.4

2.6

2.8

3

3.2

Target Latency(sec)

3.4

3.6

3.8

4 10-3

Fig. 3. Successful computation probability for different target latency

From Fig. 3, we observe that the successful computation probability increases as the target latency increases. This is reflected in successful edge computing probability as shown in Fig. 5 also. From the results we see that the probability of successful edge computing is less if the application’s latency requirement is very short. As shown in Fig. 4, there is slight increase in successful communication

A Probabilistic Offloading Approach in Mobile Edge Computing

273

Successfull Communication Probability

0.99

0.988

0.986

0.984

0.982

0.98

0.978 2

2.2

2.4

2.6

2.8

3

3.2

3.4

3.6

3.8

4 10-3

Target Latency(sec)

Fig. 4. Successful communication probability for different target latency

Successfull Edge Computing Probability

1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 2

2.2

2.4

2.6

2.8

3

3.2

Target Latency(sec)

3.4

3.6

3.8

4 10-3

Fig. 5. Successful edge computing probability for different target latency

probability as target latency increases. It shows that the different target latency does not affect the successful communication probability. From Fig. 6, we observe that there is hundred percent successful computation probability for smaller data size and it decreases significantly as the data size increases. Similar pattern can also be seen in successful communication probability for varying data size as shown in Fig. 7. Since, successful edge computing probability shown in Fig. 8, is the product of successful computation and

274

B. B. Bista et al. 1

Successfull Computation Probability

0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 4

4.2

4.4

4.6

4.8

5

5.2

5.4

5.6

5.8

6

Data Size(packets/T s)

Fig. 6. Successful computation probability for different data size

Successfull Communication Probability

0.982

0.98

0.978

0.976

0.974

0.972

0.97

0.968 4

4.2

4.4

4.6

4.8

5

5.2

5.4

5.6

5.8

6

Data Size(packets/T s)

Fig. 7. Successful communication probability for different data size

communication probabilities, it also decreases as the data size increases. From the results, we notice that for a fixed target latency, the data size of a task influences the successful edge computing probability significantly. Figures 9, 10 and 11 show the successful computation, communication and edge computing probabilities for different task arrival rates. The task arrival rate is basically the number of data packets arriving at the MEC server for computation per Ts which is converted to seconds in the figures for clarity. As shown in Figs. 9 and 11, as the task arrival rate at the MEC server increases, the

A Probabilistic Offloading Approach in Mobile Edge Computing

275

1

Successfull Edge Computing Probability

0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 4

4.2

4.4

4.6

4.8

5

5.2

5.4

5.6

5.8

6

Data Size(packets/T s)

Fig. 8. Successful edge computing probability for different data size

1

Successfull Computation Probability

0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 2000

2200

2400

2600

2800

3000

3200

3400

3600

Task Arrival Rate(packets/sec)

Fig. 9. Successful computation probability for different task arrival rate

successful computation probability and successful edge computing probability decreases since tasks have to wait in the queue longer and they may not be executed within the time to satisfy the target latency. Successful communication probability, though decreases slightly, is not affected by the different task arrival rate at the MEC server.

276

B. B. Bista et al.

Successfull Communication Probability

0.9748

0.97479

0.97478

0.97477

0.97476

0.97475

0.97474 2000

2200

2400

2600

2800

3000

3200

3400

3600

Task Arrival Rate(packets/sec)

Fig. 10. Successful communication probability for different task arrival rate

1

Successfull Edge Computing Probability

0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 2000

2200

2400

2600

2800

3000

3200

3400

3600

Task Arrival Rate(packets/sec)

Fig. 11. Successful edge computing probability for different task arrival rate

6

Conclusion and Future Work

In this paper, we derived successful computation, successful communication and successful edge computing probabilities for offloading tasks in MEC. We also simulated how the probabilities change for different task size, target latency and task arrival rate at MEC server. From the results, we observe that the successful computation and successful edge computing probabilities, i.e. successful offloading, for large tasks with short target latency is low. It is also low if the task is offloaded to congested MEC server. Successful edge computing probability for a task with higher target latency and small data size is higher. In future, we

A Probabilistic Offloading Approach in Mobile Edge Computing

277

will investigate how the above probabilities can be utilized to make optimum decision for offloading task to MEC in ultra dense network and users.

References 1. Abbas, N., Zhang, Y., Taherkordi, A., Skeie, T.: Mobile edge computing: a survey. IEEE Internet Things J. 5(1), 450–465 (2018) 2. Chen, X., Jiao, L., Li, W., Fu, X.: Efficient multi-user computation offloading for mobile-edge cloud computing. IEEE/ACM Trans. Netw. 24(5), 2795–2808 (2016) 3. Deshmukh, S., Shah, R.: Computation offloading frameworks in mobile cloud computing: a survey. In: 2016 IEEE International Conference on Current Trends in Advanced Computing (ICCTAC), pp. 1–5 (2016). https://doi.org/10.1109/ ICCTAC.2016.7567332 4. Dinh, T.Q., Tang, J., La, Q.D., Quek, T.Q.S.: Offloading in mobile edge computing: task allocation and computational frequency scaling. IEEE Trans. Commun. 65(8), 3571–3584 (2017) 5. Fettweis, G.P.: The tactile internet: applications and challenges. IEEE Veh. Technol. Mag. 9(1), 64–70 (2014) 6. Hu, Y.C., Patel, M., Sabella, D., Sprecher, N., Young, V.: Mobile edge computing: a key technology towards 5G. European Telecommunications Standards Institute, France, ETSI White Paper No. 11 (2015) 7. Ko, S., Han, K., Huang, K.: Wireless networks for mobile edge computing: spatial modeling and latency analysis. IEEE Trans. Wirel. Commun. 17(8), 5225–5240 (2018) 8. Kumar, K., Lu, Y.: Cloud computing for mobile users: can offloading computation save energy? Computer 43(4), 51–56 (2010) 9. Liu, J., Mao, Y., Zhang, J., Letaief, K.B.: Delay-optimal computation task scheduling for mobile-edge computing systems. In: 2016 IEEE International Symposium on Information Theory (ISIT), pp. 1451–1455 (2016) 10. Mach, P., Becvar, Z.: Mobile edge computing: a survey on architecture and computation offloading. IEEE Commun. Surv. Tutor. 19(3), 1628–1656 (2017) 11. Mao, Y., You, C., Zhang, J., Huang, K., Letaief, K.B.: A survey on mobile edge computing: the communication perspective. IEEE Commun. Surv. Tutor. 19(4), 2322–2358 (2017) 12. Mu˜ noz, O., Pascual-Iserte, A., Vidal, J.: Optimization of radio and computational resources for energy efficiency in latency-constrained application offloading. IEEE Trans. Veh. Technol. 64(10), 4738–4755 (2015) 13. Noor, T.H., Zeadally, S., Alfazi, A., Sheng, Q.Z.: Mobile cloud computing: challenges and future research directions. J. Netw. Comput. Appl. 115, 70–85 (2018) 14. Khan, R., Othman, A.M., Madani, S.A., Khan, S.U.: A survey of mobile cloud computing application models. IEEE Commun. Surv. Tutor. 16(1), 393–413 (2014) 15. Sardellitti, S., Scutari, G., Barbarossa, S.: Joint optimization of radio and computational resources for multicell mobile-edge computing. IEEE Trans. Signal Inf. Process. Netw. 1(2), 89–103 (2015) 16. Soyata, T., Muraleedharan, R., Funai, C., Kwon, M., Heinzelman, W.: Cloudvision: real-time face recognition using a mobile-cloudlet-cloud acceleration architecture. In: 2012 IEEE Symposium on Computers and Communications (ISCC), pp. 000,059–000,066 (2012)

278

B. B. Bista et al.

17. Tao, X., Ota, K., Dong, M., Qi, H., Li, K.: Performance guaranteed computation offloading for mobile-edge cloud computing. IEEE Wirel. Commun. Lett. 6(6), 774–777 (2017) 18. Wang, S., Dey, S.: Adaptive mobile cloud computing to enable rich mobile multimedia applications. IEEE Trans. Multimed. 15(4), 870–883 (2013) 19. Wang, Y., Chen, I.R., Wang, D.C.: A survey of mobile cloud computing applications: perspectives and challenges. Wirel. Pers. Commun.: Int. J. 80(4), 1607–1623 (2015) 20. Wang, Y., Sheng, M., Wang, X., Wang, L., Li, J.: Mobile-edge computing: partial computation offloading using dynamic voltage scaling. IEEE Trans. Commun. 64(10), 4268–4282 (2016) 21. You, C., Huang, K., Chae, H., Kim, B.: Energy-efficient resource allocation for mobile-edge computation offloading. IEEE Trans. Wirel. Commun. 16(3), 1397– 1411 (2017)

Fuzzy Geocasting in Opportunistic Networks Sanjay K. Dhurandher1 , Jagdeep Singh2 , Isaac Woungang3(B) , Makoto Takizawa4 , Geetanshu Gupta5 , and Raghav Kumar5 1

Department of Information Technology, Netaji Subhas University of Technology, New Delhi, India [email protected] 2 Division of Information Technology, Netaji Subhas Institute of Technology, University of Delhi, New Delhi, India [email protected] 3 Department of Computer Science, Ryerson University, Toronto, ON, Canada [email protected] 4 Department of Advanced Sciences, Faculty of Science and Engineering, Hosei University, Tokyo, Japan [email protected] 5 Division of Computer Engineering, Netaji Subhas Institute of Technology, University of Delhi, New Delhi, India [email protected], [email protected]

Abstract. Opportunistic Networks are composed of wireless nodes opportunistically communicating with each other following the store, carry and forward mechanism. These networks are designed to operate in an environment characterized by high delay, intermittent connectivity and non-guarantee of the end-to-end path between the sender and the destination. Opportunistic networks can play a crucial role, when cellular networks are heavily stressed and where infrastructure is unavailable due to terrorist attacks, wars, or natural disasters and censorship. Geocasting, where messages are scheduled to specific regions instead of individual devices, has a large potential in real-world communication systems. In this paper, we propose a fuzzy geocasting mechanism in opportunistic networks, termed as F-GSAF. The proposed protocol employs fuzzy attributes, that are very much likely to affect a network in the real world, to determine the next hop for the message. These attributes are Movement (direction and speed), remaining energy, and remaining buffer space. Mamdani is the fuzzy controller, which has been used in this work. Obtained simulation results confirm that the proposed F-GSAF protocol is more efficient than traditional routing protocols for opportunistic networks.

1

Introduction

By design, opportunistic networks (OppNets) [1] typically incur high delay, intermittent connectivity and no assurance of the end-to-end route between the sender c Springer Nature Switzerland AG 2020  L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 279–292, 2020. https://doi.org/10.1007/978-3-030-33506-9_25

280

S. K. Dhurandher et al.

& destination node. In a real-world environment, multitudes of wireless networks interact with each other in transmission. Unlike in ad-hoc networks [2], end-toend path is absent in Opportunistic Networks [3–5]. Thus the routing of Opportunistic Networks is highly challenging. To overcome this issue, the delay-tolerant store-carry and forward mechanism is applied. In most of the cases, content dissemination is geographic based where a natural disaster or terrorist attacks took place. A part from being able to transmit the urgent information to a specific device or person, it is also crucial to be able to transmit the information to geographical locations covered by the network. This type of message transmission is known as geocasting. Effective geocasting has a large prospective in the network environment. The applications of effective geocasting are urgent location based information of emergency situations, geographical targeted advertisement and location based service discovery, etc. These locations may be pre-decided, even before a network is established or set up by the sender for each transmission. This paper presents a geocasting protocol, based on the existing GSAF protocol, which is enhanced further using fuzzy logic. According to the current GSAF protocol, a message is forwarded only on the basis of copies remaining value, i.e. the ’C’ value must be greater than zero for the message to be forwarded to the next node. The F-GSAF protocol employs fuzzy attributes, that are very much likely to affect a delay tolerant network in the real world, to determine the next hop for the message. These attributes are Movement (direction and speed), remaining energy, and remaining buffer space. Different models have been proposed to perform fuzzification and defuzzification and are also known as fuzzy controllers. These models differ in terms of the functions used to combine the fuzzified input and to de-fuzzify the output. Mamdani is one such fuzzy controller, which has been used in this work. The remainder of this paper is organized as follows. Section 2, related works are presented. In Sect. 3, the proposed F-GSAF protocol for OppNets is described. Simulation results and justifications are presented in Sect. 4. Finally, the paper is summarized in Sect. 5.

2

Related Work

Although a number of works have used fuzzy logic for various tasks such as routing, buffer scheduling and security, to the best of our knowledge none of them has addressed geocasting in opportunistic networks. In [6], Bai et al. suggested fuzzy identity cipher based access control technique, which encodes the information with malicious attribute and represents that the malicious node can not be granted to enter in the network. This leads to enhancement of the throughput and security of the network. The drawback of this scheme is that, authors have not addressed the problem of key escrow in opportunistic networks. In [7], Nabhani et al. introduced an adaptive fuzzy routing scheme in OppNets. In this paper, authors have presented a scheme to select the forwarding list based on fuzzy logic system, which takes into account the energy of the nodes, bandwidth, density, and priority of the message [8,9].

Fuzzy Geocasting in Opportunistic Networks

281

The authors have found out that this scheme takes less energy consumption per transmission. In [10], Sabeetha et al. introduced a fuzzy based routing for delay tolerant networks, which requires calculation of the possibility of prospective communication between nodes by using their history of contact duration. The fuzzy based utility computation is used for selecting a perferable node to relay messages as well as to discard the messages from buffer. In [11], Rodrigues et al. suggested a spray & wait routing scheme for DTN based of fuzzy logic. In this work, the buffer management is done by using fuzzy attributes. In [12], Ahmad et al. proposed a fuzzy-PRoPHET routing scheme. PRoPHET routing protocol is enhanced by using fuzzy attributes in this work. Fuzzy considers the values between zero and one, due to which selecting better node can become effortless. In [13], Rajaei et al. suggested an efficient and flexible geocasting for OppNets. This scheme is also termed as GSAF. The messages are forwarded only on the basis of ticket value, i.e. the ’t’ value must be greater than zero for the message to be forwarded to the next node. It works in two phases. In first phase, message is relayed on the basis of ticket value. Moreover, when the message reached the destined region, another phase is enabled and messages are disseminated within the region.

3 3.1

System Model Motivation

Fuzzy logic is used to imitate the decision making ability of humans which is not based on precise numeric values, but on perception of a value within an acceptable degree of error. The routing protocols for OppNets deal with the knowledge of various attributes of the participating nodes, which vary during the network lifetime. Routing decisions can be made on the basis of the value of these attributes, to find out if the neighbouring node is a better carrier for the message or not. These attributes can be fed to a fuzzy controller and the output may be used to decide if the neighbouring node is fit enough or not. The use of fuzzy logic will significantly help in saving the complexity without any trade off to the systems performance. This has motivated the authors in designing F-GSAF, which is an extension of GSAF in the manner that during the first stage of routing, messages are forwarded conditionally, whereas they are forwarded unconditionally in the GSAF model. We have further added to the F-GSAF model an acknowledgement scheme for minimizing the overhead. 3.2

Proposed F-GSAF Routing Protocol

According to the current GSAF protocol, a message is forwarded only on the basis of remaining message copies, i.e. the ‘C’ value must be greater than 0 for the message to be forwarded to the next node. The F-GSAF protocol employs fuzzy attributes, that are very much likely to affect a delay tolerant network in the

282

S. K. Dhurandher et al.

Fig. 1. Direction of movement of the node

Fig. 2. Remaining buffer space of the node

real world in order, to determine the next hop for the message. These attributes are movement, remaining energy and remaining buffer space. Movement is the measure of the mobility of nodes in a delay tolerant network. Mobility is one of the defining characteristic of DTNs and hence it very much affects the choice of next hop for a message. In this work, movement is the defuzzified output of a fuzzy controller that supplies the direction of movement of a node and its speed at the moment of encounter. The direction of movement of a receiver node r, that is being considered as a next hop for a given message m, is the angle between the line joining sender s to the center of the destination geocast and the line along which r is moving at the instant. The smaller is the angle, the chances of r delivering m to its destination geocast will be higher. The speed of a node influences the time that a node s will take to transmit a message m to its destined geocast region. The remaining energy is important to consider before forwarding a message m to make sure that receiver r can take m closer to its destination, than the sender s, before it dies down. Buffer space is an important influencer in a store-carry & forward paradigm. Once the buffer gets filled, depending on the scheduling policy, the router starts dropping messages to make room for new messages. Thus the space remaining in the buffer affects the duration for which a node will carry message m and hence the probability that m will be delivered to its destination geocast.

Fuzzy Geocasting in Opportunistic Networks

283

Fig. 3. Likelihood of the node delivering the message

The range for input variable direction is from 0◦ to 180◦ . The input variable direction is classified into four classes viz. very low, low, medium, and high. The range for speed is taken from 0 m/s to 15 m/s with different distributions for different group of nodes. A movement fuzzy controller takes in as input the above variables and produces an output hereafter called movement, that is divided into three fuzzy classes (low, medium, high). Movement gives a measure of how much a host is likely to move in the direction of the destination geocast and how soon will it reach there. Movement is used as an input variable for the fuzzy controller.

Fig. 4. Movement of the node

Movement lies in the range between 0 and 1. The input variable speed is classified into three classes viz. low, medium, and high. Remaining buffer space and remaining energy are also quantified between 0 and 1. The input variables energy, buffer and movement are classified into three classes viz. low, medium, and high. They are expressed as the absolute values of respective percentages. The fuzzy controller takes the above variables as input and gives an output hereafter called likelihood. The latter is an estimate of the possibility that a node can deliver a message to its destined geocast region. It is divided into five fuzzy classes very low to very high. Likelihood is used to decide if a message should be forwarded to a neighbouring node or not and by how much should the copies remaining value be reduced. Figures 1, 2, 3, 4, 5 and 6 shows the membership functions of the inputs and output of the two fuzzy controllers. It should be noted that the membership function of the output of the movement

284

S. K. Dhurandher et al.

fuzzy controller is used as input by the likelihood fuzzy controller. Tables 1 and 2 present the fuzzy rules used to define the controllers. The representations of fuzzy rule base for movement and likelihood are as follows: Very Low, Low, Medium, High and Very High are represented as VL, L, M, H, and VH, respectively.

Fig. 5. Remaining energy of the node

Fig. 6. Speed of the node

At the time of encounter between two hosts s and r, where s is carrying message m, s requests for the above mentioned attributes of r if value m, m.t ¿ 0 and r is not in the destination geocast of m as well as r does not already have m in its buffer. Here, s feeds the attributes received from r to a Mamdani fuzzy controller and gets an output, which suggests the likelihood L(m,r) of r delivering the message m to its destination geocast. Moreover s calculates its own likelihood L(m,s) of delivering m to its destination geocast. If the copies remaining value of m is greater than zero, then s forwards the message m to r if as long asL(m,r) is greater than L(m,s). The copies remaining value of the copy of message m in s and the forwarded copy of m in r depends on the fuzzy output class of L(m,r). Otherwise, s moves to the next message m in buffer. If r is located in the destination geocast of m, then s simply forwards m to r irrespective of the copies remaining value m.t and drops m from its own buffer.

Fuzzy Geocasting in Opportunistic Networks

285

Table 1. Fuzzy rule base for movement Direction Speed Movement VL

L

L

VL

M

L

VL

H

L

L

L

L

L

M

L

L

H

M

M

L

L

M

M

M

M

H

H

H

L

M

H

M

H

H

H

H

Then, r floods m within the geocast. The message forwarding mechanism is depicted in Algorithm 1. The message contains a cast definition, a pair of epoch times defining the lifetime of the message, an integer denoting the number of allowed forwards called remaining message copies (C) and the payload data. The cast definition is a set of two dimensional points that form the geographic cast. Just like [13], F-GSAF consists of two stages, that are inspired from some widely used unicast routing protocols. These unicast routing protocols use EIDs to identify individual devices, but EIDs are required to be adapted to identify casts for performing geocasting. The routing of a message from the sender to a cast takes place in two stages: (a) the message is relayed to the geocast region and (b) all the messages, which are reached to destination region, are carefully flooded to all the hosts in the region. (1) Stage 1 − Forwarding the Message to the Destination Cast: This stage follows a multi-copying spray approach which has been observed to be fast in terms of delivery time and efficient both in terms of delivery and overhead. When a message is created, it is assigned a remaining message copies (C) value, that denotes the maximum number of copies the sender can create and forward to the encountered host, if that host is outside the cast. On encountering the other host, the sender evaluates the likelihood of it delivering a message to the destination cast using fuzzy logic. If the likelihood of the encountered host delivering the message is greater than that of the sender itself, then the message is forwarded and its copies remaining value is decreased by a factor that depends on the measure of the likelihood of the encountered host delivering the message. If the likelihood calculated for the other host is less than that of the sender, then the message is not forwarded and thus the copies remaining value remains

286

S. K. Dhurandher et al. Table 2. Fuzzy rule base for likelihood Energy Buffer Movement Likelihood L

L

L

VL

L

L

M

L

L

L

H

M

L

M

L

VL

L

M

M

L

L

M

H

M

L

H

L

VL

L

H

M

M

L

H

H

H

M

L

L

L

M

L

M

M

M

L

H

H

M

M

L

L

M

M

M

M

M

M

H

H

M

H

L

L

M

H

M

M

M

H

H

H

H

L

L

L

H

L

M

M

H

L

H

M

H

M

L

L

H

M

M

H

H

M

H

VH

H

H

L

L

H

H

M

VH

H

H

H

VH

unchanged. It must be noted that the remaining copies can have a minimum value of zero. Even though the encounter host has the maximum possible likelihood of delivering the message, the sender cannot forward the message to it if is located outside the desired cast. The decrease in remaining copies value is determined by the fuzzy class of the likelihood of delivery by the encountered host. (2) Stage 2 − Flooding the Message to all the Hosts in the Region: In this state, all the messages which are reached to destination region, are carefully flooded to all the hosts in the region. When a copy of the message is transmitted to a host in the region/cast for the first time, its C value is set to zero. Hence,

Fuzzy Geocasting in Opportunistic Networks

287

Algorithm 1. F-GSAF Routing Protocol 1: Begin initialization: Hs ⇒ sender node & Hd ⇒ destination node. 2: for Encounter between any two nodes: do 3: if the node is the sender Hs then 4: Get a list of delivered messages ld from the receiver Hd & drop messages from buffer that are present in ld . 5: drop expired messages from buffer 6: for each message in the buffer do 7: if the message is already existed in the node’s buffer then 8: pass to the next message in the for loop 9: end if 10: if C > 0 then 11: if L(Hd , m) > L(Hs , m) then 12: Switch Λ (L(Hd , m)) 13: Case 1: VH : C = C-5, break; 14: Case 2: H : C = C-4, break; 15: Case 3: M : C = C-3, break; 16: Case 4: L : C = C-2, break; 17: Case 5: VL : C = C-1, break; 18: C = max(C, 0) 19: forward a copy of message to Hd 20: end if 21: else if C = 0 then 22: if node is inside the destined region then 23: forward a copy of message to another node. 24: end if 25: end if 26: end for 27: else if the node is destination node Hd then 28: for each received message do 29: decrease message TTL 30: register message in Hd block register 31: C = C-1 32: end for 33: end if 34: end for

if any host in the cast that has received a copy of the message moves out, then there will be no further forwarding of that message. When a host forwards a message to a host inside the destination cast, it receives a list of message IDs that have already been received by the cast. Using this list the sender can drop the messages that lie in its buffer but have already been delivered. Furthermore, the sender carries this list with itself throughout the network and exchanges it with all the hosts that it encounters. The hosts use this list to drop messages that have already been delivered. Using this acknowledgment mechanism, the network is kept aware of the delivered messages so that the undelivered copies are dropped from the buffers of hosts that carry them.

288

4

S. K. Dhurandher et al.

Simulation Results

In this section, the performance of the proposed F-GSAF routing protocol is analyzed using simulation analysis and compared against two competing benchmark routing protocols: GSAF [13] and GeoEpidemic [14], using the ONE simulator tool [15]. The Helsinki city map (dimensions: 4500 × 3400 m2 ) is the simulation framework. The transmission rate for all the nodes is considered 2 Mbps and transmission range is 10 m. The simulation time is 14400 s. The number of node groups are six (group classes: pedestrian, tram & car). The host density levels are 126, 189, 252, 315, 378, 441, 504, and 567. 4.1

Results and Justification

First, Fig. 7 shows the delivery probability under varying TTL (Time To Live in the network). It is observed that the delivery probability of F-GSAF, GSAF, and GeoEpidemic increases as the TTL is increased. This increase is due to the fact that the time duration allotted to each message is increased when the TTL increases, and as more number of messages get stored in the buffer of nodes, the message delivery probability decreases. The average delivery probability of F-GSAF is 0.35626, which is the highest among all the techniques. The average delivery probability for GSAF is 0.30262, and for GeoEpidemic, it is 0.29864. In addition to this, the performance of delivery probability is calculated when the TTL is varied for the stated techniques. It is found that F-GSAF is 15.05% better than GSAF and 16.17% better than GeoEpidemic, respectively.

0.6

Delivery Ratio

0.5 0.4 GeoEpidemic

0.3

GSAF 0.2

F-GSAF

0.1 0 30

60

90

120

150

180

210

240

TTL

Fig. 7. Delivery probability vs. TTL (in Minutes)

Second, Fig. 8 shows the results of the latency under varying TTL for all the scenarios. It is observed that, with the increase in TTL, the average latency also increases. This occurs due to the fact that substantial TTL value, increases the stay of the message in the node’s buffer. The mean average latency value of F-GSAF is the lowest among all the scenarios/techniques that are 5375.2161 s. In context to this the performance of F-GSAF is 2.16% better than GSAF, and

Fuzzy Geocasting in Opportunistic Networks

289

3000

Average Latency

2500 2000 GeoEpidemic

1500

GSAF 1000

F-GSAF

500 0 30

60

90

120

150

180

210

240

TTL

Fig. 8. Average latency vs. TTL (in Minutes) 0.7

Delivery Ratio

0.6 0.5 0.4 GeoEpidemic 0.3

GSAF F-GSAF

0.2 0.1 0 126

189

252

315

378

441

504

567

Number of hosts

Fig. 9. Delivery probability vs. Number of hosts 0.3

Overhead Ratio

0.25 0.2 GeoEpidemic

0.15

GSAF 0.1

F-GSAF

0.05 0 126

189

252

315

378

441

504

567

Number of hosts

Fig. 10. Overhead ratio vs. Number of hosts

3.95% better than GeoEpidemic respectively. Third, Fig. 9 shows the delivery probability under varying number of nodes. It is observed that the delivery probability of F-GSAF, GSAF, and GeoEpidemic increases as the number of hosts are increased.

290

S. K. Dhurandher et al. 0.6

Delivery Ratio

0.5 0.4 GeoEpidemic

0.3

GSAF 0.2

F-GSAF

0.1 0 5

10

15 20 Buffer Size

25

30

Fig. 11. Delivery probability vs. Buffer size (in Mega Bytes)

Fourth, overhead ratio is defined as the average number of replicas transmitted per message. The results for the overhead ratio under varying hosts are shown in Fig. 10. Fifth, the buffer capacity is varied and the effect of this change on the delivery probability and average latency is investigated. The performance of the routing schemes is captured in Figs. 11, 12. It is found that with the increase in buffer capacity, the delivery probability is also increased. This is due to the fact that, as the buffer capacity of a node gets large, the more number of messages stored in that buffer gets increased, leading to more messages getting delivered to the receiver node. In terms of delivery probability, the performance of F-GSAF is 9.91% better than that of GSAF and 7.02% better than that of GeoEpidemic. Sixth, Fig. 13 shows the results of the latency under varying number of hosts. It is observed that with the increase in hosts count, the average latency decreases. Seventh, Fig. 14 shows the relation between the number of messages forwarded and number of hosts for all the three scenarios. From the simulation result, it is found that the average number of messages forwarded for F-GSAF is the lowest among all the scenarios. The reason behind is the nodes selection

3000

Average Latency

2500 2000 GeoEpidemic

1500

GSAF 1000

F-GSAF

500 0 5

10

15

20

25

30

Buffer Size

Fig. 12. Average latency vs. Buffer size (in Mega Bytes)

Fuzzy Geocasting in Opportunistic Networks

291

2500

Average Latency

2000

1500 GeoEpidemic GSAF

1000

F-GSAF 500

0 126

189

252

315 378 441 Number of hosts

504

567

Fig. 13. Average latency vs. Number of hosts

Number of messages relayed

140000 120000 100000 80000

GeoEpidemic 60000

GSAF

40000

F-GSAF

20000 0 126

189

252

315

378

441

504

567

Number of hosts

Fig. 14. Number of messages relayed vs. Number of hosts

criteria for the best forwarder of the message, which minimizes the number of message relayed in the network.

5

Conclusion

In this work, we proposed a novel fuzzy based routing protocol for Opportunistic networks (named F-GSAF), which relies on movement (direction and speed), remaining energy, and remaining buffer space attributes. Simulation results have proved that F-GSAF outperforms GSAF and GeoEpidemic, chosen as benchmark protocols, in terms of overhead ratio, average latency, messages relayed, and delivery probability. From a practical perspective, F-GSAF does not require high resources or high computation. In the future, we aim to secure the proposed F-GSAF routing protocol and test it on real mobility traces. Acknowledgments. This work is supported in part by a grant from the National Science and Engineering Research Council of Canada (NSERC), held by the 3rd author, under reference number: RGPIN-2017-04423.

292

S. K. Dhurandher et al.

References 1. Huang, C.M., Lan, K.C., Tsai, C.Z.: A survey of opportunistic networks. In: Proceedings of IEEE International Conference on Advanced Information Networking and Applications Workshops, Okinawa, Japan, 25–28 March, pp. 1–6 (2008) 2. Mauve, M., Widmer, J., Hartenstein, H.: A survey on position-based routing in mobile ad hoc networks. IEEE Netw. 15, 30–39 (2001) 3. Teotia, A., Dhurandher, S.K., Woungang, I., Obaidat, M.S., Gupta, S., Rodrigues, J.: An altruism-based trust-dependent message forwarding protocol for opportunistic networks. Int. J. Commun. Syst. 30, 1–11 (2017) 4. Mantas, N., Louta, M., Karapistoli, E., Karetos, G., Kraounakis, S., Obaidat, M.S.: Towards an incentive-compatible, reputation-based framework for stimulating cooperation in opportunistic networks: a survey. In: IET Networks (2017). https://doi.org/10.1049/iet-net.2017.0079 5. Borah, S., Dhurandher, S.K., Tibarewala, S., Woungang, I., Obaidat, M.S.: Energy efficient prophet-PRoWait-EDR protocols for opportunistic networks. In: Proceedings of IEEE GLOBECOM 2017, Singapore, December 2017 6. Bai, Y., Xu, J.: Access control scheme based on fuzzy identity in opportunistic network. Procedia Comput. Sci. 131, 1122–1127 (2018) 7. Nabhani, P., Bidgoli, A.M.: Adaptive fuzzy routing in opportunistic network (AFRON). Int. J. Comput. Appl. 52, 7–11 (2012) 8. Sharma, D.K., Dhurandher, S.K., Obaidat, M.S., Pruthi, S., Sadoun, B.: A priority based message forwarding scheme for opportunistic networks. In: Proceedings of IEEE International Conference on CITS 2016, Kunming, China, pp. 236–240 (2016) 9. Dhurandher, S.K., Borah, S., Obaidat, M.S., Sharma, D.K., Baruah, B.: Probability based controlled flooding in opportunistic networks. In: Proceedings of 2015 IEEE International Conference on WINSYS 2015, Colmar, France, pp. 3–8, July 2015 10. Sabeetha, K., Kumar, A.V.A., Wahidabanu, R.S.D., Othman, W.A.M.: Encounter based fuzzy logic routing in delay tolerant networks. Wirel. Netw. 21, 173–185 (2015) 11. Jain, S., Chawla, M., Soares, V.N., Rodrigues, J.J.: Enhanced fuzzy logic based spray and wait routing protocol for delay tolerant networks. Int. J. Commun. Syst. 29, 1820–1843 (2016) 12. Ahmad, K., Fathima, M., Jain, V., Fathima, A.: FUZZY-PRoPHET: a novel routing protocol for opportunistic network. Int. J. Inf. Technol. 9, 121–127 (2017) 13. Rajaei, A., Chalmers, D., Wakeman, I., Parisis, G.: GSAF: efficient and flexible geocasting for opportunistic networks. In: International Symposium on A World of Wireless, Mobile and Multimedia Networks (WoWMoM), pp. 1–9. IEEE (2016) 14. Zhang, X., Neglia, G., Kurose, J., Towsley, D.: Performance modeling of epidemic routing. In: International Conference on Research in Networking, pp. 827–839. Springer (2006) 15. Keranen, A., Ott, J., Karkkainen, T.: The ONE simulator for DTN protocol evaluation. In: Proceedings of SIMUTools 2009, Rome, Italy, 2–6 March, pp. 1–9 (2009)

Digital Content Refinement by Collecting Partly Unreliable Attributes over a Network Shinji Sugawara(B) Chiba Institute of Technology, Narashino 275-0016, Japan [email protected]

Abstract. The concept of refining digital content downloaded over a network is introduced. If a content item consists of a lot of preliminarily defined attributes, then content refinement can be executed by collecting a number of partly unreliable releases of the same content item from different information sources. In the refinement process, each release of the content item is divided into attributes of signifieds or signifi´e and weighted majority voting is applied to determine the most probable signifier or significant of the attributes. In addition, results of a simple experiment are presented to show the advantages of content refinement.

1

Introduction

The communication network environment has rapidly improved, but many people connect to the Internet and exchange vast quantities of information. In particular, digital content, including documents, images, sounds, and combinations of these, are shared over a network. However, sources of the content items, such as web servers, data servers, etc., often include errors, which are released accidentally or, in some cases, intentionally. Additionally, some sources do not update or release obsolete content items. Consequently, users, especially those with poor computer literacy, have trouble retrieving content items. This situation makes it difficult for everyone to benefit from information and communication technologies. Because making technology easily accessible to all is a constant challenge, similar research has been conducted in the field of database and information retrieval to retrieve accurate information over the Internet [1–4]. However, most previous works are from the viewpoint of measuring the reliability of the information source or a semantic understanding of the content. Unlike previous works, herein we propose a simple concept of content retrieval over a large-scale network, which reduces errors included in the retrieved content items as if shared content items are delivered to the users after refinement. Additionally, the tendency of error reduction or content refinement using the proposed concept is demonstrated using simple experiments. A similar paper was released from the author [6] in 2007, however the concept of the proposed method is refined in this paper, and also the experiments are c Springer Nature Switzerland AG 2020  L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 293–302, 2020. https://doi.org/10.1007/978-3-030-33506-9_26

294

S. Sugawara

tried again with another basis of valuation and enhancing the precision of the experimental results. The remainder of this manuscript is organized as follows. Section 2 illustrates the concept of content refinement, while Sect. 3 explains conditions necessary to apply content refinement to actual communication services. In Sect. 4, simple experiments and their results are analyzed. Finally, Sect. 5 contains the conclusion and future work.

2

Concept of Content Refinement

We assume that each content item can be divided into a number of attributes of signifieds or signifi´e, which are the smallest and semasiologically non-divisible units of the content item. On the other hand, if the appropriate attributes of signifi´e are integrated in the appropriate order, a significative and independent content item can be assembled. Figure 1 illustrates the composition of a content item.

Fig. 1. Composition of a content item.

Because each attribute consists of a finite length of binary data, the number of possible sequences should be finite. If each sequence is called the value of the attribute, the number of each attribute’s possible values is also finite. If we call each of the values a “word1 ,” the number of possible words for an attribute is finite. Consequently, the words comprise a finite set. Furthermore, if the number of words for each attribute is finite, then the number of possible combinations of words corresponding to the attributes, which form the whole content item, should be finite. This means that the number of possible variations of a content item is finite. If the variations of a content item are limited and not infinite, the most accurate variation can be selected. Assuming that a large number of sources release a certain content item, we can use the weighted majority vote even if each item differs slightly due to error interfusions. Figure 2 illustrates a model of the above content refinement. A lot of samples of a content item are gathered over the network, and the samples are divided into attributes. Then the word of an attribute is refined by the weighted majority vote for each corresponding attribute from all the samples. 1

In other words, sign or signe.

Digital Content Refinement by Collecting Partly Unreliable Attributes

295

Fig. 2. Refinement of a content item.

If the attributes do not have dependent relationships, the most probable word of each attribute can be selected independently. Consequently, the collection of the most probable words for all the attributes forms the most probable content item. In this case, the applicability of the word x as the most reliable word of attribute i of content item p is expressed by the function of Fi,p (x) in Eq. (1), Fi,p (x) =

Np  

δd(i,j),

x

 · W (i, j) ,

(1)

j=1

where i: Identifier of an attribute belonging to Sp , (i.e., i ∈ Sp ) Sp : Set of identifiers of attributes that configure content item p x: Identifier of a word belonging to Sp,i , (i.e., x ∈ Sp,i ) Sp,i : Set of identifiers of words that can be the word of attribute i in content item p Np : Number of samples of content item p gathered over the network j: Identifier of a sample of content item p (1 ≤ j ≤ Np ) d(i, j): Word of attribute i in sample j δα,β : Kronecker delta  1 (α = β) δα, β = 0 (α = β), W (i, j): Weight for attribute i of sample j. The refined content item p can be obtained by the following procedure assuming that the content item p is configured by n attributes and the attributes can be identified using a number from 1 to n (i.e., attribute 1, attribute 2, · · · , attribute n). 1. For attributes 1 to n (1 ≤ i ≤ n), calculate the values of Fi,p (x) for all possible words x of the attribute.

296

S. Sugawara

2. For each attribute, recognize the word with the highest value as the most probable word of the attribute. 3. Reconfigure the content item p with the attributes of the most probable words. The weights for the attribute i of sample j are determined to reflect the difference of the reliability between samples. For example, samples released by reliable information sources should have larger weights or samples recently updated might be relatively precise and not obsolete. Section 4 shows a practical weight definition. Generally, some dependent relationships exist among the attributes within the same content item. For example, when the word of attribute A is X, the word of attribute B, which is subject to the word of attribute A, cannot be Y , etc. In this case, the number of possible variations of the content item is reduced. Additionally, if hierarchical dependent relationships exist among the attributes, the number of possible variations of the content item can be further reduced. To determine the most probable words (i.e., refine the attributes), the dependent relationships among the attributes must be considered to maintain consistency within a content item. If the most probable word is always determined independently, even in mutually related attributes, mutually conflicting words as the most probable ones are difficult to avoid, especially when the number of samples is small.

3

Some Conditions for Practical Use

To apply the content refinement illustrated above to actual communication services, it is preferable to satisfy the following conditions: • The content item to be refined should be released by a large number of independent information sources. • The content item to be refined should be divided into attributes, and a practical solution is a content item with a uniform format (e.g., a description in XML). • The number of possible words for an attribute should be small so that a general computer system can treat the attribute. • A signifier and its corresponding signified need to have a relationship of oneto-one mapping. To date, a practical method to satisfy these conditions has yet to be realized completely. The proposed content refinement can be applied to content formed with a finite number of clearly defined attributes. Actually in the field of “Collective Intelligence”, some intelligence can be extracted from digital contents such as blogs, wikis, message boards by some practical step-by-step ways [5]. An example of extracting intelligence from a webpage is as follows. 1. Tokenization (Parse the test to extract phrases)

Digital Content Refinement by Collecting Partly Unreliable Attributes

297

2. Normalize (Convert the text to lowercase) 3. Eliminate stop words (Eliminate trivial terms) 4. Stemming (Convert the terms into their stemmed form). Although this method is for extracting a major words from a webpage, the methodology might be useful for picking up attributes from a content item which consists of disorganized data.

4

Simple Experiments

4.1

Assumptions

To evaluate the effectiveness of content refinement, we conducted a simple experiment using computer simulations under assumed conditions. In this experiment, the weight for attribute i of sample j (i.e., W (i, j)) is defined as: W (i, j) =

Rj (T > tj ) T − tj

(2)

where Rj : The reliability of the information source releasing sample j T : Current time tj : Latest renewal time in the information source releasing sample j. In this case, the weight becomes large when the reliability of the information source is large and the source has been renewed recently. That is, we assume samples recently released by a dependable source are reliable. How the reliability of the information source is defined is explained later. Table 1 shows the assumptions and the corresponding parameters. • An original content item which consists of 81 attributes is assumed, and the original content continues to be updated its words of randomly selected attributes with random intervals. • There are 18 information sources (as a default) each of which observes the original content item and releases a partial copy of the original content item with reflecting its update as fresh as possible according to the observation. The number of the sources can be changed from 12 to 30 for the discussion on the impact of the number. • Each partial copy of the original content item released by an information source consists of 9 of the 81 attributes selected from the original randomly. • The number of candidate words for each attribute is 10 and one of the candidates is selected randomly and set as the word of each attribute of the original content item at initial state. • Partial copies of the original content item released by the information sources can include error words according to the error interfusion probability, which is explained later.

298

S. Sugawara

• The simulation advances according to the simulation time, which contains 1,000 time units. • For each time unit, all events such as original content’s update, observing the original by the information sources, releasing the partial copies of the original content item by the sources, and content refinement are executed in sequence. • An original content item updates the words of its attributes that are selected randomly, and the number of the updated attributes is decided according to the Poisson arrival in every time unit. • Each information source decides whether to observe the updates of the original content in each time unit according to a random probability in a normal distribution. • After observing the original content, information sources reflect the update if any to their release immediately. • The releases also include error words, which depend on the random error probability described by Eq. (3), as explained later. The equation is similar to Zipf’s Law. At the end of every time unit, the accordance rate between the original content item of current state and the result of content refinement is calculated. The average accordance rate is determined after 1,000 units of simulation time. The simulation is repeated 100 times to yield the average accordance rate. Table 1. Specific parameters (1) Parameters

Values

Number of information sources

12–30

Number of attributes in original content item

81

Number of candidate words for an attribute

10

Simulation time (number of time units) for a simulation run

1,000

Original content update frequency (average arrival rate for the Poisson distribution)

2.0–10.0 (default: 5.0)

Average of the normal distribution for information sources’ observation probability

0.5

Maximum error interfusion probability in an attribute of an information source’s release

0.1–0.5 (default: 0.3)

Number of simulation runs

100

The original content item is updated according to the Poisson arrival per time unit. The arrival rate is set between 2.0 and 10.0. The information sources’

Digital Content Refinement by Collecting Partly Unreliable Attributes

299

probabilities of observing the original content item is between 0.0 and 1.0 in 0.1 intervals, and the number of information sources with those probabilities is determined using a normal distribution. The average of the probabilities is set to 0.5, while the variance of the distribution is set to 0.3. The error interfusion probability Er (k) is defined as: Er (k) = Emax /k

(3)

k is a natural number from 1 to Np , and the Er (k) values, which are calculated by the equation, are randomly assigned to the information sources as their error interfusion probabilities. Emax represents the maximum error probability, which is set to 0.1, 0.3, or 0.5. Next, to investigate the impact of the number of the information sources, three assumptions are made. (i) The original content consists of 900 attributes. (ii) Part of the original content released by an information source consists of 9 attributes (identical to the former simulations). (iii) The average arrival rate of the Poisson distribution for content update is set to 55.0. 4.2

Comparison of Different Refinements

To verify the efficiency, four refinements were prepared: the proposed content refinement, the Reliability Oriented Refinement (ROR), Freshness Oriented Refinement (FOR), and Majority Oriented Refinement (MOR). These refinement schemes were devised by the authors for the simulations. • Reliability Oriented Refinement (ROR): For each attribute in the refining content item, all the words describing the attribute are checked, and the word from the information source with the highest reliability is selected as the most probable word corresponding to the attribute. • Freshness Oriented Refinement (FOR): For each attribute in the refining content item, all the words describing the attribute are checked, and the word with the newest renewal time among all of the gathered samples is selected as the most probable word corresponding to the attribute. • Majority Oriented Refinement (MOR): For each attribute in the refining content item, all the words describing the attribute are checked, and the most common word among all of the gathered samples is selected as the most probable word corresponding to the attribute. 4.3

Simulation Results and Analysis

To compare the effectiveness of the four refinements, parameters such as original content update frequency (arrival rate of the Poisson distribution) and maximum error interfusion probability were varied. In each simulation, one parameter was altered, while the rest were kept at their default value (see Table 1).

300

S. Sugawara 100

Accordance rate [%]

proposed FOR MOR ROR 90

80

70

60 0.05

0.1

0.15

0.2

Average Renewal Rate

Fig. 3. Accordance rate vs. update frequency

Fig. 4. Accordance rate vs. maximum reflection error probability

Figure 3 shows the impact of the update frequency on the original content. Proposed content refinement has a higher accordance rate between the words of the actual content item and the presumed one than the other refinements. MOR and ROR are also relatively high, but as the renewal frequency increases, their performances decline. Generally, the larger the frequency of the actual information renewal, the worse the accordance rate due to error interfusion. Next, we considered the impact of the information sources’ reflection errors. Figure 4 shows the relation between the accordance rate and the error interfusion probability of the information sources. The actual error probability of each information source is given according to Eq. (3), and the maximum error rate is set to 0.1, 0.3, or 0.5. For all refinements, the performance decreases as the error probability increases. FOR performs as well as MOR and is even better than ROR when the error probability is low. These results indicate that when releases from information sources contain few errors, the refinement where the most probable words are adopting according only to the freshness works effectively. However

Digital Content Refinement by Collecting Partly Unreliable Attributes

301

as the error probability increases, the efficiency of FOR decreases drastically because the newly refreshed content is not always accurate. On the whole, proposed content refinement works better than the other refinements from the viewpoint of data accordance between the original and refined content items. Additionally, the proposed content refinement maintains content freshness as well as FOR, which is designed to prioritize recent content. Finally, Figs. 5 and 6 illustrate the impact of the number of information sources. Time difference indicates the average amount of time required for the reflection of the original content item’s update to the releases of the information sources. Generally, as the number of information sources increases, the accordance rate increases, as well as the total time difference decreases gradually because the refined content tends to be more accurate as the number of information sources increases. In this simulation, the proposed content refinement works better than the other refinements with regard to both the accordance rate and total time difference.

Fig. 5. Accordance rate vs. number of information sources

Fig. 6. Total time difference vs. number of information sources

302

S. Sugawara

Note that the similar experimental results were also shown in [6] with similar conditions. Additionally, we executed simulations with sufficient number of the other parameter settings and confirmed the tendency of the results which was similar to ones shown above.

5

Conclusion

We propose a content refinement scheme that collects a number of partly unreliable attributes of a content item released by different information sources scattered over a large-scale network. The scheme applies weighted majority voting for each attribute to determine the most probable word for the attribute. The content item re-formed by the most probable words for all attributes selected in this way tends to agree with the original content item. Simple experiments confirmed that the proposed content refinement has potential to remove errors interfused in the information sources. Future work includes researching a practical method to apply content refinement to real services over a network. Acknowledgement. This work was partially supported by JSPS KAKENHI Grant Number JP17K00134.

References 1. Bowman, C.M., Danzing, P.B., Manber, U., Schwartz, M.F.: Scalable internet resource discovery: research problems and approaches. Commun. ACM 37(8), 98– 107 (1994) 2. Gil, Y., Ratnakar, V.: Trusting information sources one citizen at a time. In: Proceedings of the 1st International Semantic Web Conference, June 2002 3. Takehara, M., Nakajima, S., Sumiya, K., Tanaka, K.: A Trust value calculation method for web searching based on blogs (in Japanese). DBSJ Letters 3(1), 101– 104 (2004) 4. Kato, Y., Kurohashi, S., Emoto, H.: Credibility of information content: concepts and evaluation technology. (in Japanese), JSAI Technical Report, SIG-SWO-A602-01, November 2006 5. Alag, S.: Collective Intelligence in Action. Manning Publications, New York (2008) 6. Sugawara, S., Sonehara, N.: An efficient information retrieval from plural independent databases partially unreliable. In: Proceedings of the Third IASTED European Conference on Internet and Multimedia Systems and Applications, EuroMSA 2007, pp. 148–153 (2007)

Web Version of IntelligentBox (WebIB) and Its Extension for Web-Based VR Applications - WebIBVR Yoshihiro Okada(&) Innovation Center for Educational Resources (ICER), Kyushu University Library, Kyushu University, Fukuoka, Japan [email protected]

Abstract. This paper treats a 3D graphics software development system called IntelligentBox and its web version called WebIB. Originally, IntelligentBox was implemented as a development system for desktop 3D graphics applications. It provides various 3D software components called boxes each of which has a unique functionality and a 3D visible shape. IntelligentBox also provides a dynamic data linkage mechanism called slot connection that allows users to develop interactive 3D graphics applications only by combining already existing boxes through direct manipulations on a computer screen. Ten years ago, the author extended IntelligentBox system to make possible the development of web-based 3D graphics applications. This extended version of IntelligentBox is called WebIB. Furthermore, this time, the author extended WebIB to make possible the development of web-based VR (Virtual Reality) applications. This version of IntelligentBox called WebIBVR. In this paper, the author explains several new functionalities of WebIBVR and introduces use cases of web-based VR applications. Keywords: 3D graphics contents
Virtual reality


Development systems
Component ware
Web

1 Introduction Advances in 1990s’ computer hardware technology had made possible the development of interactive 3D graphics applications, and then 3D graphics software had become in great demand although its development had been still laborious work than 2D software development. Those years, we proposed a 3D graphics software development system called IntelligentBox [1] and its special purpose software component for collaborative applications [2]. Application fields of IntelligentBox include 3D-CG animation creation [3, 4], Virtual Reality (VR) system development [5–9], interactive visualization tool development [10–15], and so on. However, the developed 3D graphics applications could not be available on the web because originally IntelligentBox was realized as a development system for desktop 3D graphics applications. If 3D graphics applications developed using IntelligentBox were available on the web, the usefulness of IntelligentBox would become higher than ever. Then, we extended © Springer Nature Switzerland AG 2020 L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 303–314, 2020. https://doi.org/10.1007/978-3-030-33506-9_27

304

Y. Okada

IntelligentBox in order to make possible the development of web-based 3D graphics applications [16–18]. This is called WebIB as the web version of IntelligentBox. On the other hand, advances in recent computer hardware technology including display and sensor technologies have made possible the development of VR applications and they have become popular. Especially, VR applications available on the web have become more useful. If WebIB can be used for the development of web-based VR applications, its usefulness would become higher than now. Then, this time, we extended WebIB in order to make possible the development of web-based VR applications. This extended version of WebIB is called WebIBVR. In this paper, we explain essential mechanisms of WebIBVR besides IntelligentBox and WebIB, and show their several use cases. The remainder of this paper is organized as follows: Sect. 2 describes related work about development tools and systems of 3D graphics applications. We explain essential mechanisms of IntelligentBox and its extended mechanisms of WebIB in Sect. 3, and introduces several web-based 3D graphics contents of WebIB in Sect. 4. In Sect. 5, we explain extended functionalities of WebIBVR for the development of web-based VR applications, and Sect. 6 introduces several applications. In Sect. 7, we discuss about the significance of WebIBVR as a development environment for web-based VR applications. Finally, we conclude the paper in Sect. 8.

2 Related Work Our research purpose is to propose a software architecture that makes it easier to develop 3D graphics applications including interactive web-based 3D contents and VR applications. Its related systems are 3D graphics toolkit systems like Unreal Engine [19] and Unity 3D [20], and programming libraries like Open Inventor [21] and Coin3D [22]. Unreal Engine and Unity 3D are very popular game engines. Open Inventor is an OpenGL based object oriented programming library. Coin3D is also library very similar to Open Inventor. Some of them provide an authoring tool that enables to design 3D graphics contents. Even if using such authoring tools, it is not easy to develop 3D graphics applications because developers have to write text-based programs for that. As for development tools for interactive web 3D contents, there are library systems like Papervision3D [23], WebGL [24] and Three.js [25]. Papervision3D is Flash-based 3D graphics library. WebGL (Web Graphics Library) is JavaScript API for rendering interactive 3D graphics and 2D graphics within any compatible web browser without the use of plug-ins. Three.js is also a library for the development of WebGL based 3D graphics contents. These are library systems so that the user has to write text-based programs for developing web-based 3D graphics contents. Our research system IntelligentBox and its web version WebIB provide various 3D software components called boxes represented as visible, manually operable, and reusable functional objects. Furthermore, they provide a dynamic data linkage mechanism called slot connection. These features make it easier for even end-users to develop 3D graphics applications including web 3D contents. This is the main difference of IntelligentBox and WebIB from the others. In this paper, we also propose new framework of IntelligentBox for the development of web-based VR applications by

Web Version of IntelligentBox (WebIB) and Its Extension

305

extending WebIB framework. This new framework is called WebIBVR mainly proposed in this paper.

3 Essential Mechanisms of IntelligentBox and WebIB WebIB employs the same essential mechanisms of IntelligentBox. This section explains those mechanisms, i.e., Mode-Display Object structure and Dynamic data-linkage mechanism. After that, we explain extended mechanisms for WebIB including cases of multiple users. 3.1

Model-Display Object (MD) Structure

IntelligentBox provides software components called boxes. Figure 1 shows the internal structure of each box that consists of two objects, a model and a display object, called MD (Model-Display object) structure. A model holds state values of a box those are stored in variables called slots. A display object defines the appearance of the box that means how the box appears on a computer screen, and also defines the reaction of the box that means how the box reacts to user operations. Figure 1 shows an example of RotationBox that has a slot named ‘ratio’ that holds a double precision number used as a rotation angle. Through direct manipulation on the box using a mouse device, its slot value is changed. Furthermore, its visual image is also changed simultaneously, i.e. RotationBox rotates. In this way, each box reacts to the user’s manipulation according to its dedicated functionality.

Fig. 1. An MD structure of a box and its internal messages.

3.2

Dynamic Data-Linkage Mechanism Called Slot Connection

IntelligentBox also provides a dynamic data-linkage mechanism called slot connection. Figure 2 illustrates this data linkage concept among three boxes. As described above, each box has multiple slots. Its one slot can be connected to one of the slots of other box as shown in the figure. This connection is called a slot connection. The slot connection performs through the following three standard messages, i.e., (1) a set message, (2) a gimme (give me) message and (3) an update message, when there is a parent-child relationship between two connected boxes:

306

Y. Okada

(1) Parent box set . (2) Parent box gimme . (3) Child box update. A represents any value, and a represents a user-selected slot of the parent box that receives these two messages (1) and (2), a set message writes a child box slot value into its parent box slot and a gimme message reads a parent box slot value and sets it into its child box slot. Update messages are issued from a parent box to all of its child boxes to tell them that the parent box slot value has changed. By these three messages, the two slots of a child box and its parent box are connected and their two functionalities are combined.

Fig. 2. Standard messages between a child and its parent box.

3.3

Essential Mechanisms of WebIB

Figure 3 shows essential mechanisms of WebIB framework. Original IntelligentBox system uses OpenGL 3D graphics library which can generate off-screen rendered images. As shown in the figure, WebIB system runs on a web server, and its rendered image of a 3D scene generated on the web server is transferred to a web browser through the Internet. On the web browser, besides a HTML program, a JavaScript program runs to manage user operation events, i.e., a mouse move, a mouse button click and so on. Such user operation events are transferred to the web server using XMLHTTP request message through a CGI-program written in Perl running on the web server. The CGI-program once receives the user operation events and applies them to WebIB system. And then, WebIB system will generate next off-screen rendered image of the 3D scene to be updated according to the received operation event. In this way, the user can interactively manipulate 3D contents of WebIB system that runs on the web server with looking at his/her web browser. Since WebIB supports most web browsers. WebIB is available on any mobile device like iOS and Android OS device. However, when using iOS or Android OS device, the touch interface should be supported. WebIB system also supports it. In Fig. 3, the touch interface image appears in the right lower part of the web browser. When the user do the tap on the image, the

Web Version of IntelligentBox (WebIB) and Its Extension

307

position/direction of the viewpoint will be changed according to the type of an image, i.e., there are three types for the rotation, zoom and pan. Their details will be explain in the later section.

Fig. 3. Extended mechanisms of WebIB.

Fig. 4. Data-linkage among one WebIB and its two clients (left), and data-linkage among two individual WebIBs and two dedicated clients (right).

3.4

Mechanisms of WebIB for Multiple Users

As shown in the left part of Fig. 4, WebIB framework can provide multiple users with a web-based collaborative virtual environment based on the same mechanism of Fig. 3. In this case, only one WebIB system runs on the web server. The WebIB system has a System ID No. (SID No.), and by specifying it, each client user can access its corresponding WebIB that has the same SID No. through his/her web-browser. Using this data-linkage, communications between a teacher and a learner become possible in the cases of educational 3D contents. Viewpoints’ positions/directions of the two web browsers are always the same. This means collaboration that the student sees the same display image as that of the teacher always. On the other hand, the right part of Fig. 4 shows another case that each of multiple users individually manipulates his/her own WebIB. Each of multiple WebIBs runs on a

308

Y. Okada

common web server, and using different System ID No. (SID No.), each user can access his/her own WebIB through his/her own web browser.

4 Web-Based 3D Graphics Contents Realized Using WebIB This section introduces actual examples of interactive web 3D contents realized using WebIB. In this paper, we especially introduce education contents because they are regarded to be practical contents. We also introduce visualization tools because they are useful applications actually developed using IntelligentBox. 4.1

Education Contents

The left figure of Fig. 5 shows an example 3D model for students to learn a brain and its some parts represented as web 3D contents using WebIB on a web browser. A student can operate interactively for the change of his/her viewpoint, i.e., rotation, zoom and pan, and can point out any parts by a mouse device click to display their part names of the brain. As described in the previous section, using the same SID No., these operations are shared among other users. If one user is a teacher and other users are students, the all students can see the teacher’s operations and they can understand part names pointed out by the teacher. This collaboration case is shown as Fig. 6. Each student can look at the brain model on his/her own smartphone. Although the teacher can also use his/her smartphone, note PC is used in the case of Fig. 6.

Fig. 5. Education contents of medical field.

The right figure of Fig. 5 is another example, a heart model. Similarly to the left figure, each student can learn the structure of a heart and its part names individually on the web at any time and at any location using any browser on his/her portable device as shown in the right part of Fig. 6.

Web Version of IntelligentBox (WebIB) and Its Extension

309

Fig. 6. Collaboration case of a medical education content.

4.2

Visualization Tools

One of the popular topics of recent researches is analysis of big data for data science field. As one of the analysis methods, visualization is useful. Indeed, we have already developed two types of visualization tools using IntelligentBox. The first one is for browsing multimedia data in a file system including 3D model and motion data. This is called Treecube. The left figure of Fig. 7 shows a screen image of such visualization tool. See the papers [14, 15] for its detail. The other one is for the analysis of multi-dimensional/multi-attributes data because data to be collected from the physical world are various types and have various attributes. We have already proposed one information visualization tool called Timetunnel for the analysis of time-series numerical data and extended it by adding a visualization functionality similar to Parallel Coordinates. The right figure of Fig. 7 shows a screen image of its web content. See the papers [10–13] for its detail.

Fig. 7. Information visualization contents as web services.

310

Y. Okada

5 Essential Mechanisms of WebIBVR for Web-Based VR A head mounted display (HMD) is the most important feature for VR applications and the most portable/cheapest HMD is a set of a goggle and a smartphone as shown in Fig. 8. To use the set as HMD, a stereo view support is required. Therefore, we extended WebIB to support the stereo view on a web browser. The left part of Fig. 8 is an image of a stereo view on a smartphone’s web browser, the center is VOX + 3DVR goggle and the right is Poskey blue-tooth gamepad.

Fig. 8. The stereo view of WebIBVR on a smartphone’s browser (left), VOX + 3DVR goggle (center) and Poskey blue-tooth gamepad (right).

Besides stereo view support, we extended WebIB to support touch interfaces and device orientation/motion events as follows. (1) Stereo view support Three.js [25], one of the most popular WebGL based 3D graphics library, supports a stereo graphics. However, JavaScript part of WebIBVR framework does not use any 3D graphics library. Therefore, a stereo view image of side-by-side should be generated by WebIBVR system runs on a web server. Originally, IntelligentBox system supports a stereo view so that it generates a left eye image and a right eye image separately, and then WebIBVR merges them into one side-by-side stereo image. This stereo image is transferred to the web browser on a smartphone. Then, the user can see the stereo image like the left figure of Fig. 8. (2) Touch interfaces Originally, HTML5 supports JavaScript functions for touch interfaces like ‘touchstart’, ‘touchend’, ‘touchmove’ events and you can access x and y positions of your touch fingers by event.touches[*].pageX and event.touches[*].pageY, here, * means the index of your finger. As already explained in Subsect. 3.3, the touch interface part is included in the JavaScript program of WebIBVR framework and it provides the touch interface image that appears in the right lower part of a web browser. When the user do the tap on the image, the position/direction of a viewpoint is changed according to the type of the image, i.e., there are three images for rotation, zoom and pan as shown in

Web Version of IntelligentBox (WebIB) and Its Extension

311

Fig. 9. X and Y values treated as mouse device moving values are calculated by the quadratic equation of the distance from the center position to the tapped position. (3) Device orientation/motion interfaces The device orientation/motion interfaces are simple but useful in a JavaScript program of HTML5 because there are ‘deviceorientation’ and ‘devicemotion’ events, and you can access the device orientation/motion of your smart phone by event.alpha, event.beta, event.gamma, event.acceleration.x, event.acceleration.y and event.acceleration.z, respectively. The JavaScript program of WebIBVR framework supports device orientation/motion events and corresponding values are sent to WebIBVR system runs on a web server to change the position/direction of a viewpiont.

Fig. 9. Three images for the view controls, rotation, zoom and pan.

Fig. 10. Dental training VR supporting a haptic device (phantom).

312

Y. Okada

6 Web-Based VR Example Realized Using WebIBVR There have been several VR applications actually developed using IntelligentBox so far because originally IntelligentBox supports various types of VR devices by their dedicated software components, i.e., special purpose boxes. One of the VR applications is a dental training system [9] that supports a phantom device, one of the haptic devices, as a 3D input device. Using a phantom device, the user can feel force-feedback as operation sensation. HDM had not been supported as a display device in the original dental training system. As explained in the previous section, using WebIBVR framework, we can use a simple HDM, i.e., a set of a smartphone and a goggle, as a display device of any 3D graphics applications already developed using IntelligentBox. So, this time, we tried to use our simple HDM as the display device of the dental training system as shown in Fig. 10. A side-by-side stereo graphics image of teeth and gum 3D models appears in the web browser on a smartphone. A trainee wears a goggle a smartphone put inside, then he/she can see teeth and gum 3D models. Through IEEE 1394 interface, a phantom device is connected to the web server on which WebIBVR system runs. The trainee can operate teeth 3D model using the phantom device by looking at the teeth and gum 3D models using HDM with high immersion.

7 Discussion There are many services on the web. However, there are not so many services of 3D graphics contents on the web. One of its reasons is that 3D graphics contents need 3D model data but they are usually not allowed to be distributed through the Internet due to their user-licenses (copyright law). Even if the distribution of 3D model data is allowed, there is another problem that the transmission time of them become very long when the data size is very large. On the other hand, our WebIBVR works as a SaaS (Software as a Service) application does not have the above problems because only offscreen rendered images of 3D scenes are transferred from a web server to a client web browser as explained in Subsect. 3.3. So, interactive 3D contents already developed as applications of IntelligentBox can be re-used as web 3D contents those work as SaaS applications. Its MashUp with other SaaS applications are also possible. As a result, the usefulness of WebIBVR has become higher than that of IntelligentBox. Indeed, we have been developing Web version of IntelligentBox called IB for Web [26]. This version is native application of JavaScript program with several JavaScript libraries, i.e., Three.js and so on. Already we spent around two years for the development and we need more time until its completion. Even the development will be completed, the problems mentioned above will still remain. Therefore, WebIBVR framework introduced in this paper must be useful.

Web Version of IntelligentBox (WebIB) and Its Extension

313

8 Conclusions This paper introduced software architecture that makes it easier to develop 3D graphics applications including interactive web 3D contents. Our research system IntelligentBox provides various software components each of which corresponds to each of required functionalities. Combining these software components by direct manipulations on a computer screen enables users to develop 3D graphics applications without writing any text-based program. This feature is important for end-users who do not have any programming knowledge. Furthermore, we enhanced the usefulness of IntelligentBox by extending it to make possible the development of web 3D graphics applications like web 3D educational contents. This is called WebIB, the web version of IntelligentBox. In this paper, we showed some interactive web 3D contents to clarify the availability of WebIB and mainly we proposed WebIBVR, extension of WebIB to make possible the development of web-based VR applications. The extension includes stereo graphics support, touch interface support and device orientation/motion event support. We also showed a dental training VR application as one of the web-based VR applications to insist the usefulness of WebIBVR. As future works, we will improve the transfer mechanism of an off-screen rendered image to reduce the required bandwidth and to improve the performance. For that, we will use WebRTC (Web Real-Time Communication). Also, we will develop more practical web-based VR applications to clarify the usefulness of WebIBVR. Acknowledgments. This research was partially supported by JSPS KAKENHI Grant No. JP17H00773.

References 1. Okada, Y., Tanaka, Y.: IntelligentBox: a constructive visual software development system for interactive 3D graphic applications. In: Proceedings of Computer Animation 1995, pp. 114–125, 1995 2. Okada, Y., Tanaka, Y.: Collaborative Environments of IntelligentBox for Distributed 3D Graphics Applications. Vis. Comput. 14(4), 140–152 (1998) 3. Okada, Y.: Real-time character animation using puppet metaphor. In: Nakatsu, R., Hoshino, J. (eds.) IFIP First International Workshop on Entertainment Computing (IWEC2002). Japan Entertainment Computing Technologies and Applications, 14–17 May 2002, Makuhari, pp. 101–108. Kluwer Academic Publishers (2003) 4. Okada, Y.: Real-time motion generation of articulated figures using puppet/marionette metaphor for interactive animation systems. In: Proceedings of the 3rd IASTED International Conference on Visualization, Imaging, and Image Processing (VIIP03), pp. 13–18. ACTA Press, Benalmadena, September 2003 5. Okada, Y., Shinpo, K., Tanaka, Y., Thalmann, D.: Virtual input devices based on motion capture and collision detection. In: Proceedings of Computer Animation 1999, pp. 201–209. IEEE CS Press, Geneva, May 1999 6. Okada, Y.: 3D visual component based approach for immersive collaborative virtual environments. In: ACM SIGMM 2003 Workshop on Experiential Telepresence (ETP 2003), pp. 84–90, Berkeley, November 2003

314

Y. Okada

7. Okada, Y.: 3D visual component based approach for effective telepresence systems. In: Proceedings of ACM SIGMM 2004 Workshop on Effective Telepresence: Toward Seamless Remote Interaction and Experience (ETP04), Demo paper, pp. 46–47, New York, October 2004 8. Okada, Y., Ogata, T., Matsuguma, H.: Component-based approach for prototyping of Tai Chi-based physical therapy game and its performance evaluations. ACM Computers in Entertainment 14(1), 4:1–4:20 (2016) 9. Yuuta, K., Okada, Y.: 3D visual component based development system for medical training systems supporting haptic devices and their collaborative environments. In: Proceedings of the 4th Int. Workshop on Virtual Environment and Network Oriented Applications, VENOA-2012 of CISIS-2012, pp. 687–692. IEEE CS Press, 4–6 July 2012 10. Akaishi, M., Okada, Y.: Time-tunnel: visual analysis tool for time-series numerical data and its aspects as multimedia presentation tool. In: Proceedings of 8th International Conference on Information Visualization (IV04), pp. 456–461. IEEE CS Press, London, July 2004 11. Notsu, H., Okada, Y., Akaishi, M., Niijima, K.: Time-tunnel: visual analysis tool for timeseries numerical data and its extension toward parallel coordinates, computer graphics, imaging and visualization. In: Proceedings of CGIV 05, pp. 167–172. IEEE CS Press, Beijing, July 2005 12. Okada, Y.: Network data visualization using parallel coordinates version of time-tunnel with 2D to 2D visualizaion for intrusion detection. In: IEEE 27th International Conference on Advanced Information Networking and Applications Workshops, WAINA 2013, pp. 1088– 1093, 25–28 March 2013 13. Okada, Y.: Parallel coordinates version of time-tunnel (PCTT) and its combinatorial use for macro to micro level visual analytics of multidimensional data. In: Modelling and Processing for Next Generation Big Data Technologies and Applications, p. 27. Springer (2014) 14. Tanaka, Y., Okada, Y., Niijima, K.: Treecube: visualization tool for browsing 3D multimedia data. In: Proceedings of 7th International Conference on Information Visualization (IV03), pp. 427–432, IEEE CS Press, London, July 2003 15. Tanaka, Y., Okada, Y., Niijima, K.: Interactive interfaces of treecube for browsing 3D multimedia data. In: Proceedings of ACM The 7th International Working Conference on Advanced Visual Interfaces (AVI 2004), pp. 298–302, Gallipoli, Italy, May 2004 16. Okada, Y.: IntelligentBox as development system for SaaS applications including web-based 3D games. In: Proceedings of the 9th Annual European GAMEON Conference, pp. 22–26 (2008) 17. Okada, Y.: Web version of IntelligentBox (WebIB) for development of web 3D educational contents. In: Proceedings of IADIS International Conference of Mobile Learning 2011, pp. 251–255 (2011) 18. Okada, Y.: Web version of IntelligentBox (WebIB) and its integration with Webble World. In: Webble Technology as Proceedings of First Webble World Summit (WWS 2013), CCIS series 372, pp. 11–20, 3–5 June 2013. (ISSN 1865-0929, ISBN 978-3-642-38835-4) 19. Unreal Engine, August 2019. https://www.unrealengine.com/ 20. Unity 3D, August 2019. https://unity3d.com/ 21. Open Inventor, August 2019. https://www.openinventor.com/ 22. Coin3D, August 2019. https://bitbucket.org/Coin3D/coin/wiki/IntroductionToCoin3D 23. Papevision3D, August 2019. https://github.com/Papervision3D/Papervision3D 24. WebGL, August 2019. https://www.khronos.org/webgl/ 25. Three.js, August 2019. https://threejs.org/ 26. Noguchi, K., Okada, Y.: IntelligentBox for web: a constructive visual development system for interactive web 3D graphics applications. CISIS 2019, 757–767 (2019)

Enemy Attack Management Algorithm for Action Role-Playing Games Tianhan Gao1(B) and Qingwei Mi2(B) 1

Engineering Research Center of Security Technology of Complex Network System, Ministry of Education, Liaoning Research Center of Safety Engineering Technology in Industrial Control, Northeastern University, Shenyang, China [email protected] 2 Northeastern University, Shenyang, China [email protected]

Abstract. Enemy management in the game is one of the research hotspots in the field of game AI. Enemy attack management algorithms are especially important in action role-playing games, of which the main body is combat interaction. The paper proposes an enemy attack management algorithm suitable for action role-playing games. Traditional enemy attack management algorithms have many limitations in the process of game implementation with poor adaptability, which affects the player experience. The algorithm has high adaptability, which can avoid the problems above effectively. The goal of the algorithm is to provide an efficient and complete solution for the game designers to implement the enemy attack management system, shorten the development cycle, and then complete the game with appropriate game mechanism, art and music design. Finally, the extension, implementation and application effects of the algorithm are prospected.

1

Introduction

In recent years, research in the field of game AI has continued to deepen, and its importance and priority in game design and development process is also increasing [1]. The goal of game AI is to promote the challenge of the game [2], enhance the fun of the player [3], and support the player experience [4]. The player experience refers to the total effect of the game’s feelings, thoughts, emotions and behaviors exerted on the player [5], which is the state that the player presents during the interaction with the game [6]. The most important part is the game engagement experience and the demand satisfaction experience [7], which determines the game’s favorable rating and player retention rate. In all game genres, role-playing games have the longest history. Action role-playing games are one of the most representative branches among them. In action role-playing games, the combat interaction between the player and the enemy is the main source of the game challenge. The main body of the combat is attack, and the enemy attack management algorithm will directly affect the game development progress and the player experience. c Springer Nature Switzerland AG 2020  L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 315–326, 2020. https://doi.org/10.1007/978-3-030-33506-9_28

316

T. Gao and Q. Mi

At present, the three most popular enemy attack management algorithms are the Kung-Fu Circle Algorithm, the Beyond Kung-Fu Circle Algorithm and the Belgian AI Algorithm. The process of the Kung Fu Circle Algorithm is that the enemy attacks the players in turn [8], it is not suitable for fast-paced action roleplaying games. The Beyond Kung-Fu Circle Algorithm allows multiple enemies to attack players at the same time [9], but it does not consider the combination relationship between the type of enemy and its attack type. The Belgian AI Algorithm is based on rectangular grids and uses the same rules for each decision. The enemy attack management system in “Kingdoms of Amalur: Reckoning” is based on the Belgian AI Algorithm [10]. Although it’s obviously that the Belgian AI Algorithm has more advantages than the Kung-Fu Circle Algorithm and the Beyond Kung-Fu Circle Algorithm, it still has some problems, such as the number and type of enemy that can be managed are few, the judgment mechanism has flaws, etc. The paper studies the design ideas and execution process of the Kung-Fu Circle Algorithm, the Beyond Kung-Fu Circle Algorithm and the Belgian AI Algorithm, determines the imperfections of them, proposes a new enemy attack management algorithm which is suitable for action role-playing games and makes simple implementation of the four algorithms above. Compared with the three traditional algorithms, the algorithm proposed in the paper is more adaptive and can give players better game experience. As an overall support of the game development, it helps game designers to implement the enemy attack management system for action role-playing games, improve R&D efficiency, and provide reference for the design of future enemy management systems.

2

Related Work

In the game, the player is the most important entity [11]. The enemy attack management system should be designed around the player. As the Kung-Fu Circle Algorithm, the Beyond Kung-Fu Circle Algorithm and the Belgian AI Algorithm are important references for designing enemy attack management algorithms, the paper analyzes the above algorithms in detail to determine the problems and causes, so that to avoid the same problems when designing the enemy attack management algorithm for action role-playing games. 2.1

The Kung-Fu Circle Algorithm

The Kung-Fu Circle Algorithm is one of the earliest algorithms used to manage enemy attacks. It is suitable for traditional slow-paced turn-based role-playing games. The schematic diagram of its specific execution process is shown in Fig. 1.

Enemy Attack Management Algorithm for Action Role-Playing Games

317

Fig. 1. The specific execution process of the Kung-Fu Circle Algorithm is that the enemy attacks the players in turn, that is, when an enemy attacks the player, the other enemies are waiting by default. When the enemy completes the attack, the others will continue to attack the player in a specific order.

Because of the fast pace of action role-playing games, the Kung-Fu Circle Algorithm is not applicable. 2.2

The Beyond Kung-Fu Circle Algorithm

The Beyond Kung-Fu Circle Algorithm is designed based on the Kung-Fu Circle Algorithm and is suitable for fast-paced action role-playing games. The schematic diagram of its specific execution process is shown in Fig. 2.

Fig. 2. Compared to the Kung-Fu Circle Algorithm, the Beyond Kung-Fu Circle Algorithm allows multiple enemies to attack the player at the same time, that is, execute the attack logic of each enemy at the same time, and the enemy attacks do not affect each other.

Since most of the action role-playing games have more types of enemy and attack types, the Beyond Kung-Fu Circle Algorithm cannot set the combination relationship effectively, resulting in a large limitation.

318

2.3

T. Gao and Q. Mi

The Belgian AI Algorithm

The Belgian AI Algorithm is based on rectangular grids and is suitable for fast-paced action role-playing games as well. It has a more complete design mechanism than the Kung-Fu Circle Algorithm and the Beyond Kung-Fu Circle Algorithm. It accommodates attacking enemies through eight rectangular grids around the player, each grid can accommodate one enemy. The Belgian AI algorithm sets grid capacity and attack capacity for the player and sets grid weight and attack weight for the enemy. The grid capacity limits the number of enemies that can attack the player at the same time and the attack capacity determines the attack type that the enemy can use. The grid weight is the grid capacity occupied by the enemy and the attack weight is the attack capacity possessed by the attack type. The sum of the grid weight of the attacking enemies and the sum of the attack weight cannot be greater than the initial grid capacity and attack capacity of the player. The AI used to handle enemy positioning in the Belgian AI Algorithm is called the stage manager. When an enemy is authorized by the stage manager to enter the grid, it gains the attack right, and the player reduces the grid capacity of the value of its grid weight. Then it will select the attack type which has higher attack weight than others. When an enemy is authorized by the stage manager to use an attack type to attack the player, the player reduces the attack capacity of the value of the attack weight. When the number of enemies that have gained the attack right changes, the attack capacity of the player will reset to the initial value, after which the enemies above reselects the attack type. When the enemy that has gained the attack right dies, the player increases the grid capacity of the value of its grid weight. If there are enemies that are refused to enter the gird by the stage manager, they will re-request to attack the player. Here’s an example to explain the specific execution process of the Belgian AI Algorithm. In the example, the initial grid capacity and attack capacity of the player are both 8, and the enemies only use melee attacks. Suppose the player enters the attack range of a junior soldier with a grid weight of 2. The soldier is ready to attack the player. It has two attack types, which are chop and stab with attack weights of 1 and 3. Before the attack, it will ask for the stage manager. Then the stage manager will process the request. Since the grid weight of the soldier is less than the grid capacity of the player, the stage manager will agree to the request, assigning it the closest available grid position, and then the soldier gains the attack right and the grid capacity of the player reduces to 6 at once. The soldier prefers the stab and asks for the stage manager to attack the player by this attack type. Then the stage manager will process the request. Since the attack weight of the stab is less than the attack capacity of the player, the stage manager will agree to the request and the attack capacity of the player reduces to 5 at once. This combat example is shown in Fig. 3. During the combat, the player enters the attack range of a senior soldier with a grid weight of 6. The senior soldier is ready to attack the player. It has two attack types, which are chop and whirlwind with attack weights of 1 and 7. It

Enemy Attack Management Algorithm for Action Role-Playing Games

319

Fig. 3. The combat scene after the player entering the attack range of the first soldier.

will ask for the stage manager. Since the grid weight of the senior soldier is equal to the grid capacity of the player, the stage manager will agree to the request, assigning it the closest available grid position, and the grid capacity of the player reduces to 0 at once. As the number of enemies that have gained the attack right increases, the attack capacity of the player resets to 8. Then the senior soldier and the junior senior will select the attack type in turn. The senior soldier prefers the whirlwind and asks for the stage manager to attack the player by this attack type. Since the attack weight of the whirlwind is less than the attack capacity of the player, the stage manager will agree to the request and the attack capacity of the player reduces to 1 at once. Then the junior soldier prefers the stab and asks for the stage manager to attack the player by this attack type. Since the attack weight of the stab is greater than the attack capacity of the player, the stage manager will reject the request. After which the junior soldier prefers the chop and asks for the stage manager to attack the player by this attack type. Since the attack weight of the chop is equal to the attack capacity of the player, the stage manager will agree to the request and the attack capacity of the player reduces to 0 at once. This combat example is shown in Fig. 4. During the combat, the player enters the attack range of another junior soldier. The soldier is ready to attack the player. It will ask for the stage manager. Since the grid weight of the soldier is greater than the grid capacity of the player, the stage manager will reject the request and will not assign it a grid position. Therefore, the soldier cannot enter the grid and must wait outside the grid. This combat example is shown in Fig. 5. The research and experimental results of the combat example show the three main problems of the Belgian AI algorithm: 1. Each grid can only accommodate one enemy. It is not possible to adaptively adjust according to the game mechanism. 2. It can only be used to manage enemies which use melee attacks and cannot manage enemies which use ranged attacks.

320

T. Gao and Q. Mi

Fig. 4. The combat scene after the player entering the attack range of the second soldier.

Fig. 5. The combat scene after the player entering the attack range of the third soldier.

3. When the number or position of enemies which have gained the attack right changes, the assigned position where the enemy which has gained the attack right before the change may not be the closest to the player after the change. It may also occur that the enemy which is not assigned position is already in the grid after the change, but the position is in conflict with the assigned position of the enemy which has gained the attack right before the change or it has not gained the attack right yet.

3

Methods

Due to the inapplicability of the Kung-Fu Circle Algorithm and the limitations of the Beyond Kung-Fu Algorithm, the paper proposes a new attack management algorithm for action role-playing games based on the definition of grid capacity,

Enemy Attack Management Algorithm for Action Role-Playing Games

321

attack capacity, grid weight, attack weight and the stage manager in the Belgian AI Algorithm. The algorithm is based on rings and accommodates enemies through the player-centric Four-Ring Model. Four circles with radius r1 , r2 , r3 and r4 are formed from the inside out, forming four ring areas which are named Melee Area, Ranged Area, Buffer Area and Pursuit Area for accommodating enemies which have gained the attack right using melee attacks, enemies which have gained the attack right using ranged attacks, enemies which has not gained the attack right, and enemies which are chasing the player respectively. The ring widths of the Melee Area, the Ranged Area and the Buffer Area are defined as w2 , w3 , and w4 . Divide the three ring areas into eight sector areas equally, and then complete the construction of the Four-Ring Model. The schematic diagram is shown in Fig. 6.

Fig. 6. The player-centric Four-Ring Model with four circles formed from the inside out.

Among the radii of the four circles, r1 , r2 and r3 are fixed variables. The algorithm defines that the number of enemies that can be accommodated in each ring area of the Four-Ring Model is proportional to its area. Without limiting the number of enemies that can be accommodated in the Pursuit Area, the ratio of the number of enemies that can be accommodated in the Melee Area, the Ranged Area and the Buffer Area is shown in formula 1. r12 : r22 − r12 : r32 − r22

(1)

The number of enemies that can be accommodated in each sector area of the Melee Area is set to 1, then use the rounding up method to get the ratio of the number of enemies that can be accommodated in each sector area of the three ring areas, as shown in formula 2. 1:

r2 − r2 r22 − r12  :  3 2 2 2 r1 r1

(2)

322

T. Gao and Q. Mi

Define r4 as a dynamic variable. Since r3 is a fixed variable, w4 is a dynamic variable as well. In the algorithm, the w4 value corresponding to each enemy in the player’s current combat are independent of each other. Formula 3 is used for calculating its initial value w0 . Formulas 4 and 5 are the real-time calculation formulas for the w4 value. Define t as the time elapsed since the enemy started pursuing the player. The unit of it is seconds and its initial value is 0. w0 = r4 − r3 w4 = (1 + 0.2t) · w0

(3) (4)

w4 = (1 − 0.2t) · w0

(5)

Compared to the Belgian AI Algorithm, the algorithm reinforces the control of the stage manager. The stage manager calculates the distance between the player and the enemy in each frame and detects the t value. When the enemy is pursuing the player, if the distance value of the current frame is less than the previous frame, the formula 4 should be used to increase the w4 value, thus expanding the Pursuit Area to adapt to the tendency to approach the player. If the distance value of the current frame is greater than the previous frame, the formula 5 should be used to reduce the w4 value, thus shrinking the Pursuit Area to adapt to the tendency to stay away from the player. If the distance value of the current frame is equal to the previous frame, maintain the w4 value unchanged. If the enemy does not enter the Melee Area, the Ranged Area or the Buffer Area when t equals to 5, it will stop chasing the player immediately and reset w4 and t to the initial values. The algorithm sets an unlimited grid capacity mechanism for each enemy, which is disabled by default. When the enemy asks for the stage manager to attack the player, if the mechanism is disabled, the stage manager will process according to the rule that the sum of the grid weight of the attacking enemies cannot be greater than the initial grid capacity of the player. If the mechanism is allowed, the stage manager will allow the sum of the grid weight of the attacking enemies greater than the initial grid capacity of the player, that is, skip comparing the grid weight of the enemy with the grid capacity of the player and agree to the request directly, and then assign it a position. When the number or position of enemies which in the player’s current combat changes, the grid capacity and the attack capacity of the player will reset to the initial value. At the same time, the enemy in the combat before the change immediately stops the behavior other than pursuing. If the enemy is in the player’s current combat before the change, when its position changes or dies, it will judge according to the current position in the Four-Ring Model and execute the corresponding behavior logic. If the enemy is not in the player’s current combat before the change, when it joins the combat, it will judge according to the current position in the Four-Ring Model and execute the corresponding behavior logic. The execution logic of the algorithm is shown in Table 1.

Enemy Attack Management Algorithm for Action Role-Playing Games

323

Table 1. Execution logic of the algorithm. Position (Before)

Position (After) Melee Area Ranged Area Buffer Area Pursuit Area

Melee Area (In combat)

Aa

Ranged Area (In combat) A Buffer Area (In combat)

A

Bb

Cc

A

B

C

A

A

B

C

Pursuit Area (In combat) A

A

B

Dd

Not in combat A A B C The enemy asks for the stage manager to attack the player, allows the unlimited grid capacity mechanism, and reset w4 and t to the initial values. b The enemy asks for the stage manager to attack the player and reset w4 and t to the initial values. c The enemy asks for the stage manager to pursue the player. d The enemy continues pursuing. a

After the enemy executes the corresponding behavior logic, the stage manager will prioritize the attack requests. If the unlimited grid capacity mechanism of the enemy is disabled, compare its grid weight with the grid capacity of the player. If its grid weight is not greater than the grid capacity of the player, the stage manager will agree to the request. If the enemy uses melee attacks, the stage manager will assign the closest available position to the player in the sector area of the Melee Area which is not full for it, and then it gains the attack right and the player reduces the grid capacity of the value of its grid weight. If its grid weight is greater than the grid capacity of the player or the Melee Area is full, the stage manager will reject the request and will not assign it a position in the Melee Area, but will assign the closest available position to the player in the sector area of the Buffer Area which is not full for it. If the enemy uses ranged attacks, the stage manager will assign the closest available position to the player in the sector area of the Ranged Area which is not full for it and the player reduces the grid capacity of the value of its grid weight. If its grid weight is greater than the grid capacity of the player or the Ranged Area is full, the stage manager will reject the request and will not assign it a position in the Ranged Area, but will assign the closest available position to the player in the sector area of the Buffer Area which is not full for it. Enemies which are refused to assign positions in the ring area corresponding to the attack type that they use by the stage manager will wait in place after arriving at the assigned position in the Buffer Area. If the Buffer Area is full, it will stop the current behavior immediately. If the unlimited grid capacity mechanism of the enemy is allowed, skip comparing the grid weight of the enemy with the grid capacity of the player and agree to the request directly. If the enemy uses melee attacks, the stage manager will assign the same position as the unlimited grid capacity mechanism is disabled for it and the player reduces the grid capacity of the value of its grid weight. If its grid weight is greater than the grid capacity of the player or the Melee Area is full, the stage manager will reject the request

324

T. Gao and Q. Mi

and will assign the same position as the unlimited grid capacity mechanism is disabled for it. If the enemy uses ranged attacks, the stage manager will assign the same position as the unlimited grid capacity mechanism is disabled for it and the player reduces the grid capacity of the value of its grid weight. If its grid weight is greater than the grid capacity of the player or the Ranged Area is full, the stage manager will reject the request and will assign the same position as the unlimited grid capacity mechanism is disabled for it. Enemies which are refused to assign positions in the ring area corresponding to the attack type that they use by the stage manager will wait in place after arriving at the assigned position. If the Buffer Area is full, it will stop the current behavior immediately. After the request is processed, the stage manager will disable the unlimited grid capacity mechanism of the enemy. After the stage manager has processed all the attack requests, the enemies who have gained the attack right will select the attack type, and then the stage manager will process the attack type selection request. After the stage manager has processed all attack type selection requests, the stage manager will process the chase requests. If the Buffer Area is not full, the stage manager will agree to the request and the enemy will begin pursuing the player. If the Buffer Area is full, it will stop the current behavior immediately. After the stage manager has processed all the pursuit requests, the algorithm ends, and all enemies in the player’s current combat after the change execute their behavior according to their complete behavior logic.

4

Results and Discussion

The experiment makes simple implementation of the Kung-Fu Circle Algorithm, the Beyond Kung-Fu Circle Algorithm, the Belgian AI Algorithm and the algorithm proposed in the paper. Then, compare the adaptability parameters of the four algorithms above. The results are shown in Table 2. Table 2. Adaptability of the four algorithms. Parameters

Algorithms The Kung-Fu Circle Algorithm

The Beyond Kung-Fu Circle Algorithm

The Belgian AI Algorithm

The algorithm proposed in the paper

Yes

Yes

Yes

Set the combination rela- No tionship of types of enemy and attack types effectively

No

Yes

Yes

The number of enemy that – can be managed

–

Eight

Adaptive

The type of enemy that can – be managed

–

Melee

Melee/Ranged

The perfectness of the judg- – ment mechanism

–

Low

High

Suitable for action playing games

role No

Enemy Attack Management Algorithm for Action Role-Playing Games

325

By analyzing the experimental results, it can be concluded that the experimental results are consistent with the analysis results in the paper. The experiment shows that compared with the three traditional algorithms, the algorithm proposed in the paper is more adaptive. Therefore, it can give players better game experience.

5

Conclusion and Future Work

In order to implement the enemy attack management system suitable for action role-playing games efficiently and completely and improve the game development progress and the player experience. The paper proposes an enemy attack management algorithm with high adaptability. The algorithm uses the Four-Ring Model to accommodate enemies. The radii of the four circles in the Four-Ring Model can be adaptively set according to the game mechanism to adjust the number of enemies that can be accommodated in each sector area. The algorithm can not only manage the enemy using melee attacks, but also manage the enemy using ranged attacks. At the same time, it reinforces the control of the stage manager and has perfect management mechanism and judgment mechanism. The algorithm has high adaptability and rigorous execution logic, which can effectively avoid the problems in the traditional enemy attack management algorithms. The algorithm still has improvements in some aspects, such as the specific priority when the stage manager processing request, the monopoly problem by the same attack type, and the situation that the same attack type used by the same type of enemy. In the future work, the algorithm will be extended for the above aspects and implemented based on the popular game engine. Then compare the algorithm with the traditional enemy attack management algorithms in an all-round and multi-angle way and optimize it according to its actual running efficiency and cost in the game and the experimental data overall. So as to make the algorithm the first choice for the enemy attack management algorithm of action role-playing game designers. Acknowledgments. This paper is supported by China Fundamental Research Funds for the Central Universities under Grant No. N180716019 and Grant No. N182808003.

References 1. Ponsen, M.J., Spronck, P., Munozavila, H., Aha, D.W.: Knowledge acquisition for adaptive game AI. Sci. Comput. Program. 67(1), 59–75 (2007) 2. Shi, Z., Wang, Y., He, S., Wang, J., Dong, J., Liu, Y., Jiang, T.: Automatic game AI design by the use of UCT for dead-end. In: International Conference on Natural Computation (2010) 3. Shaker, N., Asteriadis, S., Yannakakis, G.N., Karpouzis, K.: Fusing visual and behavioral cues for modeling user experience in games. IEEE Trans. Syst. Man Cybern. 43(6), 1519–1531 (2013)

326

T. Gao and Q. Mi

4. Nareyek, A.: AI in computer games. ACM Queue 1(10), 58–65 (2004) 5. De Castell, S., Jenson, J. (Eds.).: Worlds in play: international perspectives on digital games research, vol. 21. Peter Lang (2007) 6. Francillette, Y., Abrouk, L., Gouaich, A.: A players clustering method to enhance the players’ experience in multi-player games. In: Computer Games (2013) 7. Denisova, A., Nordin, A.I., Cairns, P.A.: The Convergence of Player Experience Questionnaires. In: Annual Symposium on Computer-Human Interaction in Play (2016) 8. Rogers, S.: Level Up!: The Guide to Great Video Game Design. Wiley, Hoboken (2010) 9. Despain, W.: 100 Principles of Game Design. New Riders, San Francisco (2012) 10. Rabin, S.: Game AI Pro: Collected Wisdom of Game AI Professionals. A. K. Peters, Ltd., Natick (2013) 11. Cruz, C.A., Uresti, J.A.: Player-centered game AI from a flow perspective: towards a better understanding of past trends and future directions. Entertain. Comput. 20, 11–24 (2017)

Apply Lagrange Interpolation Based Access Control Mechanism in Personal Health Record Medical System Kuang-Yen Tai1(&), Dai-Lun Chiang2, Chun-Yen Chuang3, Tzer-Shyong Chen4, and Frank Yeong-Sung Lin1 1

3

Department of Information Management, National Taiwan University, Taipei, Taiwan [email protected] 2 Graduate Institute of Biomedical Electronics and Bioinformatics, National Taiwan University, Taipei, Taiwan Department of Electronic Engineering, Chung Yuan Christian University, Taoyuan City, Taiwan 4 Department of Information Management, Tunghai University, Taichung City, Taiwan [email protected]

Abstract. This study applied a Lagrange Interpolation based access control mechanism in the cloud Personal Health Record (PHR) system. We proposed an access control mechanism to make sure that health care information sharing is secure through the public key cryptosystem and Lagrange interpolation. Additionally, the access authority and the privacy setting have to be confined severely when we consider personal health records and private information. The PHR ought to prevent unauthorized users on the top of providing users with the access authority. As a result, we analyzed the mechanism’s security of the network attackers. On the basis of our analysis consequent, the access control and key management mechanism that we proposed can protect the health care information which can be shared among medical institutions efficiently and effectively. Keywords: Lagrange interpolation
Access control (PHR)
Key management mechanism


Personal Health Record

1 Introduction Unauthorized access request, data harmed, and privacy exposed are the common access control. The issues above reveal how important access control mechanisms are. An access control mechanism can also protect confidential contents and resources through implementing authorized predecessors access to the secured information of successors. Considering the security and privacy when designing the cloud medical information system is an important issue [1]. The previous application of hierarchical access control in information systems is apply to manage the file system. This system is wildly applied among military communications fields, private corporations, and government © Springer Nature Switzerland AG 2020 L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 327–337, 2020. https://doi.org/10.1007/978-3-030-33506-9_29

328

K.-Y. Tai et al.

departments for a long time which are database management systems [2], computer networks [3] and operating systems [4]. Therefore, people can witness the importance of the application of access control. Table 1 reveals the method of the application. Table 1. Various application domains of access control

Application

Wireless communication [5]: Transferring messages

Database management systems (DBMS) [6]: Cells, rows, tables, views

Online social networks (OSNs) [7]: Messages, databases

Government institutions, Commercial corporations[8]: Documents, Emails

For patients, it is critical that secure data sharing becomes a sufficient collaborative treatment and care choices [9]. When people face the emergence, in order to increase the chance of providing appropriate medicine, it is necessary that the medical personnel should have some necessary elementary and valuable health information about the patient. Not everyone can access to PHR which includes the physician because the system of PHR is sensitive and confidential. Transferring data among medical institutions such as pre-defined access control policies decreases when there is an emergence since the emergency medical team are not allowed to access the patients’ health record. Patients are not given the authority to manipulate the access control for providing his/her PHR [10]. As for the patient’s security and privacy, two things are crucial that caring patients and storing their health record. There is also another issue during an emergency condition. After accessing to the PHR, the personal information might be abused and be exploit illegally. The purpose of this study is to propose an integrated medical information access system allowing PHR users accessing the system under the secure environment through the high-security and efficient information security mechanism. Security threat including interruption of health care information, stolen patients’ data, and loss of finance and personal private information can be prevented. Based on management of the hierarchical structure, the idea of public-key cryptosystems and public encryption, as well as Lagrange interpolation polynomial, are applied in this study, such the key

Apply Lagrange Interpolation Based Access Control Mechanism

329

management and access control mechanisms, to ensure the medical information sharing security and confidentiality and guarantee the privacy security of personal medical information.

2 Mathematical Backgrounds A. Lagrange Interpolation In order to figure out the regular pattern, we formulate the with numerical analysis to discuss the cross-referencing among affairs and objects. From the regular patterns derived from functions, however, we can not definitely express and confirm the relationship because the functions are various [11]. With interpolation in mathematics, we obtain the function notation of Lagrange interpolation polynomial from the finite set on an x-y plane discussed in this study. Through Lagrange interpolation polynomial, functions are allowed covering multipoint coordinates in the finite set on the x-y plane. For example, we give n + 1 points on the x-y plane, e.g. ðx1 ; y1 Þ; ðx2 ; y2 Þ; . . .; ðxn ; yn Þ, the n-th degree of polynomial is established through Lagrange, where lj(x) is defined as the interpolation basis function. n Y x  xi x  xi i¼0;i6¼j j        x  x0 x  xj1 x  xj þ 1 x  xn ¼





; 1jn xj  x0 xj  xj1 xj  xj þ 1 xj  xn

‘j ðxÞ ¼

ð1Þ

lj(x) presents the characteristics that taking the value of xj as 1 and the value of other points xi (i 6¼ j) as 0, lj(x) could be acquired as below.  ‘j ðxÞ ¼

0; i 6¼ j 1; i ¼ j

ð2Þ

In this case, the interpolation polynomial is express as below. LðxÞ ¼

n X

yj ‘j ðxÞ

ð3Þ

j¼0

In general, the form of Lagrange will not expand but it is complicated while its regularity can be quickly formed into a high-order polynomial. If the solution is demanded, it would be an easier and faster method than the violent method or Isaac Newton. It is also easier to implement in a programming language. B. Secure Dynamic Access Control Scheme of PHR Methodology Personal Health Record (PHR) is a medical information records. It provides a comprehensive health care. Some clients are allowed to get into different private documents; the data on the framework can be included, evaluated, updated, and

330

K.-Y. Tai et al.

requested by those distinctive clients. [12]. Because of the large quantity of the confidential files and users, the authority for each user has to be complicated. The following steps present the establishment of the access control system. Step1: Build a list of the system user. With the partially ordered sets, we establish the access relationship of this system. (S; 4) represents for a partially ordered set; 4 represents the reflexivity, antisymmetric, and the symbol to deliver binary data in set S, where different users are denoted Si in set S. The access rights are set according to the identity, and Hi stands for the individual key of user Si. Given set S = {S1, S2, …, Sn} and H = {H1, H2, …, Hn}, the system would then construct the user list with n users. Step2: Establish the associative array of system users and data files as well as the access function. In consideration of security, all data available for users’ access are first encrypted by the system to construct the set file = {file1, file2, …, file, m} combined with m files. Each file u (for u = 1, 2, …, m) has the correspondent decryption key DKu. The file encrypt with the key will not be randomly accessed by internal or external users without authority. Step3: When user Si is authorized the decryption key of an encrypted file, Si is express as below: Si ¼ fu: u is the file index value of Si with access rightsg

ð4Þ

where i = 1, 2, …, n, and n 2 N. For set ðS, 4Þ, Sj 4 Si ði, j; 2 NÞrepresents that user Si is authorized to access the encrypted file. For example, when Sj = {6, 7} and Si ¼ f6; 7; 8g; f6; 7g4f6; 7; 8g that Sj4Si . In this case, Si is authorized to acquire the decryption key for accessing the encrypted files file1 and file2. Through an access control matrix, Formula (5), the system presents a user’s confidential file access, where the numerical value 1 stands for the user with access rights, while the numerical value 0 represents without access rights. As the example in Formula (5), user S2 is authorize to access file2, file3, and file4, but not file1. Users S1, S2…S4 have the exclusive private keys H1, H2…H4, and confidential files file1, file2…file4 require separate decryption keys DK1, DK2…DK4 that the decryption key of a file needs to be acquired for the access. S1 S2 S3 S4

file 2 1 1 61 6 41 0

file2 1 1 1 0

file3 1 0 0 1

file34 1 17 7 05 0

ð5Þ

An indicator formula I (x, y) is utilize for defining and expressing the possession of access rights, whether a user is authorized to acquire DKu and access file u. With the example in Fig. 2, I (1, 3) = 1 reveals user S2 could use the private key H2 to acquire DK2 for accessing file2. On the contrary, user S2 is not authorized to acquire DK2 for the decrypted file file2 when I (1, 3) = 0.

Apply Lagrange Interpolation Based Access Control Mechanism

 Iðx; yÞ ¼

1; if user x has access to file y 0; otherwise

331

ð6Þ

Step4: According to the system user key to generate relevant functions for the establishment, the system would provide the user with the decryption key DKu. As the example of access control matrix in Fig. 2, 4 users and 4 files are assumed. First, the indicator function under the definition of the system is used for authenticating the user’s key with access rights. If so, the result 1 reveals the user has the use right of the key; otherwise, the user does not have the right to use.  IfH1 ;...;Hn g ðxÞ ¼

1; if x 2 fH1 ; . . .; Hn g 0; o:w:

ð7Þ

3 Research Method Based on a Lagrange interpolation polynomial, we establish a secure and efficient information security mechanism for cloud medical information system of this study. Each authorized member of the system can access to the distinct confidential files. Under this pattern, we can guarantee the personal medical information to be private and secure. If there is a change in the system, e.g. patient referral, change of attending physician, rotation of caring nurses, or the employment of family physician, the system will not show those problems on the information security for the member adding and removal or the change of access rights and even the update of a confidential document. A. Usage Scenarios of Medical System on Cloud The application of Lagrange interpolation is able to solve variables and conditions of using effectively and efficiently. It can optimize variables and limit conditions and transform data into the solution with variables. Since the key is designed with Lagrange interpolation, each user’s private key is randomly generated. In addition, the correspondent relationship does not exist in private keys which would lead to the difficulty of cracking the private key system. Each user could use the private key for document modification such as add, delete, and demand as well as share files with specific users for the access. Figure 1 shows the usage scenarios on a PHR cloud medical system. Users can share all information on the cloud medical information system. It can also make several users access to it at the same time. In order to deal with this situation, we set the different authority for distinct access rights. Furthermore, for improving the efficiency and security of accessing confidential data, we suppose to consider multiple users accessing at the same time. Because the records are patient’s privacy, we have to ensure the user’s right and also prevent the illegal access. Therefore, we propose a practicable and secure method to prevent the system from being hacked.

332

K.-Y. Tai et al.

Fig. 1. Usage scenarios on a PHR cloud medical system

4 Research Implement A. Authority Setting of Users and File Access Rights We store patient’s medical information distinctively in a medical cloud system of the cloud system. It can provide the medical system members the use of it without being restricted to time and space. We control multiple users’ access right too files which means that a user should be authorized to access to the file and the private key. Furthermore, the system will exam the user’s private key to ensure the security and to prevent the attack. Through the access control matrix established by the certification authority (CA), the system can record all the access control. The matrix can record all the files and control authority. As it is shown in Formula (8), the numerical value 1 presents for a user with the file access rights, while the numerical value 0 represents without the access rights. S1 attending physician ðH1 Þ S2 physician ðH2 Þ S3 head nurse ðH3 Þ S4 ðH4 Þ S5 patient ðH5 Þ S6 family ðH6 Þ

file 2 1 1 61 6 60 6 60 6 41 0

file2 1 0 0 0 0 1

file3 1 1 0 1 0 0

file4 1 1 0 0 1 0

file5 1 1 1 1 0 0

file36 1 17 7 17 7 07 7 05 0

ð8Þ

(1) Method Establishment CA would calculate new polynomials Aj(x) and Bi(y) for each system user, according to the access control matrix.

Apply Lagrange Interpolation Based Access Control Mechanism

333

Step1: Polynomial Aj(x) is establish as below Ai ð xÞ ¼

Y



1  k  n;k6¼i

x  Hk þ ðx  Hi Þ Hi  Hk



ð xÞ

 IfHi g ; for i ¼ 1; 2;


; n ^ x 2 R ð9Þ

ðxÞ



where IfHi g ¼

1; if x 2 fH1 ;


; Hn g is used for authenticating the legality of private 0; o:w:

key Hj. Step2: The following conditions should be established. If Hj is a legal private key, Hi ðxÞ ¼ 1; if not, Hi ðxÞ ¼ 0, denoted as below: ð A ð xÞ Þ Hi ð xÞ ¼ If1gi ; Hi ðHi Þ ¼ 1; Hi ð6¼ Hi Þ ¼ 0:

ð10Þ

Step3: A private polynomial Bi(y) is established. h i ðiÞ ðiÞ ðiÞ ð yÞ Bi ðyÞ ¼ bm1 ym1 þ


þ b1 y þ b0  IJi ; y 2 R

ð11Þ ð12Þ



1; if y 2 Ji is used for authenticating the user’s legal decryption key DKu. 0; o:w: Step4: The complete decryption polynomial is established by CA, with the following expansion. IJi ðyÞ ¼

ðx; yÞ ¼

G

Xn i¼1

Ai ð xÞBi ð yÞ ^ x; y 2 R

ð13Þ

(2) Security Test of Decryption Polynomial ðxÞ

ðyÞ

Excluding the effects of IfHi g and IJi , let y = 0, the private key Hk would not be cracked, as following proof.  x  Hk ðxÞ þ ð x  H Þ  I fH i g ; for i ¼ 1; 2;


; n ^ x 2 R i 1  k  n;k6¼i H  H i k h i ðiÞ ðiÞ ðiÞ ð yÞ Bi ð yÞ ¼ bm1 ym1 þ


þ b1 y þ b0  IJx ; y 2 R

Ai ðxÞ ¼

Y



ð14Þ

334

K.-Y. Tai et al.

Ai ð xÞB Þ o ninð yQ h i h i o ð xÞ ðiÞ ð iÞ ð iÞ ð yÞ xHk m1 þ


þ b1 y þ b0  IJi ¼ 1  k  n;k6¼i Hi Hk þ ðx  Hi Þ  IfHi g  bm1 y o h nnQ h i io ð iÞ ðiÞ ð iÞ ð xÞ ð yÞ xHk m1 þ


þ b1 y þ b0 ¼  IfHi g  IJi 1  k  n;k6¼i Hi Hk þ ðx  Hi Þ  bm1 y

ð15Þ Let Ai ð xÞBi ð yÞ ¼

Y

h  i x  Hk ðiÞ ðiÞ ðiÞ þ ðx  Hi Þ bm1 ym1 þ


þ b1 y þ b0 Hi  Hk



1  k  n;k6¼i

ð16Þ ð xÞ

ð yÞ

Ai ð xÞBi ð yÞ ¼ Ai ð xÞBi ð yÞIfHi g IJi

ð17Þ

When the polynomial expansion is acquired, the effect of I(Hi)(x)IJi(y) could be ignored. ðrÞ ðrÞ Replace Ai ðxÞBi ðyÞ with Ai(x)Bi(y) and let y = 0, the following equation could be acquired. Ai ð xÞBi ð0Þ ¼

Y



1  k  n;k6¼i

x  Hk þ ðx  H i Þ Hi  Hk



ðiÞ

ð18Þ

b0

According to the equation above, it could not be further decomposed that Qn k ¼ 1 ðx  Hk Þ could not be acquired, as the previous formulae, revealing that the k 6¼ i private key Hk could not be solved. The system security is therefore ensured. B. Dynamic Access Control of Users and Files (1) Add Users In case of adding a new user in the system, the system CA will establish the new user’s access rights and update the existing public polynomial G(x, y). The user adding steps are listed as below. Step1: CA will generate the exclusive private key Hn+1 for adding a member Sn+1 to the system. Step2: CA updates the polynomial An+1(x) and sets it as private as well as updates ðxÞ the authentication indicator IfHn þ 1 g as following. ðrÞ A n þ 1 ð xÞ

¼

Y 1  k  n þ 1;k6¼n þ 1



x  Hk þ ðx  Hn þ 1 Þ Hn þ 1  Hk



ð xÞ

 IfHn þ 1 g

ð19Þ

Step3: When the private key Hn+1 is legal, An+1(Hn+1) = 1; otherwise, it would be 0.

Apply Lagrange Interpolation Based Access Control Mechanism

ðA ðxÞÞ hn þ 1 ð xÞ ¼ If1gn þ 1 ; hn þ 1 ðHn þ 1 Þ ¼ 1; hn þ 1 ð6¼ Hn þ 1 Þ ¼ 0:

335

ð20Þ

Step4: CA updates the polynomial B+1(y) and sets it as private as well as updates the authentication indicator Ijn+i(y) as below. ðr Þ Bn þ 1 ð yÞ

¼

X u2Jn þ 1

DKu

Y

ðy  t Þ t¼1;t6¼u ðu  tÞ m

  IJn þ 1 ð yÞ; y 2 R

ð21Þ

Step5: Update the existing public polynomial G(x, y) to new G ðx; yÞ From the user adding steps above, it is discovered that the system CA will conduct the polynomial An+1(x), Bn+1(y), and Jn+1 for the new user Sn+1 and update the ðxÞ ðr Þ authentication indicators IfHn þ 1 g , hn þ 1 ðxÞ, and IJn+i(y). Updating such new information G(x, y) to complete the change procedure. In the entire update process, merely some algorithms are used for adding the polynomial information of user Sn+1, and simple operation is applied to the updated G(x, y). The operation cost is therefore reduced. (2) Remove Members When a user of the medical cloud system needs to change the access rights, CA would adjust the access matrix and modify the parameters, as following steps. Assuming that a user Sn should be removed, there are two methods in CA. First, the polynomials A(x) and B(y) of Sk are removed from the public polynomial. G ðx; yÞ ¼ G ðx; yÞ  An ð xÞBn ð yÞ

ð22Þ

Second, the system delete the user’s access rights completely and the set Jn = {} is an updated one to complete the removal. (3) Modify Users’ Access Rights When a user of the medical cloud system requests to change the access rights, CA would adjust the access matrix and modify the parameters, as following steps. Step1: In advance, CA will first check the user’s access rights and the correspondent decryption key DKu and then reset the authentication indicator Ji. ð23Þ 0

Ji is the modified access rights of user Si and also the new set after CA recalculates the member access matrix. Step2: Since the indicator Ji is mutually associated with the polynomial Bi(y), CA has to simultaneously update the indicator Ji to Ji and the polynomial Bi(y) to Bi ðyÞ. Finally, they are update to the public polynomial. Gðx; yÞ ¼ Gðx; yÞ  Ai ðxÞBi ðyÞ þ Ai ðxÞBi  ðyÞ With the steps above, the user’s access rights can be updated.

ð24Þ

336

K.-Y. Tai et al.

5 Conclusion Based on the secure dynamic access control scheme in PHR methodology, we proposed this reformed method. PHR system can store data in the cloud computing environment and also be protected by the solid security mechanism. The system we proposed can arrange access among all different kinds of aspects and deliver the qualified protection efficiently. The compatibility of the access control management and system becomes an evitable and important issue in exchanging data of medical records. As a result, a secure and stable system for data management and sharing will be beneficial to the medical community. Through the analysis of four common attacks including external collective attack, reverse attack, collusion attack, and equation breaking attack, we have proved the security of the access control of this system. This is a reliable system in providing secure and qualified services. This study presents an effective method for optimizing the use of medical resources and promoting the quality of medical services. Regarding to the secure encryption method and the stable characteristics of the PHR system, the adoption of the system will bring great benefit to the medical community by delivering the qualified medical services to patients and the aid for practitioners making accurate decisions in real time. The composition of the access control mechanism will guarantee the safety of sensitive and private data in the exchange across the medical community.

References 1. U.S. Department of health and human services personal health records and the HIPAA privacy rule (2008). http://www.hhs.gov 2. Shete, S.S., Kulkarni, C.S.: Database security using role-based access control system. Int. J. Eng. Sci. Comput. 6(6), 8047-5053 (2016) 3. Karimi, F., Esmaeilpour, M.: A dynamic media access control protocol for controlling congestion in wireless sensor network by using fuzzy logic system and learning automata. Int. J. Comput. Sci. Inf. Secur. 14(4), 445–460 (2016) 4. Onarliogluamd, K., Robertson, W.: Overhaul: input-driven access control for better privacy on traditional operating systems. In: International Conference on Dependable Systems and Networks, pp. 443–454 (2016) 5. Xiao, P., He, J.H., Fu, Y.F.: Distributed group key management in wireless mesh networks. Int. J. Secur. Appl. 6(2), 115–120 (2012) 6. Bogaerts, J., Lagaisse, B., Joosen, W.: Idea: supporting policy-based access control on database systems. In: International Symposium on Engineering Secure Software and Systems, pp. 251–259 (2016) 7. Rathore, N.C., Tripathy, S.: A trust-based collaborative access control model with policy aggregation for online social networks. Soc. Netw. Anal. Min. 7, 7 (2017) 8. Kongsgard, K.W., Nordbotten, N.A., Mancini, F., Engelstad, P.E.: Data loss prevention based on text classification in controlled environments. In: International Conference on Information Systems Security, pp. 131–150 (2016) 9. Zhang, P., et al.: FHIRChain: applying block chain to securely and scalable share clinical data. Comput. Struct. Biotechnol. J. 16, 267–278 (2018)

Apply Lagrange Interpolation Based Access Control Mechanism

337

10. Thummavet, P., Vasupongayya, S.: Privacy-preserving emergency access control for personal health records. Int. J. Sci. Technol. 9(1), 108–120 (2015) 11. Mehrabian, A.R., Khorasani, K.: Distributed formation recovery control of heterogeneous multiagent euler-lagrange systems subject to network switching and diagnostic imperfections. IEEE Trans. Control Syst. Technol. 24(6), 2158–2166 (2016) 12. Liu, Y.C.: Leaderless consensus for multiple euler-lagrange systems with event-triggered communication. In: IEEE International Conference on Systems, Man, and Cybernetics (SMC), October 2018

Analysis of the Relationship Between Psychological Manipulation Techniques and Personality Factors in Targeted Emails Kota Uehara1, Hiroki Nishikawa1,2, Takumi Yamamoto2, Kiyoto Kawauchi2, and Masakatsu Nishigaki1(&) 1

2

Shizuoka University, 3-5-1 Johoku, Naka, Hamamatsu, Shizuoka 432-8011, Japan [email protected] Mitsubishi Electric Corporation, 5-1-1, Ofuna, Kamakura, Kanagawa 247-8501, Japan

Abstract. The damage from targeted email attacks continues to be an acute issue for Internet users. Several recent studies have demonstrated that psychological manipulation techniques (e.g. Cialdini’s principles) are used effectively in phishing mails, the susceptibility to Cialdini’s principles correlates with the personality factors (so-called Big Five), and the use of AI can serve to facilitate the assessment of the personality factors based on social media information. Based on the results outlined in the aforementioned studies, this paper considers the possibility of a new type of attack that uses open source intelligence (OSINT) tools to obtain social media information about the target and then misuse personality estimation tools and psychological manipulation techniques to create malicious emails with a highly effective level of psychological manipulation for each specific target. In this paper, to estimate the possibility of such attack, investigation and analysis in relation to such questions as whether Cialdini’s principles work in targeted emails, and whether the effectiveness of Cialdini’s principles in targeted email correlates to personality factors was performed through conducting a user experiment.

1 Introduction In recent years, the damage due to targeted e-mail attacks have increased rapidly. Targeted e-mail attacks are a typical example of social engineering causing damage (for example, exploiting information and money or manipulating PCs illegally) to the targets by deceiving them. To accomplish a targeted e-mail attack, it is essential to convince the target that the malicious e-mail is a regular e-mail, and consequently, the attacker attempts to collect sensitive information about the target. Currently, it is a common practice for companies and users to send information about themselves over the Internet, through owned media and social media. It has been reported that the web and social media is overflowed with information pertaining to companies and individuals. Moreover, personally identifiable information and privacy information can be obtained by consolidating publicly available personal information [1, 2]. This method of gathering information is called Open Source Intelligence © Springer Nature Switzerland AG 2020 L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 338–351, 2020. https://doi.org/10.1007/978-3-030-33506-9_30

Analysis of the Relationship Between Psychological Manipulation Techniques

339

(OSINT), and there are several tools that facilitate OSINT activities. Attackers can use OSINT tools to obtain information related to the identified targets (e.g., their organizations, direct managers, the names of their friends, email addresses, related events, and personal interests) and then, to incorporate this information in creating specialized emails for each target with a high level of the mimicry precision (that is, a malicious email whose legitimacy cannot be easily determined by a receiver) [3, 4]. Moreover, a recent study has revealed that information obtained from social media can be used to estimate the psychological tendencies of users, and that tools are actually being used to estimate the personality related to a users’ personality factors (Big Five [5]) based on information extracted from their tweets and blog posts [6]. As further research in conducted this field, it is expected that OSINT will become capable of estimating other psychological characteristics in addition to personality factors. It is evident that human behavior is susceptible to psychological manipulation to some extent. Experiments have shown that Cialdini’s principles [7] can be used to improve the effectiveness of psychological manipulation of phishing emails [8]. It has also been demonstrated that based on each user’s personality factors, Cialdini’s principles can differently affect the ease with which a user is influenced by psychological manipulation [9]. Hence, it is important to forearm ourselves against new types of attacks that may use OSINT tools to obtain social media information on the target and then, misuse personality estimation tools and psychological manipulation techniques to create malicious emails for each target with a highly effective level of psychological manipulation (that is, creating targeted emails that easily influence targets). To estimate the possibility of such an attack, a user experiment was performed during this study to investigate and analyze, whether Cialdini’s principles work in targeted email creation, and whether the effectiveness of Cialdini’s principles in targeted emails correlates to personality factors, with the goal of anticipating the danger arising from such attacks and defending against them. The user experiment revealed that Cialdini’s principles are effective when used within the text of targeted emails, but did not allow revealing correlation between the effects of Cialdini’s principles and the human personality factors. With regard to the latter, the paper of Egelman and Peer demonstrates that behavioral characteristics (rather than personality factors) tend to have a stronger effect on human ideas on privacy [10]. Therefore, it may be possible to obtain an insight on which types of people could be easily influenced by psychological manipulations using Cialdini’s principles in targeted emails by analyzing both personality factors and behavioral characteristics.

2 Related Work A brief overview of the previous studies related to the targeted email attacks using OSINT is provided in this section from the perspective of mimicry precision and the effectiveness of psychological manipulation. With regards to the former, it was demonstrated by Ball et al., [3] that the threats to social engineering using OSINT have been increasing dramatically. Additionally, Uehara et al. in their study [4], proposed a model of the state transition diagram to demonstrate how attackers make use of OSINT

340

K. Uehara et al.

tools during the creation of targeted emails with high-precision mimicry to obtain targeted individual information in a state transition manner. The latter is broadly classified by the previous studies into the following main groups: the character estimation of the target using OSINT; the effect of psychological manipulation according to the character; and the deterioration of social engineering due to psychological manipulation. Earlier study related to ascertaining the personality of targeted individuals using OSINT include active research of a user’s five major personality factors (Big Five) extrapolated from information posted to the social media. The remarkable results of this research have already been implemented in IBM Watson’s Personality Insights [6]. The Personality Insights tool employs a user’s tweets published on Twitter and their blog posts to calculate their Big Five score. Several related studies show the presence of correlation between the Big Five of the psychological test subjects and the text written by those subjects. Personality Insights is an AI-based tool that employs machine learning techniques to identify the presence of correlation between the results of a psychological test (questionnaire) on the Big Five administered to several thousand users, and social media texts and posts (tweets or blog posts). Personality Insights is publicly available as an API module, hence, just by uploading social media texts of a user, it calculates a Big Five score for this individual. Previous study on the effect of psychological manipulation based on personality factors has demonstrated that certain personality factors are more easily affected by Cialdini’s principles [9]. Cialdini’s principles (proposed by Dr. Robert Cialdini) are the psychological principles that make another party susceptible to external influence [7]. Alkış et al. clarified that Cialdini’s principles are effective on people of all types, and performed experiments demonstrating that a user’s personality factors affect “the degree of influence” of Cialdini’s principles. Specifically, they administered psychological tests (questionnaires) related to the Big Five proposing subject answer questionnaires to determine the response rate to Cialdini’s principles, and then investigated the correlation between the specific Big Five score of each user and various Cialdini’s principles. The investigation showed several cases of significant correlation between the Big Five scores and users’ response rates for each of Cialdini’s principles. Furthermore, considering the existing studies on psychological manipulation, Wright et al. [8] and Akbar [11] in their papers illustrated the results of the misuse of Cialdini’s principles in phishing emails. Wright et al. sent a phishing email incorporating the six Cialdini’s principles and an email without incorporating them to a group of university student test subjects, and then, compared the difference in response rates (rate of students who followed the instructions in the phishing email) [8]. The results of the experiment showed a higher response rate from subjects that received the phishing email incorporating any of Cialdini’s principles (at least one) compared to those who received an email created without applying the principles, which demonstrated the effectiveness of Cialdini’s principles in phishing emails. Akbar developed a flow chart to identify whether or not an email text incorporated Cialdini’s principles, and investigated how frequently phishing emails actually employ Cialdini’s principles [11]. The results showed that 96.1% of phishing email data sets studied by Akbar incorporated “Authority”, and 41.1% used “Scarcity” principles, and that a high percentage of phishing emails also complied with other principles.

Analysis of the Relationship Between Psychological Manipulation Techniques

341

Akbar’s study demonstrated that currently Cialdini’s principles are widely used in phishing emails, and it can be inferred that Cialdini’s principles will work effectively in targeted malicious emails. Therefore, it should be taken into consideration that new types of attacks may be invented to make use of OSINT tools to obtain social media information on the target and thereafter, to misuse personality estimation tools and psychological manipulation techniques seeking to automatically create specialized emails for each target with a highly effective level of psychological manipulation (that is, targeted emails that easily influence targets).

3 Targeted Emails with a Highly Effective Level of Psychological Manipulation by Using OSINT 3.1

Big Five

The Big Five, as proposed by Goldberg is a type of personality evaluation scale, also known as the “five major factor personality model” [5]. It is considered a comprehensive, clear model that provides an understanding of personalities, and is often used in many fields such as healthcare and consumer preference surveys. The Big Five is based on the following five elements used to evaluate personalities: • • • • • 3.2

Openness; Conscientiousness; Extroversion; Agreeableness; Neuroticism. Obtaining Big Five Score Using OSINT

Recent developments in AI have made it possible to extrapolate the Big Five of individuals from their social media information (tweets and blog posts). Machines have no human biases and therefore, are able to extrapolate the Big Five objectively. Representative examples include IBM Watson’s Personality Insights [6]. Personality Insights is openly available as an API in the IBM Watson Developer Cloud. The API was used to build a program that analyzes the user’s tweets to estimate the user’s Big Five scores. 3.3

Cialdini’s Principles

Cialdini’s principles are the psychological principles proposed by Dr. Cialdini that coerce a person to be susceptible to external influence [7]. Cialdini’s principles consist of the following six elements outlined below: • Liking This psychological principle explains that people actively respond to requests made by those they like. The person does not necessarily have to know the other party, but the principle of Liking includes a “likable personality” or a “polite tone.”

342

K. Uehara et al.

• Reciprocation This psychological principle explains that people want to repay favors to others (feel like they have to repay favors). Even if the favor is forced upon the person in a one-sided fashion, they feel obligated to return the favor. In other words, the person receiving the favor is psychologically motivated to return the favor whether they are happy to or not. • Social Proof This psychological principle explains that people want to follow what other people around them are doing. This psychological motivator arises from the desire to use the behavior of others (third parties) as judgment criteria for decision making in different situations. • Commitment and Consistency This psychological principle explains that people want their behavior to be consistent (want to become consistent). People tend to justify their decisions. In other words, when faced with similar situations from the past, people execute the same actions that were effective before. The principle of Commitment and Consistency also says that people attempt to keep the promises they have made. • Authority This psychological principle explains that people trust those in positions of authority with titles or experience. It results in the psychological motivation to follow those who are or seem to be above one’s station or experts in a certain field. • Scarcity This psychological principle explains that the rarer something is, the more value is placed on it. It results in the psychological motivation to obtain something quickly before it is finished; this usually refers to things with pressing time limits or limited quantities. 3.4

Relationship Between Big Five and Cialdini’s Principles

Although Cialdini’s principles apply to all people, it has been demonstrated that the degree of susceptibility to these principles is differently affected by one’s personality [9]. For example, the persons with a high Big Five extroversion score are more susceptible to Reciprocation, Scarcity, and Liking, whereas the persons with a high conscientiousness score are more susceptible to reciprocation, liking, and commitment and Consistency, but less susceptible to Liking as shown in the experiments. Therefore, the attacker can use the methods explained in the previous section to obtain the Big Five of the target, thereby, learning, about the exposure to Cialdini’s principles most effective for the targeted individual. Afterwards, the attacker can use the method, which is explained in the next section to incorporate the principles into an email seeking to create a targeted email with a highly effective level of psychological manipulation for the target. 3.5

Application of Cialdini’s Principles in Targeted Emails

As demonstrated by Uehara et al. in their study [4], attackers can employ OSINT tools to increase the mimicry precision of targeted emails. Figure 1 shows a sample targeted email attack perpetrated by using OSINT to obtain the target’s email address and

Analysis of the Relationship Between Psychological Manipulation Techniques

343

organization from their name. The email is spoofed to appear as if it originated from the technical department of the organization where the target works, recommending that the target install an antivirus software, and prompting the user to click a malicious URL. Figures 2 and 3 show emails to which the attacker applied the Cialdini’s principles Social Proof and Authority, respectively. As explained in Sect. 3.3, the principle of Social Proof is a psychological principle that causes people to “want to follow what others around us are doing.” This is misused by the attackers to insinuate that “other employees have already installed this antivirus software.” The principle of Authority is a psychological principle, in which people trust those in ‘authority’ with titles or experience. This can be misused by attacker through using the name of a manager in the organization. Because many companies currently list investor relations (IR) information on their websites, it is comparatively easy to obtain the names of executives using OSINT. It is thought that this information included in the text of an email will result in the high possibility of a user following the instructions.

Fig. 1. Example of targeted email

Fig. 2. Example of targeted email using social proof

Fig. 3. Example of targeted email using authority

Other Cialdini’s principles aside from Authority can be incorporated into the text to utilize other human psychological principles to create targeted emails with a highly effective level of psychological manipulation. The attacker’s cost effectiveness is

344

K. Uehara et al.

considered extremely high as a result of adding just a small chunk of text that has a certain extent of effectiveness in psychological manipulation. 3.6

Research Questions

Sections 3.1 through 3.5 describe new possible types of attacks that may use OSINT tools to obtain social media information on the target and misuse personality estimation tools and psychological manipulation techniques to create targeted malicious emails for each target with a highly effective level of psychological manipulation (that is, targeted emails that easily influence targets). To evaluate the possibility of such an attack, the user experiment was performed to investigate and analyze “whether Cialdini’s principles work in targeted emails” and “whether the effectiveness of Cialdini’s principles in targeted emails correlates to personality factors.” The research questions addressed in this paper are defined as follows. RQ1: Can Cialdini’s principles be applied in writing the text of a targeted email to easily convince a recipient to open the targeted email? RQ2: Would a Cialdini’s principle that is effective on an individual show different trends based on personality factors in targeted emails?

4 User Experiment To answer the RQ1 and RQ2, this section describes the user experiment conducted to investigate the correlation between the response rate (i.e., the ratio at which targets followed instructions written in the targeted emails) to targeted emails that utilized each of Cialdini’s principles and personality factors. OSINT tools and Personality Insights enable a smooth estimation of personality factors. However, to ensure participation of a large number of subjects and because of concerns over handling personal information on social media, subjects filled out personality survey questionnaires instead. 4.1

User Experiment Overview

In our user experiment, subjects took personality surveys and afterwards, received pseudo targeted emails that did not employ Cialdini’s principles and pseudo targeted emails that included phrases corresponding to each of Cialdini’s principles. Subjects then indicated on a scale from one to five how likely they were to obey the instructions within each email. Finally, the effectiveness of Cialdini’s principles in the targeted email and the correlation between Cialdini’s principles and personality factors were both statistically analyzed.

Analysis of the Relationship Between Psychological Manipulation Techniques

4.2

345

User Experiment Process

The LimeSurvey [12] web questionnaire system was used along with the Lancers [13] crowdsourcing platform to recruit test subjects, and the user experiment was then performed according to the following procedure: i. Last name input Subjects input their own last names but their surnames were only used for embedding in the address of the email text used for the experiment. ii. Attribute information survey Subjects entered information such as their age range, occupation, duties, and type of IT usage. iii. Personality survey via a personality test (Big Five) The personality test proposed by Namikawa et al., [14] was performed by subjects. iv. Questionnaire on the effects of the Cialdini’s principles using a pseudo targeted email Subjects selected the degree of obedience to instructions included in the email. The subjects responded to the questionnaire after the preliminary explanation about information that we were going to collect was provided and their consent was received. The explanation also included the following: Because the experiment was designed for the company employees, non-company employees are not supposed to take part. However, the Lancers system does not have a specification to accurately identify the occupation of participants. Therefore, it should be noted that the results of this user experiment may include responses from non-company employee participants. 4.3

Questionnaire for Personality Test

Japanese language versions of the Big Five questions have been preciously analyzed by Wada [15]. Based on the study [15] and applying methods that take into consideration the burden on users, a shortened version of a personality survey consisting of 29 questions was administered by Namikawa et al. [14]. In our user experiment, the Big Five scores were measured for each user using the shortened version of questions. Subjects were asked to respond to each question on a scale from one to five: “1: Disagree,” “2: Disagree a little,” “3: Neither agree nor disagree,” “4: Agree a little,” and “5: Agree.” Then the score for each of the Big Five was calculated based on the responses obtained. 4.4

Questionnaire on the Effects of Cialdini’s Principles

4.4.1 Procedures for Creating Pseudo Targeted Emails The following procedures were used to create a pseudo targeted email data set that includes phrases corresponding with each Cialdini’s principle. i. The pieces of text from 21 types of different emails (called “original email”), which were shown to the public at the Japan Cybercrime Control Center (JC3)

346

ii. iii.

iv.

v.

vi. vii.

4.4.2

K. Uehara et al.

[16] and had been used in the actual targeted email attacks in Japan, were randomly selected. Then three sets of original email data set consisting of seven kinds of original emails were created. One set (seven kinds of emails) was selected from the three original email data sets. Phrases incorporating Cialdini’s principles were deleted in the original emails to create “plain email” data sets. Plain emails are pseudo targeted emails which did not incorporate any of Cialdini’s principles. Seven kinds of plain emails were created. By embedding phrases corresponding to each Cialdini’s principle in the email text, six different types of pseudo targeted emails were created for each plain email (called “Cialdini email”). This resulted in targeted email data sets consisting of 49 pseudo targeted emails including plain emails. The text of the pseudo targeted emails was reviewed by the experimenters (the four authors). If even a single experimenter felt that the email seemed slightly unclear, the matter was collectively deliberated and the text was corrected. The phrases were appropriately corrected based on Cialdini’s principles to comprise of natural-sounding Japanese on the premise that the meaning of the original emails would not be changed. Another set was then selected from the remaining original email data sets and steps iii, iv, and v, were executed. The procedure was complete when targeted email data sets were created from all the three original email data sets. Experiment Procedures

i. One set was selected from the three targeted email data sets. ii. Seven individual emails were chosen randomly including plain emails and emails incorporating Cialdini’s principles, so that the content was not duplicated. Duplicated content was removed owing to the fact that the order effect (the first time a text is viewed influences the second response) occurs if the same type of text (content) appears two or more times. iii. The seven emails were then placed in a random order on a single webpage so that they could be scrolled. Test subjects then responded with a five-grade evaluation for each email regarding the “Degree of obedience to instructions provided in an email (response score)” in the following manner: “1: Definitely wouldn’t obey,” “2: Wouldn’t obey,” “3: Can’t say either way,” “4: Would obey,” “5: Definitely would obey.” The subjects provided responses to all seven emails before proceeding further by clicking the “Next” button. iv. Another set was then selected from the remaining targeted email data sets and steps ii and iii were executed. v. The experiment was concluded once responses were submitted for all three targeted email data sets.

Analysis of the Relationship Between Psychological Manipulation Techniques

347

4.4.3 Test Subjects For this user experiment, one-hundred test subjects were recruited using Lancers [13], which is a crowdsourcing service. The average response time among test subjects was 7 min 40 s, and the shortest response time was 2 min 52 s, while the longest response time was at 22 min 34 s. A quartile response time was adopted to remove outliers corresponding to overly large and small values from the distribution of test subject response times. The values that were 1.5 times smaller than the first interquartile range, and the values that were 1.5 times larger than the third quartile range were treated as outliers. When the outliers were removed, the results showed 96 effective responses. It should be noted that as the user experiment was administered using a web questionnaire, it was not possible for the test administrators to adequately control the test subjects.

5 Experimental Results and Analysis To investigate the internal consistency of the scale used in this experiment, Cronbach’s alpha for the subscales within the data set was calculated (Table 1). Each resulted in a value of at least 0.7, showing some degree of validity. Then, the results of the user experiments RQ1 and RQ2 were analyzed. 5.1

Analysis 1: Analysis for RQ1

To answer RQ1 (“Can Cialdini’s principles be applied in writing the text of a targeted email to easily convince a recipient to open the targeted email?”), analysis was performed to investigate if using Cialdini’s principles to create targeted emails would result in a higher tendency of recipients opening such targeted emails, similar to that of phishing emails. A one-sided test was administered using the following null hypothesis and alternative hypothesis to analyze if there was a significant difference between the response scores for the plain emails and the Cialdini emails. The significance level was set to 5%. Null hypothesis: Response scores for targeted emails will not change if Cialdini’s Principles are applied. Alternative hypothesis: Response scores for targeted emails will increase if Cialdini’s Principles are applied. As each subject was required to provide their response scores for both the plain emails and Cialdini emails (for each principle), the two response scores are comparable (that is, data is obtained from corresponding samples). Therefore, a Wilcoxon signedrank test was implemented for verification. Table 2 illustrates the results of the test. A significant difference could not be confirmed for the principles of consistency and social proof, and hence, the null hypothesis cannot be rejected. A significant difference (p < 0.01, p < 0.05) could be confirmed for the principles of scarcity, reciprocation, authority, and liking. Therefore, the null hypothesis could be rejected, and the alternative hypothesis was adopted. This reveals that some of Cialdini’s principles are effective at convincing recipients to open targeted emails, and that RQ1 can be partially established.

348

K. Uehara et al. Table 1. Reliability scores Scale Big Five

Subscale Extroversion (Ext) Agreeableness (Agr) Conscientiousness (Conc) Neuroticism (Neu) Openness (Ope) Cialdini’s principles Reciprocation (Rec) Scarcity (Sca) Authority (Aut) Social Proof (SP) Liking (Lik) Consistency (Cons)

Cronbach’s alpha Number of items 0.850 5 0.807 6 0.771 7 0.784 5 0.811 6 0.719 3 0.811 3 0.832 3 0.829 3 0.800 3 0.826 3

Table 2. Test results of response scores between plain email and Cialdini email p-value Reciprocation 0.0411* Scarcity 0.0306* Authority 0.0038** Social proof 0.444 Liking 0.0014** Consistency 0.37 † p < 0.10, *p < 0.05, **p < 0.01

Table 3. Correlation between Big Five and Cialdini’s principles Rec Sca Aut SP Lik Ext 0.024 0.073 0.108 0.015 0.011 Agr −0.167 −0.143 −0.17 −0.138 −0.035 Conc −0.169 −0.103 −0.058 −0.143 −0.057 Neu 0.068 0.017 −0.005 0.072 0.053 Ope 0.01 0.046 0.096 0.005 0.03

5.2

Cons −0.079 −0.122 0.013 0.057 −0.029

Analysis 2: Analysis for RQ2

To answer RQ2 (“Would a Cialdini’s principle that is effective for a single individual show different trends based on personality factors in targeted emails?”), the correlation between the scores of each Big Five factor and the response scores for individual Cialdini’s principles was considered, and analysis was performed whether any of Cialdini’s principles notably influenced subjects with respect to each personality factor. Table 3 shows the presence of correlation values between the Big Five and Cialdini’s principles. The Spearman’s rank correlation coefficient between each pair of

Analysis of the Relationship Between Psychological Manipulation Techniques

349

Big Five factors measured using the questionnaires, and the response scores for Cialdini’s principles were determined for the analysis. The analysis was unable to determine a significant correlation between Cialdini’s principles and any of the Big Five factors. Therefore, unlike the conclusions from the experimental results of the previous study by Alkış et al., this study did not succeed in confirming that the extent of influence of Cialdini’s principles differs based on the user’s personality factors.

6 Discussion of Analysis Results 6.1

Discussion on the Results of Analysis 1

In this section, the Cialdini’s principles of consistency and social proof, for which no significant difference could be observed and confirmed, are discussed. As described in the sections above, the principle of consistency is a psychological principle that states that people want their behavior to be consistent (want to become consistent). However, only a single email employing this principle was taken for analysis, although phrases suggesting that the recipient “did something similar before” were inserted when creating the targeted email, it is likely that the principle of consistency was not conveyed adequately in this experimental environment. For example, it may have been possible to make effective use of the consistency principle if the content of the targeted email suggested that the recipient “actually did this same thing before.” As noted earlier, the principle of social proof is a psychological principle stating that we want to follow what others around us are doing. Therefore, during this experiment, phrases were inserted in targeted emails asking the recipient to perform the following actions: “please open the attachment like everyone else”, or “everyone else has already checked the attachment.” However, the former could have been interpreted as follow: “please check the attachment using the same procedure/method as everyone else,” while the latter could be interpreted as “someone else has already checked it for you.” Subjects may therefore have misunderstood these emails, which meant that the principle of social proof did not work at a significant level. 6.2

Discussion on the Results Analysis 2

The research by Alkış et al. [9] showed the presence of correlation between personality factors and statistics related to the extent to which subjects are influenced in terms of all Cialdini’s principles. This section examines why the results of the current experiment varied from those of Alkış et al. Alkış et al. used the susceptibility to persuasive strategies scale (STPS) developed by Kaptein et al. to measure the extent, to which subjects were persuaded using Cialdini’s principles [17]. When using STPS, researchers prepare in advance questions related to persuasion, and then, ask the subjects if they would obey. For example, one question on reciprocation is “when a family member does me a favor, I am very inclined to return this favor.” Therefore, questions asked using STPS are not limited to targeted emails,

350

K. Uehara et al.

but also include general requests and direct questions. In contrast, we, in this paper, included phrases corresponding to each of Cialdini’s principles in the body of targeted emails and asked whether subjects would intentionally obey the text in these emails to evaluate responses within an experimental scenario that more closely resembled a targeted email. Elements such as the intrinsic suspicion people have of targeted emails, may have a direct effect on the judgment of the subjects, and hence, this may explain the differences in the obtained results compared with those of Alkış et al. However, the results of this experiment suggest that some human personality factors may affect Cialdini’s principles used in targeted emails. There is the existing research pertaining to this and illustrating that behavioral characteristics (rather than personality factors) tend to have a stronger effect on human ideas on privacy [10]. Hence, it may be possible to gain an insight on “which types of people would be more easily influenced by psychological manipulation using Cialdini’s principles” in targeted email rather than analyzing both personality factors and behavioral characteristics.

7 Conclusion This study considered the hypothesis that “an attacker may use OSINT tools to obtain social media information on the target and then, to misuse personality estimation tools and psychological manipulation techniques seeking to create targeted emails aimed for each target with a highly effective level of psychological manipulation based on the existing studies on correlation between personality estimation tools (which estimate an individual’s personality from social media information), and psychological manipulation techniques (Cialdini’s principles). To test this hypothesis, an experiment, which confirmed the effectiveness of Cialdini’s principles in composing targeted emails, was performed involving 100 users. However, the results of this experiment were unable to demonstrate the presence of correlation between human personality factors and the effect of using Cialdini’s principles in targeted emails. In future research, analysis performed from the perspective of both personality characteristics and behavioral characteristics should be considered. Further research will also be required to assess if the same effect is obtained for English language emails, as the experiment in the present study was performed using the emails written in Japanese.

References 1. Acquisti, A., Gross, R., Stutzman, F.: Face recognition and privacy in the age of augmented reality. J. Priv. Confid. 6(2), 1–20 (2014) 2. Rainie, L., Kiesler, S., Kang, R., Madden, M., Duggan, M., Brown, S., Dabbish, L.: Anonymity, privacy, and security online. Pew Research Center (2013) 3. Ball, L.D., Ewan, G., Coull, N.J.: Undermining-social engineering using open source intelligence gathering. In: Proceedings of 4th International Conference on Knowledge Discovery and Information Retrieval (KDIR), pp. 275–280. SciTePress-Science and Technology Publications (2012)

Analysis of the Relationship Between Psychological Manipulation Techniques

351

4. Uehara, K., Mukaiyama, K., Fujita, M., Nishikawa, H., Yamamoto, T., Kawauchi, K., Nishigaki, M.: Basic study on targeted E-mail attack method using OSINT. In: International Conference on Advanced Information Networking and Applications, pp. 1329–1341 (2019) 5. Goldberg, L.R.: An alternative “description of personality”: the big-five factor structure. J. Person. Soc. Psychol. 59(6), 1216–1229 (1990) 6. IBM: Personality Insights. https://www.ibm.com/watson/services/personality-insights/. Accessed 25 Mar 2019 7. Cialdini, R.B.: Influence, vol. 3. A. Michel, Port Harcourt (1987) 8. Wright, R.T., Jensen, M.L., Thatcher, J.B., Dinger, M., Marett, K.: Research note—influence techniques in phishing attacks: an examination of vulnerability and resistance. Inf. Syst. Res. 25(2), 385–400 (2014) 9. Alkış, N., Temizel, T.T.: The impact of individual differences on influence strategies. Pers. Individ. Differ. 87, 147–152 (2015) 10. Egelman, S., Peer, E.: The myth of the average user: improving privacy and security systems through individualization. In: Proceedings of the 2015 New Security Paradigms Workshop, pp. 16–28 (2015) 11. Akbar, N.: Analysing persuasion principles in phishing emails. Master’s thesis, University of Twente (2014) 12. LimeSurvey. https://www.limesurvey.org/. Accessed 24 Mar 2019 13. Lancers. https://www.lancers.jp/. Accessed 24 Mar 2019, (In Japanese) 14. Namikawa, T., Tani, I., Wakita, T., Kumagai, R., Nakane, A., Noguchi, H.: Development of a short form of the Japanese Big-Five Scale, and a test of its reliability and validity. Jpn. J. Psychol. 83(2), 91–99 (2012). (In Japanese) 15. Wada, S.: Construction of the Big Five Scales of personality trait terms and concurrent validity with NPI. Jpn. J. Psychol. 67(1), 61–67 (1996). (In Japanese) 16. Japan Cybercrime Control Center (JC3). https://www.jc3.or.jp/. Accessed 20 Mar 2019, (In Japanese) 17. Kaptein, M.: Personalized persuasion in ambient intelligence. J. Ambient Intell. Smart Environ. 4(3), 279–280 (2012)

Gait-Based Authentication Using Anomaly Detection with Acceleration of Two Devices in Smart Lock Kazuki Watanabe1 , Makoto Nagatomo1 , Kentaro Aburada2 , Naonobu Okazaki2 , and Mirang Park1(B) 1

2

Kanagawa Institute of Technology, 1030 Shimo-Ogino, Atsugi, Kanagawa 243-0292, Japan [email protected] University of Miyazaki, 1-1 Gakuen-Kibanadai-Nishi, Miyazaki 889-2192, Japan

Abstract. Currently, authentication in Smart locks is performed by fingerprint or face authentication. However, these authentications are inconvenient for smart locks because they require the user to stop for several seconds in front of the door and remove certain accessories (e.g., gloves, sunglasses). In this paper, we propose a user authentication method based on gait features. We propose a system model of gait-based authentication method using accelerometers in a smartphone and a wearable device (i.e., smartwatch), that is robust for unknown data using anomaly detection by machine learning. In addition, we conduct experiment to confirm the authentication rate of the proposed gait-based authentication. As a result, when using Isolation Forest as the anomaly detection algorithm, the average FAR (False Acceptance Rate) was 8.3%, the average FRR (False Rejection Rate) was 9.5%. Furthermore, we found that the better algorithm of anomaly detection of FAR and FRR is different depending on the subjects.

1

Introduction

Recently, the number of Internet of Things (IoT) products is increasing and will reach about 40 billion by 2020 [1]. IoT products can communicate and interact with each device. Furthermore, wearable devices are also increasing year by year, and are expected to increase to about 453 million in 2022 [2]. Therefore, it is predicting that many people will wear wearable devices. Various smart lock products, which can be electronically locked and unlocked, have been released, including Qrio [3], August [4], Kwikset [5], those smart locks using some authentication (e.g., PIN, password, fingerprint, face authentication). However, authentication for smart locks has several problems. For instance, an attacker can guess passwords, and the user needs to remember the password, which is memory burden for the user. In biometric authentication, which uses human physical and behavioral features, has been developed. Fingerprint authentication requires the user to stop for several seconds in front of the door and cannot be used with gloves. c Springer Nature Switzerland AG 2020  L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 352–362, 2020. https://doi.org/10.1007/978-3-030-33506-9_31

Gait-Based Authentication Using Anomaly Detection in Smart Lock

353

Face authentication requires the user to stop for several seconds in front of the door and cannot be used with sunglasses. Therefore, we focus on an authentication method based on gait features using acceleration from a smartphone and a wearable device (e.g., smartwatch). This method authentication can start at a place far from the smart lock and ends at a place near the smart lock. Hence, authentication is performed automatically. In addition, the memory burden is low for the user. However, authentication methods that use behavioral features have the possibility of spoofing attacks. For this problem, Muaaz et al. [6] have confirmed whether a gait-based authentication method based on acceleration data acquired from a smartphone is effective against spoofing attacks. In the experiment, an acting student spoofed the registrant’s walking and confirmed whether the authentication was successful. As a result, in most cases, the spoofing does not succeed, showing that gait-based authentication is effective against spoofing attacks. Previous studies have proposed gait-based authentication using smartphone accelerometers [7–9]. These methods require the smartphone to be fixed at in a specific inclination during authentication and have low identification accuracy. Mondal et al. proposed a gait authentication method based on many wearable sensors [10]. Although identification accuracy was higher than that using a single sensor, this method requires the user to wear many devices. This is inconvenient. In this paper, we propose a system model of gait-based authentication method that uses accelerometers in a smartphone and a wearable device (i.e., smartwatch), which is robust for unknown data using anomaly detection by machine learning. We calculate composite acceleration because authentication without being conscious of devices inclination. Furthermore, we extract features from the composite acceleration data and calculate accuracy for various machinelearning algorithms. As a result, when using Isolation Forest as the anomaly detection algorithm, the average FAR (False Acceptance Rate) was 8.3%, the average FRR (False Rejection Rate) was 9.5%. Therefore, we confirm FAR and FRR of each subject, and found that the anomaly detection algorithm in which FAR and FRR become different depending on the subject. The rest of this paper is organized as follows. Section 2 describes related work on gait-based authentication. Section 3 presents the proposed system model for gait-based authentication based on two devices and shows the processing of acceleration data used for authentication and feature extraction. Section 4 describes the implementation and experiments. Section 5 shows the accuracy of the proposed method and the features that contribute to identification accuracy. Section 6 gives the conclusions and future work.

2 2.1

Related Work Gait Authentication Using Machine Learning

Hou et al. [7] proposed a gait-based authentication method that uses x-, y-, and z-axis accelerations acquired from a smartphone accelerometer. This method

354

K. Watanabe et al.

classifies users using machine-learning based on the extracted acceleration features (e.g., average value, standard deviation, mean absolute deviation, average composite acceleration, time between peaks, bottle distribution). The FAR 0.6% and the FRR was 8.7% when using Decision Tree J48 as the classification algorithm; they were 0.3% and 3.8%, respectively, when using a neural network. The average values of x-, y-, and z-axis accelerations, respectively, contributed to identification accuracy, whereas the average combined acceleration and the time between peaks did not. Furthermore, the accuracy was calculated using 56 classification algorithms in Weka, a collection of machine-learning algorithms for data mining tasks. It was found that the accuracy greatly depends on the classification algorithm. However, this method does not consider the inclination of the device. Konno et al. [8] proposed a gait-based authentication method that uses two sensors, namely the accelerometer and gyroscope in a smartphone. This method calculates the distance between the sensor data and reference data. The user is identified by a classifier based on the distance. However, it is necessary to authenticate in the same inclination of the device as at registration because uses accelerometers and gyroscope. Iwamoto et al. [9] estimated gait state (stationary, walking, running) and smart device position (front pocket, back pocket, chest pocket, watch device screen, shake arm) and user. Experiments were conducted on five test subjects. Accuracies of 99.7%, 99.4%, and 97.0% were obtained for gait estimation, device position estimation, and user identification, respectively. From the result of the smart device position estimation, it was found that wearing sensors at multiple position on the body improved accuracy. 2.2

Gait Authentication with Multiple Wearable Sensors

Mondal et al. [10] attached eight rotation sensors on subjects’ bodies (right shoulder, left shoulder, right arm, left arm, right hip, left hip, right foot, left foot) and asked the subjects to walk. Using machine-learning, this method was able to discriminate users with an accuracy of 100%. However, since many sensors must be worn, this method is inconvenient. In the present study, we achieve high-accuracy gait-based authentication without considering the inclination of the device by placing a wearable device (smartwatch) and a smartphone at different positions on the body. This method is practical because many people have such devices.

3 3.1

Proposal Method System Model

The proposed method performs smart lock authentication using anomaly detection by machine learning based on acceleration data obtained from two devices. Figure 1 shows the system model. The proposed system model consists of two

Gait-Based Authentication Using Anomaly Detection in Smart Lock

355

Fig. 1. Proposed system model

devices (e.g., a smartphone, a wearable device equipped with an accelerometer) and a smart lock. The acceleration data obtained from two device are used to authenticate the user. The smartphone and wearable device use proximity detection based on the iBeacon protocol (based on Bluetooth). iBeacon uses three levels of proximity, namely immediate (less than about 2 cm), near (about 2 cm to about 1 m), and far (about 1 m to about 50 m). The proposed method uses the near and far levels. The authentication procedure is as follows: (1) (2) (3) (4) (5) (6) (7) (8) (9)

The two devices detect the iBeacon of smart lock at the far level. The two devices send ID each. The smart lock checks ID of the two devices. The two devices obtain acceleration data. The two devices detect the iBeacon at the near level, and stop the acceleration measurement. The two devices send the acceleration data to the smart lock. The smart lock processes acceleration data from received two acceleration data. The smart lock extracts features. The smart lock performs authentication using anomaly detection, and classify correct data and anomaly data.

The smart lock stores reference user gait data in advance and registers device ID of each device in registration phase. Authentication is performed using the reference data. In step (7), the smart lock processes the acceleration data and extracts features. The times at which acceleration data are acquired from the two devices are expected to be different due to differences in communication delay and proximity detection. Therefore, we extract features from obtained composite acceleration data, which are not influenced by acquisition time such as the mean and the maximum. In step (9), we use anomaly detection by machine learning to strengthen robustness for unknown data. For example, we use anomaly detection algorithm such as Elliptic Envelope, GMM (Gaussian Mixture Model), Isolation Forest, KDE (Kernel Density), LOF (Local Outlier Factor), One Class SVM.

356

3.2

K. Watanabe et al.

Acceleration Data Processing

We use composite acceleration which is processed from the acceleration data, it can calculate the same value even if the inclination of the device changes. We represent the composite acceleration ri1 , ri2 of the smartphone and wearable device as follows. The obtained x-, y-, and z-axis acceleration data are represented as x1i , yi1 , zi1 for the smartphone and x2i , yi2 , zi2 for the wearable device.  ri1 = (x1i )2 + (yi1 )2 + (zi1 )2 (1)  (2) ri2 = (x2i )2 + (yi2 )2 + (zi2 )2 In addition, d1 and d2 are shown below for the set of composite acceleration data acquired from the two devices. The times at which the i-th data point was acquired (with respect to the start of measurement) are t1i and t2i for the smartphone and wearable device, respectively. n1 and n2 are the numbers of data points acquired during measurement.    (3) d1 = (t1i , ri1 )|i ∈ 1, ..., n1     d2 = (t2i , ri2 )|i ∈ 1, ..., n2 (4) Since the numbers of acceleration data points for the two devices are different due to differences in performance and measurement time, the number of data points are adjusted. For example, when n1 > n2 , the set d1 , d2 of data when combining the sample numbers are as shown below.    1 (5) d1 = (t1 , r )|i ∈ 1, ..., n 2 i i     2 d2 = (t2 (6) i , ri )|i ∈ 1, ..., n2 1 Here, t1 i , ri are calculated as follows: 1 t1 i = t arg

min (|t1j −t2i |)

(7)

ri1 = r1 arg

min (|t1j −t2i |)

(8)

j∈{1,...,n1 }

j∈{1,...,n1 }

Since the variation of acceleration data is small in the stationary state and large in the walking state, a section of the acceleration data with a high standard deviation is extracted as the walking state. Noise removal is then applied to the acceleration data using a low-pass filter. Figure 2 shows the waveforms obtained after processing the acceleration data from the two devices (smartphone in the right pocket and smartwatch worn on the left wrist).

Gait-Based Authentication Using Anomaly Detection in Smart Lock

357

Fig. 2. Composite acceleration after acceleration processing

3.3

Feature Extraction

In order to perform gait-based authentication using anomaly detection by machine learning, our method extracts features based on acceleration data obtained from two devices. We extract the difference between the features of the two devices that cannot acquire by one device. The features are shown below. • • • • • • • • • • •

Mean (Device1, Device2) Difference of mean STD: standard deviation (Device1, Device2) Difference of STD Maximum (Device1, Device2) Difference of maximum Minimum (Device1, Device2) Difference of minimum Local maximum interval (Device1, Device2) Local minimum interval (Device1, Device2) SSD: Sum of Squared Difference SSD =

n−k 1  1 2 (ri − ri+k )2 k∈{1,..., 3 n} n − k i=1

min 2

• NCC: Normalized Cross Correlation N CC =

max

k∈{1,..., 23 n}



 n−k i=1

n−k 1 2 i=1 ri ri+k

(ri1 )2



n−k 2 2 i=1 (ri+k )

(9)

(10)

358

K. Watanabe et al.

• DTW: Dynamic Time Warping DT W = f (n, n)

⎧ ⎨ f (i − 1 , j − 1 ) f (i, j) = |ri1 − rj2 | + min f (i − 1 , j ) ⎩ f (i , j − 1 )

(11)

The number of samples is represented as n, and i, j ∈ {1, ..., n}. In addition, the SSD and NCC use the highest similarity as the features when shifting two acceleration data in the time axis as the features, in order to use the features whether the two devices move similarly. Since the number of samples for comparing the degree of similarity decreases if you shift the number of samples, we shift the two acceleration data by k ∈ {1, ..., 23 n}. Based on the above features, perform authentication using anomaly detection. In addition, we confirm FAR and FRR by selecting features to improve authentication accuracy.

4

Implementation and Experiments

To acquire acceleration data from a device, we developed a Java program. The smartphone was a Sony Xperia XZs and the wearable device was a Sony SmartWatch 3. We use server instead of a smart lock, and developed program in Python to receive acceleration data and classify registered user or Non-registered user. Bluetooth was used to transmit and receive data. Acceleration measurement is conducted by transmitting an acceleration measurement command and

Fig. 3. Experiment conditions

Gait-Based Authentication Using Anomaly Detection in Smart Lock

359

a measurement end command to each device. For machine learning, we used the Python module scikit-learn [11]. We calculate FAR and FRR of Elliptic Envelope, GMM, Isolation Forest, KDE, LOF, One Class SVM. We conducted experiments to confirm the effectiveness of the proposed gaitbased authentication method. As shown in Fig. 3, subjects put the smartphone in their right pocket and wore the smartwatch on their left wrist. Each subject walked in a corridor for about 10 m. The 11 subjects conducted the experiment 50 times, of which 30 times is used as training data, and 20 times as test data. Also, the 10 other subjects walk normally 20 times as anomaly data. The subjects were students from Kanagawa Institute of Technology. The experiment procedure was as follows: (1) (2) (3) (4) (5)

The The The The The

two devices started acceleration measurement. subject stopped for 5 s. subject walked 10 m in the corridor. subject stopped for 5 s. two devices ended acceleration measurement.

Steps (2) and (4) were used to facilitate the extraction of the walking state in step (3). The user was identified using the acceleration data. We evaluated the identification accuracy.

5

Evaluation

Based on the acceleration data obtained in the experiment, we calculate FAR and FRR using anomaly detection of machine learning, such as Elliptic Envelope, GMM, Isolation Forest, KDE, LOF, and One Class SVM. We create each classifier based on 11 subjects 30 times as training data, and confirm FAR and FRR by distinguishing the correct data of 11 subjects and the anomaly data of 10 subjects. In order to improve the authentication rate by features selection, features divide into 13 groups (mean, difference of mean, STD, difference of STD, maximum, difference of maximum, minimum, difference of minimum, local maximum interval, local minimum interval, SSD, NCC, DTW) and calculate FAR and FRR for all combinations. For example, the mean group is consisted of the mean of device 1 and the mean of device 2. Table 1 shows the combination of features with the highest authentication accuracy for each machine learning algorithm, and the average FAR and average FRR of all the subjects. Here, the machine learning parameters are adjusted so that the average FAR and the average FRR became smaller. As a result, it was confirmed that the optimum combination of features differs for each anomaly detection algorithm, and that there is a difference in authentication rate. Furthermore, FAR and FRR decreased when used GMM and Isolation Forest, FAR is 13.6% and FRR is 3.6% when used GMM, FAR is 8.3% and FRR is 9.5% when using Isolation Forest. In addition, in the proposed method, it is expected that the authentication rate is different individually due to the way users walk. Therefore, we confirmed FAR and FRR for

360

K. Watanabe et al. Table 1. FAR and FRR of the proposed system using anomaly detection Machine learning Features

FAR

Elliptic Envelope Difference of STD, maximum, difference of 7.8% maximum, difference of minimum, DTW, local maximum interval

FRR 12.2%

GMM

Mean, difference of mean, STD, difference 13.6% 3.6% of STD, maximum, minimum, difference of minimum, SSD, NCC, local maximum interval

Isolation Forest

Mean, STD, difference of STD, maximum, 8.3% minimum, difference of minimum, SSD, NCC, DTW, local minimum interval

KDE

Mean, difference of mean, STD, difference 15.2% 17.7% of STD, minimum, difference of minimum, NCC, DTW, local maximum interval

LOF

Difference of STD, maximum, difference of 12.4% 13.1% maximum, minimum, difference of minimum, local maximum interval, local minimum interval

One Class SVM

Mean, difference of mean, STD, maximum, 6.4% difference of maximum, minimum, difference of minimum, SSD, NCC, local maximum interval

Table 2. FAR and FRR of each subjects Machine learning GMM Isolation Forest FAR FRR FAR FRR Subject A 6.0%

5.0%

4.0%

10.0%

Subject B 42.5% 0.0%

16.5% 0.0%

Subject C 28.5% 0.0%

6.5%

Subject D 0.0%

10.0% 0.0%

25.0% 5.0%

Subject E 21.5% 0.0%

3.5%

5.0%

Subject F 2.5%

0.0%

7.5%

5.0%

Subject G 0.0%

5.0%

0.0%

0.0%

Subject H 0.5%

0.0%

13.0% 10.0%

Subject I

37.5% 15.0% 23.0% 20.0%

Subject J 8.5%

5.0%

9.5%

25.0%

Subject K 3.0%

0.0%

8.5%

0.0%

9.5%

17.2%

Gait-Based Authentication Using Anomaly Detection in Smart Lock

361

each subjects of GMM and Isolation Forest. Table 2 shows FAR and FRR for each subjects. Here, it was confirmed that authentication rate was different by the subjects even using the same method and classifier having highest accuracy are different even if the same subjects. From the above, we found that authentication rate to improve by classifying anomalies based on the results of several anomaly detection algorithms.

6

Conclusion

In this paper, we proposed a system model of gait-based authentication using anomaly detection based on two devices (a smartphone and a wearable device) acceleration. In the proposed method, we acquire acceleration data from two devices worn on the body, and extract features using their composed acceleration. As an experiment, we acquire acceleration data of 11 subjects as learning data and correct data, and acquire acceleration data of another 10 subjects as anomaly data. We created each classifier from the acquired learning data using the machine learning anomaly detection algorithm, and confirmed FAR and FRR by making classify on correct data and anomaly data. In addition, in order to improve the authentication rate, we selected features by dividing the features into groups and calculating the average FAR and the average FRR for the combinations of the groups. As a result, FAR is 13.6% and FRR is 3.6% when used GMM as the anomaly detection algorithm, and FAR is 8.3% and FRR is 9.5% when using Isolation Forest. We also confirmed the FAR and FRR for each subjects in the GMM and Isolation Forest, and confirmed the anomaly detection algorithm are lowering the FAR and FRR differs for each subjects. From this, it expected that authentication rate is improved by classifying anomalies based on the results of each anomaly detection algorithms. In future work, we plan to extract features of 3-axes each from acceleration data obtained wearable devices that do not need to consider the orientation of devices such as smartwatches, and to confirm whether authentication rate is improved. In addition, we will evaluate attack to our proposed method. Acknowledgements. This work was supported by JSPS KAKENHI Grant Numbers JP17H01736, JP17K00139.

References 1. Ministry of Internal Affairs and Communications: 2018 White Paper on Information and Communication in Japan (2018). http://www.soumu.go.jp/ johotsusintokei/whitepaper/ja/h29/pdf/n3300000.pdf 2. Gartner: Gartner Says Worldwide Wearable Device Sales to Grow 26 Percent in 2019. https://www.gartner.com/en/newsroom/press-releases/2018-11-29gartner-says-worldwide-wearable-device-sales-to-grow-. Accessed 30 July 2019 3. Qrio: Qrio Smart Lock. https://qrio.me/smartlock. Accessed 30 July 2019 4. August: August Smart Lock. https://august.com. Accessed 30 July 2019

362

K. Watanabe et al.

5. Kwikset: Door Locks Door Hardware Smart Locks & Smart key Technology. https://www.kwikset.com. Accessed 30 July 2019 6. Muaaz, M., Mayrhofer, R.: Smartphone-based gait recognition: from authentication to imitation. IEEE Trans. Mob. Comput. 16(11), 3209–3221 (2017) 7. Hou, R., Watanabe, Y.: A Study on authentication at the time of the walk of using the acceleration sensor of smartphone. In: Computer Security Symposium, vol. 2013, pp. 21–23 (2013). (in Japanese) 8. Konno, S., Nakamura, Y., Shiraishi, Y., Takahashi, O.: Improvement of gait-based authentication by using multiple wearable sensors. IPSJ J. 57(1), 109–122 (2016). (in Japanese) 9. Iwamoto, T., Sugimori, D., Matsumoto, M.: A Study of identification of pedestrian by using 3-axis accelerometer. IPSJ J. 55(2), 734–749 (2014) 10. Mondal, S., Nandy, A., Chakraborty, P., et al.: Gait based personal identification system using rotation sensor. J. Emerg. Trends Comput. Inf. Sci. 3(3), 395–402 (2012) 11. Scikit-learn: scikit-learn machine learning in Python Scikit-learn 0.19.1 documentation. http://scikit-learn.org/stable/index.html. Accessed 30 July 2019

Accurate Online Energy Consumption Estimation of IoT Devices Using Energest Adnan Sabovic1(B) , Carmen Delgado1 , Jan Bauwens2 , Eli De Poorter2 , and Jeroen Famaey1 1

IDLab, University of Antwerp – imec, Antwerp, Belgium {adnan.sabovic,carmen.delgado,jeroen.famaey}@uantwerpen.be 2 IDLab, Ghent University – imec, Ghent, Belgium {jan.bauwens2,eli.depoorter}@ugent.be

Abstract. Minimizing the energy consumption of Internet of Things (IoT) devices is one of the biggest challenges and crucial issues for the future of a sustainable IoT vision. In order to estimate the remaining device lifetime and optimize its energy consumption, it is necessary to have an accurate online view on the consumed energy with minimal overhead. This is non-trivial, as many factors influence energy consumption, therefore requiring a generic measurement methodology. For example, the Medium Access Control (MAC) protocols have a very important influence on the energy consumption. This paper presents an accurate method for estimating the energy consumption of IoT devices using Energest. Our method combines a device-specific offline profiling phase, with a device and protocol-agnostic online energy estimation methodology. Energy measurements have been performed for different scenarios, using measured values and values from the datasheet, for Carrier Sense Multiple Access (CSMA) and Time Slotted Channel Hopping (TSCH) protocols. Results show that the accuracy of our method is very high, more than 96% for CSMA and more than 82% for TSCH, with very small overhead of 0.11%. Keywords: IoT CSMA · TSCH

1

· Energest · Energy estimation · Zolertia RE-Mote ·

Introduction

Internet of Things (IoT) is a paradigm used to connect objects to the Internet, where billions of devices cooperate and communicate with each other with the aim to simplify and improve daily life. To achieve the full potential of the IoT, it is necessary to develop new systems and technologies that are able to minimize energy consumption, as most of these objects are battery powered. Low-Power Wireless Sensor Networks (LPWSNs) are presented as one of the best possible solutions that combine low power consumption with long-range communications. Since IoT devices are usually equipped with small batteries, that are expensive and short-lived, the reduction of their energy consumption is one of the c Springer Nature Switzerland AG 2020  L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 363–373, 2020. https://doi.org/10.1007/978-3-030-33506-9_32

364

A. Sabovic et al.

biggest challenges and crucial issues for the future of a sustainable IoT vision. Energy efficiency has become one of the key criteria for designing Wireless Sensor Networks (WSNs), first because of the ecological aspect, but also to ensure the functionality of the sensor nodes for a long period without recharging or replacing batteries. The network’s lifetime directly depends on the energy consumption of sensor nodes and minimizing and modeling of energy consumption are the main objectives for designing energy-efficient WSNs [4]. Several mechanisms to measure and estimate energy consumption online have been proposed that are based on the accumulated time values of MCU and radio usage. For example, Energest is a software-based energy estimation mechanism implemented as a Contiki NG1 module, that provides functions to measure the accumulated time the sensor nodes spend in different MCU (active, low power, etc.) and radio (TX, RX, LISTEN, OFF, etc.) states [9]. In order to optimize the lifetime of IoT devices, it is necessary to get an accurate view on their consumed energy while they are online. This would allow battery lifetime prediction, energy-aware operations, etc. Traditionally, Energest, and other similar frameworks, rely on datasheet energy consumption values to transform timing information into energy consumption estimates. We show this leads to inaccurate results, and instead propose a generic device profiling methodology. In this paper, we propose an accurate method for estimating the energy consumption of IoT devices using Energest. Through our experiments, we show the accuracy of our method, where we combine Energest with real power consumption values from a power analyzer. We also evaluate the overhead of using Energest in terms of energy consumption.

2

Related Work

WSNs are composed of a large number of interconnected sensor nodes which usually use batteries as their main source of energy. Therefore, the design and development of low-power and energy-efficient WSNs has become a major challenge for the new wave of IoT technologies. Power consumption of a sensor node depends on many factors, such as the states of the radio, microcontroller (MCU), Light-emitting diodes (LEDs) and other components [10]. Different mechanisms have been used before to estimate energy consumption on small IoT devices such as Contiki-NG’s built-in functionality, named Energest2 . It is a time-based estimation mechanism implemented as a collection of functions and macros that runs directly on the sensor nodes and measures the accumulated time the sensor node spends in different MCU and radio states [9,10]. Time-based energy estimation is very easy to implement and configure with existing applications, but the accuracy of the obtained results is not always good enough. Possible problems about the accuracy of the results are shown 1 2

https://github.com/contiki-ng/contiki-ng. https://github.com/contiki-ng/contiki-ng/wiki/Documentation:-Energest.

Accurate Online Energy Consumption Estimation of IoT Devices

365

by Steinfeld et al. [10]. For example, the CPU mode is activated whenever the node is active, including the period when the radio transmits or receives some data, but not for all the time needed for a transmission. After it makes the radio ready to transmit data, the MCU can switch to the LPM mode. It is very hard to know the accurate period of the CPU and the LPM state of the MCU during one transmission and almost impossible to estimate the energy consumption of the MCU and the radio separately. Steinfeld et al. [10] propose a hardwarebased approach, named Smart Coulomb Counter (SCC), that easily adds to a sensor node the capability of measuring its energy consumption. Also, they compare their solution with Energest and get very similar results between SCC and Energest in the OFF state, but in the ON state, the consumption values differ greatly (more than 85% of relative error) because of the LED current consumption. In contrast, we use Energest to get time values and estimate the energy consumption using actual measured current values. This greatly improves the accuracy without a need for expensive hardware. Daneels et al. [6] present an accurate energy consumption model for devices, using both the 868 MHz and 2.4 GHz frequency bands, using the TSCH protocol. They identify all network-related CPU and radio state changes, providing a precise representation of the device behavior and an accurate prediction of its energy consumption. Their approach is protocol-specific, while we propose a general methodology. There are a few works where Energest has been used as the main software for calculating power consumption. For example, Schandy et al. [9] presented a simple approach for the analysis of the average power consumption of a sensor node according to the node states and network protocols that have been used. Kharce and Pawar [8] analyzed the energy consumption in a 6LoWPAN network for real and emulated Zolertia Z1 motes using the Energestbased module Powertrace to record each of the possible states of the radio. Some authors such as Dunkels et al. [7] and Wu et al. [12], used Powertrace to estimate the total energy consumption of the system. Powertrace tracks the state of the device to estimate the power consumption for individual tasks that are captured in energy capsules. Powertrace is based on Energest and records the energy consumption of the activity by opening an energy capsule when the activity starts and closing when it ends [7]. Wu et al. [12] used values from Powertrace and the datasheet of the MCU and radio to get the energy consumption from every possible state. Ahmad et al. [5] measured the energy consumption for different states of the radio and the MCU depending on the number of nodes in the network using Powertrace with datasheet values, and performing simulations for more network scenarios. Both Energest and Powertrace can estimate the energy consumption of the radio and the MCU, by recording how long each component spends in a specific state. In contrast to these existing Energest-based methodologies, we use more accurate device profiling, rather than datasheet values. Moreover, we show that our method is agnostic to the used MAC protocol, and evaluate its overhead.

366

3

A. Sabovic et al.

Methodology

In this section, the basics of the used software, hardware, MAC protocols and how our measurements have been performed are described. 3.1

Software

3.1.1 Contiki NG Contiki-NG3 is an open-source, cross-platform operating system for the next generation of IoT devices that is focused on enabling reliable and secure lowpower communication using standard protocols, such as Internet Protocol version 6 (IPv6), Constrained Application Protocol (CoAP) or IPv6 Routing Protocol for Low-Power and Lossy Networks (RPL). This is a new version of the Contiki project that provides an RFC-compliant, low-power IPv6 communication stack and enables Internet connectivity [2]. Contiki-NG can be used for building a lot of new WSN programs and it supports many hardware platforms, such as Zolertia Zoul, Tmote Sky/TelosB, OpenMote cc2538, etc. 3.1.2 Energest Energest4 is a lightweight, software-based energy estimation mechanism for resource-constrained IoT devices that provides functions to measure the time the device is in different states. This is a Contiki-NG’s built-in functionality implemented as a collection of macros and functions. The macros are used to tell the Energest module when a component changes its mode or to return the total time the Energest module has been tracking. The functions are usually used to initialize Energest, update the total time for all types that are currently turned on or to return the total time for the specified state of the device. There are five Energest modes that all Contiki-NG platforms support: CPU, LPM, Deep LPM for the microcontroller (MCU) and TRANSMIT, LISTEN, OFF for radio. Not all MCUs support both LPM and Deep LPM and in that case, the unused type will always report 0 seconds as time. Note that RECEIVE is not considered as a separate radio state, and reported as LISTEN. 3.2

Hardware

3.2.1 Zolertia RE-Mote Platform The Zoul5 is a core module developed by Zolertia6 that provides a flexible and compatible module solution to be integrated with most existing products and solutions. The Zoul module can be used for industrial and IoT projects, to ease the development of new WSNs and applications for them, from intelligent lighting systems to monitoring applications in Smart Homes and Cities. 3 4 5 6

https://github.com/contiki-ng/contiki-ng. https://github.com/contiki-ng/contiki-ng/wiki/Documentation:-Energest. https://github.com/Zolertia/Resources/wiki/The-Zoul-module. https://zolertia.io/.

Accurate Online Energy Consumption Estimation of IoT Devices

367

The Zolertia RE-Mote7 is an ultra-low power complete hardware development platform, based on the Zoul, designed jointly by universities and industrial partners, easy and flexible to use for most of IoT applications. The platform is based on the Texas Instruments CC2538 ARM Cortex-M3 System on Chip (SoC)8 , with an on-board 2.4 GHz IEEE 802.15.4 RF interface. It supports two radios and it is compatible with existing and trending protocols, for both indoor and long-range IoT applications with maximum range between 100 m and 20 km. Contiki-NG has been successfully ported to the Zolertia RE-Mote platform. 3.2.2 N6705B DC Power Analyzer The N6705B DC Power Analyzer9 is a device that provides the possibility for sourcing and measuring voltage and current levels of different devices. The N6705B offers flexible configuration to meet power sourcing and analysis requirements. It integrates capabilities of up to four advanced power supplies with Digital Multimeter (DMM), Scope, Arb and Data Logger features. All functions and measurements are available at the front panel and there is no need for developing or debugging programs to control something on the instrument. 3.3

MAC Protocols

To show that our methodology is MAC-protocol-agnostic, we evaluated it with two commonly used MAC protocols for WSN and IoT applications as described below. CSMA Protocol CSMA is a network protocol that listens to the channel before transmitting to check if it is idle or not [3]. When the channel is detected idle for transmission, the device can send the packet. Otherwise, it will perform a random back-off before retrying. Only one device can send packets through the channel, otherwise a collision will occur and the transmission will fail. The radio is listening to the channel all the time and never goes to the OFF state, while the MCU allows the Low Power Mode (LPM) state, but with the highest level of power consumption. Figure 1 illustrates the channel states for the CSMA protocol for one transmission between two nodes, transmitter and receiver, for a random period. Different states of the radio (LISTEN, TX and RX) and MCU (CPU and LPM) are shown. 3.3.1 TSCH Protocol TSCH mode, specified in IEEE 802.15.4e [1], is an ultra-low power and highly reliable medium access control technology [11]. It aims to achieve a high reliability of 99.99% and minimal power consumption, which are very important 7 8 9

https://github.com/Zolertia/Resources/wiki/RE-Mote. http://www.ti.com/product/CC2538. https://www.keysight.com.

368

A. Sabovic et al. Time

OFF RADIO

MCU

LISTEN

CPU

LPM

CPU

Tx/Rx

LPM

LPM

Detect Activity

Channel is busy Prepare to transmit data

CPU

LPM

CPU

LPM

Idle Channel

Listen for free channel

Wait

...

LISTEN

CPU

LPM

Detect Activity

Transmit data

Prepare to transmit data

Listen for free channel

TRANSMITTER Waiting for free channel

RECEIVER

Listen for data

Listening

...

...

Transmission

Waiting for free channel

...

Receive data

Listen for data

...

Listening

...

Reception

Fig. 1. Channel state for CSMA protocol without ACK mechanism for a transmitter and receiver node

for industrial and other challenging IoT environments [6]. In TSCH networks, all nodes are synchronized following a time-synchronized schedule, divided into time slots of generally 10 or 15ms, which instructs every node exactly when to send or receive data and how to avoid wasting valuable energy. Moreover nodes hop between the available channels in a pseudo-random manner to avoid interference. The sender’s and receiver’s radio are ON only during their assigned slots. After that, both nodes can switch their radios to OFF and go to the sleep mode for few milliseconds before repeating the same procedure in order to transmit/receive an acknowledgment (ACK) [11]. A timeslot template for a transmitter and receiver node, with different MCU (CPU, LPM, and Deep LPM) and radio states (LISTEN, Tx, Rx and OFF), is shown in Fig. 2. Radio and MCU spend most of the time in OFF and Deep LPM states, as such saving a large amount of energy. 3.4

Measurement Methodology

For energy estimation, we used Energest as a timing mechanism together with energy consumption values that have been measured with the power analyzer. We have performed four types of measurements with two different MAC protocols, CSMA and TSCH. The energy consumption Etot in Joule is calculated as follows:   Es = ts × Is × V (1) Etot = s∈S

s∈S

where S is the set of MCU and radio states, ts is the time spent in the state s in seconds (as reported by Energest), Is is the current consumption of s in Ampere and V is the operating voltage in Volts. The measurement setup is shown in Fig. 3. In order to increase the accuracy of the energy estimations, we have measured the energy consumption of the

Accurate Online Energy Consumption Estimation of IoT Devices

369

Time Slot Lenght

MCU

CPU

LPM

RADIO

CPU

DEEP LPM

CPU

CPU

LPM

TRANSMIT/LISTEN/RECEIVE

OFF

Prepare to transmit data

TRANSMITTER

DEEP LPM

LPM

Transmit data

DEEP LPM CPU

OFF

DEEP LPM

CPU

LISTEN

Prepare to receive ACK

DEEP LPM

LPM

RECEIVE/TRANSMIT

Listen for ACK

Receive ACK

OFF

Sleep

AGT

TxOffset PGT

Prepare to receive data

RECEIVER

Listen for data

Receive data

Prepare to transmit ACK

Transmit ACK

Sleep

RxOffset

Start of timeslot

End of timeslot

Fig. 2. TSCH time slot template for a transmitter and receiver node RE-Mote

Zolertia

RGB LED LED1, EXT_WDT, PD5 01 LED2, UART1.CTS, PD4 02 LED3, UART1,.RTS, PD3 03 UART0.RX, PA0 04

OUTPUT +

UART0.TX, PA1 05 SHUTDOWN_ENABLE, PD1 06 RTC.SDA, I2C.SDA, PC2 07 RTC.SCL, I2C.SCL, PC3 08

KEYSIGHT

DGND 09

N6705B DC POWER ANALYZER

D+3.3 10 USD.CS, AIN7, PA7 11 SHUTDOWN_ENABLE, PDO 12 UART1.RX, PC1 13 UART1.TX, PC0 14

OUTPUT -

Cable connection DGND D+3.3 16 DGND 17 +VBAT 18

.. .. .. .. .. .. ..

Cable connection

.

.

PROGRAM & DEBUG USB

.. .. .. .. .. .. .. .. .

18 PC6, SPI1.MISO, USD.MISO 17 PC5, SPI1.MOSI, USD.MOSI 16 PC4, SPI1.SCLK, USD.SCLK 15 PA3, RTC_INT1, BUTTON.USER 14 RESET, JTAG.RESET, BUTTON.RESET 13 DGND 12 D + 3.3 11 PA5, AIN5, ADC1 10 PA4, AIN4, ADC2 09 DGND 08 D + 5,1 07 PA2, AIN2, ADC3 06 JTAG.TMS 05 JTAG.TCK 04 PB7,JTAG.TDO 03 PB6,JTAG.TDI 02 PS + EXT 01 DGND

USB 2.0 CONNECTOR USER BUTTON RESET BUTTON

Fig. 3. Energy consumption measurement setup

Zolertia RE-Mote MCU and radio states with the power analyzer. We removed the resistor to enable the direct voltage supply of 3.3 V for the chip (MCU and radio) which decreased the total energy consumption of the device. As such, our methodology consists of using measured current values for each Is rather than values obtained from datasheets.

4 4.1

Results Experimental Setup

We have created an application for two Zolertia RE-Mote devices that operate in the 2.4 GHz frequency band using the CC2538 radio. The duration of our experiments is 60 s with packet transmission interval of 5 s, unless stated otherwise.

370

A. Sabovic et al.

Two different packet sizes have been tested, 8 and 64 bytes. We performed the experiments using two different MAC protocols, CSMA and TSCH. For all the cases, we used the same operating voltage of 3.3 V. We repeated each experiment ten times and our results are averages over these iterations. As a baseline, we compare to Energest with datasheet values and use the power analyzer to measure the total device current consumption over the entire experiment duration to get the real consumed energy. 4.2

Device Profiling

Table 1 shows the differences between measured and datasheet values for different MCU and radio states. As expected, the measured current consumption is much higher for the LPM and Deep LPM states compared to the datasheet. This can be explained because there are other components that are affecting energy consumption, which are not taken into account in the datasheet values. In Sect. 4.3 we use both the datasheet and profiled values for Is in combination with Energest. Table 1. Comparison between values from the datasheet and N6705B DC power analyzer State

CC2538 datasheet Device profiling

CPU

20 mA

LPM

0.6 mA

15.35 mA 9.59 mA

Deep LPM 0.0013 mA

2.58 mA

LISTEN

24 mA

28.32 mA

Rx

27 mA

30.14 mA

Tx

34 mA

31.12 mA

Figure 4 shows the current drawn for the TX and RX states using the CSMA protocol and the current drawn for TX and RX time slots using TSCH as MAC protocol, both when using the CC2538 radio. 4.3

Accuracy of Energest

Table 2 shows the average power consumption of the Zolertia RE-Mote platform for the two different MAC protocols and for two different packet sizes. The first column (Datasheet + Energest) has been calculated using Eq. 1, using for Is the current consumption from the datasheet. For the second scenario, we combined Energest with measured values from the power analyzer to obtain more accurate values for Is . The two last experiments show the actual total energy consumed by the device, with Energest active and inactive respectively. It shows that the impact of Energest is 0.11%, which means that it doesn’t consume a lot of energy.

Accurate Online Energy Consumption Estimation of IoT Devices

(a) CSMA TX state

(b) CSMA RX state

(c) TSCH TX time slot

(d) TSCH RX time slot

371

Fig. 4. Measured current consumption for TX and RX states using CSMA and TSCH protocol Table 2. Power consumption of Zolertia RE-Mote platform Node

MAC model

Transmitter CSMA

Packet size 8B

Receiver Transmitter CSMA

64B

Receiver Transmitter TSCH

8B

Receiver Transmitter TSCH Receiver

64B

Datasheet + Real values + Power analyzer + Power analyzer + Energest Energest Energest ON Energest OFF 82.7 mW

93.9 mW

90.7 mW

90.6 mW

82.7 mW

93.9 mW

90.5 mW

90.5 mW

82.8 mW

94.1 mW

90.8 mW

90.6 mW

82.7 mW

93.9 mW

90.6 mW

90.6 mW

9.8 mW

15.5 mW

18.7 mW

18.7 mW

9.7 mW

15.4 mW

17.9 mW

17.8 mW

9.9 mW

15.5 mW

18.9 mW

18.9 mW

9.8 mW

15.4 mW

18.3 mW

18.2 mW

Table 2 shows the accuracy of using Energest with datasheet values and real measured values for different MCU and radio states. In case of CSMA, the accuracy of using Energest with datasheet values is 91% for both an 8B and 64B payload. When using real measured values with Energest instead the datasheet, the accuracy increases to more than 96%, which is a significant improvement. For TSCH, the accuracy of using Energest with datasheet values for an 8B and 64B payload is less than 53%. But, when using real measured values the accuracy increases to more than 82% for both payloads.

372

A. Sabovic et al.

The device consumes more energy when using the CSMA protocol compared to TSCH, with differences between the protocols of more than 70mW. Using CSMA, the device never goes to the Deep LPM state, but spends most of the time in the LISTEN state using a lot of energy. In contrast to CSMA, when using TSCH, the MCU spends most of the time in the Deep LPM state, so it saves a large amount of energy. Also, the radio turns off when the device is not in a transmission or reception slot, which also significantly reduces the power consumption of the device. The two last experiments show the impact of Energest that is 0.11% which means that it doesn’t consume a lot of energy.

(a) CSMA measurements

(b) TSCH measurements

Fig. 5. Power consumption for different packet transmission intervals for 64 bytes

We measured the power consumption for different packet transmission intervals i.e. 0.2, 0.5, 1 and 5 seconds, for both used protocols, as shown in Fig. 5. Different packet transmission intervals also have an impact on energy consumption, as they affect the number of transmissions. The results also show that when using measured current values the accuracy of Energest with actual measured values is not affected by the transmission interval. However, when using datasheet values it is, and accuracy decreases as the transmission interval grows.

5

Conclusions

In this paper, an accurate method for energy estimation of Zolertia RE-Mote devices using Energest has been presented. We have analyzed four different scenarios for different MAC protocols (CSMA and TSCH) and different packet sizes (8 and 64 bytes). We used current consumption values from the datasheet, but also measured values for every possible state of the device. Our results showed that the values used from the datasheet are not accurate and the current consumption for some states of the device is much higher because of the impact of other device components. We have shown that the power consumption depends on the used MAC protocol, packet size, device configuration and packet transmission interval. Our results show that using accurate state current measurements,

Accurate Online Energy Consumption Estimation of IoT Devices

373

Energest has an accuracy of more than 96% for CSMA, and more than 82% for TSCH. Thus is a great improvement compared to the use of datasheet values, where the accuracy is 91% for CSMA, and less than 53% for TSCH. Moreover, Energest has an overhead in terms of power consumption of only 0.11%, making it on highly suitable for online energy estimation in low-power IoT devices. Acknowledgments. Part of this research was funded by the Flemish FWO SBO S004017N IDEAL-IoT (Intelligent DEnse and Long range IoT networks) project.

References 1. IEEE standard for local and metropolitan area networks–part 15.4: Low-rate wireless personal area networks (LR-WPANs) amendment 1: Mac sublayer. IEEE Std 802.15.4e-2012 (Amendment to IEEE Std 802.15.4-2011), pp. 1–225 (2012). https://doi.org/10.1109/IEEESTD.2012.6185525 2. Practical Contiki-NG. Apress (2018) 3. Abdulrazzak, H.N., Hussein, A.A.: Performance analysis of CSMA/CA based on D2D communication. Int. J. Comput. Sci. Eng. 652–657. (2019) 4. Abo-Zahhad, M., Farrag, M., Ali, A.: Modeling and minimization of energy consumption in wireless sensor networks. In: 2015 IEEE International Conference on Electronics, Circuits, and Systems (ICECS), pp. 697–700 (2015). https://doi.org/ 10.1109/ICECS.2015.7440412 5. Ahmad, H., Ali, J., Waqar Shah, S., Nayab, D.: Optimizing computations in intermittently powered wireless sensor nodes. Int. J. Eng. Works 5(3), 50–55 (2018) 6. Daneels, G., Municio, E., Van de Velde, B., Ergeerts, G., Weyn, M., Latre, S., Famaey, J.: Accurate energy consumption modeling of IEEE 802.15.4e TSCH using dual-band openmote hardware. Sensors 18(2), E437 (2018) 7. Dunkels, A., Eriksson, J., Finne, N., Tsiftes, N.: Powertrace: network-level power profiling for low-power wireless networks. SICS Technical Report (2011) 8. Kharche, S., Pawar, S.: Node level energy consumption analysis in 6LoWPAN network using real and emulated zolertia z1 motes. In: 2016 IEEE International Conference on Advanced Networks and Telecommunications Systems (ANTS), pp. 1–5 (2016). https://doi.org/10.1109/ANTS.2016.7947870 9. Schandy, J., Steinfeld, L., Silveira, F.: Average power consumption breakdown of wireless sensor network nodes using IPv6 over LLNs. In: 2015 International Conference on Distributed Computing in Sensor Systems, pp. 242–247 (2015). https:// doi.org/10.1109/DCOSS.2015.37 10. Steinfeld, L., Oreggioni, J., Bouvier, D., Fernandez, C.A., Villaverde, J.: Smart coulomb counter for self-metering wireless sensor nodes consumption. J. Low Power Electron. 11(2), 236–248 (2015) 11. Vilajosana, X., Wang, Q., Chraim, F., Watteyne, T., Chang, T., Pister, K.S.J.: A realistic energy consumption model for TSCH networks. IEEE Sens. J. 14(2), 482–489 (2014). https://doi.org/10.1109/JSEN.2013.2285411 12. Wu, X., Steinfeld, R., Liu, J., Rudolph, C.: An implementation of access-control protocol for IoT home scenario. In: 2017 IEEE/ACIS 16th International Conference on Computer and Information Science (ICIS), pp. 31–37 (2017). https://doi.org/ 10.1109/ICIS.2017.7959965

Comparison of LoRa Simulation Environments Christos Bouras1,2(&), Apostolos Gkamas3, Spyridon Aniceto Katsampiris Salgado2, and Vasileios Kokkinos2 Computer Technology Institute & Press “Diophantus”, Patras, Greece [email protected] Computer Engineering and Informatics Department, University of Patras, Patras, Greece 3 University Ecclesiastical Academy of Vella, Ioannina, Greece

1

2

Abstract. Internet of Things (IoT) is used more and more in our everyday life, connecting different wireless devices, and their distance can vary from some centimeters to many kilometers. New IoT technologies such as Long Range (LoRa) are emerging enabling power efficient wireless communication over very long distances. Simulation of LoRa networks is quite important, because can be used for the design and the evaluation of LoRa based application without the need of costly implementations or before to proceed to the actual implementation of the system. Choosing the right parameters of the systems like spreading factor can improve the energy consumption of the wireless devices. This paper presents the most important LoRa simulation environments available in the literature and after that, we present a comparative evaluation of LoRa simulation environments. The benefits, the disadvantages and the highlights of each LoRa simulation environment is presented.

1 Introduction Nowadays, Internet of Things (IoT) has been established in our everyday life, as it offers a number of capabilities. So, more and more devices and systems are being created that aim to offer solutions that need technologies that can interconnect wireless devices over long distances. Two candidates that try to solve this are LoRa technology [1], proposed by the LoRa Alliance, offering a low power, wide area network protocol for IoT devices and Narrow Band Internet of Things (NB-IoT) technology as part of 5G networks [2]. So, it is necessary to study this kind of technologies, because the IoT market is gaining exponential popularity introducing more solutions, and our lives can be improved in various ways. LoRa and NB-IoT have been evaluated and compared in a high extent, using metrics such as latency performance Quality of Service (QoS), range, coverage, battery life, cost efficiency [3], as these two technologies are very promising LPWAN (Low Power Wide Area Network) technologies. LoRa could be a good choice to integrate into IoT applications. Choosing the right combination of the spreading factors and bandwidth can reduce the transmission time, increase the data rate, thus improving the energy consumption of the IoT devices [3, 4]. Despite this, LoRa is an asynchronous protocol, so the engineers and scientists are able to choose the periods of sleep mode © Springer Nature Switzerland AG 2020 L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 374–385, 2020. https://doi.org/10.1007/978-3-030-33506-9_33

Comparison of LoRa Simulation Environments

375

that the devices can enter, in contrast to the NB-IoT that needs synchronization which leads to additional power consumption. In conclusion, LoRa should be used to send small data packets over large distances. LoRa [17] is an emerging new technology which enable power efficient wireless communication over very long distances. LoRa is a spread spectrum modulation technique derived from chirp spread spectrum (CSS) technology. LoRa Technology enables smart IoT applications that solve some of the biggest challenges facing our planet: energy management, natural resource reduction, pollution control, infrastructure efficiency, disaster prevention, and more. LoRa Technology has amassed over 600 known uses cases for smart cities, smart homes and buildings, smart agriculture, smart metering, smart supply chain and logistics, and more. With 97 million devices connected to networks in 100 countries and growing, LoRa Technology seems to be a very important technology for implementation of IoT, Networks. Devices typically communicate directly to a sink node which removes the need of constructing and maintaining a complex multi-hop network. LoRa provides a range of communication options (center frequency, spreading factor, bandwidth, coding rates) from which a transmitter can choose. Simulation of LoRa networks is very important, because can be used for the initial design and evaluation of LoRa based applications without the need of costly implementations. All these LoRa simulators have been used for testing various scenarios, but there is not, to the best of our knowledge, any a comparison between them. For example, PhySimulator has been used for the evaluation of link-level of LoRa, showing that despite the theoretical point of view that spreading factors can be considered as orthogonal, in reality the inter-spreading factor collisions are problem to LoRa [5]. FloRa simulator has been used for the evaluation of the performance of LoRa using Adaptive Data Rate (ADR) mechanism, showing that ADR is an effective way to increase the delivery ratio in an energy efficient way [9]. LoRaSim has been used to test the scalability of the LoRa Low-Power Wide-Area networks [6]. Last but not least, the LoraWAN Module for ns-3, was created in order to provide to the community a powerful tool for real networks, instead of using a simplified MAC protocol. So, it implements in addition to the LoraSim the acknowledgments, so as to give the ability to the users to test a network where the spreading factor can change according to given feedback from the gateway [11]. The aim of this paper is to study and examine the most common LoRa simulators that are available in the literature, provide a detailed comparison featuring their main advantages, disadvantages and showcasing the various highlights. The above comparison provides to the academic community the required data to help choose the necessary and more suitable simulator depending on the needs, the condition, the knowledge and programming experience of the user. Such features will be the operating systems that are supported, the programming language that is used, the existence and quality of the graphical user interface, the community support etc. The rest of this work is organized as follows: Next section describes the basics of LoRa technology. We briefly discuss the basic network simulation process in Sect. 3. The Sect. 4 presents the features of the LoRa simulation environments available in the literature. After that, the Sect. 5 presents a comparative evaluation of these LoRa

376

C. Bouras et al.

simulation environments. Finally, Sect. 6 presents the conclusions and Sect. 7 discusses the future work of our paper.

2 LoRa Technology LoRa technology can be considered as two parts, the first one is the LoRa and the second one is the LoRaWAN. LoRa is a proprietary wireless technology owned by Semtech, derived from the chirp spread spectrum modulation (CSS), while LoRaWAN is an open standard network protocol [1]. The physical layer of LoRa aims to provide the ability to communicate in an energy efficient way for energy constrained devices over long distances, supposedly coverage capability over 15 km Line of Site (LoS). The fact that is similar to chirp spread spectrum modulation allows to provide to the network a trade-off between the data rate for sensitivity within a channel bandwidth. Moreover, this fact helps in maintaining the low power features of the frequency shifting keying, while increasing the coverage, with low cost. But the fact that physical layer of the LoRa is proprietary means that we don’t have a lot of information available and the documentation it is not freely available to the scientific community [1, 4]. In Fig. 1. is shown the LoRa stack, depicting the layers of LoRa technologies.

Fig. 1. The LoRa stack [1].

In contrast to the physical layer of LoRa, LoRAWAN is promoted by the LoRa Alliance that is consisted by various companies such as Semtech, IBM, etc. It defines the communication protocol of the network and is the MAC (Medium Access Control) layer protocol. Its primary goal is to determine in the best way possible the battery lifetime of the device the network capacity, the QoS (Quality of Service) etc. The LoRaWAN protocol categorizes the devices-nodes into 3 district classes A, B and C. All devices must support at least the Class A functions. The other two classes are optional and depending on the hardware can be certified for the other classes. So, class A enables the communication of the end devices with the server, where each end device sends uplink transmission and follows two small downlink windows. It is the most power efficient class. Class B adds to class A the ability to the end devices to open extra

Comparison of LoRa Simulation Environments

377

downlink transmissions with fixed latency. Last but not least, Class C considers that the end devices have continuously open receive windows. From the three classes, class A is the most power efficient and Class C needs the biggest amount of energy in order to operate.

3 Network Simulation Network design with specific parameters is a challenging task, because the choice of the different parameters is not unique but should be suitable for the specific condition or application. For this reason, it is important to use a simulator.

Fig. 2. Flowchart describing the process of the simulation.

Generally, a simulation software should provide the user the ability to define the network topology, to specify the characteristics and features of each node, the link between them and the traffic model and the packet routing algorithms that can be used. Moreover, the user should get the performance metrics for the simulated network and if it is available to the simulation software to get some visualization of these metrics. These metrics for a LoRa simulation could be Data Extraction Rate (DER), Network

378

C. Bouras et al.

Energy Consumption (NEC) [6] and bit error rate etc. There are simulators that provide a good graphical interface, and others that can provide only some graphical representations through plots, or even command line like outputs. In general, simulators can be free open source, or commercial, but here we will examine only free simulators for academic purposes or open-source, and until now there are not priced LoRa simulators. Another important aspect of a simulator is its performance. First of all, as Fig. 2. shows, the users should define clearly the problem that is to be solved, in order to define correctly the system’s requirements and the environment. Then they can proceed to the modeling of this network, so to define the details in a greater extent and set the necessary parameters. The next step is to execute this simulation, so as to get the results. Finally, after the analysis of these information can repeat again all these steps with the given feedback to the simulation software or stop the process and analyze the results.

4 LoRa Simulation Environments In this section we present the following LoRa simulators: • • • •

PhySimulator [7] FLoRA [8] Ns-3 module [11] LoRaSim [12].

Fig. 3. An example of plot the BER-SIR.

Comparison of LoRa Simulation Environments

4.1

379

Physimulator

The first simulator is called PHY Simulator and aims to implement the link level of LoRa. PhySimulator is written in MATLAB. The aim of this simulator is to test the reception of two overlapping-interfering LoRa transmissions that have been modulated with different spreading factors. After each run of the program, 8 figures are generated, showing the packet, symbol and bit error rate. Specifically, the output is the packet, symbol and bit error rate for each spreading factor, interfered with any other spreading factor. This simulator gives the ability to the user to edit various parameters (changing the values of the variables on the code). For example, you can change the bandwidth, the payload bits and the max trials per step etc. All these parameters cannot change through a graphical interface, but the user has to edit them changing directly the MATLAB code. Next it is presented a plot showing the bit error rate and the SIR using the physimulator. In Fig. 3. we present an example of plot using physimulator. 4.2

FLoRa

The FLoRa simulator, is a simulation framework, using the well-known OMNeT++ discrete event simulation library that is distributed under the Academic Public License, so it is free for non-profit or academic use. In spite of OMNeT++ framework, FLoRa, is based also on the INET Framework, which is an open-source library for OMNeT++ and its purpose is to help the process of experimentation for different network protocols. FloRa is written in C++.

Fig. 4. An example of FLoRa [8].

380

C. Bouras et al.

In Fig. 4. we present an example of the GUI result of FLoRa simulator. It allows the creation of LoRa nodes, gateways and a network server [8]. Furthermore, its modules aim to simulate the physical layer, and the LoraWan MAC protocol [9]. In contrast to the other simulation software, this provides a very good graphical interface, because it is based on the OMNeT++ and a graphical representation of the network. It offers an accurate model for the LoRa physical layer. It emphasizes to giving statistics for the energy consumption of the network as well. The features of FloRa simulator according to [8] include the following: • • • •

Accurate model of LoRa physical layer (including collisions and capture effect) Simulations with one (or more) gateways in the network End-to-end simulations, including accurate modeling of the backhaul network Statistics of energy consumption in network.

After each run, a number of files containing the statistical data are created. Also, the FLoRa provides some sample scenarios. 4.3

NS-3 Module

This simulation software is not actually a distinct simulator, but a module that plugs in the ns-3, a free discrete event network simulator, that is designed for academic and research purposes. It supports a big range of protocols and networks including IP and non-IP networks, and wireless simulations [10]. Ns-3 is designed with modularity in mind and provides the ability to work in both graphical interface and command line. It is written in C ++ and python. This particular module it is compliant with class A of the LoRaWAN 1.0 specifications. This means that it simulates the case where the devices send only uplink transmission and the server sends only downlink transmission. This class is the most energy efficient end-service system, in contrast to the other two available classes of LoRa. The physical layer, the MAC layer and the Transport and Application has been created, trying to give an agile and highly configurable solution. It gives the ability to integrate new algorithms on the server side to this ns-3 module [11]. Moreover, this proposed module contains an extra class in order to track the power consumption named LoRaRadioEnergyModel, tracking the energy of the states of the physical layer of the LoRa protocol. Furthermore, it was tested and evaluated by its creators using three scenarios. • The first one considers a circular topology with one gateway that is located in the center, without the need of acknowledgments • The second one uses more gateways, assuming circular topology too. • The last scenario considers a network like the first scenario with the difference that the messages now need confirmation. This scenario according to the authors of [11] shows that is impossible to acknowledge all the packets. 4.4

LoRaSim

Finally, the last tool examined is the LoRaSim, is a discrete event simulator which aim is to analyze the ability of scalability of a LoRa network and the collisions. It allows us to place the LoRa nodes in a 2-dimensional grid. The LoRaSim is written in python

Comparison of LoRa Simulation Environments

381

ver. 2.7, something that is considered as a drawback because only version 3 of python is updated. Also, it uses NumPy, matplotlib and SimPy python libraries. This tool has four different simulations, each one for different properties of the network and the nodes. More specifically, there are simulations for single base station and one that can simulate up to 4 base stations. Another categorization of the simulations is the type of the antennae, as there are files that simulate nodes with directional antennae.

Fig. 5. An example of LoRaSim plot, using one sink, 50 nodes (blue: nodes, green sink).

In this simulator, you can change a number of parameters, such as the number of the nodes, the number of the base stations to simulate. Furthermore, it is possible to choose between full and simplified check of the collisions, the number of the LoRa networks, the time that the simulation will run and the distance between two base stations. LoraSim, when executed provides the user some plots if the variable graphics is set to 1, but the majority of the information is presented through the command line and it exports these details to a file called expX.dat. Apart from the visualizations through the plots there is not a graphical interface, and the user has to work through the command line. Below in Fig. 5. there is an example of the plot that is created, in the case of one sink and fifty nodes.

5 Comparative Comparison In this section, we present an overall comparison of the features of the examined simulators. Some of the features we have evaluated are the operating system support, the license type, the GUI, the availability of the statistics of energy consumption and the available documentation.

382

C. Bouras et al. Table 1. Comparison of The Simulators

Features Event License type

PhySimulator Discrete Free

Language Operating system GUI Energy consumption statistics Documentation Number of published papers Website Community support

Matlab Windows, Linux, MacOS Only plots No

FLoRa Discrete Open source (study and research) C++ Windows, Linux, MacOS Yes Yes

Ns-3 module Discrete Open source

C++, python Linux, MacOS, Windows through virtualization Yes Yes

LoRaSim Discrete Creative common attributes 4.0 python Windows, Linux, MacOs Only plots No

Ok 2

Ok 1

Average 1

Ok 2

Yes Good

Yes Limited

No Very good

Yes Limited

The reasons we chose to evaluate all these features are the following: it is very important for the user to know in what platform should work, what programming languages should know for the specific problem that has to be solved and especially for the scientific community and the researchers to know if it is open source. This is necessary in order to expand the existing tools taking into consideration additional scientific aspects, novelties and as happened with the ns-3 module to use the existing protocols and “insfrastracture” for other protocols such as LoRa. In this framework we examine the existance of related publications, website and the community that support each software. In addition to this, in order to understand and evaluate the results of the simulation so as to check your system model, it is also necessary to have some visualized data and statistics, for this reason this feature is examined too. The comparison of the features of each simulation software is shown in Table 1. As we see in the table where there are all the features of the examining simulators side by side, and we can say that all available simulators are quite good. First of all, all four simulators are discrete event. This means that they are modelling the system as a sequence of discrete events in the time domain. This allows the simulators to move to the next event, assuming that there is no change in the system between two consecutive events, thus there is no need to track the system continuously. Concerning the programming languages which have been used in the implementation of the simulators, all the simulators are based on well know programming environments with a great support community. The above is very important because a researcher can easily expand the simulator capabilities by implementing new modules

Comparison of LoRa Simulation Environments

383

(for example support new network protocols or integrate additional tools on the existing available environment etc.). In summary, PhySimulator is implemented in Matlab, FLoRa is implemented in C++, Ns-3 module is implemented in C++ and python and LoRaSim is implemented in python. The FLoRa through OMNeT++ and Ns-3 through NetAnim have a more extended graphical interface compared to the other simulators. PhySimulator and LoraSim give only some plots. All the examined simulators have been published to the scientific community and FloRa, ns-3 module have 1 related publication and PhySimulator and LoRaSim have 2, and all simulators have their own website except the Ns-3 module. Nevertheless, Ns-3 is an open source project that has a great and big community that supports it [13]. FloRa, Physimulator, LoRasim have both related publications and a related website in contrast to the Ns-3 module that has only related publication. For this reason, the three simulators have more detailed information about the process of installation and how to use the tools. The Ns-3 module is open source and the code is available on Github. LoRA is a technology for the implementation of IoT applications where the devices are powered usually by a battery, so the optimization of energy consumption is a key factor for the deployment of IoT applications. As result, the investigation of energy consumption through Energy Consumption statistics is a very important feature of a LoRA simulator. From the simulators which we study in this paper only FLoRa and Ns-3 module supports Energy Consumption statistics. Another feature that was examined is the license of each simulator. FloRa is free for academic and study use. LoRaSim on the other hand is provided under the license Creative Commons Attribution 4.0. This means that is free to use, share and change the code, giving the appropriate credit to the authors and provide link with the license. All simulators can be run in all operating systems in one way or other. PhySimulator can be executed wherever Matlab can be executed, so the supported operating systems include Linux MacOs and Windows. In the same way, LoraSim can be used in whatever operating system can support python. So LoraSim is used in Windows, Linux and MacOs. The ns-3 module, because it is based on ns 3, it can be executed on Windows only by using a virtual machine. It supports Linux and MacOS natively. As we have mentioned before, one very important aspect is the energy consumption. The FloRa simulator and the Ns-3 module focus on providing the user this kind of information. One other important factor which plays an important role in the usability of the simulator, is the detailed documentation of the simulator operation. Except of Ns-3 module (which offers an average documentation), all the other simulators provide a detailed documentation something which is very important. Also, there is a number of tutorials for the ns-3 and OMNeT++ simulators that can be used in order to understand how each simulator works. If we discuss for the community support, FLoRa and Ns-3 module seems to have the better community support, because both the underline network simulators, Ns-3 and OMNet++ respectively, have a strong user community. Ns3 being open source means that there are less maintainers to respond to questions or fix reported bugs and abnormalities. However, it is extremely widespread and used by many students, scientists and academics that the online community can help and offer great support.

384

C. Bouras et al.

More specifically, according to [14] a Google Scholar search of the ‘ns-3 simulator’ results since 2017 (excluding patents and citations) yields over 2000 links (with some false positives). In addition, the IEEE digital library lists 145 ns-3 publications for 2017, and the ACM digital library lists 2579 publications matching the search term ‘ns3’ in 2017. In addition, there are organized Workshops on Ns-3 and the related proceedings are published in the ACM digital library [15]. The above facts ensure the important acceptance of ns-3 simulator as network research tool. In addition to Ns-3, also OMNeT++ has an active community which have organized 5 OMNeT++ Community Summits until 2018 [16]. As result, if we compare the above simulators in terms of research community support seems that Ns-3 and OMNeT++ have the most active research community which organize relative workshops about the evolution of the simulation software. This seems reasonable based on the fact that both Ns-3 and OMNet++ can be obtained at no cost.

6 Conclusions As we have said, LoRa is an ideal solution for long range communication of the IoT. It is known, that choosing the right variables such as spreading factor is very crucial, in order to maintain an acceptable level of power consumption of the device, to have greater coverage, good QoS etc. So, in order to find the best tradeoff between the above, it is helpful to use a simulator, as it can help the developers and scientists to choose the right parameters, executing the simulations with low risk and without the need of expensive implementations and investment. Moreover, it is possible to compare different protocols and technologies using simulators and evaluate the advantages and disadvantages of each technology. In this paper, we tried to give to the scientific community a general overview of the most common simulators available to the literature for LoRa technologies and a comparison between them.

7 Future Work Future work is the study and implementation of energy efficient mechanisms and systems for IoT devices using LoRa technology, with emphasis to search and rescue systems with various health data and medical monitoring sensors. So, the use of simulators is a key aspect of this process and a helpful tool, before we start the implementation of such systems, study the feasibility of such actions. In the context of our future work we plan to use LoRa simulators in order to optimize energy efficient algorithms in search and rescue systems with various health data and medical monitoring sensors. Acknowledgments. The authors acknowledge that the starting point and the drive of this paper have been generated in the context of the project ‘WeSAR (Wearable based Search And Rescue system)’. In this framework, we would like to thank all the project partners and employees of Econais AE, Yodiwo AE and CTI (Computer Technology Institute & Press “Diophantus”) for their personal copy information and experience on this project. This research has been co-

Comparison of LoRa Simulation Environments

385

financed by the European Union and Greek national funds through the Operational Program Competitiveness, Entrepreneurship and Innovation, under the call RESEARCH - CREATE INNOVATE (project code: T1EDK-01520).

References 1. LoRa alliance official site. https://lora-alliance.org/. Accessed 9 July 2019 2. Mobile IoT in the 5G future. https://www.gsma.com/iot/wp-content/uploads/2018/05/ GSMA-5G-Mobile-IoT.pdf. Accessed 9 July 2019 3. Sinha, R.S., Wei, Y., Hwang, S.H.: A survey on LPWA technology: LoRa and NB-IoT. ICT Expr. 3(1), 14–21 (2019) 4. Bouras, C., Kokkinos, V., Papachristos N.: Performance evaluation of LoraWan physical layer integration on IoT devices. In: Global Information Infrastructure and Networking Symposium, GIIS 2018, Thessaloniki, Greece (2018) 5. Croce, D., Gucciardo, M., Mangione, M., Santaromita, G., Tinnirello, I.: Impact of LoRa Imperfect Orthogonality: Analysis of Link-Level Performance. IEEE Commun. Letters 22 (4), 796–799 (2018). https://doi.org/10.1109/LCOMM.2018.2797057 6. Georgiou, O., Raza, U., Low Power Wide Area Network Analysis: Can LoRa Scale? In: IEEE Wireless Communications Letters, vol. 6(2), pp. 162–165, April 2017. 10.1109/L. WC.2016.2647247 7. PhySimulator site. http://lora.tti.unipa.it/. Accessed 9 July 2019 8. FloRa official site. https://flora.aalto.fi/. Accessed 9 July 2019 9. Slabicki, M., Premsankar, G., Di Francesco, M.: Adaptive configuration of lora networks for dense IoT deployments. In: 2018 IEEE/IFIP Network Operations and Management Symposium, NOMS 2018 , Taipei, 2018, pp. 1–9. https://doi.org/10.1109/noms.2018. 8406255 10. NS3 simulator official site. https://www.nsnam.org/about/. Accessed 9 July 2019 11. Reynders, B., Wang, Q., Pollin., S.: A LoRaWAN module for ns-3: implementation and evaluation. In: Proceedings of the 10th Workshop on ns-3, WNS3, pp. 61–68. ACM, New York (2018). https://doi.org/10.1145/3199902.3199913 12. LoRaSim site. https://www.lancaster.ac.uk/scc/sites/lora/. Accessed 9 July 2019 13. Kabir, M., Syful I., Hossain, Md., Hossain, S.: Detail comparison of network simulators. https://doi.org/10.13140/rg.2.1.3040.9128 14. NS-3 Statistics. https://www.nsnam.org/about/statistics/. Accessed 9 July 2019 15. NS-3 Workshops. https://www.nsnam.org/research/wns3/. Accessed 9 July 2019 16. OMNeT++ Community Summit 2018. https://summit.omnetpp.org/archive/2018/. Accessed 9 July 2019 17. Semtech LoRa Technology Overview | Semtech, https://www.semtech.com/lora. Accessed 9 July 2019

Proactive Network Slices Management Algorithm Based on Fuzzy Logic System and Support Vector Regression Model Amal Kammoun1,2(B) , Nabil Tabbane1 , Gladys Diaz2 , Nadjib Achir2 , and Abdulhalim Dandoush3 1

MEDIATRON Laboratory, University of Carthage, Sup’Com, Aryanah, Tunisia [email protected] 2 L2TI Laboratory, University of Paris 13, Paris, France [email protected], {gladys.diaz,nadjib.achir}@univ-paris13.fr 3 ESME, Paris, France [email protected]

Abstract. Software Defined Networks (SDN ), Network Function Virtualization (NFV ) and Network Slicing are the key technologies for future network implementation. Their aggregation allows more flexibility for the networks by provisioning network slices according to specific use cases requirements. However, in order to ensure these requirements during all the slice execution time, a management module has to be implemented. In this paper, we present our considered architecture for the management of network slices. We detail especially the network controller components. Moreover, we propose a proactive dynamic approach which forecasts the future workload behavior of network slices. Based on the actual and predicted load state, the management algorithm, which is based on a fuzzy logic system (FLS), will determine the adequate management decision for the deployed slices. Based on real network traces, an evaluation of the efficiency of our algorithm is presented.

1

Introduction

The Fifth Generation of Mobile Networks (5G) is proposed to address the future network context. This context is characterized by an everywhere-connected society where not only persons will be connected to the network but also clothes, vehicles and smart objects. Therefore, 5G has to serve an expanding number of emerging services. These services, such as e-health, autonomous driving, IoT, etc. request different and stringent constraints mainly in terms of latency, reliability, and availability that cannot be ensured by current network implementations [1]. SDN, NFV and network slicing are introduced to support the new requirements of network services. A network slice is a customized virtual network running on top of a shared infrastructure. Each slice is composed of a set of connectivity links and programmable resources. Slice resources will embed specific Virtual Network Functions (VNFs) to meet the needs of network applications and use-cases. c Springer Nature Switzerland AG 2020  L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 386–397, 2020. https://doi.org/10.1007/978-3-030-33506-9_34

Proactive Network Slices Management Algorithm

387

While deploying network slices, the management of their resources is a crucial aspect which allows to maintain the required Quality of Service (QoS) for users. In [2], ISG NFV presents some mechanisms required for the management of NFV environment. According to their paper, a management system involves three features which are failures prevention, failures detection and failures remediation. Each module should cooperate with the rest of the network entities in order to report and diagnose performance and real-time resource consumption. Failure prevention is the ability to block the occurrence of failures in the system. It takes effect through the control of the quality offered by the system. Online failure prediction is considered as the adequate approach to enhance the reliability and availability in NFV environment. The auto-scaling is the ability to dynamically scale the resources [3]. In the literature we identify two types of auto-scaling which are the vertical and the horizontal auto-scaling defined as the following [4]: • Vertical auto-scaling: it will increase (scaling-up) or decrease (scaling-down) the resources of the virtual machine (VM). In this case, the VM will be respectively more or less powerful. Thus vertical auto-scaling focuses on the efficiency of VMs. • Horizontal auto-scaling: it consists on the addition (scaling-out) and removal (scaling-in) of virtual machines (VMs). The scaling-out and scaling-up mechanisms allow the system to redress the predicted congestion in the system and thus to avoid future possible failures. Moreover, the auto-scaling allows also to perform an energy saving by scalingdown network resources. In this paper, we interest on the management of network slices based on their load state and the predicted future state. We propose a proactive and dynamic management algorithm which will at first predict the state of slice resources and then apply the convenient management decision based on fuzzy logic system. Fuzzy controllers are characterized by their ability to analyze unclear data behavior such as the variation of network load [5]. The remainder of this paper is organized as follows: Sect. 2 presents the related works. Section 3 presents our considered network architecture and details our control and management module. In Sect. 4, simulations and results of our proposed algorithms are presented. The last section concludes the paper and proposes some perspectives.

2

Related Works

Several works interest on the management of network and cloud resources. Among these works, some papers introduce the forecasting of network resources state in order to better apply management actions. The prediction of resources utilization rate is a robust approach for determining their future requirements in advance. In [6], authors present a survey on the prediction schemes with their main advantages and challenges.

388

A. Kammoun et al.

In [7] authors propose an energy-effective prediction algorithm for the cloud environment in order to identify future requirements of cloud resources based on Multilayer Perceptron (MLP) model. In [8] authors propose Wavelet Support Vector Machine (WSVM) algorithm for the prediction of data centers behaviour. In [9], authors interest on application-aware SDN solutions. They propose an Autoregressive Integrated Moving Average (ARIMA) based model to forecast large data transfers. Their model integrate an automated parameters estimation module and a module for parameters verification and re-estimation. Authors in [10] interest on the context of Massively Multiplayers Online Gaming (MMOG). They propose a provisioning approach that forecasts future workload of an MMOG application to allocate the adequate resources using a Seasonal ARIMA (SARIMA) model. For management algorithms based on fuzzy logic controllers we cite the works in [11] and [12]. [11] proposes an auto-scaling mechanism based on fuzzy controller for fog environment. [12] presents a dynamic learning strategy for selfadaptive controller. It is based on two reinforcement learning approaches which are Fuzzy SARSA learning and Fuzzy Q-learning. In the context of NFV and SDN, Open Baton project [13] proposes an autoscaling module however the decision adopts only the linear threshold method. This method is not suitable for rapid changeable service context.

3

System Model

Our considered architecture, as depicted in Fig. 1, is composed of three layers: (i) the virtual infrastructure which encompasses servers and forwarding nodes (ii) a set of slices deployed to serve network use cases and forward users traffics and (iii) a Slice Resource Management module that commands the creation of network slices and performs their management. The Slice Resource Management module is composed of many sub-modules as depicted in Fig. 1. In the following we define each sub-module and we present its functionalities. • Load Monitor Module: This module collects the CPU, disk, and memory utilization rates of the deployed slices and network resources. Thereafter, It stores these measures in the Status Database. • Status Database: This database contains information about the state of network slices and virtual machines. The computational capacities of each network slice is then updated periodically by the Load Monitor Module. • Workload Analyzer : This module analyzes the Information data base in order to compute the load state of each network slice future slice state. The results of this module are stocked in the Predicted and Historical Information Database and will be used by the Resource Usage Predictor module in order to predict the future state of the corresponding slice. • Resource Usage Predictor : This module will forecast for each slice the resource utilization rate of VNFs which are running on top of virtual machines. It will predict the future load state based on historical information about the load

Proactive Network Slices Management Algorithm

389

Fig. 1. Network slice management architecture

state. In this paper, the prediction module is designed using the Support Vector Regression (SVR) model. By deploying a proactive solution, we can anticipate some problems that may occur in the slice. For instance, we will overcome the increase of the processing delay and avoid unnecessary resources migration. • Predicted and Historical Information Database: This database contains historical and predicted information about the total load state of each network slice. These information will be used by the Optimization module and the Dynamic Handler Framework (DHF) in order to better optimize and manage network slices. • Optimization Framework : This module is out of the scope of this paper. Its details are depicted in our previous work [14]. Mostly, it decides about the creation of a new slice in accordance with user request description. It aims to select the best target network resources among available schemes in order to assure slice requirements. • The Dynamic Handler Framework (DHF): DHF deals with unexpected events like the congestion of the slice, the failure of VMs, energy waste, etc. For this purpose, it examines periodically the Predicted and Historical Information Database and detects the trigger events for slices management (e.g. addition, deletion or modification of network resources). Thus, DHF studies the slices

390

A. Kammoun et al.

performances and decides what management action is needed to be exercised on each slice. • The Decision Execution Module: This module receives commands from the DHF and the Optimization Framework. It is responsible for the execution of the taken decision on the considered slices.

4

Proposed Algorithm

In order to perform the management of network slices, several phases will be performed by the proposed Slice Resource Management Module as depicted in Fig. 2. Details of each phase are presented in the following subsections.

Fig. 2. Network slice management algorithm

4.1

Information Collecting Phase

According to our system, the Load Monitor module is responsible for the gathering of the information about network slices and virtual resources. In this paper we consider three metrics for the monitoring module which are the CPU, memU M em and Ldisk ory, and disk utilization rates. We define LCP i,k , Li,k i,k as respectively the CPU load, the memory load and the disk load of the virtual machine i of the slice k. 4.2

Resource Usage Predictor

Load and utilization rate of virtual network resources are characterized by their high variability. Thus, the computational performances of the network slice are not stationary all the time. This makes hard to satisfy the slice requirements during its execution time. Therefore, the load prediction of network slices resources is fundamental.

Proactive Network Slices Management Algorithm

391

Several methods exist for the prediction of future application state such as Queuing Network (QN) models, Linear Regression (LR), ARIMA, etc. In our work, we opt for the SVR prediction model to forecast the stationarity of the slice state [15]. SVR algorithm has presented better Root Mean Squared Error (RMSE) measure compared to LR and ARIMA algorithms. In fact, prediction algorithms are evaluated through Mean Squared Error (MSE) and RMSE which are equals to: N

1  ∗ (A − P )2 N 1 √ RM SE = M SE M SE =

(1) (2)

where A is the actual value; P is the predicted value and N is the size of the prediction set. In this paper, Radial Basis Function (RBF) is used as the kernel function for the SVR model. This model is shown in Eq. 3. 2

K(ti , tj ) = exp(−0.5 ti − tj  /σ 2 )

(3)

where K(ti , tj ) is the kernel function, ti and tj are the input vectors and σ is the value of sigma in the Gaussian kernel. The SVR model is also defined by C parameter which is the trade-off between the empirical risk and the model flatness and also by  which is the value of epsilon in the insensitive loss function. For our simulation in Sect. 5, we fixed the value of C at 1e3 and the value of σ to 1e2. After the prediction step, we will test the stationarity of the future state of the network slice in order to anticipate management actions. We consider that the predicted values are stationary if they have constant mean and variance and also the auto-covariance does not depend on time. 4.3

Decision of the Management Action

The DHF module detects sudden events in the slices and determines the accurate moment to trigger the management action based on the fuzzy logic system. Its decision takes into consideration the actual and predicted resources utilization rates. This module was developed using the MATLAB fuzzy logic toolbox. We used the Mamdani fuzzy inference system which includes four functional blocks: the fuzzifier, the set of fuzzy rules, the fuzzy inference engine and the defuzzifier. The next subsections detail those blocks. 4.3.1 The Fuzzifier The input parameters for the fuzzifier module are the information collected during the Information Collecting Phase which are CPU, memory and disk utilization rate as well as the predicted load state. The fuzzifier converts the numerical

392

A. Kammoun et al.

collected information into linguistic variables using the membership functions. In our fuzzy model, we define for each input variable a membership function with several membership degrees defined as the following: • X utilization rate (X refers to CPU, memory and disk): denoted as Xocp . It provides a measure of the X utilization rate on all the VMs of the network slice. We define the fuzzy set of Xocp as the following: Low, Moderate and High. • Predicted load state: denoted as Pt . We define the fuzzy set of Pt as the following: stationary, variant. 4.3.2 The Fuzzy Inference System After the fuzzification of the input variables, results are evaluated by the Fuzzy Inference System (FIS). In fact, FIS analyses fuzzy sets according to the inputs variables and fuzzy rules which are based on a set of If(condition)-THEN(action) rules. Depending on these rules, the algorithm will decide which strategy should be adopted for the management of network slice resources. The considered strategy can be either scale-down, scale-up, scale-out or no slice modification.

Fig. 3. Flowchart of the management action decision

In the following we present our considered rules. Figure 3 presents the flowchart of the different management decisions.

Proactive Network Slices Management Algorithm

393

• When the utilization rate of the CPU, memory or disk is low and the predicted load state is stationary, we will scale down slice resources in order to decrease the computing performances of those resources. We opt for the scaling-down strategy instead of resource migration in order to maintain the required QoS and avoid the additional cost of resource migration. • If the predicted load state is variable and the resource utilization rate is low, no management decision will be executed to prevent redundant scaling down actions. • If the resource utilization rate is moderate and the predicted load state is stationary, the slice will keep its actual situation. However, if the resource utilization rate is moderate and the predicted load state is variable, the strategy will be to scale up the hot-spot nodes of this slice. • When the resource utilization rate is high, we have to increase slice capacities in order to guarantee the required QoS. If the predicted load state is stationary, the strategy will be to scale up existing resources on the slice and so to increase the computing resources assigned to VMs. If the predicted load state is variable, the strategy will be achieving efficiency by a scaling-out action (adding resources to the slice). The scaling out of resources is required in order to assign more resources to VMs which are expected to be hot-spots. 4.3.3 The Defuzzifier The last step is the defuzzification. It takes the final decision for the initiation of the management action. In fact, by applying the fuzzy inference rules, a linguistic decision about the management strategy can be generated based on fuzzy input. The defuzzifier converts the decision sets into precise value to determine the management decision action. We define the fuzzy set strategies as the following: scaling-down, scaling-up, scaling-out or no slice modification. 4.4

Execution of the Management Action

The process of the slice management ends with the execution of the decision taken by the Dynamic Handler Framework module. Based on this decision, the Decision Execution Module will perform the required auto-scaling action. Therefore it will either increase the resources (scaling-up), decrease the allocated resources (scaling-down) or create new VMs (scaling-out) for the network slices.

5

Simulation and Results Analysis

In this section, our objective is to study the performances of the Resource Usage Predictor module and the Dynamic Handler Framework. We consider real data-set traces in order to validate the prediction module. In fact, prediction tests have been performed on a trace dataset provided by Alibaba which was released on September 2017 [16]. This data set details the performances of 11089 online service jobs and 12951 batch jobs running on top

394

A. Kammoun et al.

of 1300 machines during 12 hours. The analysis of this public trace is performed in [17]. Figure 4 presents the results of the prediction module based on the alibaba dataset for machines CPU utilization rate. It shows that this prediction allows to forecast the variability of the CPU utilization rate. Despite the fact that the predicted data is not equal to real data, we can predict the variability of the utilization rate. If we consider an increasing values of CPU utilization rate, as shown in Fig. 5, our predicted values have exactly the same variation and thus the forecast of the variability of the resources utilization rate will be possible.

Fig. 4. Prediction of the CPU utilization rate of machines during execution time

After the validation of our prediction module, we consider a network slice scenario characterized by a variable load state as shown in Fig. 6. We suppose that the considered slice has the capacity to serve at most 1000 users. We suppose also that, at time t=0, all the VMs of this slice have the same computational capacities. Given resources utilization rate and predicted slice state, the fuzzy decision maker gives the management decision to execute. According to our DHF, several actions can be executed such as scaling-in, scaling-out and scaling down. Figure 7 illustrates the progress in time of the processing delay of each slice under different load state. The processing delay is defined as the time it takes VMs to process incoming tasks. For the considered slice, at time t=1 the DHF decides to increase the number of allocated resources in order to satisfy the pic of users at this time. Then, as the load at this slice decreases over the time, it decides to decrease the number of allocated resources and perform a scale-down action. The processing delay when our algorithm was used is lower than the generated delay when no

Proactive Network Slices Management Algorithm

395

Fig. 5. Prediction of resources utilization rate for an increasing dataset

Fig. 6. Load state of the considered network slice

Fig. 7. Progress in time of the processing delay

management decision was implemented. In fact, for each slice, our proposed algorithm will consider the predicted values of resources utilization rates and decides according to the actual and predicted state the best management decision to execute.

396

6

A. Kammoun et al.

Conclusion

Network Slices management is an important aspect that aims to ensure the required QoS for end users during the slice run-time. For this purpose, we proposed in this paper a proactive algorithm which predicts the future resources behaviour based on SVR and then triggers the management action depending on the slice condition and the result of the Prediction Module based on FLS. In this paper, we consider the auto-scaling actions as management strategies. According to results, applying the FLS with the consideration of the prediction values enhances the processing delay in the network slices. As future works, we will study other management decisions such as resource migration and load balancing. Furthermore, we aim to study the enhancement of our algorithm on the utilization rate of network resources. We manage also to improve the considered SVR parameters in order to have better predictions.

References 1. Omnes, N., Bouillon, M., Fromentoux, G., Grand, O.L.: A programmable and virtualized network it infrastructure for the internet of things: How can NFV SDN help for facing the upcoming challenges. In: 2015 18th International Conference on Intelligence in Next Generation Networks, pp. 64–69, February 2015 2. ETSI GS NFV-REL 001, Network Functions Virtualisation (NFV); Resiliency Requirements, V1.1.1 (2015) 3. Mell, P., Grance, T.: NIST special publication 800-145: The NIST definition of cloud computing (2011). https://nvlpubs.nist.gov/nistpubs/Legacy/SP/ nistspecialpublication800-145.pdf. Accessed 20 June 2019 4. Carella, G.A., Pauls, M., Grebe, L., Magedanz, T.: An extensible autoscaling engine (ae) for software-based network functions. In: 2016 IEEE Conference on Network Function Virtualization and Software Defined Networks (NFV-SDN), pp. 219–225, November 2016 5. Dotcenko, S., Vladyko, A., Letenko, I.: A fuzzy logic-based information security management for software-defined networks. In: 16th International Conference on Advanced Communication Technology, pp. 167–171, February 2014 6. Amiri, M., Mohammad-Khanli, L.: Survey on prediction models of applications for resources provisioning in cloud. J. Netw. Comput. Appl. 82, 93–113 (2017) 7. Ahammad, T., Kumar Acharjee, U., Hasan, M.M.: Energy-effective service-oriented cloud resource allocation model based on workload prediction. In: 2018 21st International Conference of Computer and Information Technology (ICCIT), pp. 1–6, December 2018 8. Zhong, W., Zhuang, Y., Sun, J., Gu, J.: The cloud computing load forecasting algorithm based on wavelet support vector machine. In: Proceedings of the Australasian Computer Science Week Multiconference, ACSW 2017, pp. 38:1–38:5. ACM, New York (2017) 9. Nadig, D., Ramamurthy, B., Bockelman, B., Swanson, D.: Large data transfer predictability and forecasting using application-aware SDN, pp. 1–6, December 2018

Proactive Network Slices Management Algorithm

397

10. Dhib, E., Zangar, N., Tabbane, N., Boussetta, K.: Impact of seasonal ARIMA workload prediction model on QoE for massively multiplayers online gaming. In: 2016 5th International Conference on Multimedia Computing and Systems (ICMCS), pp. 737–741, September 2016 11. Tseng, F., Tsai, M., Tseng, C., Yang, Y., Liu, C., Chou, L.: A lightweight autoscaling mechanism for fog computing in industrial applications. IEEE Trans. Ind. Inform. 14, 4529–4537 (2018) 12. Arabnejad, H., Pahl, C., Jamshidi, P., Estrada, G.: A comparison of reinforcement learning techniques for fuzzy cloud auto-scaling. In: Proceedings of the 17th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing, CCGrid 2017, pp. 64–73. IEEE Press, Piscataway (2017) 13. “Open baton”. http://openbaton.github.io/. Accessed 25 Mar 2018 14. Kammoun, A., Tabbane, N., Diaz, G., Dandoush, A., Achir, N.: End-to-end efficient heuristic algorithm for 5G network slicing. In: 2018 IEEE 32nd International Conference on Advanced Information Networking and Applications (AINA), pp. 386–392, May 2018 15. Cortes, C., Vapnik, V.: Support-vector networks. Mach. Learn. 20, 273–297 (1995) 16. “Alibaba cluster trace program”. https://github.com/alibaba/clusterdata. Accessed 31 July 2019 17. Lu, C., Ye, K., Xu, G., Xu, C., Bai, T.: Imbalance in the cloud: an analysis on alibaba cluster trace. In: 2017 IEEE International Conference on Big Data (Big Data), pp. 2884–2892, December 2017

An Optimal Route Recommendation Method for a Multi-purpose Travel Route Recommendation System Chen Yuan(&) and Minoru Uehara Graduate School of Information and Science and Arts, Toyo University, Kawagoe, Japan {s3b101800023,Uehara}@toyo.jp

Abstract. With the rapid development of tourism, the demand for travel is becoming increasingly personalized. Travelers are increasingly traveling to places that they have not visited previously. When travelers decide to visit unfamiliar scenic spots, they need to spend a great deal of time making relevant travel plans. Therefore, we consider a system that is specifically designed to make travel plans for travelers when they visit a country or city for the first time. Simultaneously, the optimal path result is obtained using a genetic algorithm. This system can provide travelers with highly satisfying travel paths that only require the traveler to enter the degree of destination and time constraints.

1 Introduction The Japanese population is currently declining, which is causing a stagnant Japanese economy because of the declining birth rate and aging population. Additionally, tourism requirements are diversifying because tourists’ sightseeing preferences have changed from traveling in groups to individual travel. The Tokyo Metropolitan Tourism Industry Promotion Plan [1] observed that “for our country to grow economically, the number of foreign tourists visiting Japan must increase. Therefore, to expand foreign tourists’ domestic consumption while they visit Japan, it is important to capture their vigorous international demand.” Thus, the quality of sightseeing information recommendations is very important, and a recommendation system for sightseeing spots and tourist routes is necessary. According to an investigation of the behavioral characteristics of foreign tourists in 2017 [2], the number of foreign tourists visiting Japan is rapidly increasing. Additionally, the proportion of individual tourists is increasing annually. Because of the decision to host the Tokyo Olympic Games in 2020, more foreign tourists will travel to Japan individually. However, personal travel is not easy for some foreign travelers, and they may need to prepare their journey itinerary before they travel by surveying information about a variety of tourism sites. Thus, we should help foreign tourists to determine their sightseeing plans more conveniently using a travel recommendation program.

© Springer Nature Switzerland AG 2020 L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 398–408, 2020. https://doi.org/10.1007/978-3-030-33506-9_35

An Optimal Route Recommendation Method

399

We consider recommending a route with a high satisfaction rate for travelers who are planning to travel to Japan for the first time. Even if the user does not have any information, such as place names, type of location, or the distance to a travel site, an appropriate travel plan can be obtained. When the algorithm considers tourist sites using the user’s preferences, it can provide information about the tourism site route using the user’s input, including travel time, time at the tourism site, other users’ evaluations, and the user’s preferences for different tourism categories, to recommend a route with a high satisfaction rate for individual users. In [8], we developed a system for recommending travel plans for first-time visitors to Japan. It provided users with highly satisfying travel plans so that they did not need to spend time determining a destination. We found that there were still problems in terms of computational efficiency for our optimal travel route recommendation system based on a basic algorithm. For example, when the number of scenic spots increases to a relatively large range, the system cannot calculate the corresponding results smoothly. Additionally, the search process within the system is limited because the number of selected attractions cannot exceed a certain range. If users want to visit more than eight or nine scenic spots per day, then the system takes a long time to calculate a path. Thus, in this paper, we solve this problem using a genetic algorithm (GA). Additionally, we compare the basic algorithm and GA to determine which is more efficient in the system. Therefore, in this paper, we use a GA to effectively eliminate the local optimal problem and search effectively in the case of a large path population, and combine [8] it with an objective function to constrain the path fitness. In Sect. 2, we review relevant research. The main content the paper considers the principle, characteristics, and shortcomings of the travel recommendation system for tourists visiting Japan for the first time. In Sect. 3, we explain how to use the GA algorithm to establish the recommendation system model. In Sect. 4, we present the results of the system compared with various methods for the basic algorithm and GA. Finally, in Sect. 5, we summarize the study and describe tasks for future work.

2 Related Work There are many travel recommendations that we believe need to be improved. Therefore, we developed a travel recommendation system for tourists visiting Japan for the first time. However, the system still has some problems. In this section, we briefly describe the principles and features of the proposed system and what we believe needs to be improved. Then we introduce a GA and its applications. 2.1

Tourist Recommendation System

Tourist recommendation systems (TRS) are very useful for tourists that want to search for relevant travel information. TRSs have been developed to a degree that allows tourists use a computer to organize the entire trip instead of only obtaining a message about particular spots. With the development of tourism, the trend of personalized tourism is growing, and the points of interest for tourists have become difficult to determine.

400

C. Yuan and M. Uehara

Early versions of TRSs recommend sightseeing routes with a high similarity to users based on the attraction degree of a sightseeing spot and ambient tourist attractions. However, these TRSs cannot assess the sightseeing time and preferences of the advanced traveler, and whether they are suitable for the user. According to [3], many users post travel information using services such as social networking services (SNS) as a result of the spread of smartphones. Therefore, in that study, SNS location information was used to extract tourist information and sightseeing routes in Foursquare. Sightseeing tweets were collected and tourist routes were extracted, and then classified into three categories: “meal,” “landscape,” and “action.” The system recommends tourist routes using tourist tweets classified as extracted tourist routes. However, it is limited to recommending the sightseeing order of sightseeing spots according to the type of sightseeing spot. It cannot assess whether the visiting time at the tourist spot and the user’s preference match. In [4], the system used the tourist’s Facebook profile to suggest the next tour destination. It considered that not all travelers like the same destinations. However, it did not consider specific issues, such as the evaluation of the attraction and time. The system in [5] considers different routes according to different weather conditions, but the destination of the determined path may not satisfy users’ preferences. In [6], a model was built that allows users to plan a trip for a few days under limited circumstances, such as opening hours, and breaks, such as lunch time. The study was limited to time-related limitations and did not include more realistic limitations; hence, the proposed model does not provide a highly satisfactory path to users. Therefore, in this paper, we consider an algorithm based on the algorithm in [8] to solve this problem. We aim to make recommendations to travelers that have many types of travel purpose, and satisfactory travel plans for firsttime foreign visitors to Japan, even in circumstances in which they do not know the spot of Japan well. The system considers an attraction according to the user’s preference, and only needs the user to input the travel time and the degree of interest in the attraction type. We provide the user with a satisfactory path (it is influenced by the attraction time, the attraction evaluation, the user’s preference for different types of attractions, etc.) selected under factor conditions. In [7], users were provided with a complete travel plan, which could be changed according to their preferred destination. However, we want to offer travel plans for a special group of people, that is, those who are traveling to a location for the first time. For first-time visitors to Japan, it can be difficult to plan a trip without being familiar with the destination. In the following, we briefly describe the system [8]. 2.2

Optimal Travel Route Recommendation System

In [8], the Optimal Travel Route Recommendation System for Tourists’ First Visit to Japan (AOTRRS) was designed to recommend travel plans for people visiting Japan for the first time. AOTRRS [8] used a proposed improved algorithm for personalized route recommendations. The main feature of the algorithm was to prevent foreign tourists from experiencing any difficulties in locating their destination when they visit Tokyo for the first time. The degree of purpose was used to determine the best fit for user travel preference tourism routes. The cosine similarity algorithm was used to determine routes with the same

An Optimal Route Recommendation Method

401

degree of purpose. After the same rate of tourism sites was found, the scores for each route were increased or decreased using a fitness function, which included restrictions on time, meals, amount spent, return to hotel, and travel time. Thus, our recommended tourism route had a relatively high satisfaction rate and was reasonable for the user’s experience. The results showed that users were relatively satisfied with the recommended spots. The program-recommended route was faster and more reasonable than the user-defined route because it met the information restrictions entered by the user. Additionally, we screened out many unreasonable routes. However, AOTRRS [8] still has several shortcomings. For example, the total number of spots in this system is 11. If we filter attractions from more candidate sets, then the system computing time may not be ideal. Additionally, AOTRRS [8] was developed for a day’s travel time; hence, the number of spots in the path is generally seven to eight. However, if the user wants to plan a trip that lasts more than three days, 20 to 30 spots are required. At present, the algorithm in AOTRRS cannot solve this problem reasonably. 2.3

Genetic Algorithms

A GA has no constraints regarding solving the optimal problem, and the search process is very flexible because of its evolutionary characteristics. The GA is commonly used to generate high-quality solutions to optimization and search problems [9–13]. The ergodic nature of GA evolution factors at different time periods enables a GA to provide a large number of independent heuristics of a compound structure for various special problems, which makes it effective for searching. There are many excellent applications of GA for solving the path planning problem. In [14], a GA was used to solve public transportation planning problems. The results demonstrated that the GA was robust, and provided good solutions within computational time constraints for the chosen problem. In [15], GA was used to determine the best escape route and evacuate people as quickly as possible from a highrise building in the case of a disaster, such as a fire. In [15], a conventional method was used in the large solution space and the calculation time increased, whereas GA obtained reasonable results in a shorter time. Therefore, in the present paper, we use GA to solve a problem in a large space, which is conducive to determining the optimal result, and the computing time efficiency is high. First, we briefly describe the structure of the GA. There are a number of strategies used to represent a chromosome within a population. A certain number of individuals make up each population, and each individual is a chromosomal entity with characteristics. After the generation of the initial population, according to the principle of the survival of the fittest, the inferior individual gradually evolves and finally the optimal approximate solution is determined. Using the genetic operator of natural genetics, a group of new solution sets is selected according to the fitness of individuals in each problem domain. This is the process of determining good genes from the primary population and producing results that are increasingly close to the optimal solution. Compared to traditional search and optimization procedures, such as calculus-based and enumerative strategies, the GA is robust, global and generally in [16]. In [17], a genetic algorithm with interval valued fitness function is proposed.

402

C. Yuan and M. Uehara

3 System Design The process of GA includes the initial population, selection, crossover, and mutation. First, we set up a population of scenic spots and randomly generate a path population. Using the natural selection rule of GA, the optimal solution is searched from a large number of alternate paths. Next, we explain how to use GA to build a system to search for the global optimal path using the fitness condition. We use a method called maximizing the problem of an objective function. Genetic genes are design variables that take all the zero or one bits, and the target function is the sum of all the individual genes. The general process of the system is divided into three steps: create an initial population, establish a fitness function, and explore the maximum objective function. 3.1

Algorithm Flow

The first step is to define the population. A population contains individuals, each with a set of chromosomes. First, we randomly generate the initial population. A random number object is given a seed value to generate a random sequence of numbers. Generally, an initial population of size N is randomly generated. In this paper, the cases are set to 10 scenic spots and 20 scenic spots. We use the matrix method to calculate the distance between two scenic spots. We set the population to 300 and individuals to 100 in the system, and then search in this state. The flow chart of the proposed algorithm is shown in Fig. 1.

Fig. 1. The flow chart of Genetic Algorithm

An Optimal Route Recommendation Method

3.2

403

Fitness Function and Structural Steps

We evaluate each path using the fitness function. We use the six fitness functions in [8] to screen the aggregate population: the degree of purpose, total travel time, total travel cost, scenic spot evaluation, hotel return time, and gourmet scenic spot time. We choose the path with a high satisfaction rate according to these constraints. We attach weight values to the fitness function according to the preference degree for each factor. Thus, each path in the population has its own fraction value. A GA is used to determine the path with the maximum fraction value. Next, we explain how to use the GA to explore the maximum objective function. First, the symbols defined in this research are represented as follows. fun(i) is the sum score of the fitness function of each route. Weight is the weight value of each path limiting factor. Sim is a similarity because it obtains the same result route as the goal level. T is travel time. Return Time is the time to return to the hotel from the last sight. Start Time is the time of departure from the hotel to the first sights. Sum score of each fitness function: funðiÞ ¼ t  TimPðiÞ þ c  CosPðiÞ þ r  RatPðiÞ þ v  ExpPðiÞ þ b  BacPðiÞ þ m  MeaP(iÞ Weight ¼ ½t; c; r; v; b; m

ð1Þ Search for degree of purpose difference: ExpPðiÞ ¼ simðxij ; yij Þ Pi ¼ ½ðxij ; yij Þ ðj  0Þ

ð2Þ

Time constraints: 0  T  Return Time  Start Time T ¼ departureðiÞ time þ duringðiÞ time þ moveðiÞ time

ð3Þ

In the above formulae, Eq. (1) is the total score for each route, which is obtained by summing the individual fitness functions. Equation (2) is the cosine similarity algorithm is used to calculate the degree of similarity between the rate of the degree of purpose input by the user and the degree of purpose for each route. Thus, the difference between them can be known. If the cosine value is closer to 1, the angle is closer to zero and the result is the more similar value between the two. Equation (3) is a constraint for time. In this paper, we use the DEAP tool to implement the GA. The structural steps are as follows: 1. Determine the types–creator creation and search for the path using the fitness function maximum method. 2. Initialize the individual, population, and fitness functions of the initialization– toolbox registration path.

404

C. Yuan and M. Uehara

3. Iterate the operator: selection, crossover, mutation, and evolution to make the local search a global search whose goal is to approach the optimal solution. 4. Obtain the results sorted by the size of the fitness score. The obtained path result meets the number of spots for which the user searches.

4 Evaluation In this paper, a GA is used to solve the problem that arises when using the basic algorithm. In our previous system [8], we used the basic algorithm to obtain the optimal path to recommend to the user. To verify that the calculated results met the user requirements, we conducted tests. The following is a detailed numerical example to compare the results between the GA and basic algorithm for the recommended sightseeing system for foreigners who come to Japan for the first time. First, we compare the satisfaction rate of the recommended results. 4.1

Recommended Path Satisfaction Test

To reflect whether the system’s recommendation results meet the needs of users, we tested the basic algorithm and GA. Previously, we asked volunteers to test the satisfaction of the recommended path for AOTRRS [8], and the test results showed that the satisfaction rate was 85%. For the tourism route recommendation system based on the GA, we asked three volunteers to test the suitability and satisfaction rate of the recommended path in each case for various degrees of purpose. First, we considered 52 cases, where 47 path results were satisfactory and five were unsatisfactory. Then 100 tests were conducted with different degrees of purpose, and the results were that 93 paths were satisfactory and seven were unsatisfactory. The recommended path for the GA had a 93% satisfaction rate. We tested the path satisfaction of the basic algorithm and GA at the same degree of purpose, and the GA had the highest path satisfaction rate. The GA took advantage of it to create larger spaces for evolutionary selection and determine paths more efficiently. The cases were tested on different populations and individual states. The number of individuals in the population and generation were (100, 50), (300, 100), (500, 300), and (1000, 800). The four degrees of purpose were scenery, gourmet, shopping, and experience. The degrees of purpose for the four classifications were set to (1, 0, 0, 0), (0.8, 0.2, 0, 0), (0.6, 0.4, 0, 0), (0.6, 0.2, 0.2, 0), (0.4, 0.4, 0.2, 0), (0.4, 0.2, 0.2, 0.2). When we analyzed the results of the unsatisfactory recommendation paths for the GA, all cases occurred where the population and individual were (100, 50). From these data, we can see that, for the GA, the larger the population setting, the higher the accuracy of the search path. 4.2

Time Comparison for the Two Algorithms

In this section, we compare the calculation time of the basic algorithm and GA in the case of the same number of attractions. The basic algorithm was tested by determining the calculation time of the path result calculated in [8]. For the GA, the candidate

An Optimal Route Recommendation Method

405

me/s

populations were generated randomly through the collection of spots. After the steps of crossover, mutation, etc., the fitness function was used to calculate the score value of each individual, and the individual with the largest score was displayed. This series of steps was iterated again to obtain the optimal results. We analyzed the time taken to obtain the optimal results. The efficiency of the calculation of the two algorithms was different. We analyzed whether the full permutation algorithm or GA was more advantageous for search problems under large amounts of data by testing the computation time of the permutation algorithm and GA. Figure 2 shows the comparison of the calculation time for the two algorithms.

3000 2500 2000 1500 1000 500 0

1

2

3

4

5

6

7

8

9

10

spot number basic version

GA version

Fig. 2. Comparison of the calculation time for various numbers of sightseeing spots for the basic algorithm and GA

The blue polyline represents the basic algorithm and the orange polyline represents the GA. From the figure, we can clearly see that the calculation time greatly increased as the number of scenic spots increased in the case of the full permutation algorithm, whereas the GA had no obvious change. To make the comparison between the two methods clearer, in Fig. 2, we show the calculation time of both algorithms for more specific data, that is, the calculated time variation trend for a number of different scenic spots. We tested the program calculation time under the condition that the number of attractions increased. The number of spots refers to the number of attractions in the recommended route, which means the number of spots that can be visited during one day. In the next test, we searched the combined paths of one to ten scenic spots under the condition that the number of scenic spots was [1, 11], and found the optimal path using the fitness function. From Fig. 3, we can see that if the number of scenic spots in the range was 1  S  5, then the basic algorithm was clearly faster than the GA, where S is the number of spots. Additionally, the time difference was huge. However, as the number of scenic spots increased, the calculation time of the basic algorithm increased rapidly, and it was impossible to calculate a path in a larger scope. By contrast, the calculation time of the GA increased at a steady rate. The data proves that the calculation time of the basic algorithm was more advantageous when the search data range was small. However, as the data range increased, the GA was more advantageous in terms of

406

C. Yuan and M. Uehara

computational efficiency. To improve the system, we increased the number of spots to 20 and determined which algorithm was more efficient in the case in which the number of spot candidates increased. Figure 4 shows the test results.

Fig. 3. Comparison of the calculation time for the basic algorithm and GA

Fig. 4. Comparison of the calculation time when the spot number is increased for the basic algorithm and GA

From Fig. 4, we can see that the basic algorithm was not efficient and took a great deal of time under the condition that the search path contained more than three scenic spots in the range of 20 candidate spots. If the search was for more than six spots, then the general algorithm almost reached its limit, that is, it was not reasonable to search for a highly satisfactory path under a large number of scenic spot data using the basic algorithm. By contrast, the GA results showed that it was relatively stable. As the data for the candidate spots increased, the calculation time was relatively reasonable.

An Optimal Route Recommendation Method

407

Additionally, this did not affect the search for the optimal path results using the GA. The results show that the GA was superior to the common algorithm in the case in which the number of scenic spots that could be visited expanded within one day, or when the number of candidate attractions increased. Additionally, as the number of scenic spots increased, the calculation time of GA did not increase substantially. Therefore, it is not necessary to enter the destination search condition in a system that only uses various fitness functions and weight adjustments. The GA is more suitable for applications.

5 Summary In this paper, we proposed a sightseeing recommendation system for tourists visiting Japan for the first time. In a previous study, we used a basic algorithm. First, all candidate scenic spots were arranged and then the composed path set was screened. The results of our tests showed that the basic algorithm had limitations. Additionally, it was not suitable for searching for optimal results under a large amount of data. Thus, we used a GA to solve this problem. The test results showed that the GA solved the problem and searched for the best results quickly for any number of scenic spots that could be visited in one day. However, the basis algorithm was faster in the case of a small number of scenic spots. As the number of scenic spots increased, the GA became more efficient than the basic algorithm, and the increase in the number of candidate sites did not greatly affect its calculation speed. The system provided the optimal path for the user more intelligently using the GA. However, the weighting that our system set for each restriction was caused by the effect of each factor on the path. Therefore, the weight value of each factor could not be satisfied for different users. The proposed method still requires improvement. The next task is how to adapt the weight value of each factor to the different needs of different users, and recommend a tourism plan with a high satisfaction rate according to the user’s pursuit of individual personalized travel. Acknowledgments. We thank Maxine Garcia, PhD, from Edanz Group (www.edanzediting. com/ac) for editing a draft of this manuscript.

References 1. Tokyo Tourism Industry Promotion Action Program (2017). http://www.metro.tokyo.jp/ tosei/hodohappyo/press/2016/05/documents/70q5u101.pdf 2. Japan Tourism Agency Ministry of Land, Infrastructure, Transport and Tourism. Revision of “Basic Plan for Promotion of Tourism Nation” (Established April 25, Heisei 29). http:// www.mlit.go.jp/common/001182997.pdf 3. Nakajima, Y., Niitsuma, H., Ohta, M.: Travel route recommendation using tweets with location information. Res. Rep. Database Syst. (DBS) 158(28), 1–6 (2013) 4. González-Vélez, H.: Tourist Destination Recommendation System Based on User Facebook Profile National College of Ireland Project Submission Sheet – 2016/2017 School of Computing

408

C. Yuan and M. Uehara

5. Arts, M., Yoshihiro, M., Naoki, S., Minoru, I.: Algorithm for composing satisfactory tour schedules for fickle weather. Res. Rep. Math. Model. Probl. Solving (MPS) 75(3), 1–6 (2009) 6. Vansteenwegen, P., Souffriau, W., Berghe, G.V., Van Oudheusden, D.: The city trip planner an expert system for tourists. Expert Syst. Appl. 38(6), 6540–6546 (2011) 7. Chiang, H.-S., Huang, T.-C.: User-adapted travel planning system for personalized schedule recommendation. Inf. Fusion 21(1), 3–17 (2015) 8. Chen, Y., Minoru, U.: An optimal travel route recommendation system for tourists’ first visit to Japan. In: Barolli, L., Takizawa, M., Xhafa, F., Enokido, T. (eds.) Advanced Information Networking and Applications, AINA 2019. Advances in Intelligent Systems and Computing, vol. 926, pp. 872–882. Springer, Cham (2020) 9. Basu, A., Vanajakshi, L.: Transportation Research Record, Transportation Research Board, National Research Council, Washington, D.C., 15 November 10. Mitchell, M.: An Introduction to Genetic Algorithms. MIT Press, Cambridge (1996). ISBN 9780585030944 11. Abbaspour, R.A., Samadzadegan, F.: Itinerary planning in multimodal urban transportation network. J. Appl. Sci. 9(10), 1812–5654 (2009) 12. Deng, Y., Hu, S.: Route optimization of multi-modal travel based on improved genetic algorithm. In: 2011 International Conference on Transportation, Mechanical, and Electrical Engineering (TMEE) 16–18 December, Changchun, China (2011) 13. Sun, X., Wang, J., Wu, W., Liu, W.: Genetic algorithm for optimizing routing design and fleet allocation of freeway service overlapping patrol. Sustainability 10, 4120 (2018). https:// doi.org/10.3390/su10114120 14. Johar, A., Jain, S.S., Garg, P.K.: Transit network design and scheduling using genetic algorithm – a review. Int. J. Optim. Control: Theor. Appl. 6(1), 9–22 (2016). © IJOCTA ISSN 2146-0957 eISSN 2146-5703. https://doi.org/10.11121/ijocta.01.2016.00258. http://www.ijocta.com 15. Dogan, S., Yilmaz, H.: A multi-objective route planning model based on genetic algorithm for cuboid surfaces, 06 August 2018 16. Baker, E.J.: Reducing bias and inefficiency in the selection algorithm. In: Proceedings of the Second International Conference on Genetic Algorithms, Cambridge, MA, USA, 28–31 July 1987, pp. 14–21 (1987) 17. Chipperfield, A., Fleming, P., Pohlheim, H., Fonseca, C.: Genetic Algorithm Toolbox for Use with MATLAB; Department of Automatic Control and System Engineering, University of Sheffield: Sheffield, UK (1994)

Artificial Intelligence Technique for Optimal Allocation of Renewable Energy Based DGs in Distribution Networks Zia Ullah1(&), M. R. Elkadeem1,2, and Shaorong Wang1 1

Key Laboratory of Advanced Electromagnetic Engineering and Technology, Huazhong University of Science and Technology, Wuhan 430074, China {ziaullah,elkadeem}@hust.edu.cn, [email protected] 2 Electrical Power and Machines Engineering Department, Faculty of Engineering, Tanta University, Tanta 31521, Egypt

Abstract. This paper proposes the artificial intelligence technique based on hybrid optimization phasor particle swarm optimization and a gravitational search algorithm, called PPSO-GSA for optimal allocation of renewable energybased distributed generators (OA-RE-DGs), particularly wind and solar power generators, in distribution networks. The main objective is to maximize the techno-economic benefits in the distribution system by optimal allocation and integration of RE-DGs into distribution system. The proposed PPSO-GSA is implemented and validated on 94-bus practical distribution system located in Portuguese considering single and multiple scenarios of RE-DGs installation. The results reveal that optimizing the location and size of RE-DGs results in a substantial reduction in active power loss and yearly economic loss as well as improving system voltage profile and stability. Moreover, the convergence characteristics, computational efficiency and applicability of the proposed artificial intelligence technique is evaluated by comparative analysis and comparison with other optimization techniques. Keywords: Artificial intelligence technique
Renewable energy
Distributed generators
Phasor particle swarm optimization
Gravitational search algorithm

1 Introduction Worldwide, increasing electricity demands enabled a significant escalation in electric power production and such a load expansion that causes to influence the economies of developed countries to be motivated towards optimal planning of electric power generation sources and its utilization. Recently, a major part around 75% of electrical power is producing via fossil fuel-based energy sources [1] and it is directly related to natural reserves which is expected to be reduced day by day [2] also increase in the prices [3]. The production of electricity using conventional methods are causes to produce carbon emission [4]. Moreover, the large power plants are commonly located at long distance from the load centers which lost about 15% of power in transmission lines which affect the system performance and customers appliances [5]. © Springer Nature Switzerland AG 2020 L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 409–422, 2020. https://doi.org/10.1007/978-3-030-33506-9_36

410

Z. Ullah et al.

Hence, it is worth mentioning that the optimal planning and utilization of renewable energy sources (RE-DGs) such as wind turbine, solar photovoltaic has been considered globally to tackle and meet the high electricity demands at reasonable price [6]. The integration of RE-DGs units in distribution system plays an important role in term of active loss minimization, but still it has a challenge of optimal planning that needs to be addressed thoroughly [7, 8]. The optimal planning and integration of RE-DGs assist to maximize the technical and economic benefits of the power system. Generally, the objectives of optimal planning and integration of RE-DGs in distribution networks include power loss minimization, improving voltage profile, system stability, and operational cost minimization [9–14]. Recently, optimal planning and integration of RES-DGs in distribution networks are addressed and proposed the solution by using different optimization techniques that have been reviewed in [15–18], where different distribution systems are considered along with objective functions. Furthermore the objective functions are formulated for optimal allocation and sizing of RE-DG units and the work done is focused on strengthening the technical and economic implications in the distribution networks in terms of key objectives; power loss minimization, voltage profile improvement, energy loss, and smoothing buses voltages. In this paper, a novel artificial intelligence technique namely PPSO-GSA is developed to solve the problem of OA-RE-DGs in distribution networks. The PPSOGSA algorithm is employed to enhance the techno-economic benefits in distribution system by employing RE-DGs integration into the distribution system. Technoeconomic benefits include, minimizing active power loss, improving buses voltage, voltage stability and yearly economic loss. Different scenarios of PV and WT integration, considering single and multiple DG units are examined on practical Portuguese RDS and results are compared with the existing literature. The rest of the paper is organized as follows: Sect. 2 defines the problem formulation of the OA-RE-DGs. Proposed PPSO-GSA approach, and its implementation procedure is described in Sect. 3. Section 4 presents the simulation results and comparison with literature; finally, the conclusions are given in Sect. 5.

2 Problem Formulation The problem of optimal placement and sizing of renewable energy based DG units (OA-RE-DGs) can be considered as a constrained, non-linear, discrete optimization problem [19]. The problem is formulated as an optimal power flow model [20], which has a single or multi-objective optimization function. The considered objectives should be properly optimized while satisfying the system operational constraints. 2.1

Objective Function

The key objective of this study is to maximize the techno-economic benefits in the distribution system by integration of RE-DG units into distribution networks. Several performance indicators should be taken into account while solving the problem of OARE-DGs. Therefore, minimizing active power losses, optimizing the annual profit as

Artificial Intelligence Technique for Optimal Allocation

411

well as improving voltage and voltage stability index of system buses are considered as objectives of this study. This multi-objective function can be expressed using a weighted sum method [5, 21, 22].
 ObjF ¼ min x1  Ploss þ x2  VDT þ x3  ð1=OVSIÞ þ x4  TAES

ð1Þ

where, Ploss, VDT, OVSI, and TAES represents the reduction of the total active power losses, improvement of buses voltage, strengthening of buses stability, and saving in yearly economic loss, respectively. In addition, x1 ; x2 ; x3 and x4 are weighting factors, in which the total sum of absolute values of the weights associated with each objective should equal to 1.0. The four objectives studied in this work can be expressed mathematically as: Ploss ¼

NBR X

Ploss;b

ð2Þ

b¼1


 Ploss;b ¼ gij;b  ½Vi2 þ Vj2  2  Vi  Vj  cos hi  hj 

ð3Þ

where, Ploss;b is the active power loss of branch bth, gij;b is the conductance of branch b connecting the ith and jth buses, Vi ; Vj ; hi ; and hj are the voltage magnitudes and voltage angles at bus i and j, respectively. VDT ¼

NB   X  ref  Vi  Vi 

ð4Þ

i¼1

where, Vi is the magnitude of the voltage of the bus number i, while vref i denoting the magnitude of the reference voltage at the same bus, which is usually equal to 1 p.u. OVSI ¼

NB X

jVSIðjÞj

ð5Þ

i¼2

VSIðjÞ ¼ jVðiÞj4 4  ½Pij;b  xij;b  Qij;b  rij;b 2  4  ½Pij;b  rij;b þ Qij;b  xij;b   jVðiÞj2 ð6Þ where, OVSI is the overall voltage stability index of the whole power network, while VSI(j) is the stability index of jth bus connected with the ith bus through a branch bth of a resistance rij;b and reactance xij;b . Also, Pij;b and Qij;b respectively are the active and reactive power flow from ith bus to jth bus. Where the bus with least VSI, corresponds to the most sensitive bus. TAES ¼ AELT;noDG  AELT;DG

ð7Þ

412

Z. Ullah et al.

where, TAES is the total annual economic saving gained when RES is integrated into the power network. AELT;noDG and AELT;DG are the total annual economic loss without and with DG and can be expressed as follows: AELT;noDG ¼ PlossnoDG  CE  8760 AELT;DG ¼ PlossDG  CE  8760 þ CDG 

N DG X

ð8Þ !

PDG;m =TDG

ð9Þ

m¼1

where, CE is the average cost of energy loss per kWh. CDG and TDG are the cost of injected power via DGs per kW and the total DG lifetime in years, respectively. 2.2

Equality Constraints

The power balance in the distribution network has considered as equality constraints. Pslak þ

N DG X

PDG;m ¼

m¼1

Qslak þ

N DG X

NB X

PDG;i þ

i¼1

QDG;m ¼

m¼1

NB X

NBR X

PLoss;b

ð10Þ

QLoss;b

ð11Þ

b¼1

QDG;i þ

i¼1

NBR X b¼1

where, Pslak , Qslak , PDG;i , and QDG;i are the active and reactive power injected by the slack (i.e. swing) bus and ith DG, respectively. PDG;m and QDG;m are the active and reactive power demand, where PLoss;b and QLoss;b are the active and reactive power loss of bth branch. 2.3

Inequality Constraints

• The voltage limits must be as follow: Vmin  Vi  max

i 8 NB

ð12Þ

• The apparent power flow limits need to be as below: Sb  Smax b

b 8 NBR

ð13Þ

• The DG capacity constraints; i.e. DG penetration level (l) and the maximum size of individual units. Also, the DG power factor (PFDG;m ) should be within safe limits as expressed below: N DG X m¼1

SDG;m  l

NB X i¼1

SD;i

ð14Þ

Artificial Intelligence Technique for Optimal Allocation

SDG;m  Smax DG;m

m 8 NDG

min max PFDG;m  PFDG;m  PFDG;m

SDG;m ¼

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi P2DG;m  Q2DG;m

SD;i ¼

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi P2D;i  Q2D;i

413

ð15Þ ð16Þ ð17Þ ð18Þ

3 Proposed Method In this study the artificial intelligence technique based on combined metaheuristic optimization approach PPSO and GSA algorithm, namely PPSO-GSA algorithm is proposed [23]. The proposed model is able to optimally allocate the RES-DG units in distribution networks which has extensive properties of PPSO [24] and GSA [25]. In particular the PPSO-GSA considering the phase angle (h) and being represented by trigonometric functions (i.e., sine and cosine) associated with phase angle h [26]. Furthermore, it uses the stochastic search method and search agents are considered as vectors with control variables (xi ) in n dimension. Utilizing stochastic search operators on the current population, the proposed AI technique create a new population using a successive iterative correction scheme [27]. The following equations have been considered while program problem optimization. vi ðiter þ 1Þ ¼ r1  vi ðiter Þ þ r2  C1 ðiterÞ  ai ðiter Þ þ r3  C2 ðiterÞ  ðgbestðiter Þ  xi ðiter ÞÞ

ð19Þ

C1 ðiterÞ ¼ jcos hi ðiter Þj2sin hi ðiterÞ

ð20Þ

C2 ðiterÞ ¼ jsin hi ðiter Þj2cos hi ðiterÞ

ð21Þ

xi ðiter þ 1Þ ¼ xi ðiter Þ þ vi ðiter þ 1Þ

ð22Þ

where, the acceleration of agent ai , being modified utilizing the equations given in [25] and r1, r2, and r3 are random numbers in the range of [0, 1]. hi ðiter þ 1Þ ¼ hi ðiter Þ þ jcos hi ðiter Þ þ sin hi ðiter Þj  2p

ð23Þ

The program execution of the proposed PPSO-GSA technique follows many steps and procedures as per given flow chart illustrated in Fig. 1.

414

Z. Ullah et al.

Fig. 1. Flow diagram of PPSO-GSA implementation for OA-RE-DGs

4 Simulation Results and Discussion 4.1

Simulation Strategies

In this paper, PPSO-GSA is implemented on practical distribution system of 94 bus located in Portuguese, the proposed model considering the allocation of renewable energy sources particularly, PV and WT, with different generation capacities. The input data and study parameters are shown in Table 1. Moreover, two cases, including single and multiple installation of PV and WT units are examined to evaluate the performance of the proposed methodology and its positive impacts on system indices.

Artificial Intelligence Technique for Optimal Allocation

415

Table 1. Input parameters for simulation Description Set value(s) PPSO-GSA parameters: 100 Niter,max Npop 50 Nruns 50 System inequality constraints: Bus voltage limits (p.u.) ±10% RE-DG size limits (MVA) 0  SDG  5 RE-DG PF limits [28] 1  PFPV  1 & 0.65  PFWT  1 Max. No of RE-DG units/bus 2 Cost data [21]: CDG ($/kW) 30 TDG (years) 10 CE ($/kWh) 0.05 *PV inject active power only *WT inject active and reactive power *Each bus can integrate one DG only

4.2

Practical Portuguese 94-Bus RDS

In order to evaluate the effectiveness of the artificial intelligence technique the proposed PPSO-GSA has implemented on Practical 94-bus shown in Fig. 2. The line and load data of considered system has obtained from [29], and the load model of the distribution system is assumed as a constant power load [30]. Using the base case without DGs integration and 100% loading conditions the results are presented in the Table 2. It interprets the system specifications and initial power flow results without any RE-DG unit contribution. Moreover, employing the proposed model considering RE-DG units integration along with different scenarios (i.e. single and multiple RE-DG units) the results obtained are shown in Table 3. It can be seen that integrating the PV and WT generators into the system at optimal locations, the voltage profile has significantly improved and the stability index increased dramatically as compared to the base case without RE-DGs as presented in Figs. 3 and 4, respectively.

Fig. 2. Single line diagram of practical 94-bus distribution system located in Portuguese

416

Z. Ullah et al. Table 2. Test System results without RE-DGs Description Portuguese 94-bus system System specifications: NB 94 Nbr 93 Vsys (kV) 15 Base MVA 100 Initial power flow results: 4.797 + j 2.323 Sload (MVA) Ploss (kW) 362.86 Qloss (kVAr) 504.04 Vmin,bus (p.u.) 0.5183,92 Vmax,bus (p.u.) (without slack bus) 0.9804,2 VDT (p.u)* 9.126 OVSI 62.265 AELT,noDG ($) 158932.68 CPU time (s) 6.831223

Fig. 3. Voltage profile without and with RE-DG unit’s integration

Fig. 4. Bus VSI profile without and with RE-DG unit’s integration.

Artificial Intelligence Technique for Optimal Allocation

417

Table 3. Results obtained by proposed PPSO-GSA for optimal allocation of RE-DGs Description

Single RE-DG (DGs No/Type) 1/PV 1/WT

Multiple RE-DGs (DGs No/Type) 2/PV

2/WT

3/PV

3/WT

Ploss (kW) Vmin/bus (p.u.) Vmax/bus (p.u.) DG_MVA/PF, Bus No

132.395 0.9301/66 0.9968/2 2.636/1,19

81.269 0.9522/66 0.9981/19 2.9683/0.893,19

79.255 0.9332/92 0.9974/2 1.978/1,20 1.726/1,58

21.719 0.9802/92 0.9997/20 2.189/0.898,20 1.905/0.898,58

OVSI VDT (p.u) AELT,DG ($) TAES ($) CPU time (s)

76.384 4.491 57996.939 34412.301 13.96653

86.372 1.712 35603.76 56805.48 13.906

78.814 3.790 34719.53 57689.70 20.25375

90.445 0.636 9519.01 82890.22 16.241

72.932 0.9442/92 0.9974/2 1.467/1,19 0.557/1,25 1.6987/1,58 79.558 3.570 31949.47 60459.76 23.038

14.856 0.9892/42 0.9993/19 1.62/0.899,19 0.61/0.899,25 1.87/0.895,58 91.286 0.420 6508.49 85900.74 17.982

In the first scenario, considering single RE-DG unit integration, the optimal location has identified as bus number 6 for PV and WT installation. Also optimized the sizes of PV and WT at power factor of 0.8239. The proposed optimization reduces the high power loss to 132.95 kW and 81.269 kW by suggesting the PV and WT, respectively. Moreover, Fig. 5 shows the branch power loss variation with and base case without the integration of PV and WT units, where the current flow and power loss in branches 1 to 19 is decreased drastically after PV or WT unit’s penetration. This optimal planning reduces the yearly economic loss to 57996.939 $ and 35603.761 $ from 158932.68 $ for the base case without RESs.

Fig. 5. Branch active power loss without and with RE-DG units

418

Z. Ullah et al.

Furthermore, the minimum voltage magnitude has improved to 0.9301 p.u and 0.9522 p.u using PV and WT respectively, together with increasing the VSI of all system buses. Thus, strengthening system power quality and stability. It should be mentioned that the optimal planning of RE-DGs in active distribution network has offline implementation nature; thus, the time of processing is not considered a concern [30]. Nevertheless, the PPSO-GSA consumes a small value of 13.966 s and 13.906 s of CPU time with PV and WT, respectively. In the second scenario, the integration of multiple RE-DG units (i.e., two and three units) is considered, and its implications are investigated. For the case of two PVs, buses 20 and 58 are selected as optimal locations for PV unit integration with a capacity of 1.9785 MW and 1.7267 MW, respectively. On the other hand, the optimal locations for three PV units are the buses 19, 25 and 58 and the PVs capacities are 1.4679, 0.5571 and 1.6987 MW, respectively. It is worth mentioning that the active power loss has significantly reduced to 79.255 kW and 72.932 kW by two PV and three PV units, respectively. On the same line, the yearly savings enhanced to 57689.709 $ and 60459.762 $ are achieved with two PV and three PV units, respectively. Furthermore, while optimal planning of two WT units, buses 20 and 58 are also designated as optimal locations with optimal WTs power capacities of 2.1898 MVA, 1.9054 MVA with power factors equals 0.8985 and 0.8983, respectively. As a result, the network losses are minified to 21.719 kW. Also, the optimal integration of three WT units at buses 19, 25 and 58 intensively decreased the power loss to only 14.856 kW. Moreover, the obtained results also show that the annual economic savings are increased to 82890.229 $ and 85900.741 $, respectively. In general, due to the capability of WTs to supply reactive power, it gives better voltage profile and noticeably enhances system stability compared to PV (realize Figs. 5 and 6), where the minimum voltage magnitudes are achieved 0.9802 p.u. at bus 20 and 0.9892 p.u at bus 42, together with a considerable increase in the OVSI of 90.445 and 91.286 by the contribution of 2 and 3 WTs, respectively. Clearly, it can be seen from Table 3 that whenever the number of PVs and WTs (i.e., the penetration level of RESs) is increased, a considerable improvement in the techno-economic performance is achieved, which is presented by the bar graph given in Fig. 6. Table 4 arranges the numerical results of the proposed PPSO-GSA method and comparison with other existing techniques reported in the literature for 94-bus RDS. The fair comparison shows that the proposed algorithm provides global optimal solutions and better results than other methods and verify the capability of PPSO-GSA to solve the planning problem and improve the system performance.

Artificial Intelligence Technique for Optimal Allocation

419

Table 4. Comparison of PPSO-GSA results with other methodologies DGs No/Type 1/PV

1/WT 2/PV

3/PV

BSOA [28] KHA [31] SKHA [31] PPSO-GSA BSOA [28] PPSO-GSA KHA [31]

Ploss (kW) 153.56 132.3957 132.3957 132.395 85.13 81.269 86.6475

Vmin (p.u.) 0.9276 0.9301 0.9301 0.9301 0.9519 0.9522 0.9284

Vmax (p.u.) 0.9967 0.9968 0.9968 0.9968 0.998 0.9981 0.9974

SKHA [31]

79.2549

0.9301

0.9968

PPSO-GSA

79.25

0.9332

0.9974

KHA [31]

74.4907

0.934

0.9976

SKHA [31]

73.1022

0.9437

0.9974

PPSO-GSA

72.932

0.9442

0.9974

Methodology

RE-DG data (kVA/PF, bus) 2398.5/1,21 2636.0175/1,19 2636.018/1,19 2636/1,19 2398.5/0.532,18 2968.3/0.894,19 1.940/1,56 1.752/1,83 1.7260/1,58 1.978/1,20 1978.5/1,20 1726.7/1,58 0.955/1,10 1.833/1,20 1.285/1,58 0.498/1,25 1.575/1,19 1.638/1,58 1.4679/1,19 0.5571/1,25 1.6987/1,58

CPU time (s) 118.00 22.570 21.170 13.966 234.01 13.906 20.626 20.298 20.253 19.249

19.046

23.038

Fig. 6. Techno-economic indices achieved by RE-DG units integration in 94-bus RDS

420

4.3

Z. Ullah et al.

Performance Evaluation of the Proposed Algorithm

In general, the meta-heuristic algorithms are characterized by its randomness. Therefore, many trials have been made to prove the robustness of the proposed PPSO-GSA with 20 independent runs. Samples of the optimization objective convergences of the Ploss is given in Fig. 7. The results elucidate that the PPSO-GSA accelerates to the near-optimal solution smoothly, robust, and having steady convergence characteristics.

Fig. 7. Convergence rate of PPSO-GSA algorithm

5 Conclusions In this paper, the artificial intelligence technique based on PPSO and GSA, namely PPSO-GSA has utilized for the RE-DG unit’s allocation in the distribution system. The developed model optimizes the sizing and locations of RE-DG units with the main objective of techno-economic binifit maximation in distribution system, The key highlights from the obtained results are: • The PPSO-GSA provides better solution to the planning problem of RE-DGs either in single or multiple scenarios of DGs integration. • The contribution of PV and WT power generators in the distribution system and its optimal allocation enhances the system performance and power quality in term of power loss reduction and voltage profile improvement. • The more increase in the penetration of RE-DGs, the more melioration in technical and financial indices of the distribution system, especially with the case of WTs. • In comparison with the literature, thank to the artificial intelligence technique that provides better results and high-quality solutions in all cases. Acknowledgments. The authors would like to thank the State Key Laboratory of Advanced Electromagnetic Engineering and Technology, Huazhong University of Science and Technology for providing the essential facilities.

Artificial Intelligence Technique for Optimal Allocation

421

References 1. IRENA. Global Energy, a Roadmap to 2050 by Irena 2018. http://www.irena.org 2. Dawoud, S.M., Lin, X., Okba, M.I.: Hybrid renewable microgrid optimization techniques: a review. Renew. Sustain. Energy Rev. 82, 2039–2052 (2018) 3. Djelailia, O., Kelaiaia, M.S., Labar, H., Necaibia, S., Merad, F.: Energy hybridization photovoltaic/diesel generator/pump storage hydroelectric management based on online optimal fuel consumption per kWh. Sustain. Cities Soc. 44, 1–15 (2019) 4. Quek, T.Y.A., Ee, W.L.A., Chen, W., Ng, T.S.A.: Environmental impacts of transitioning to renewable electricity for Singapore and the surrounding region: a life cycle assessment. J. Clean. Prod. 214, 1–11 (2019) 5. Poornazaryan, B., Karimyan, P., Gharehpetian, G.B., Abedi, M.: Optimal allocation and sizing of DG units considering voltage stability, losses and load variations. Int. J. Electr. Power Energy Syst. 79, 42–52 (2016) 6. Kosai, S.: Dynamic vulnerability in standalone hybrid renewable energy system. Energy Convers. Manag. 180, 258–268 (2019) 7. Xia, S., Bu, S., Wan, C., Lu, X., Chan, K.W., Zhou, B.: A fully distributed hierarchical control framework for coordinated operation of DERs in active distribution power networks. IEEE Trans. Power Syst. 8950 (2018) 8. Wang, L., Yuan, M., Zhang, F., Wang, X., Dai, L., Zhao, F.: Risk assessment of distribution networks integrating large-scale distributed photovoltaics. IEEE Access 7, 59653–59664 (2019) 9. Xiang, Y., et al.: Optimal active distribution network planning: a review. Electr. Power Compon. Syst. 44(10), 1075–1094 (2016) 10. Qazi, A., Hussain, F., Rahim, N.A.B.D., Member, S.: Towards sustainable energy: a systematic review of renewable energy sources, technologies, and public opinions. IEEE Access 7, 63837–63851 (2019) 11. Razavi, S.E., et al.: Impact of distributed generation on protection and voltage regulation of distribution systems: a review. Renew. Sustain. Energy Rev. 105, 157–167 (2019) 12. Mararakanye, N., Bekker, B.: Renewable energy integration impacts within the context of generator type, penetration level and grid characteristics. Renew. Sustain. Energy Rev. 108, 441–451 (2019) 13. Ghadi, M.J., Ghavidel, S., Rajabi, A., Azizivahed, A., Li, L., Zhang, J.: A review on economic and technical operation of active distribution systems. Renew. Sustain. Energy 104, 38–53 (2019) 14. Carvalho, P.C.M., Costa, R.M.: Implementation and evaluation of the first renewable energy systems technical course in Brazil. IEEE Access 7, 46538–46549 (2019) 15. Mahmoud, P.H.A., Huy, P.D., Ramachandaramurthy, V.K.: A review of the optimal allocation of distributed generation: objectives, constraints, methods, and algorithms. Renew. Sustain. Energy Rev. 75, 293–312 (2017) 16. Nduka, O.S., Pal, B.C.: Quantitative evaluation of actual loss reduction benefits of a renewable heavy DG distribution network. IEEE Trans. Sustain. Energy 9(3), 1384–1396 (2018) 17. Sultana, U., Khairuddin, A.B., Aman, M.M., Mokhtar, A.S., Zareen, N.: A review of optimum DG placement based on minimization of power losses and voltage stability enhancement of distribution system. Renew. Sustain. Energy Rev. 63, 363–378 (2016) 18. Ehsan, A., Yang, Q.: Optimal integration and planning of renewable distributed generation in the power distribution networks: a review of analytical techniques. Appl. Energy 210, 44– 59 (2018)

422

Z. Ullah et al.

19. Mithulananthan, N.: Optimal allocation of distributed generation using hybrid grey wolf optimizer. IEEE Access 5, 14807–14818 (2017) 20. El-Fergany, A.: Multi-objective allocation of multi-type distributed generators along distribution networks using backtracking search algorithm and fuzzy expert rules. Electr. Power Compon. Syst. 44(3), 252–267 (2016) 21. Kumar, S., Mandal, K.K., Chakraborty, N.: Optimal DG placement by multi-objective opposition based chaotic differential evolution for techno-economic analysis. Appl. Soft Comput. J. 78, 70–83 (2019) 22. Ullah, Z., Wang, S., Radosavljevic, J., Lai, J.: A solution to the optimal power flow problem considering WT and PV generation. IEEE Access 7, 46763–46772 (2019) 23. Mirjalili, S., Zaiton, S., Hashim, M.: A new hybrid PSOGSA algorithm for function optimization, no. 1, pp. 374–377 (2010) 24. Ghasemi, M., Akbari, E., Rahimnejad, A., Ehsan, S., Sahand, R., Li, G.: Phasor particle swarm optimization: a simple and efficient variant of PSO (2018) 25. Rashedi, E., Nezamabadi-pour, H., Saryazdi, S.: GSA: a gravitational search algorithm. Inf. Sci. (NY) 179(13), 2232–2248 (2009) 26. Maleki, A., Khajeh, M.G., Ameri, M.: Optimal sizing of a grid independent hybrid renewable energy system incorporating resource uncertainty, and load uncertainty. Int. J. Electr. Power Energy Syst. 83, 514–524 (2016) 27. Radosavljevic, J.: Metaheuristic optimization in power engineering. Institution of Engineering and Technology (2018) 28. El-Fergany, A.: Optimal allocation of multi-type distributed generators using backtracking search optimization algorithm. Int. J. Electr. Power Energy Syst. 64, 1197–1205 (2015) 29. Pires, D.F., Antunes, C.H., Martins, A.G.: NSGA-II with local search for a multi-objective reactive power compensation problem. Int. J. Electr. Power Energy Syst. 43(1), 313–324 (2012) 30. El-Fergany, A.: Study impact of various load models on DG placement and sizing using backtracking search algorithm. Appl. Soft Comput. J. 30, 803–811 (2015) 31. ChithraDevi, S.A., Lakshminarasimman, L., Balamurugan, R.: Stud Krill herd algorithm for multiple DG placement and sizing in a radial distribution system. Eng. Sci. Technol. Int. J. 20(2), 748–759 (2017)

Impact of Sharing Algorithms for Cloud Services Management Lidia Ogiela1(&), Makoto Takizawa2, and Urszula Ogiela1 1

2

Pedagogical University of Krakow, Podchorążych 2 Street, 30-084 Kraków, Poland [email protected], [email protected] Department of Advanced Sciences, Hosei University, 3-7-2, Kajino-cho, Koganei-shi, Tokyo 184-8584, Japan [email protected]

Abstract. In this paper will be presented a classification of sharing algorithms dedicated to secure services management in Cloud Computing. Especially will be described an impact of sharing processes in service protocols, realised in the Cloud. Service management procedures can be used at different levels, so the impact of service management processes will be analysed at all of them. Also will be presented the impact of service management and cloud data sharing protocols, on decision-making processes in management-support systems, using linguistic and biometric threshold schemes. Keywords: Sharing algorithm
Cloud Computing
Data management process
Data management levels
Linguistic threshold scheme
Biometric threshold scheme

1 Introduction Data sharing protocols are very useful in various data security tasks [1, 3, 6, 8, 10–12]. The main features of data sharing protocols have been described in the papers [2, 4, 5, 7, 9], where their advantages and disadvantages are discussed in details. The main advantage of data sharing protocols are implementation of independent algorithms of data sharing and distribution its parts. The choice of implemented protocol can also be random. Random selection of the implemented protocol significantly increases the security of the entire protocol, due to the fact that the breakability of such algorithm is associated with the prediction of all possible solutions. Random selection of the implemented protocol refers to solutions, in which there is more than one possible implementation of the protocol. However, this solution has no restrictions about the maximum number of possible solutions.

© Springer Nature Switzerland AG 2020 L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 423–427, 2020. https://doi.org/10.1007/978-3-030-33506-9_37

424

L. Ogiela et al.

It is the result of the possibilities of various solutions, such as: • • • •

selection of the optimal data sharing protocol, choice of the distribution of parts of the shared secret – equal or privileged, choice of the method of privileged distribution of the part of the shared secret, choice of type of data sharing protocol: – classic, – linguistic, – biometric, – mixed linguistic-biometric, • choice of the method of linguistic shadow management: – with division process – equal or privileged, – without division processes, • choice of biometrics for shared secret shadows: – marking by the same type of biometric, – marking by the different biometrics, • in mixed protocols, the choice of how to combine possible solutions. The variety of secret sharing protocols creates wide possibilities of using individual solutions in their most favorable configurations.

2 Classification of Sharing Algorithms in Cloud Computing The general classification of data sharing protocols dedicated to Cloud Computing processes is as follows: • (m, n)-threshold schemes that perform data encryption tasks through their distribution of secret parts between all participants in the protocol – in this class there may be a protocol of equal (one or more secret parts) distribution of shadows between all participants of the protocol, or privileged shadow distribution (depending on the algorithm defined) [1, 10–12], • linguistic threshold schemes – taking into account the linguistic description (using grammar descriptions – sequential, tree or graph grammars) of a shared data used to describe the meaning of encrypted secret and the multilevel process of their reproduction using of knowledge (at various knowledge levels – basic, expert and specialist) [2–4, 6, 8], • biometric threshold schemes – performing tasks of unambiguously assigning part of the secret to its owner by biometric data (by standard or non-standard biometrics, by the same type for all protocols’ participants, or different) [5, 7], • linguistic-biometric threshold schemes – combining the possibilities of linguistic and biometric solutions in one protocol [8, 9]. The presented classes of threshold schemes schematically presents Fig. 1.

Impact of Sharing Algorithms for Cloud Services Management

425

Fig. 1. The general threshold schemes classification.

All of the above solutions can be used in data management tasks both within a given structure (organization, enterprise, etc.) and in structures external to a given entity. A special case of such solution is the structure of the Cloud. From the data management processes point of view, this is an example of a multilevel structure, in which the basic level is the entity (organization, enterprise, etc.), the higher level refers to the fog level, and the highest level is attributed to the Cloud. This multi-level structure is characterized by the ability to implement data management tasks (including a data, information, services, secrets, etc.) both independently at each of its levels, and concurrently at all of them. This solution makes it an example of a universal and at the same time the most complex data management structure. The universality of the used solutions refers to the assessment of their impact and importance for increasing the efficiency of existing data management processes.

3 Impact of Sharing Algorithms in Cloud Computing Assessment of the impact of proposed solutions in supporting data management processes in multi-level structures (with cloud structure) includes analysis of all types of data sharing schemes discussed in the previous section. The main characteristic of this impact presents Table 1. Presented new classes of data sharing algorithms, including linguistic and biometric algorithms, provide the best opportunities in supporting data management processes. They allow the implementation of these processes from different levels of multi-level structures, while maintaining the possibility of independent implementation of data management processes at each of the selected levels. High computational complexity also guarantees the security of processed and encrypted data.

426

L. Ogiela et al. Table 1. The main characteristic features of data sharing algorithms. Linguistic threshold schemes

Biometric threshold schemes

– + – –

Classic (m, n)threshold schemes + + + +

+ + + +

+ + + +

Linguisticbiometric threshold schemes + + + +

–

+

+

+

+

–

+

+

+

+

–

–

+

–

+

–

–

+

–

+

– –

– –

– –

+ +

+ +

–

–

–

+

+

–

+

+

+

+

–

+

+

+

+

–

+

+

+

+

–

+

+

+

+

Classic cryptographic protocols Data management Data security Data sharing Equal distribution of secret parts Privileged distribution of secret parts Possibility of using different data description structures Linguistic data description Description of the secret at various levels of knowledge Biometric data marking Use of standard biometrics Use of non-standard biometrics Division of data in the entity Division of data in the fog level Division of data in the cloud level Implementation of protocols at various levels of multi-level structures

4 Conclusions This paper presents the general classification of currently know secret sharing protocols dedicated to securing different types of data as well as services, information and secrets. The main idea of this paper was to assess the capabilities of all classes of protocols used to securing data, especially in multi-level structures. Especially was described an impact of sharing stage in management processes, realised in the Cloud. Comparative analysis were carried out on new classes of data sharing protocols such as biometric and linguistic protocols and their combined version. Undoubtedly,

Impact of Sharing Algorithms for Cloud Services Management

427

the optimal solution is the use of linguistic-biometric data sharing protocols due to the fact that they provide both the possibility of meaningful description of a secret and personal marking of its individual parts. Acknowledgments. This work has been supported by the National Science Centre, Poland, under project number DEC-2016/23/B/HS4/00616. This work was supported by JSPS KAKENHI grant number 15H0295.

References 1. van Menezes, A., Oorschot, P., Vanstone, S.: Handbook of Applied Cryptography. CRC Press, Waterloo (2001) 2. Ogiela, L.: Intelligent techniques for secure financial management in cloud computing. Electron. Commer. Res. Appl. 14(6), 456–464 (2015) 3. Ogiela, L.: Advanced techniques for knowledge management and access to strategic information. Int. J. Inf. Manage. 35(2), 154–159 (2015) 4. Ogiela, L.: Cryptographic techniques of strategic data splitting and secure information management. Pervasive Mob. Comput. 29, 130–141 (2016) 5. Ogiela, L., Ogiela, M.R.: Bio-inspired cryptographic techniques in information management applications. In: IEEE 30th International Conference on Advanced Information Networking and Applications (IEEE AINA), Switzerland, 23–25 March 2016, pp. 1059–1063 (2016) 6. Ogiela, L., Ogiela, M.R.: Insider threats and cryptographic techniques in secure information management. IEEE Syst. J. 11, 405–414 (2017) 7. Ogiela, L., Takizawa, M.: Personalized cryptography in cognitive management. Soft. Comput. 21, 2451–2464 (2017) 8. Ogiela, M.R., Ogiela, U.: Secure information management in hierarchical structures. In: Kim, T.H., Adeli, H., Robles, R.J., et al. (eds.) 3rd International Conference on Advanced Science and Technology (AST 2011), Jeju Island, South Korea, 15–17 June 2011. Communications in Computer and Information Science, vol. 195, pp. 31–35 (2011) 9. Ogiela, U., Takizawa, M., Ogiela, L.: Visual captcha application in linguistic cryptography. Concurr. Comput.: Pract. Exp. J. 30(2), 1–8 (2018). https://doi.org/10.1002/cpe.4362 10. Schneier, B.: Applied Cryptography: Protocols, Algorithms, and Source Code in C. Wiley, Hoboken (1996) 11. Shamir, A.: How to share a secret. Commun. ACM 22, 612–613 (1979) 12. Tang, S.: Simple secret sharing and threshold RSA signature schemes. J. Inf. Comput. Sci. 1, 259–262 (2004)

Application of Cognitive Protocols in Transformative Computing Marek R. Ogiela1(&) and Lidia Ogiela2 1

Cryptography and Cognitive Informatics Research Group, AGH University of Science and Technology, 30 Mickiewicza Avenue, 30-059 Kraków, Poland [email protected] 2 Department of Cryptography and Cognitive Informatics, Pedagogical University of Krakow, Podchorążych 2 Street, 30-084 Kraków, Poland [email protected]

Abstract. Transformative computing is a new computational paradigm, which join wireless technologies with artificial intelligence, and edge computing. In this paper we describe possible applications of different cognitive protocols for such computational areas, especially focused on security technologies, and considering personal characteristics, and environmental features. Also will be presented the impact of cognitive approaches and cryptographic algorithms for securing distributed systems and distributed infrastructures. Keywords: Cognitive cryptography computing


Security protocols
Transformative

1 Introduction Recently, a new computational paradigm has appeared that allows combining wireless technologies with artificial intelligence and edge computing algorithms. This is Transformative Computing, which significantly extends previous Ubiquitous Computing technologies with the use of artificial intelligence techniques. Therefore, it seems that ubiquitous computers will not only allow for quick and efficient communication, but also in combination with sensors from mobile devices, intelligent home or wearable sensors etc., will allow for performing computational tasks using advanced inference techniques and data analysis. This technology seems to be very promising and can be used in many different areas of human activity, science and technology. In this work, it will be discussed in a slightly different context, namely in combination with cognitive systems [1, 2], and the possibilities of integrating this approach with cognitive inference, proposed by the authors in previous works and focused on the task of computer understanding of visual patterns [3], as well as cognitive cryptography solutions [4]. In particular, a new area of Transformative Computing will be discussed here, in which sensory signals will not only be analyzed using artificial intelligence techniques, but also analyzed by cognitive information systems based on the inference model using cognitive resonance. Such connection allow to define a new branch called cognitive transformative computing. © Springer Nature Switzerland AG 2020 L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 428–432, 2020. https://doi.org/10.1007/978-3-030-33506-9_38

Application of Cognitive Protocols in Transformative Computing

429

2 Cognitive Algorithms in Transformative Computing Communication technologies of 5G transmission in connection with sensors used for data acquisition allow to make calculations related to the user’s activity, but also to devices recording broad range of sensor signals. Their quantity is so enormous for fast analysis, so it requires the application of very advanced methods of analytical analysis, which are usually associated with AI techniques. Currently, artificial intelligence methods are related with neural networks, and deep learning approaches, which allows to train neural network to solve complex problems. At this point, it is worth noting the possibility of using perception techniques for data analysis, which are used in cognitive information and vision systems. Cognitive systems operate based on cognitive resonance, bases on the application of experiences and knowledge from the analysis of previous patterns or solved tasks. It is based on inference, in which the analyzed features or patterns are compared with certain expectations generated on the basis of previous experience. This inference model is also known from research on human perception models, and allows to analyze very complex patterns or situations [5, 6]. These features mean that it can also be introduced to Transformative Computing techniques as a kind of extension of traditional artificial intelligence algorithms, towards methods that allow semantic inference and meaning interpretation of collected data. This combination of cognitive systems with wireless technologies and sensors collecting data, allows defining a new paradigm, i.e. Cognitive Transformative Computing. Cognitive transformative computing as a new computational paradigm can be applied in different areas. The most important among them are smart communication, real-life pattern understanding, IoT, advanced Cloud Computing services analysis, and personal and global security areas.

3 Application of Cognitive Transformative Computing in Security Areas In this section will be presented one of the most challenging application of transformative computing technologies based on cognitive systems. This application is connected with security areas oriented for creation of global and personal security protocols, which use personal features or cognitive characteristics. Such solutions are connected with cognitive cryptography [7, 8], which is oriented on application of perceptual features or visual thresholds, which can be applied for user-oriented authentication protocols. In fact such approaches are very unique and allow to focus security protocols on particular person or group of users [9, 10]. Application of such techniques with relation to global communication technologies, 5G, IoT, edge sensors, and smart communication devices, allow to extend cognitive cryptographic solutions for general application, which will be oriented for global security solutions, and not especially applicable for particular users or persons. The main idea of creation security protocols based on cognitive transformative computing lays on fast and secure evaluation of unique identifiers (or features) describing local

430

M. R. Ogiela and L. Ogiela

context of nets connections, or environmental features, which can be evaluated or extracted using cognitive information systems. This allow to considerably extend application of personal security protocols for general application, which depend not only from user but also from context or situation in which this particular user should work. Such security protocol can quickly adapt to different circumstances and reevaluate security identifiers depending on the changing context. Such changes will be registered by edge sensors and wireless communication devices, and finally evaluate by cognitive systems, which will be able to extract informative parameters describing the context in unique manner. For example it is possible to create authentication protocol oriented for particular user, which consider unique personal features or cognitive abilities, but also consider different user activities with computer equipment. Such protocol should allow to authenticate particular persons in different manners for different activities like accessing to databases, cloud services, computing infrastructure, banking, shopping, and communication devices. In each activities beside personal characteristics, an analysis of acquired data can be performed using cognitive systems, and additional unique identifier can be evaluated. Such identifier next can also be used for authentication to further tasks, and change over the time, or along different activities. Additional feature of such global cognitive transformative computing solutions is possible connection with real external environment, in which particular user works. As was mentioned such protocols can be oriented on application of personal, unique features, which are characteristic only for particular user (like behavioral features, biometric patterns etc.), but also in such protocols can be involved context data describing external environment, in which user works. Such features are available thanks to application of cognitive systems which are able to analyze complex patterns and multi-object scenes. Application of cognitive information systems as an integral part of transformative computing solutions make also possible to realize two-directional flow of information, not only from sensors to AI modules, but also from cognitive systems to communication devices and edge sensors. This make possible to establish a control procedure on the whole environment. This allow to create security procedures oriented not only on particular users, but also considering external features and context situations. Such procedures can be adjusted for different persons and different external connections. The above mentioned solutions, in which we can replace traditional AI algorithms by cognitive information and vision systems, considerably extend the functionality, and range of application of transformative computing technologies. It also allow to create a new cognitive transformative computing paradigm, which will play important role in future Internet, and ambient world [11–13].

4 Conclusions In this paper was presented a new computational paradigm called cognitive transformative computing. It is a new functional extensions of known transformative computing technologies, joining wireless communication, AI and edge sensors. The idea of cognitive transformative computing lays on application of cognitive systems oriented for semantic evaluation of complex pattern or situations. Cognitive systems allow

Application of Cognitive Protocols in Transformative Computing

431

additionally to send back information from cognitive modules to edge sensors to control further actions or adjust computing approaches. Cognitive systems allow also to use sensor data in more general ways, and adjust user actions to external environment or circumstances. This new technology seems to be very promising and applicable in many areas connected with 5G technology IoT, smart and ambient worlds etc. It is also very promising for creation of new security procedures oriented for particular users, and having adaptation features, which consider external real-world features and context situation [14]. So even simple user authentication procedures can be dependent not only on personal user features but also on external features or devices used by users. Acknowledgments. This work has been supported by the National Science Centre, Poland, under project number DEC-2016/23/B/HS4/00616. This work has been supported by the AGH University of Science and Technology research Grant No 16.16.120.773.

References 1. Ogiela, M.R., Ogiela, L.: On using cognitive models in cryptography. In: IEEE AINA 2016 The IEEE 30th International Conference on Advanced Information Networking and Applications, Crans-Montana, Switzerland, 23–25 March, pp. 1055–1058 (2016) 2. Ogiela, M.R., Ogiela, L.: Cognitive keys in personalized cryptography. In: IEEE AINA 2017 - The 31st IEEE International Conference on Advanced Information Networking and Applications, Taipei, Taiwan, 27–29 March, pp. 1050–1054 (2017) 3. Ogiela, M.R., Ogiela, U., Ogiela, L.: Secure information sharing using personal biometric characteristics. In: Kim, T.-H., et al. (eds.) Computer Applications for Bio-technology, Multimedia and Ubiquitous City. CCIS, vol. 353, pp. 369–373. Springer, Heidelberg (2012) 4. Ogiela, M.R., Ogiela, L., Ogiela, U.: Biometric methods for advanced strategic data sharing protocols. In: The Ninth International Conference on Innovative Mobile and Internet Services in Ubiquitous Computing (IMIS 2015), Blumenau, Brazil, 8–10 July, pp. 179–183 (2015) 5. Ogiela, L., Ogiela, M.R.: Bio-inspired cryptographic techniques in information management applications. In: IEEE AINA 2016 - The IEEE 30th International Conference on Advanced Information Networking and Applications, Crans-Montana, Switzerland, 23–25 March, pp. 1059–1063 (2016) 6. Ogiela, U., Ogiela, L.: Linguistic techniques for cryptographic data sharing algorithms. Concurr. Comput. Pract. E 30(3), e4275 (2018). https://doi.org/10.1002/cpe.4275 7. Ogiela, L., Ogiela, M.R.: Insider threats and cryptographic techniques in secure information management. IEEE Syst. J. 11, 405–414 (2017) 8. Ogiela, M.R., Ogiela, U.: Secure information management in hierarchical structures. In: Kim, T.-H., et al. (eds.) AST 2011. CCIS, vol. 195, pp. 31–35. Springer, Heidelberg (2011) 9. Ogiela, L., Ogiela, M.R., Ogiela, U.: Efficiency of strategic data sharing and management protocols. In: The 10th International Conference on Innovative Mobile and Internet Services in Ubiquitous Computing (IMIS 2016), Fukuoka, Japan, July 6–8, pp. 198–201 (2016) 10. Ogiela, L.: Advanced techniques for knowledge management and access to strategic information. Int. J. Inf. Manag. 35(2), 154–159 (2015) 11. Meiappane, A., Premanand, V.: CAPTCHA as Graphical Passwords - A New Security Primitive: Based on Hard AI Problems. Scholars’ Press (2015)

432

M. R. Ogiela and L. Ogiela

12. Osadchy, M., Hernandez-Castro, J., Gibson, S., Dunkelman, O., Perez-Cabo, D.: No bot expects the DeepCAPTCHA! Introducing immutable adversarial examples, with applications to CAPTCHA generation. IEEE Trans. Inf. Forensics Secur. 12(11), 2640–2653 (2017) 13. Easttom, Ch.: Modern Cryptography: Applied Mathematics for Encryption and Information Security. McGraw-Hill Education, New York (2015) 14. Schneier, B.: Applied Cryptography. Wiley, Indianapolis (2015)

Analyzing Mobile Cycling Applications for Monitoring Workouts Fabricio Landero Cristobal(B) , Miguel A. Wister, and Pablo Payro Campos Academic Division of Information Technology and Systems, Juarez Autonomous University of Tabasco, Cunduacan, Tabasco, Mexico [email protected], {miguel.wister,pablo.payro}@ujat.mx

Abstract. This paper analyzes three mobile bike applications that compare different measurements in this sport. For cyclists, it is crucial to know the power of pedaling, several computer systems estimate or calculate this variable instead of measuring. There are power meters, but several models give different measurements. This paper tries to show that some mobile applications for cycling supply different measurements to each other, as well as the power obtained by estimation. We showed by means three experimental rides that sometimes the power measurements are not proportional to the speed produced by the cyclist, so we propose to build a mobile bike application that integrates data from power meters, speedometers, and wireless sensor network to synchronize power and speed for delivering it to the cyclist in real time.

1

Introduction

Previously heart rate monitors had been the basis for planning, quantifying, and analyzing bicycle training. The pulsations per minute are a straightforward parameter to understand, simply quantifying how fast our heartbeats, expressed in beats per minute. For this reason, training methods based on physiological thresholds were developed, always taking as a reference to the pulse of each cyclist. Pulse or heart rate has problems. First, it beats per minute suffer alterations due to fatigue, stress, illness, dehydration, etcetera. The second problem is that the heart takes several seconds to adjust and stabilize the cardiac rhythms against short-term stimuli, so making short and high-intensity series with a heart rate monitor is almost impossible. After appearing power meters, all this has changed significantly. The power generated by legs can be measured almost instantaneously, so quantifying the effort in real time is much simpler. Now, there are several kinds of power meters. In the literature reviewed, all these devices give different measurements to each other. Recently, there have been countless mobile applications for cycling that provide different measurements such as speed, distance, time, average speed, maximum speed, some applications with the measurements obtained from the c Springer Nature Switzerland AG 2020  L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 433–444, 2020. https://doi.org/10.1007/978-3-030-33506-9_39

434

F. L. Cristobal et al.

smartphone-based on GPS, are able to deliver an estimate of the power, similarly to power meters, these mobile applications also give us very different measurements to each other. When the concept of the Internet of Things (IoT) appears, it may be possible to reduce the problem described above, since the IoT interconnects different devices that combined and integrated properly, it is feasible to obtain better training from cycling. The IoT is the new trend of the 21st century, which took force at the time that there were many more objects connected to the Internet than people. This connectivity, interconnection between devices, and the exchange of information reinforced this concept. Bicycle is as such an object and several people have made studies integrating the Internet into it. For example, [8] proposes a Health Navigation System application for smartphone, based on a network of low cost, miniaturized wireless mobile monitoring system that can be easily embedded on a bike frame. Makes available an accurate pollution urban map that bikers can use to determine the healthiest route. In the cycling is used applications to obtain a relation of variables of power measurement. Smartphone or wearables are increasingly essential when it comes to cycling. From the cardiac measurement, the routing of new routes, to panic buttons; There is an application for every cyclist, professional or amateur. There are applications of practice cycling in a hard way or just for short tours in the city such as Bike Gear1 , MTB project2 , Bike repair3 , MyFitnessPal4 , and so on. An exercise professional or a trainer do not replace a mobile application for cyclists; however, the applications greatly help to show the tracking of activities, routes, analyze performance and calculate the best ratio for speeds. There are many applications such as E-Tube, Wahoo Fitness [20], Runtastic [14], Bike Citizens [2], Strava [17], MapMyRide [11], Citymapper, BKool [3], Endomondo [5], Moovit. For our case study and tests, we used only three mobile apps: Strava, Endomondo, and Runtastic. Cyclo-computer is used to get measurements like velocity, distance, and time. This cyclo-computer is connected to the wheel’s fork; it uses a sensor to collect signal each time when passing a magnet. There are different models depending on the brand manufacturer like Mio, Garmin, iGPSPORT, and Polar. Some mechanical speedometers installed on a wheel to transmit movements towards a device that is graduated to lift a needle that marks the distance that is traveled depends on the bike speed. i.e.: iGPSPORT [9], SIGMA [16], POLAR [13]. Applications for cycling can calculate time, distance, duration, pace, calories, cadence, temperature, velocity, altitude, average speed, heart rate, maximum speed, and power by GPS. There is an error range that adds meters while the

1 2 3 4

Bike Gear: http://gears.mtbcrosscountry.com/#26I1I1. MTB project: https://www.mtbproject.com/. Bike repair: https://www.bikerepairapp.com/. MyFitnessPal: https://www.myfitnesspal.com/es/.

Analyzing Mobile Cycling Applications for Monitoring Workouts

435

distance traveled is longer. It works by drawing small points that are counted every second and that, at the end of the race, trace a straight line. To measure the pedaling power is a used power meter, that is a meter capable of measuring the force applied by cyclists at a specific time or period. The formulas for calculating power are shown as follows:   distance P ower = f orce × velocity Power measures the amount of work needed by the bike in a given time interval. Cadence is the rhythm or the number of pedals per minute (rpm). Force is the intensity with which the pedal is pushed and the time used. Distance is a pre-programmed value that can be calculated according to the bicycle assembly of the bike, that is, the plate and the crank. Power meters can be in different areas of a bike on the rear wheel hub, bottom bracket/spindle, chainrings and crank spiders, crank arm, pedals. In this paper, we will compare measurements such as distance, power, time, and speed from three mobile cycling applications and speedometer, and using three different workouts.

2

Related Works

Several papers exist about cycling, many works dealing with the subject of pedaling and power, mainly works about health and mobility related to routes and maps of cycling routes; experiments about tests applied to cyclists in laboratories and indoor controlled scenarios. Studies about pedaling movement to maximize the performance and minimize risks of injuries; it also about mobile sensing system for mapping cyclist experiences, and so on. Below we will briefly describe some related works to cycling. The authors present in [12] a study of the accuracy of a system concerning the gold standard motion capture system. It measures the knee and ankle angles, which influence the performance as well as the risk of overuse injuries during cycling. In [21] introduces a bicycle record system of ground conditions based on IoT, which is combining smartphone and embedded system, providing various types of real-time information for cyclists to help them achieve their desired exercise results. Other authors [10] intend to determine the effect that different pedaling techniques have on gross efficiency during cycling in steady-state and mechanical effectiveness, cranks with power meters and pedals with force sensors should be used to determine if a mechanically effective pedaling technique can achieve greater efficiency. A work found in [15] presents a mobile sensing system applied to cycle, which collects performance data using both smartphones with sensors integrated and several wireless sensor nodes. The users can share data, consult graphs, and access past routes on a map. A study made by [6] researches challenges and requirements of an IoT implementation based on GPS trackers from a technological as well as a consumer perspective. The results suggest a high interest of

436

F. L. Cristobal et al.

users in data from a connected e-bike but also indicate that technical restrictions still exist, e.g., concerning the completeness of collected data. BikeNet is a sensing system built leveraging the MetroSense architecture to provide people-centric sensing into the real-world and mapping the cyclist’s experience. Is a prototype of the system architecture based on small sensors and a mobile phone; it infers cyclist performance and the cyclist environment [4]. Another paper [18] uses a low-cost ultrasonic distance sensor attached on a bicycle for monitoring the road surface conditions in the front area, allowing detect the 223 cm away obstacle on the road in the front area of moving the bicycle. Ueberham [19] presents a novel open-source task automation solution and its evaluation in a personal exposure study with cyclists. It was designed as an automation script for data collection, management, and storage of acoustic noise, geolocation, light level, timestamp, and qualitative user perception. Current techniques are based on the assembly of bicycles and measurements with laboratory tests. These techniques do not allow to evaluate the cyclist’s kinematics in real scenarios during training and competition when fatigue can alter the ability of cyclist to apply forces to the pedals and thus induce a poorly adapted load on the joint.

3

Cycling Performance

Equipment For our tests, we used the following equipment: 1. One Smartphone - Huawei Y6. Size 6.09”, screen resolution of 720 × 1560 pixels and Android v9.0 (Pie) operating system. Quad-core, 2 GHz, Cortex A53, processor 2 GB of RAM. Battery 3020 mAh. 2. One Bike Computer - Brand Sigma BC 12.12. 12 functions. 3. One Bicycle - Brand Specialized model Crave comp 29. M4 Premium Aluminum, XC 29 Geometry. 29 × 2.1”. Specialized Stout SL Disc 29 rims. Weight 11.91 Kg. 4. Three Mobile Cycling Applications: a. Strava: This mobile app tracks the fitness activity, record our run, map a cycling route, and analyze our training with all the stats. Is a free digital service accessible through both a mobile application and a web page. The website tracks us via GPS. b. Endomondo: It allows real-time track workouts, it measures distance, speed, altitude, and location, because it uses GPS and Google maps. It has functions such as: follow a workout with GPS while riding a bicycle, check duration, speed, distance, or caloric expenditure. c. Runtastic: It records fitness and cycling activities using GPS technology. It also plans cycle route, records rides, monitors workouts. Offers some features as follows: Track GPS, measure distance, duration, speed, rhythm, burned calories, view map, speed, elevation, heart rate, generates training history, create a table for lapses.

Analyzing Mobile Cycling Applications for Monitoring Workouts

4

437

Performance Analysis Comparison

The main objective of our experiments is to compare the performance of the measurements collected by mobile applications and the bike computer. To this objective, we focus on the tests on the following tasks: (a) Collect all the raw data from all mobile application to a web application. (b) Compare the results obtained with the bike computer to the ones supplied by the dataset from the mobile apps. Some applications calculate the power measurements getting the cyclist weight and the bicycle weight. It also three principles forces, that the cyclist must win to advance. • Gravity: if a cyclist is cycling uphill, he will be fighting against gravity, but if he is cycling downhill, gravity works for him. • Rolling resistance: friction between the tires and the road surface slows down. If the tires and tube have a good quality, less friction will experience the cyclist. • Aerodynamic drag: as the cyclist cycle through the air, the bike and body need to push the air around. Due to the air exerts a force on the cyclist while he rides. At a higher speed, more force will push the air against the cyclist. 4.1

Participant

We invited to a cyclist for participating in our experiments. This person was a male, 50 years old, height 1.70 m, and weight 84 Kg. All tests were performed riding the same bicycle. Our experiments consisted of three rides or workouts. The cyclist completed each trip in approximately 52 min, 20.5 km, average speed 25 km/h, and 600 Kcal. 4.2

Experimental Setting

In our experiments, we used three different cycling mobile cycling applications to evaluate their accuracy to collect data from speed, distance, time, and power. Our experiments were carried out on three different workouts rode on June 30, July 6, and July 7, it was also necessary a smartphone were installing the mobile cycling applications and a cyclist. One mountain bike was used to carry out our experiments. The technical specifications are described above. 4.3

Protocol

We rode on three workouts, the first one was on June 30 finishing approximately 20.60 km, the second route was made on July 6 riding 21.68 km, and finally, the third route was on July 7 pedaling about 20.55 km. To record data from the three workouts, firstly it was reset and started the three mobile cycling applications (Strava, Endomondo, and Runtastic) on the smartphone. After resetting and starting for deleting the current values in speed,

438

F. L. Cristobal et al.

max speed, time, and distance from the bike computer. Cyclist starts pedaling and rides on the route for the three different days. Once it finishes pedaling, cyclist stops the three mobile cycling applications and the bike computer. It is necessary to open the three mobile cycling applications on its web page and export the three datasets (June 30, July 6, and July 7) in .gpx format and save them in the computer. On Strava web page is uploaded all dataset and select each one to see the data analysis generated by Strava and visualize two graphics. First, we can analyze distance and speed, and then it is possible to analyze speed versus the power, placing the cursor on-axis speed or axis power. 4.4

Scenario

In Fig. 1, we include a map that contains the points traveled in the workouts. This ride was carried out in an area near to Villahermosa, Tabasco, Mexico, temperature 36◦ celsius, very early in the morning, on asphalt, flat ground with few slopes in ascent and descent, and no wind.

Fig. 1. Map of the ride on July 7, 2019

5

Results

Our experiments were summarized in Tables 1, 2, and 3. Several measures Analysis of Variance (ANOVA) was conducted considering max speed, average speed, time, distance, cadence, and calories. 5.1

Distance Analysis

Our first workout completed on June 30. About the traveled distance, the applications recorded: 20.65 km (Strava), 20.63 km (Endomondo), and 20.51 km (Runtastic). In Table 1 it is noticed slight differences of some meters in the same applications. In Table 1 is described the previously mentioned. The workout on July 6. The cycle computer recorded 21.65 km, while the mobile applications logged 21.76 km (Strava), 21.73 km (Endomondo), and 21.57 km (Runtastic). Table 2 contain values where is found slight differences of a few meters. There are also some differences between data collected by the mobile applications and data collected by the cycle computer. The summary results are shown in Table 2.

Analyzing Mobile Cycling Applications for Monitoring Workouts

439

Finally, the workout on July 7, the cycle computer recorded 20.53 km, the mobile applications registered 20.61 km (Strava), 20.57 km (Endomondo) and 20.47 km (Runtastic). In Table 3 we can see slight differences in some meters in the same applications. We found some differences between data collected by the mobile apps and data collected by the cycle computer. Table 3 shows all results previously described. 5.2

Speed Analysis

Figure 2 was obtained by joining three graphs and analysis of the three applications used in the workouts. It was plotted and analyzed by Strava to standardize all collected data; these graphs show speed, time, and power only. The x-axis represents time, and the y-axis represents speed. Average speed recorded on June 30 was: Strava recorded 24.40 km/h, Endomondo marked 24.26 km/h, and Runtastic recorded 24.00 km/h. On July 6 the average speed on the cycle computer recorded 23.64 km/h, Strava recorded 23.80 km/h, Endomondo recorded 23.57 km/h, and Runtastic recorded 23.34 km/h. On July 7 the average speed on the cycle computer recorded 25.07 km/h, Strava recorded 25.20 km/h, Endomondo recorded 24.70 km/h and Runtastic recorded 34.80 km/h. There are slight differences between data collected by the cycle computer and data collected by mobile applications. See Fig. 2 for the compared values. As an example and to save space in this paper, we only include the graph obtained in training on July 7.

Fig. 2. General information workout. Date: July 7 2019. Three mobile cycling apps graphs. The plots show the comparison between speed, power, and distance computed by Strava, Endomondo, and Runtastic.

440

F. L. Cristobal et al.

Table 1 shows maximum speed, average speed, time, distance, cadence, and calories obtained by the three applications on June 30. Table 2 shows the values recorded on July 6. In Table 3, we see the values obtained on July 7. Table 1. Workout summary. Total distance 20.60 km. Date June/30/2019. Application Devices Max speed Avg speed Time Distance Cadence Calories Km/h Km/h Minutes Km Min/Km Kcal Strava App Endomondo App Runtastic App

42.10 41.29 40.15

Mean 41.18 Standard deviation 0.9796 Variance 0.9597 Median 41.29

5.3

24.40 24.26 24.00

50.53 51.02 51.16

20.65 20.63 20.51

24.22 0.2030 0.0412 24.26

50.90 20.60 0.3308 0.0757 0.1094 0.0057 51.02 20.63

– 2.28 2.30

417 717 496

2.29 0.0141 0.0002 2.29

Power Analysis

In Table 4 summarizes the July 7 workout, Strava estimates different values for the power, e.g., if speed is 25.6 km/h different power values are calculated, that is, 222 w and 400 w, while speed is 25.2 km/h, the power is 184w.

6

Discussion

As we have seen, the analyzed applications are not very accurate; these present very slight differences. In the three workouts analyzed, the differences in distances are from 10 to 15 m because these are obtained using a GPS receiver embedded in smartphones. However, with speed values recorded by these applications are very similar to those obtained by the speedometer. Another important aspect to highlight is the discrepancy in some data, e.g., in the route on July 7 the three applications show the same speed on a flat road with the same altitude but different power measurement. Table 4 contains the values above mentioned. We made a comparison to obtain power measurement using Strava application, Gribble calculator of Steve Gribble5 and Bike calculator of Curt Austin6 when speed is 29.2 km/h (see Fig. 3) and the difference was more significant. Using the Strava application, the power measurement was 971 w (see Fig. 4), using the Gribble calculator [7] the power measurement was 144.88 w and 5 6

Steve Gribble - Cycling power and speed. https://www.gribble.org/cycling/ power v speed.html. Curt Austin - Bike Calculator - http://bikecalculator.com/.

Analyzing Mobile Cycling Applications for Monitoring Workouts

441

Table 2. Workout summary. Total distance 21.68 km. Date July/06/2019. Application

Devices Max speed Avg speed Time Distance Cadence Calories Km/h Km/h Minutes Km Min/Km Kcal

Sigma BC12.12 Sensor App Strava App Endomondo App Runtastic

27.48 29.90 36.07 27.82

23.64 23.80 23.57 23.34

54.57 55.22 55.20 55.27

21.65 21.76 21.73 21.57

Mean Standard deviation Variance Median

30.32 3.9814 15.8515 28.86

23.59 0.1910 0.0365 23.61

55.07 21.68 0.3313 0.0854 0.1098 0.0073 55.21 21.69

– – 2.33 2.34

– – 778 494

2.34 0.0071 0.0000 2.34

Table 3. Workout summary. Total distance 20.55 km. Date July/07/2019. Application

Devices Max speed Avg speed Time Distance Cadence Calories Km/h Km/h Minutes Km Min/Km Kcal

Sigma BC12.12 Sensor App Strava App Endomondo App Runtastic

44.22 43.60 46.15 42.42

25.07 25.20 24.70 34.80

49.09 49.07 49.59 49.31

20.53 20.61 20.57 20.47

Mean Standard deviation Variance Median

44.10 1.5588 2.4298 43.91

27.44 4.9096 24.1039 25.14

49.27 20.55 0.2424 0.0597 0.0588 0.0036 49.20 20.55

– – 2.26 2.25

– – 703 529

2.26 0.0071 0.0000 2.26

Table 4. Power estimated by Strava. Workout July 7, 2019 Point on the route Speed (Km/h) Power (w) 13 Km 8.2 Km 11.8 Km

25.6 25.6 25.2

222 400 184

using Bike Calculator [1] the power measurement was 166 w. Each calculator and/or applications use different parameters, methods, and formulas to calculate power(w).

7

Mobile Bike Application Proposed

We propose developing a mobile bike application (see Fig. 5) integrating a pedal power meter and heart rate strap, the measurements obtained are sent via ANT+

442

F. L. Cristobal et al.

Fig. 3. Gribble Calculator. Web page dedicates cycling power and speed. It explore the relationship between the cycling power (wattage) and speed. Moving the cursor over the graph, or tap on it, we can explore specific points.

Fig. 4. Power in Strava application. This graph shows power vs speed

Fig. 5. Mobile bike application proposed

or Bluetooth protocol to a smartphone. We collect route and points of the workout; these data are upload to the cloud computing for performing workout analysis data. It is stored into a database to keep the record of cyclists. Finally, a smartphone’s dashboard displays these processed data.

8

Conclusions

This article evaluated the relationship between values of speed, distance, time, and power. We found differences compared with the speedometer. Although we used the same smartphone in the three mobile cycling applications, all applications started and stopped at the same time. The results obtained in different

Analyzing Mobile Cycling Applications for Monitoring Workouts

443

workouts demonstrate that mobile bike applications based on GPS present differences in distance, time, and power. We have observed that calculators and/or apps use different parameters, methods, and formulas to calculate power since these data analyzed are different from each other. If a cyclist knows in real time the accurate measurements of the force that uses when pedaling, then to allow to regulate the power available and to avoiding exceeding certain limits, to the detriment of the improvisation and the sensations of the cyclist himself.

9

Future Works

As result of differences found in the data collected by the web page BikeCalculator [1], Steve Gribble [7], and Strava application, we propose to develop a mobile bike application integrating data from power meters, speedometers, and wireless sensor network to synchronize and obtain measurements of power cadence, heart rate, ambient temperature, distance, speed, altitude, time as well as routes and points of training or competition. This proposal to be stored and analyzed, it going to visualize these variables in a dashboard in real time through a smartphone. Acknowledgements. This paper was supported by Programa de Fortalecimiento de la Calidad Educativa (PFCE) 2019. Number: P/PFCE-2019-27MSU0018V-11. We would also like to express our gratitude to the Universidad Juarez Autonoma de Tabasco for supporting the academic resources needed for this research.

References 1. 2. 3. 4.

5. 6.

7. 8.

9. 10.

11.

Austin, C.: Bike calculator (2019). http://bikecalculator.com/about.html BikeCitizens: Bike citizens official web (2019). https://www.bikecitizens.net/ Bkool: Bkool official web (2019). https://www.bkool.com/es/app-para-ciclismo Eisenman, S.B., Miluzzo, E., Lane, N.D., Peterson, R.A., Ahn, G.S., Campbell, A.T.: BikeNet: a mobile sensing system for cyclist experience mapping. ACM Trans. Sen. Netw. 6(1), 6:1–6:39 (2010). https://doi.org/10.1145/1653760.1653766 Endomondo: Endomondo official web (2019). https://www.endomondo.com/ Fl¨ uchter, K., Wortmann, F.: Implementing the connected e-bike: challenges and requirements of an IoT application for urban transportation (2014). https://doi. org/10.1145/2666681.2666682 Gribble, S.D.: The computational cyclist (2019). https://www.gribble.org/ Guerriero, A., Guaragnella, C., Martines, C., Castellaneta, A.: A distributed health navigation system based on opportunistic mobile WSN, pp. 1–6 (2012). https:// doi.org/10.1109/ICITeS.2012.6216596 iGPSPORT: Igpsport official web (2019). http://global.igpsport.com/ Korff, T., Romer, L., Mayhew, I., Martin, J.C.: Effect of pedaling technique on mechanical effectiveness and efficiency in cyclists (2007). https://doi.org/10.1249/ mss.0b013e318043a235 MapMyRide: Map my ride official web (2019). http://www.mapmyride.com/

444

F. L. Cristobal et al.

12. Marin-Perianu, R., Marin-Perianu, M., Havinga, P., Taylor, S., Begg, R., Palaniswami, M., Rouffet, D.: A performance analysis of a wireless body-area network monitoring system for professional cycling (2013). https://doi.org/10.1007/ s00779-011-0486-x 13. Polar: Polar official web (2019). https://www.polar.com/mx-es 14. Runtastic: Runtastic official web (2019). https://www.runtastic.com/ 15. Oliveira, D.S., Afonso, J.: Mobile sensing system for georeferenced performance monitoring in cycling. In: World Congress on Engineering 2015, vol. 1 (2015) 16. SIGMA: Sigma official web (2019). https://www.sigmasportstore.com/product-p/ 02120.htm 17. Strava: Strava official web (2019). https://www.strava.com/?hl=es 18. Taniguchi, Y., Nishii, K., Hisamatsu, H.: Evaluation of a bicycle-mounted ultrasonic distance sensor for monitoring road surface condition. In: 2015 7th International Conference on Computational Intelligence, Communication Systems and Networks, pp. 31–34 (2015). https://doi.org/10.1109/CICSyN.2015.16 19. Ueberham, M., Schmidt, F., Schlink, U.: Advanced smartphone-based sensing with open-source task automation. Sensors 18(8), 2456 (2018). https://doi.org/10.3390/ s18082456. https://www.mdpi.com/1424-8220/18/8/2456 20. WahooFitness: Wahoo fitness official web (2019). https://www.wahoofitness.com/ 21. Zhao, Y., Su, Y., Chang, Y.: A real-time bicycle record system of ground conditions based on internet of things. IEEE Access 5, 17525–17533 (2017). https://doi.org/ 10.1109/ACCESS.2017.2740419

Road State Information Platform Based on Multi-sensors and Bigdata Analysis Yoshitaka Shibata1(&), Goshi Sato2, and Noriki Uchida3 1

2

Regional Corporate Research Center, Iwate Prefectural University, 152-52 Sugo, Takizawa, Iwate, Japan [email protected] National Institute of Information and Communications Technology, 2-1-3 Katahira, Aoba-ku, Sendai, Miyagi, Japan [email protected] 3 Department of Information and Communication Engineering, Fukuoka Institute of Technology, 3-30-1 Wajiro-higashi, Higashi-ku, Fukuoka, Japan [email protected]

Abstract. In this paper, in order to keep safe and secure driving, a new generation wide area road surface state information platform based on crowd sensing and V2X Technologies is introduced. In crowd sensing, various environmental sensors including accelerator, gyro sensor, infrared temperature sensor, quasi electrical static sensor, camera and GPS are integrated to precisely detect the various road surface states and determine the dangerous locations on GIS. Those road information are transmitted the neighbor vehicles and road side server in realtime using V2X communication network. In V2X communication on the actual road, both the length of communication distance and the total size of data transmission must be maximized at the same time when vehicle are running on the road. The conventional single wireless communication such as Wi-Fi, IEEE802.11p, LPWA, cannot satisfy those conditions at the same time. In order to resolve such problems, N-wavelength cognitive wireless communication method is newly introduced in our research. Multiple next generation wireless LANS including IEEE802.11ac/ad/ah/in addition to the current popular LANs with different wavelengths are integrated to organize a cognitive wireless communication. The best link of the cognitive wireless is determined by SDN. Driver can receive the road surface status information from the vehicle in opposite direction or road side server and eventually pay attentions to his/her driving before encountering the danger location. This technology can also apply for automatic driving car.

1 Introduction In our social life, mobility is the most important means for economic activity to safely and reliably carry persons, loads, feeds and other materials around world. Huge number of vehicles are being produced day by day and their qualities are developed year and year in the developed countries. Furthermore, in recent, self-driving cars have been emerged and running on highways and public roads in well developed countries. © Springer Nature Switzerland AG 2020 L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 445–454, 2020. https://doi.org/10.1007/978-3-030-33506-9_40

446

Y. Shibata et al.

However, on the other hand, the road infrastructures in many developing countries are not well maintained compared with improvement of vehicles. In particular, as shown in Fig. 1, the road conditions in developing countries are so bad and dangerous due to luck of regular road maintenance, falling objects from other vehicles, overloaded trucks or buses. Therefore, in order to maintain safe and reliable auto driving, the vehicles have to detect those obstacles in advance and avoid when they pass away.

Fig. 1. Road condition in highway and local roads

Second, in the cold or snow countries, such as Japan and Northern countries, most of the road surfaces are occupied with heavy snow and iced surface in winter and many slip accidents occurred even though the vehicles attach snow specific tires. In fact almost more 90% of traffic accidents in northern part of Japan is caused from slipping car on snowy or iced road as seen in Fig. 2.

Fig. 2. Road condition in snow country

In those cases, traffic accidents are rapidly increased. Therefore, safer and more reliable road monitoring and warning system which can transmit the road condition information to drivers and new self-driving method before passing through the dangerous road area is indispensable. Furthermore, the information and communication environment in local areas, is not well developed and their mobile and wireless communication facilities are unstable along the roads compared with urban area. Thus, once a traffic accidents or disaster occurred, information collection, transmission and sharing are delayed or even cannot be made. Eventually the resident’s lives and reliabilities cannot be maintained. More robust and resilient information infrastructure and proper and quick information services with road environmental conditions are indispensable.

Road State Information Platform Based on Multi-sensors and Bigdata Analysis

447

In order to resolve those problems, we introduce a new generation wide area road surface state information platform based on crowd sensing and V2X technologies. In crowd sensing, many data from many vehicles with various environmental sensors including accelerator, gyro sensor, infrared temperature sensor, quasi electrical static sensor, camera and GPS are integrated to precisely detect the various road surface states and determine the dangerous locations on GIS. Those road information are transmitted to the neighbor vehicles and road side server in realtime using V2X communication network. In V2X communication on the actual road, both the communication distance and the total size of data transmission must be maximized at the same time when vehicles are running on the road. The conventional single wireless communication such as Wi-Fi, IEEE802.11p, LPWA, cannot satisfy those conditions at the same time. In order to resolve such problems, N-wavelength cognitive wireless communication method is newly introduced in our research. Multiple next generation wireless LANS including IEEE802.11ac/ad/ah/ai/ay in addition to the current popular LANs with different wavelengths are integrated to organize a cognitive wireless communication. The best link of the cognitive wireless is determined by Software Defined Network (SDN). In the following, general system and architecture of Road Surface State Information Platform are explained in section two. The sensing system and its functions with various sensors are precisely shown in section three. The V2X communication system and its function are explained in section four. The prototype system to evaluate function and performance of the proposed system is explained in section five. Finally conclusion and future works are summarized in section six.

2 Road Surface State Information Platform In order to resolve those problems in previous session, we introduce a new generation wide area road surface state information platform based on crowd sensing and V2X technologies as shown in Fig. 3. The wide area road surface state information platform mainly consists of multiple road side wireless nodes, namely Smart Relay Shelters (SRS), Gateways, and mobile nodes, namely Smart Mobile Box (SMB). Each SRS or SMB is furthermore organized by a sensor information part and communication network part. The sensor information part includes various sensor devices such as semielectrostatic field sensor, an acceleration sensor, gyro sensor, temperature sensor, humidity sensor, infrared sensor and sensor server. Using those sensor devices, various road surface states such as dry, rough, wet, snowy and icy roads can be quantitatively decided. On the other hand, the communication network part integrates multiple wireless network devices with different N-wavelength (different frequency bands) wireless networks such as IEEE802.11n (2.4 GHz), IEEE802.11ac (5.6 GHz), IEEE802.11ah (920 MHz) and organizes a cognitive wireless node. The network node selects the best link of cognitive wireless network depending on the observed network quality by Software Defined Network (SDN). If none of link connection is existed, those sensing data are locally and temporally stored until approaches to another mobile node or road

448

Y. Shibata et al.

side node, and starts to transmit sensor data by DTN Protocol. Thus, data communication can be attained even though the network infrastructure is not existed in challenged network environment such as mountain areas or just after large scale disaster areas. In our system, SRS and SMB organize a large scale information infrastructure without conventional wired network such as Internet. The SMB on the car collects various sensor data including acceleration, temperature, humidity and frozen sensor data as well as GPS data and carries and exchanges to other smart node as message ferry while moving from one end to another along the roads. On the other hand, SRS not only collects and stores sensor data from its own sensors in its database server but exchanges the sensor data from SMB in vehicle nodes when it passes through the SRS in road side wireless node by V2X communication protocol. Therefore, both sensor data at SRS and SMB are periodically uploaded to cloud system through the Gateway and synchronized. Thus, SMB performs as mobile communication means even through the communication infrastructure is challenged environment or not prepared. This network not only performs various road sensor data collection and transmission functions, but also performs Internet access network function to transmit the various data, such as sightseeing information, disaster prevention information and shopping and so on as ordinal public wide area network for residents. Therefore, many applications and services can be realized.

Fig. 3. Road surface state information platform

3 Sensing System with Various Sensors In order to detect the precise road surface conditions, such as dry, wet, dumpy, showy, frozen roads, various sensing devices including accelerator, gyro sensor, infrared temperature sensor, humidity sensor, quasi electrical static sensor, camera and GPS are integrated to precisely and quantitatively detect the various road surface states and

Road State Information Platform Based on Multi-sensors and Bigdata Analysis

449

determine the dangerous locations on GIS in sensor server as shown in Fig. 4. The sensor server periodically samples those sensor signals and performs AD conversion and signal filtering in Receiver module, analyzes the sensor data in Analyzer module to quantitatively determine the road surface state as shown in Fig. 5 and learning from the sensor data in AI module to classify the road surface state as shown in Fig. 6. As result, the correct road surface state can be quantitatively and qualitatively decided. The decision data with road surface condition in SMB are temporally stored in Regional Road Condition Data module and mutually exchanged when the SMB on one vehicle approaches to other SMB. Thus the both SMBs can mutually obtain the most recent road surface state data with just forward road. By the same way, the SMB can also mutually exchange and obtain the forward road surface data from road side SRS.

Fig. 4. Sensor server system

Fig. 5. Analyzer module

Fig. 6. AI module

450

Y. Shibata et al.

4 V2X Communication System In Fig. 7 shows V2X communication method between the SMB of a vehicle and the SRS of road side server. First, one of the wireless networks with the longest communication distance can first make connection link between SMB and SRS using SDN function. Through the this connection link, the communication control data of other wireless networks such as UUID, security key, password, authentication, IP address, TCP port number, socket No. are exchanged. As approaching each other, the second wireless network among the cognitive network can be connected in a short time and actual data transmission can be immediately started. This transmission process can be repeated during crossing each other as long as the longest communication link is connected. This communication process between SMB and SRS is the same as the communication between SMB to other SMB except for using adhoc mode.

Fig. 7. V2X communication method between the SMBS and SRS

On the other hand, the V2X communication between vehicle and the global cloud server on Internet shows in Fig. 8. The sensor data from SMB is transmitted to the SRS in road side server. Then those data are sent to the gateway function unit and the address of those data from local to global address and sent to the global cloud server through Internet. Thus, using the proposed V2X communication protocol, not only Intranet communication among the vehicle network, but also Intranet and Internet communication can be realized.

Road State Information Platform Based on Multi-sensors and Bigdata Analysis

451

Fig. 8. V2X communication method between the SMB

5 Prototype System and Evaluation In order to verify the effects and usefulness of the proposed system, a prototype system is constructed and those functional and performance are evaluated. The prototype system for two-wavelength communication is shown in Fig. 9. We call this prototype system as Smart Mobility Base station (SMB) for mobility and Smart Rely Shelter (SRS) for roadside station. We currently use WI-U2-300D of Buffalo Corporation for Wi-Fi communication of 2.4 GHz as the prototype of two-wavelength communication, and OiNET-923 of Oi Electric Co., Ltd. for 920 MHz band communication respectively. WI-U2-300D is a commercially available device, and the rated bandwidth in this prototype setting is 54 Mbps. On the other hand, the OiNET-923 has a communication distance of 1 km at maximum and a bandwidth of 50 kbps to 100 kbps.

Fig. 9. Realtime road surface data exchange system

452

Y. Shibata et al.

Next, data transmission of V2X communication between the SMB and SRS was evaluated. The Fig. 10 shows the field test where a vehicle with SMB runs several on the public road in the cold and snow area from the start point to the end point by transmitting to the road side server in SRS which is allocated at the middle of the distance by changing the speed of the vehicle from 20–50 km. The maximum communication distance and the RSSI and total data transmission were measured where the communication between the vehicle and road side server is maintained.

Fig. 10. V2X field experimentation

Two networks transmission cases including single Wi-Fi network communication and a combined communication with Wi-Fi and Sub-Giga band with 920 MHz considered to compare their performance. It is known that Wi-Fi transmission provides higher transmission data rate such as 54 Mbps for IEEE802.11n, but it communication distance between the vehicle and the roadside server is limited only up to 60 m at most. On the other hand, Sub-G band provides longer communication distance with more than 1–2 km, but the transmission data rate is limited only 100K bps for LORA. Therefore, in order to provide both long distance communication and higher total transmission rate while driving the vehicle, both Wi-Fi is combined. The combined network, the UUIDs of both servers, SSID and its key, Socket ID for TCP and metadata of sensor DB are priori exchanged through the Sub-Gia band. When the vehicle approaches enough to the road side server to communicate by Wi-Fi network, the larger data can be immediately exchanged in a short time. Eventually a larger total data transmission can be attained. The results of the RSSI and throughput of data transmission for both cases when the transmission distance, which is equivalent to the relative time, between the road side server and the vehicle varies are shown in Figs. 11 and 12. In case of Wi-Fi only network, since the possible transmission distance is very short, about 60 m and its possible communication time is about 5.4 s when the vehicle runs at 40 km/h, the throughput during this time period is low, as result, the total transmission data is small as shown in Fig. 11.

Road State Information Platform Based on Multi-sensors and Bigdata Analysis

453

Figure 12 shows the total transmission data for nine times of filed experiments. From this experimental result, the combination network provided very stable total data transmission of 5.0 MB while Wi-Fi network provide unstable total data transmission of average 3.0 MB. Thus, the combination network can attain 50% better performance in data transmission. In this time, only two different wireless networks were combined into a cognitive wireless network in this experiment. Therefore, by combining with n (n > 2) numbers of wireless networks with different wavelength, better performance such as, larger total transmission data with longer communication distance can be expected.

Fig. 11. Total amount of data transmission

Fig. 12. Comparison of communication distance

6 Conclusions and Remarks In order to recover the problems of poor road conditions and challenged information communication infrastructure in aging society, IoT based mobility information network is proposed. Road side wireless nodes and mobile nodes with various sensors and

454

Y. Shibata et al.

different wireless communication devices are introduced to organize a large scale information infrastructure without conventional wired network and with very economic and low priced mobility network environment can be realized by V2V and V2X communication protocols. The system concept and architecture are precisely explained. The expected application and services for residents are also discussed. As future works, we will design and build a prototype system of our proposed mobility Information infrastructure in rural road between a city and a town in coast areas. We will design and implement Realtime Road Surface Data Exchange System, Dig Data Road Condition Sharing system and Tourist and Disaster Information to evaluate their functions and performance. Acknowledgement. The research was supported by Strategic Information and Communications R&D Promotion Program Grant Number 181502003 by Ministry of Affairs and Communication.

References 1. A 2016 Aging Society White Paper. http://www8.cao.go.jp/kourei/whitepaper/w-014/zenbun/ s1_1_1.html 2. A 2016 Declining Birthrate White Paper. http://www8.cao.go.jp/shoushi/shoushika/ whitepaper/measures/english/w-2016/index.html 3. Ito, K., Hirakawa, G., Shibata, Y.: Experimentation of V2X communication in real environment for road alert information sharing system. In: IEEE AINA 2015, pp. 711–716, March 2015 4. Otomo, M., Sato, G., Shibata, Y.: In-vehicle cloudlet computing based delay tolerant network protocol for disaster information system. In: Advances on Broad-Band Wireless Computing, Communication and Application Applications. Lecture Notes on Data Engineering and Communications Technologies, vol. 2, pp. 255–266, October 2016 5. Hirakawa, G., Uchida, N., Arai, Y., Shibata, Y.: Application of DTN to the vehicle sensor platform CoMoSe. In: WAINA 2015, pp. 24–27, March 2015 6. Kitada, S., Sato, G., Shibata, Y.: A DTN based multi-hop network for disaster information transmission by smart devices. In: Advances on Broad-Band Wireless Computing, Communication and Application Applications. Lecture Notes on Data Engineering and Communications Technologies, vol. 2, pp. 601–611, October 2016 7. Goto, T., Sato, G., Hashimoto, K., Shibata, Y.: Disaster information sharing system considering communication status and elapsed time. In: IWDENS 2017, March 2017

A New Discounting Approach to Conflict Information Fusion Using Multi-criteria of Reliability in Dempster-Shafer Evidence Theory Jin Zhu(&) Department of Computer Science and Engineering, Shanghai JiaoTong University, No. 800, Dongchuan Road, Minghang, Shanghai 200240, China [email protected]

Abstract. Dempster-Shafer evidence theory (DSET) is an important tool to combine uncertain and imprecise information from multiple sources. However, when combining information with highly conflict, it will lead counterintuitive results. A lot of research has been done to resolve the problem. In this paper, we focus on the approach to revise the basic probability assignment of information (evidence) through discount factors. Two methods are proposed to computer discount factors by multi-criteria of reliability measurement. Then we combine multi-source information in the improved Dempster’s rule. Finally, some numerical examples are used to illustrate the efficiency of our proposed methods.

1 Introduction With the development of network technology, information fusion techniques combine data from multiple sources, sensors, institutes, etc., to achieve more specific and more accurate inference than single data source. The role of fusion is to resolve such inconsistency, possibly by using some other kind of information such as reliability of sources. The rapid growth of quantity of information makes that uncertain and erroneous information also increase quickly. Dempster-Shafer evidence theory (DSET) [1, 2] is one of the most famous method for high-level information fusion. The advantage of this method is that it considers both information imprecision and uncertainty in multi-source information fusion. Though DSET has been applied in many fields, Zadeh [3] raised some questions as to the appropriateness of this combination rule. When fusing highly conflict information with Dempster’s rule of combination, it will lead counterintuitive results. Many scholars have made some efforts to solve the issue and these researches can be divided into two classes: (1) Alternatives to Dempster’s combination rules. Most of such approaches [4–6] are ac hot so that they are not in common use for all types of information. (2) Adjustment of basic probability assignment of information evidence through the degree of sources’ reliability. Murphy firstly proposed a weighted average evidence correction model [7] to effectively deal with evidence conflicts. But this © Springer Nature Switzerland AG 2020 L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 455–467, 2020. https://doi.org/10.1007/978-3-030-33506-9_41

456

J. Zhu

model considered the reliability of all information sources as the same. Actually, for multi-source information fusion, the reliability of information sources is different. Information generated by sources with high reliability are more important than information generated by sources with low reliability. At present, most scholars proposed objective measurement of reliability of information sources, subjecting to the “Majority Principle”, which means that most proportions of similar information sources can be considered reliable. A small number of sources that are dissimilar, or have outliers are unreliable. Some researchers [8–10] use a single criterion to evaluate the reliability of information sources. However, it is unreliable to use a single criterion because the single criterion is unilateral to measure the reality. Recently, multi-criteria approaches were developed. Frikha [11] proposed a novel approach for determining the discounting operator of the information provided by a set of experts based on multiple criteria using the PROMETHEE II method. de Oliveira Silva [12] analyzed conflict within DSET by using a multi-criteria analysis, for which the Multi Criteria Decision Making (MCDM) method considered was ELECTRE TRI. Sarabi-Jamab [13] selected a set of nine criteria and proposed an aggregation method to estimate the reliability degree. Although multiple criteria are more comprehensive to estimate the reliability, the process of multi-criteria approaches is very complex and the cost of computation is rather larger. Therefore, instead of choosing the most comprehensive criteria, it is better to propose a more efficient approach to combine fewer criteria. We have noticed that few scholars unified multiple criteria by distinguishing internal and external criteria. In this paper, we propose two approaches to measure the reliability of information sources based on multi-criteria. The first is to unify the criteria of internal uncertainty of information sources and the external conflict based on TOPSIS [17]. Furthermore, we proposed our first method by distinguishing the internal criteria from the external criteria, and combining them in a new way to evaluate the reliability of sources. This paper is organized as follows: In Sect. 2, we introduce the basics of DempsterShafer evidence theory and criteria of reliability in DSET. In Sect. 3, we combine three internal criteria and three external criteria to measure the reliability based on TOPSIS. After that, we propose a discounting method to combine multi-source information in Sect. 4. Numerical experiments are done to illustrate the efficiency of our proposed methods in Sect. 5. Finally, conclusion and future work is drawn in Sect. 6.

2 Preliminaries In 1967, Dempster firstly proposed a general framework for constructing uncertain inference models, and transformed the uncertainty problem of propositions into a set of uncertain problems [1]. Later, it was developed by his student Shafer in 1976, to deal with the uncertainty information, which is called Dempster-Shafer evidence theory (DSET) [2]. Evidence theory is an extension of classical probability theory. In addition, it can be considered as the generalization of the Bayesian inference to process the uncertain data associated with no exclusive hypotheses.

A New Discounting Approach to Conflict Information Fusion

2.1

457

Dempster-Shafer Evidence Theory

Assume that H ¼ fh1 ; h2    hn g is a frame of discernment (FOD), and 2H ¼ fAjA
Hg ¼ f/; fh1 g; fh1 [ h2 g; fh1 [ h3 g    Hg is a set of all the subsets on H. The subset A includes as special cases the empty set ; and the full set H. It represents a statement or proposition that the truth lies in A. A basic probability assignment (BPA) is a function m: 2H ! ½0; 1, satisfying as followed: P

mðAÞ ¼ 1 mð/Þ ¼ 0

ð1Þ

A2H

If mðAÞ [ 0, A will be called a focal element, and the union of all of the focal elements is called as the core of the mass function. For hypothesis above, a belief function Bel is defined as the sum of the basic probabilities of all subsets in A: BelðAÞ ¼

X

mðXÞ

ð2Þ

XA

Corresponding to the belief function, a plausibility function Pl is defined as: X   PlðAÞ ¼ mðXÞ ¼ 1  Bel A ð3Þ X \ A6¼;

Suppose m1 ; m2 are two sets of BPA in the frame of discernment H, they can be combined as follows: m1  m2 ðAÞ ¼

X 1 m1 ðA1 Þm2 ðA2 Þ 1  K A \ A ¼A 1

ð4Þ

2

P K ¼ A1 \ A2 ¼; m1 ðA1 Þm2 ðA2 Þ measures the total amount of conflict. Formally, we call the logarithm of the renormalization constant the weight of conflict: W ¼  log

X

fm1 ðA1 Þm2 ðA2 ÞjA1 \ A2 6¼ ;g



ð5Þ

With the incremental evidence combination, the computational complexity   increases exponentially, which is O 22jHj . 2.2

Criteria of Reliability in DSET

Pervious works on the criteria of reliability have been proposed based on two classes of criteria: (1) credibility weights estimation from the information (imprecision and amount of uncertainty) and (2) conflict estimation between information (dissimilarity

458

J. Zhu

and disparity). In the following, several criteria are presented to evaluate sources (bodies of evidence, BOE) and estimate their reliability. (1) Yager [14] proposed a entropy with the plausibility function. Em ¼ 

X

mi ðAÞ log PlðAÞ

ð6Þ

AH

(2) Dubois and Prade [15] proposed the following measure of non-specificity. Im ¼

X

mi ðAÞ logðjAjÞ

ð7Þ

AH

(3) Klir and Parviz [16] introduced Stðmi Þ to express the contradiction among disjunctive set-valued statements. Stðmi Þ ¼ 

X

" mi ðAÞ log

AH

X BH

# jA \ B j mi ðBÞ jAj

ð8Þ

(4) Frikha [11] defined the conflict between the BOE i and the artificial BOE a (majority opinion). conf ðmi ; ma Þ ¼  logð1  KÞ; ma ¼

M X 1 mj M  1 j¼1;j6¼i

ð9Þ

(5) Based on Jousselme’s distance [18] which measure distance of two sets, Martin [19] then proposed an average distance between one BOE and all other BOEs.

Average dJou

M X 1 ¼ M  1 j¼1;j6¼i

rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 1  ! !T  ! ! mi  mj D mi  mj 2

ð10Þ

Where D ¼ jjAA \[ BBjj (6) Liu [20] developed a distance between two basic belief assignments, called pignistic probability function (BetP).

A New Discounting Approach to Conflict Information Fusion

Average difBetPðAÞ ¼

M X   1 max BetPi ðAÞ  BetPj ðAÞ M  1 j¼1;j6¼i AH

459

ð11Þ

  jA \ B j jBj Our research focuses on the combination of multiple criteria rather than the choice of multiple criteria. For the reason that criteria are not unique and there will be more and more comprehensive criterion as research progressing. In next sections, we just select the six typical criteria above, which are the most widely used, to evaluate the reliability of information sources. Where BetPi ðAÞ ¼

P

BH mi ðBÞ

3 New Multi-criteria Methods of Reliability Measurement TOPSIS (Technique for Order Preference by Similarity to an Ideal Solution) [17] is one of the multi-attribute decision making methods. This method selects the most satisfactory scheme by constructing the ideal solution and the negative ideal solution of the multi-attribute problem. Then the best schemes are closest to the ideal solution and most far away from the negative ideal solution scheme. Our multi-criteria method bases on TOPSIS. 3.1

Our Method

Suppose that there are n sources of information (BOEs) fm1 ; m2 ; m3 ;    mn g and six criteria fEm ; Im ; St; conf; Average dJou ; Average difBetPg. A criteria matrix is constructed as:

where gij represents the j-th criterion of i-th BOE. Since dimension of criteria is different, it is necessary to normalize the criteria matrix for uniform dimension. Normalization Critij ¼ Critij =

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffi Xm 2 Crit kj ; i ¼ 1;    n; j ¼ 1    6 k¼1

ð12Þ

460

J. Zhu

From the viewpoint of inter-criteria analysis, we also need to determine the weights of criteria fw1 ; w2 ;    w6 g, which show the relative importance between the criteria. The most important criteria are those have the highest discrimination power regard to all sources. Shannon entropy is usually used to measure the information quality that the information with less entropy is more certain. We calculate the entropy of criteria to measure the importance which is given by Eq. (13). Entropy(Crit)j ¼ 

n   1 X gij log gij ; j ¼ 1    6 logð2nÞ i¼1

ð13Þ

The entropy of criterion is less, the more important the criterion will be. So we calculate the weight wj wj ¼ 1  Entropy(Crit)j ; j ¼ 1    6 The final weight matrix W is consist of normalized wj qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi P6 2 wj nomal ¼ wj = j¼1 wj .

ð14Þ nomal

where

After that, we get weighted normalization criteria matrix V ¼ W  Normalization Crit, which is the final matrix to estimate the reliability. Next, the discount factor ai is calculated by using the following steps:

Step 1: Let A þ ¼ V1þ ;    Vnþ be ideal value of weighted normalization criteria



 matrix where Vjþ ¼ max Vij ; i ¼ 1;    n . And let A ¼ V be the 1 ;    Vn negative ideal value of

weighted normalization criteria matrix, where V j ¼ min Vij ; i ¼ 1;    n . Step 2: Calculation of the distance from each alternative to the positive ideal value and the negative ideal value. Diþ

rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 2ffi Xn  Xn   þ  ; Di ¼ ¼ Vij  Vj Vij  Vj j¼1 j¼1

ð15Þ

The source is more reliability with the higher values of positive distance and lower values of negative distance. Step 3: Calculation of discount factors ai which is a representation of the closeness with the negative ideal value. ai ¼

D i ;i ¼ 1n Diþ þ D i

ð16Þ

A New Discounting Approach to Conflict Information Fusion

461

Algorithm 1 shows the process of TOPSIS to obtain discount factors. In addition, one advantage of TOPSIS is that we can take the closeness index of evidence and ideal solution as the discount factor. Comparing with other methods, extra steps are no need to recalculate the discount factor. 3.2

Our Improved Method

In the previous subsection, we unify internal uncertainty criteria and external conflicts criteria using TOPSIS to get discount factors. Here, we improve the combination method that distinguish internal and external criteria using TOPSIS separately in two steps to get the final criteria matrix. Firstly, criteria matrix is divided into internal criteria matrix and external criteria matrix as follows: 2

Em1 6 .. Crit in ¼ 4 . Emn

Im1 .. . Imn

3 2 conf1 St1 6 .. .. 7 . 5 and Crit ex ¼ 4 . Stn confn

djou1 .. .

3 difBetP1 7 .. 5 .

djoun

difBetPn

Secondly, calculation of weighted normalization external criteria matrix. Thirdly, using followed equation Crit in  Weighted Normalization Crit ex to combine internal and external criteria as the final matrix.

462

J. Zhu

At last, the calculation of discount factors using TOPSIS is the same as the previous subsection. Table 1 shows a numerical example of five sources information. Given that the second evidence m2 is conflict with other four evidences. Table 1. Five sources of information for the conflict numerical example m1 fh1 g 0.7 fh2 g 0 fh3 g 0.3

m2 0 0.9 0.1

m3 0.5 0.1 0.4

m4 0.6 0.3 0.1

m5 0.4 0.4 0.2

Table 2 shows discount factors of four methods. Among the four methods other than Martin’s method, the most reliable evidence is m1 and the least reliable evidence is m4 . For the same multi-criteria, our two methods consider m2 is much more reliable than Frikha’s method. The results prove that our methods are more objective and do not depend on the “majority principle” which means that most proportions of similar information sources must be more reliable than few proportions of outliers. Table 2. Discount factors of five information sources Martin Frikha Our method Our improved method

m1 0.91 1 0.98 1

m2 0.76 0.25 0.65 0.54

m3 0.94 0.03 0.05 0

m4 0.95 0.01 0.05 0.22

m5 0.95 0.1 0.24 0.26

4 Improved Dempster’s Combination Rule The discounting process is defined as follows:  mjaj ðAÞ

¼

1  aj þ aj mj ðAÞ; A ¼ H aj mj ðAÞ; AH

ð17Þ

When n pieces of evidence is collected, the weighted discount evidence is blended n − 1 times using Dempster’s combination rule (Fig. 1). Using the numerical example in 3.2 shows the effectiveness of the proposed combination rule.

A New Discounting Approach to Conflict Information Fusion

463

Fig. 1. Process of improved Dempster’s combination rule

Table 3. Results of combination 2i¼1 mi Dempster mðh1 Þ ¼ 0 mðh2 Þ ¼ 0 mðh3 Þ ¼ 1 Murphy mðh1 Þ ¼ 0:48 mðh2 Þ ¼ 0:39 mðh3 Þ ¼ 0:13 Martin mðh1 Þ ¼ 0:47 mðh2 Þ ¼ 0:18 mðh3 Þ ¼ 0:29 Frikha mðh1 Þ ¼ 0:69 mðh2 Þ ¼ 0 mðh3 Þ ¼ 0:31 Our method mðh1 Þ ¼ 0:69 mðh2 Þ ¼ 0 mðh3 Þ ¼ 0:3 Our improved mðh1 Þ ¼ 0:7 mðh2 Þ ¼ 0 mðh3 Þ ¼ 0:3

3i¼1 mi

4i¼1 mi

5i¼1 mi

mðh1 Þ ¼ 0 mðh2 Þ ¼ 0 mðh3 Þ ¼ 1

mðh1 Þ ¼ 0 mðh2 Þ ¼ 0 mðh3 Þ ¼ 1

mðh1 Þ ¼ 0 mðh2 Þ ¼ 0 mðh3 Þ ¼ 1

mðh1 Þ ¼ 0:57 mðh1 Þ ¼ 0:6 mðh1 Þ ¼ 0:65 mðh2 Þ ¼ 0:33 mðh2 Þ ¼ 0:32 mðh2 Þ ¼ 0:29 mðh3 Þ ¼ 0:1 mðh3 Þ ¼ 0:08 mðh3 Þ ¼ 0:06 mðh1 Þ ¼ 0:56 mðh1 Þ ¼ 0:69 mðh1 Þ ¼ 0:6 mðh2 Þ ¼ 0:07 mðh2 Þ ¼ 0:04 mðh2 Þ ¼ 0:0 mðh3 Þ ¼ 0:29 mðh3 Þ ¼ 0:08 mðh3 Þ ¼ 0:03 mðh1 Þ ¼ 0:69 mðh1 Þ ¼ 0:69 mðh1 Þ ¼ 0:7 mðh2 Þ ¼ 0 mðh2 Þ ¼ 0 mðh2 Þ ¼ 0 mðh3 Þ ¼ 0:31 mðh3 Þ ¼ 0:31 mðh3 Þ ¼ 0:3 mðh1 Þ ¼ 0:71 mðh2 Þ ¼ 0 mðh3 Þ ¼ 0:28 method mðh1 Þ ¼ 0:72 mðh2 Þ ¼ 0 mðh3 Þ ¼ 0:28

mðh1 Þ ¼ 0:72 mðh1 Þ ¼ 0:73 mðh2 Þ ¼ 0 mðh2 Þ ¼ 0 mðh3 Þ ¼ 0:27 mðh3 Þ ¼ 0:27 mðh1 Þ ¼ 0:75 mðh1 Þ ¼ 0:77 mðh2 Þ ¼ 0 mðh2 Þ ¼ 0 mðh3 Þ ¼ 0:25 mðh3 Þ ¼ 0:23

Table 3 summarizes the combination results obtained by Dempster’s, Murphy’s, Martin’s and Frikha’s rules as well as our two approaches proposed in this study. It also shows the evolution of the reliability of each source when sources are added sequentially. Dempster’s rule has generated counterintuitive results because of the conflict evidence m2 . Murphy’s method does not consider discount facts. The results of our two methods are most efficient in combining with the conflict evidences.

464

J. Zhu

5 Experiment In this section, we use two different experiments to illustrate the effect of our two methods. The first experiment shows the performance of our improved method by the discount factors. The second experiment verifies the impact of multiple criteria on combination results generally, comparing with Dempster’s rule, Murphy’s approach, Deng’s approach [21] and Martin’s approach. 5.1

Discount Factors Examination in Our Methods

We assume the frame of discernment H ¼ fh1 ; h2 ; h3 g with five focal elements ffh1 g; fh2 g; fh3 g; fh1 ; h2 g; fh1 ; h2 ; h3 gg. The example is simulated by 50 iterations. In each iteration, 25 random information sources are generated sequentially through the following structure: (1) The first six information sources are generated in favor of fh1 g mðfh1 gÞ ¼ 0:45 þ r  0:1; mðfh2 gÞ ¼ 0:15  s  0:1; mðfh3 gÞ ¼ 0:15  r  0:1; mðfh1 ; h2 gÞ ¼ 0:15 þ s  0:15; mðfh1 ; h2 ; h3 gÞ ¼ 0:1  s  0:05 (2) The next eight information sources are generated in favor of fh3 g mðfh1 gÞ ¼ 0:15  s  0:1; mðfh2 gÞ ¼ 0:15  r  0:1; mðfh3 gÞ ¼ 0:45 þ r  0:1; mðfh1 ; h2 gÞ ¼ 0:15 þ s  0:15; mðfh1 ; h2 ; h3 gÞ ¼ 0:1  s  0:05 (3) The last eleven information sources are generated in favor of fh1 g, which is same as the first six sources. Figure 2(a) shows that is the reliability of the first six sources decrease when we add the next eight information sources because they are conflicting. On the other hand, it increases when last eleven sources are added, which are also support fh1 g. Figure 2(b) shows that the reliability of next eight sources is very low at first, which is nearly close to zero, then raises a little as the number of sources increases gradually. After the last eleven sources join in, it is close to zero again. Figure 2(c) shows that the reliability of the last eleven is very high since they are not conflict with most sources. 5.2

General Numerical Example

We apply our method to more general numerical examples. As in the previous subsection, we also assume the frame of discernment H ¼ fh1 ; h2 ; h3 g with five focal elements ffh1 g; fh2 g; fh3 g; fh1 ; h2 g; fh1 ; h2 ; h3 gg. There are ten information sources (bodies of evidence) fm1 ; m2 ; m3 ;    m10 g. The BPA of each source is given a random uniform number in the intervals of (0, 1). A set of randomly generated BPAs is shown in Fig. 3 that m7 is an obvious outlier.

A New Discounting Approach to Conflict Information Fusion

465

Fig. 2. The evolution of the reliability of each BPA by the sequentially. (a) the first six sources, (b) the next eight sources, (c) the last eleven sources

Fig. 3. Random BPA of ten information sources by simulation.

466

J. Zhu

Fig. 4. Combination results of different discounting approaches. (a) Dempster, (b) Murphy, (c) Deng, (d) Martin, (e) our method, (f) our improved method.

Figure 4 shows the combination results of four different methods as well as our methods. Inaccurate results are obtained in Murphy’s and Deng’s approach. When m7 is added, the combination results of both Dempster’s rule and Martin’s approach have large fluctuation. It can be seen that our methods, especially our improved method, obtain the most stable results and can be least affected by conflict sources. After all, it can fully verify that our methods adapt to the general situation.

6 Conclusion and Future Work This paper studies the method to make comprehensive estimation with multi-criteria, rather than choice of the set of criteria. Our multi-criteria methods select six criteria to measure the reliability of sources and TOPSIS is used to aggregate these criteria to calculate discount factors. We also proposed two methods at the same time. The latter one separates internal and external conflicts to measure reliability of sources. Using the discount factors, we improved Dempster’s rule to combine conflict information. Through some different experiments, our methods obtain more objective discount factors and best combination results. Furthermore, it can be extended to general situation to combine various types of information. The accuracy of basic probability assignment (BPA) directly affects the fusion results. Therefore, the method of generating BPA is very important. We will further study to quantify the conflict of evidence in the conversion process of BPA. It would be efficient to apply our methods on the real data.

A New Discounting Approach to Conflict Information Fusion

467

References 1. Dempster, A.: Upper and lower probabilities induced by multivalued mapping. Annu. Math. Stat. 38(2), 325–339 (1967) 2. Shafer, G.: A Mathematical Theory of Evidence. Princeton University Press, Princeton (1976) 3. Zadeh, L.A.: Review of ‘‘a mathematics theory of evidence’’. AI Mag. 5(3), 81–83 (1984) 4. Dezert, J.: Foundations for a new theory of plausible and paradoxical reasoning. Inf. Secur. 9, 13–57 (2002) 5. Lefevre, E., Colot, O., Vannoorenberghe, P.: Belief function combination and the conflict management. Inf. Fusion 3(2), 149–162 (2002) 6. Smets, P.: The combination of evidence in the transferable belief model. IEEE Trans. Pattern Anal. Mach. Intell. 12(5), 447–458 (1990) 7. Murphy, C.K.: Combining belief functions when evidence conflicts. Decis. Support Syst. 29 (1), 1–9 (2000) 8. Xia, J., Feng, Y., Liu, L., Liu, D., Fei, L.: An evidential reliability indicator-based fusion rule for Dempster-Shafer theory and its applications in classification. IEEE Access 6, 24912– 24924 (2018) 9. Marhic, B., et al.: An evidential approach for detection of abnormal behaviour in the presence of unreliable sensors. Inf. Fusion 13(2), 146–160 (2012) 10. Huynh, V.-N.: Discounting and combination scheme in evidence theory for dealing with conflict in information fusion. In: MDAI 2009, pp. 217–230 (2009) 11. Frikha, A.: On the use of a multi-criteria approach for reliability estimation in belief function theory. Inf. Fusion 18, 20–32 (2014) 12. de Oliveira Silva, L.G., de Almeida Filho, A.T.: A multicriteria approach for analysis of conflicts in evidence theory. Inf. Sci. 346–347, 275–285 (2016) 13. Sarabi-Jamab, A., Araabi, B.N.: How to decide when the sources of evidence are unreliable: a multi-criteria discounting approach in the Dempster-Shafer theory. Inf. Sci. 448–449, 233– 248 (2018) 14. Yager, R.R.: Entropy and specificity in a mathematical theory of evidence. Classic Works of the Dempster-Shafer Theory of Belief Functions, pp. 291–310 (2008) 15. Dubois, D., Prade, H.: A note on measures of specificity for fuzzy sets. Int. J. Gen. Syst. 10 (4), 279–283 (1985) 16. Klir, G.J., Parviz, B.: A note on the measure of discord. In: UAI 1992, pp. 138–141 (1992) 17. Hwang, C.L., Yoon, K.: Multiple Attribute Decision Making. Springer, Berlin (1981) 18. Jousselme, A.-L., Grenier, D., Bossé, É.: A new distance between two bodies of evidence. Inf. Fusion 2(2), 91–101 (2001) 19. Martin, A., Jousselme, A.-L., Osswald, C.: Conflict measure for the discounting operation on belief functions. In: FUSION 2008, pp. 1–8 (2008) 20. Liu, W.: Analyzing the degree of conflict among belief functions. Artif. Intell. 170(11), 909– 924 (2006) 21. Yong, D., Shi, W., Zhu, Z., Qi, L.: Combining belief functions based on distance of evidence. Decis. Support Syst. 38(3), 489–493 (2004)

The 21th International Symposium on Multimedia Network Systems and Applications (MNSA-2019)

The Group-Based Linear Time Causally Ordering Protocol in a Scalable P2PPS System Takumi Saito1(B) , Shigenari Nakamura1 , Tomoya Enokido2 , and Makoto Takizawa1 1 Hosei University, Tokyo, Japan [email protected],[email protected], [email protected] 2 Rissho University, Tokyo, Japan [email protected]

Abstract. The P2PPS (peer-to-peer type of a topic-based PS (publish/ subscribe)) model is a distributed model where each peer subscribes interesting topics and publishes messages with topics. Messages which have some common publication topic are related with one another. In the P2PPS model, only messages related with respect to topics are required to be causally delivered to peers. In our previous studies, each message is broadcast to every peer and each peer only receives messages whose publication topics include some subscription topic. In scalable systems, it is not easy, maybe difficult to broadcast messages. The two-layered P2PPS (2P2PPS) model is proposed in our previous papers in order to efficiently publish and receive messages and causally deliver related messages in a scalable system. The 2P2PPS model is a group of peers which is composed of subgroups. Peers of each subgroup are interconnected in a local area network (LAN) and the subgroups are interconnected in a wide area network (WAN). Here, messages are broadcast to every member peer in each subgroup and messages are unicast among subgroups. We propose a TLCO (Two-Layered Causally Ordering) protocol which uses the linear time vector to causally deliver related messages in a 2P2PPS model in this paper. In the evaluation, we show the number of pairs of unnecessarily ordered messages can be reduced in the TLCO protocol than the linear time (LT) protocol. Keywords: Topic-based publish/subscribe system · P2P model · 2P2PPS (two-layered P2PPS) model · Unnecessarily ordered messages · TLCO (Two-Layered Causally Ordering) protocol · Linear time vector (LV)

1

Introduction

The publish/subscribe (PS) model [13] is an event-driven, contents-based model of a distributed system which is composed of processes which communicate with c Springer Nature Switzerland AG 2020  L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 471–482, 2020. https://doi.org/10.1007/978-3-030-33506-9_42

472

T. Saito et al.

one another by exchanging messages. In the P2PPS (P2P (peer-to-peer) type of topic-based PS) model [1,7,8], each peer process (peer) can both publish and subscribe messages in a distributed manner [2,4]. Each peer publishes a message with publication topics. Messages published by peers are brought to peers in underlying networks. A peer receives only a message, some of whose publication topics are subscription topics of the peer. Here, the peer is a target peer of the message. A pair of messages which have some common publication topic are related with each other in the topic-based PS systems. A message m1 causally precedes another message m2 (m1 → m2 ) if and only if (iff) the publication event of the message m1 happens before the message m2 according to the causality theory [5]. Each peer is required to causally deliver only related messages in the P2PPS model. In order to causally deliver messages related with respect to topics, the topic vector is proposed [8,12]. The size of the topic vector is O(tpn) for total number tpn of topics in a system. If topics are scalable, it is not easy for each message to carry the topic vector due to the message overhead. In addition, each peer is assumed to be able to broadcast messages to all the peers in a system. However, it is difficult, maybe impossible for each peer to broadcast messages to all the peers in scalable systems. In our previous paper [10], we propose a hierarchical model, i.e. the two-layered P2PPS (2P2PPS) model is proposed in order to efficiently causally deliver related messages in a scalable system. Here, the 2P2PPS model is composed of exclusive member subgroups of peers. Each member subgroup is composed of member peers. The size of each member subgroup is decided to be so small that every message published by a member peer can be efficiently broadcast to every member peer. Each member subgroup includes a gateway peer. A subgroup composed of gateway peers is a gateway subgroup. Messages published by member peers in a member subgroup are forwarded to other member subgroups by a gateway peer in the gateway subgroup with unicast communication. In order to causally deliver messages in a hierarchical P2PPS group of g (≥ 1) member subgroups, a linear time vector LV (= LV1 , ..., LVg ) is newly proposed in this paper. The tth element LVt of a linear time vector LV shows linear time [5] used in a member subgroup Gt . Each message m published by a peer in a member subgroup carries the linear time vector m.LV of the peer. Messages received by a peer are ordered in the linear time vectors of the messages. The message length is O(g) for number g of member subgroups. In this paper, we propose a T LCO (Two-Layered Causally Ordering) protocol for a 2P2PPS model, where related messages are causally delivered to target peers by using the linear time vectors. If a message m1 causally precedes a message m2 , the linear time vector m1 .LV of the message m1 is smaller than the linear time vector m2 .LV of the message m2 . On the other hand, even if m1 .LV is smaller than m2 .LV , the message m1 may not causally precede the message m2 . Here, if a peer pi receives the messages m2 but does not receive the message m1 , the peer pi has to wait for the message m1 after receiving the message m2 . After receiving the message m1 , if the peer pi delivers the message m1 before the message m2 , e.g.

The Group-Based Linear Time Causally Ordering Protocol

473

by comparing the linear time vectors of the messages, the messages m1 and m2 are unnecessarily ordered in the peer pi . We show the number of unnecessarily ordered messages can be reduced in the linear time vector compared with the linear time (LT) protocol [5] in the evaluation. In Sect. 2, we present a system model. In Sect. 3, we propose the TLCO protocol for a two-layered group of peers. In Sect. 4, we evaluate the TLCO protocol.

2

System Model

A P2PPS (peer-to-peer type of publish/subscribe) model is a distributed topicbased PS (publish/subscribe) model [9]. Let T be a set of topics tp1 , ..., tptpn (tpn ≥ 1). Each peer process (peer) subscribes interesting topics named subscription topic. A peer publishes message with publication topics which denote the contents of the message. On arrival of a message, a peer receives the message whose publication topics include some subscription topic. A pair of messages m1 and m2 are related if and only if (iff) the publication topics of the messages m1 and m2 include some common topic. A 2P2PPS group G [10] of member peers is composed of subgroups G0 , G1 , ..., Gg (g ≥ 1) as shown in Fig. 1. Here, G0 is a gateway subgroup and G1 , ..., Gg are member subgroups. Each member subgroup Gt is composed of member peers pt1 , ..., pt,lt (lt > 1) and a gateway peer pt0 (t = 1, ..., g). A gateway peer pt0 belongs to not only a member subgroup Gt but also the gateway subgroup G0 while each member peer pti belongs to only one member subgroup Gt . The gateway subgroup G0 is composed of gateway peers p10 , ..., pg0 of member subgroups G1 , ..., Gg , respectively. Each gateway peer pt0 belongs to both a member subgroup Gt and a gateway subgroup G0 . In each member subgroup Gt , every message m published by a member peer pti is broadcast to all the member peers pt1 , ..., pt,lt and the gateway peer pt0 by taking advantage of the underlying network service. For example, all the member peers pt1 , ..., pt,lt in a member subgroup Gt are interconnected in a local area network (LAN) with broadcast communication facility. Let pti .S (⊆ T ) be a set of subscription topics of the peer pti . Every message published by a member peer pti arrives at each member peer ptj in a member subgroup Gt by taking usage of the underly network. Each message m carries the publication topics m.P which denote the contents of the message m. If m.P ∩ pti .S = φ, member peer pti receives the message m. A gateway peer pt0 receives every message m sent by a member peer pti in a member subgroup Gt . On receipt of a message m from a member peer, a gateway peer pt0 forwards the message m to gateway peers of other subgroups by using the unicast communication in the gateway subgroup G0 . For example, gateway peers p10 , ..., pg0 of a gateway subgroup G0 communicate with one another by using TCP [3] in a wide area network (WAN). A gateway peer pt0 has the subscription pt0 .S which is a set of subscription topics of all the member peers in the member subgroup Gt , i.e. pt0 .S = ∪pti ∈Gt pti .S. A gateway peer pt0 knows about the subscription pu0 .S of every other gateway peer pu0 . On receipt of a message m from another gateway peer pu0 , a gateway

474

T. Saito et al.

Fig. 1. Two-layered group.

peer pt0 receives the message m if pt0 .S ∩ m.P = φ. Then, the gateway peer pt0 broadcasts the message m to every member peer in the member subgroup Gt . Here, every pair of related messages have to be causally delivered to member peers in member subgroups. We assume the underlying network of each subgroup is reliable. That is, in every subgroup, every pair of messages m1 and m2 published by a peer is delivered to every peer in the publication order without any message loss and any duplication. However, some pair of messages published by different peers may be delivered to different peers in different orders. A message m1 causally precedes a message m2 (m1 → m2 ) iff the publication event of the message m1 happens before the publication event of the message m2 according to the causality theory [5]. In the P2PPS system, even if m1 → m2 , a common target member peer of the messages m1 and m2 is not required to deliver a messages m1 before another message m2 if the messages m1 and m2 are not related with each other. A message m1 has to be delivered before another message m2 in a common target peer if m1 → m2 and the messages m1 and m2 are related. It is decided whether or not a pair of messages are related by checking the publication topics of the message.

The Group-Based Linear Time Causally Ordering Protocol

3

475

A TLCO (Two-layered Causally Ordering) Protocol

We propose a TLCO (Two-Layered Causally Ordering) protocol to causally deliver related messages to target peers in the 2P2PPS model. A 2P2PPS model G is composed of member subgroups G1 , ..., Gg and a gateway subgroup G0 as shown in Figure 1. Each member peer pti of a member subgroup Gt manipulates a linear time vector LV = LV1 , ..., LVg  and each message m carries a linear time vector LV = LV1 , ..., LVg  of a source member peer pti . The tth element LVt shows the linear time of a member subgroup Gt (t = 1, ..., g). Here, pti .LV stands for a linear time vector (LV ) variable LV = LV1 , ..., LVg  of a peer pti in a member subgroup Gt . Initially, pti .LV = 0, ..., 0 for every peer pti (i = 0, ..., lt ). The LV field m.LV of each message m shows the linear time vector LV carried by the message m. The length of each message m is O(g) where g is the number of member subgroups, since the message m carries the linear time vector field m.LV . Each member peer pti of a member subgroup Gt publishes and receives a message m by using topics and causally delivers messages by manipulating the LV variable pti .LV and the LV field m.LV . T is a set of topic tp1 , ..., tptpn in G. Each peer publishes and receives a message as follows: [Publication of a message m at pti ]. A member peer pti publishes a message m as follows. 1. m.P = publication topics (⊆ T ); 2. The tth element pti .LVt in the LV variable pti .LV = LV1 , ..., LVg  is incremented by one in a member peer pti , i.e. pti .LVt = pti .LVt + 1; 3. For the LV field m.LV of the messages m, m.LV = pti .LV (= LV1 , ..., LVg ); 4. The peer pti publishes the message m; [Gateway peer pt0 receives a message from a member peer pti ] 1. A gateway peer pt0 receives a message m published by a member peer pti in a member subgroup Gt ; 2. The gateway peer pt0 forwards the message m to every gateway peer pu0 of a member subgroup Gu such that pu0 .S ∩ m.P = φ in the gateway subgroup G0 ; [Gateway peer pt0 receives a message from a gateway peer pu0 ] 1. A gateway peer pt0 receives a message m from another gateway peer pu0 in a gateway subgroup G0 ; 2. if pt0 .S ∩ m.P = φ, the gateway peer pt0 neglects the message m; 3. pt0 .LVu = max(pt0 .LVu , m.LVu ) (for u = 1, ..., g); 4. pt0 .LVt = max(pt0 .LV1 , ..., pt0 .LVg ); 5. The gateway peer pt0 publishes the message m in the member subgroup Gt ;

476

T. Saito et al.

[Receipt of a message]. A member peer pti receives a message m as follows. 1. 2. 3. 4. 5.

A message m arrives at a member peer pti in a member subgroup Gt ; pti .LVu = max(pti .LVu , m.LVu ) (for u = 1, ..., g); pti .LVt = max(pti .LV1 , ..., pti .LVg ); if m.P ∩ pti .S = φ, the member peer pti receives the message m; Otherwise, the member peer pti neglects the message m;

The publication topics of the message m are given to the publication field m.P of the message m. Each member peer pti first increments the tth element pti .LVt of the LV variable pti .LV by one in a member subgroup Gt , pti .LVt = pti .LVt +1. Then, the LV variable pti .LV is set to the LV field m.LV, m.LV = pti .LV . The member peer pti publishes the message m. Every message published by each member peer pti arrives at not only every member peer but also a gateway peer pt0 in a member subgroup Gt . On arrival of a message m from a member peer pti , a gateway peer pt0 forwards the message m to other gateway peers in the gateway subgroup G0 by using the reliable unicast communication like TCP [3]. Each gateway peer pt0 knows the subscription pu0 .S of every other gateway peer pu0 . A gateway peer pt0 forwards the message m to only a gateway peer pu0 whose pu0 .S ∩ m.P = φ by using one-to-one communication. Messages sent by another gateway peer pu0 are forwarded to a gateway peer pt0 in a gateway subgroup G0 . On receipt of a message m from another gateway peer pu0 in the gateway subgroup G0 , a gateway peer pt0 first changes each element pt0 .LVu of an LV variable pt0 .LV with a larger one of pt0 .LVu and m.LVu , i.e. pt0 .LVu = max(pt0 .LVu , m.LVu ) for u = 1, ..., g. Then, the largest one of the elements pt0 .LV1 , ..., pt0 .LVg of the LV variable pt0 .LV is taken to be the element pt0 .LVt , i.e. pt0 .LVt = max(pt0 .LV1 , ..., pt0 .LVg ). The gateway peer pt0 broadcasts the message m in the member subgroup Gt . It is noted the LV field m.LV of the message m is not changed. Messages published by member peers in another member subgroup Gu are forwarded to a member peer pti in a member subgroup Gt by a gateway peer pt0 . In addition, a member peer pti receives messages published in the member subgroup Gt . On receipt of a message m, a member peer pti takes a larger one of pti .LVu and m.LVu as a new value of the LV variable pti .LVu , i.e. pti .LVu = max(pti .LVu , m.LVu ) for u = 1, ..., g. In addition, the LV variable pti .LVt is changed with the largest one of the elements pti .LV1 , ..., pti .LVg in the LV variable pti .LV . The size of a message m is O(g) for number g of member subgroups. In the topic vector [9,11,12], the size of a message m is O(tpn) for number tpn of topics. The number g of member subgroups is smaller than the number tpn of topics. Furthermore, the topic set T might be changed as new topics are added and obsolete topics are deleted. It is not easy to change the topic vector as the membership of the topic set T is changed. In the TLCO protocol, the publication topics m.P of each message m and subscription topics pti .S of each member peer pti are implemented in a bitmap

The Group-Based Linear Time Causally Ordering Protocol

477

to reduce the size of the message m. The bitmap is tpn bits long for number tpn of topics. The kth bit is 1 if a topic tpk is included, else 0. On receipt of messages, a member peer pti stores the messages in the receipt buffer RBti . If a peer pti receives a message m1 before another message m2 , m1 ≺≺ti m2 . Then, messages in the buffer RBti are ordered in the linear time vector LV fields. That is, for a pair of related messages m1 and m2 in the buffer RBti , m1 and m2 are ordered, i.e. exchanged if m2 ≺≺ti m1 but m1 .LV < m2 .LV . Messages in the receipt buffer RBti are delivered in the order of the LV fields, i.e. a message m1 is delivered before a message m2 (m1 ≺ m2 ) if m1 .LV < m2 .LV and the messages m1 and m2 are related. However, even if m1 .LV < m2 .LV , the related message m1 may not causally precede the related message m2 .

Fig. 2. Linear time vector (LV).

[Example 1]. In Fig. 2, there are a pair of member subgroups Gt and Gu which are interconnected in a gateway subgroup G0 . The gateway subgroup G0 is composed of a pair of gateway peers pt0 and pu0 of the member subgroups Gt and Gu , respectively. The linear time vector LV is composed of two elements LVt , LVu  since there are two member subgroups Gt and Gu . Initially, pvi .LV = 0, 0 in every member peer pvi (v = t, u). A member peer puk in the member subgroup Gu first publishes a message mu1 . Here, the element puk .LVu is incremented by one, i.e. puk .LV = 0, 1. The LV field mu1 .LV is 0, 1. A member peer puj receives the message mu1 in the member subgroup Gu . Here, since mu1 .LVu (= 1) > puj .LVu (= 0), the second element puj .LVu in the LV variable puj .LV is changed with 1. The LV variable puj .LV is 0, 1. Then, the member peer puj publishes a pair of messages mu2 and mu3 with the LV fields mu2 .LV = 0, 2 and mu3 .LV = 0, 3, respectively.

478

T. Saito et al.

On receipt of the message mu1 , a gateway peer pu0 forwards the message mu1 to the gateway peer pt0 . Here, mu1 .LV = 0, 1. The gateway peer pt0 receives the message mu1 in the gateway subgroup G0 . Here, since mu1 .LVu (= 1) > pt0 .LVu (= 0), the second element pt0 .LVu in the LV variable pt0 .LV is changed with 1, i.e. pt0 .LV = 0, 1. Then, the first element LVt in the LV variable pt0 .LV is changed with the maximum value of pt0 .LVt (= 0) and pt0 .LVu (= 1), i.e. 1. The LV variable pt0 .LV is 1, 1. Then, the gateway peer pt0 publishes the message mu1 in the member subgroup Gt . The message mu1 carries the LV field mu1 .LV = 0, 1 to target member peers in the member subgroup Gt . On receipt of the message mu1 with mu1 .LV = 0, 1, a member peer pti in another member subgroup Gt changes the LV variable pti .LV with 0, 1. Then, in a way similar to the gateway peer pt0 , the maximum value of the elements of the LV variable pti .LV , i.e. 1 is taken as the element pti .LVt . Here, pti .LV = 1, 1. Then, the member peer pti publishes a message mt1 with the linear time vectors mt1 .LV = 2, 1, i.e. the publication event of the message mu1 happens before the message mt1 . After receiving the message mu2 , the gateway peer pu0 receives the message mt1 with the LV field mt1 .LV = 2, 1 in the gateway subgroup G0 where pu0 .LV = 0, 2. The LV variable pu0 .LV is changed with 2, 2 since pu0 .LVt is 0 but mt1 .LVt is 2. The gateway peer pu0 receives the message mt1 in the gateway subgroup G0 and publishes the message mt1 in the member subgroup Gu . The member peer puk receives the message mt1 after receiving the messages mu2 and mu3 . Hence, the member peer puk receives the message mt1 with mt1 .LV = 2, 1 where puk .LV = 0, 3. The LV variable puk .LV of the peer puk is changed with 2, 3. Next, let us consider the linear time (LT) protocol [5]. Suppose member peers and gateway peers manipulate linear time (LT ) variable and each message m carries an LT field m.LT as shown in Fig. 2. Initially, pti .LT = 0 in every member peer pti . Here, since mt1 .LT (= 2) < mu3 .LT (= 3), the message mt1 is delivered before the message mu3 . On the other hand, a pair of the LV fields mt1 .LV (= 2, 1) and mu3 .LV (= 0, 3) are not comparable (mu1 .LV || mt3 .LV ). Hence, the message mt1 is not delivered before the message mu3 since the message mu3 is received before the message mt1 (mu3 ≺≺uk mt1 ). Thus, a pair of messages mt1 and mu3 are unnecessarily ordered in the LT protocol but not in the TLCO protocol.

4

Evaluation

In the TLCO protocol proposed in this paper, messages are ordered by the linear time vector (LV ). If a message m1 causally precedes a message m2 (m1 → m2 ), m1 .LV < m2 .LV . However, even if m1 .LV < m2 .LV, m1 may not causally precede m2 . Next, suppose every member peer pti manipulates a linear time (LT ) variable pti .LT [5]. Initially, pti .LT = 0. Each message m carries the linear time (LT )

The Group-Based Linear Time Causally Ordering Protocol

479

field m.LT which is linear time pti .LT of a source member peer pti . Here, each member peer pti increments the LT variable pti .LT by one and sends a message m with the LT field m.LT (= pti .LT ). On receipt of a message m, a member peer pti changes the LT variable pti .LT with max(pti .LT, m.LT ). Here, if a message m1 causally precedes m2 (m1 → m2 ), m1 .LT < m2 .LT [5]. Here, If m1 .LV < m2 , LV, m1 .LT < m2 .LT . This means, even if m1 .LT < m2 .LT, m1 .LV may not be smaller than m2 .LV . A pair of related messages m1 and m2 are unnecessarily ordered in a peer pti with a protocol iff the peer pti receives the message m2 before the message m1 (m2 ≺≺ti m1 ) and the message m1 is delivered before the message m2 (m1 ≺ m2 ) but m1 does not causally precede m2 (m1 → m2 ). Next, we evaluate the TLCO protocol in terms of number of messages unnecessarily ordered compared with the linear time (LT) protocol [5] and vector time (V T ) protocol [6]. A system includes n member peers p1 , ..., pn (n ≥ 1). A group G is composed of a gateway subgroup G0 and member subgroups G1 , ..., Gg (g > 1). Each member subgroup Gt is composed of the same number of n/g member peers. Let T be a set of topics tp1 , ...tptpn (tpn ≥ 1) in a system. Each member peer pti randomly publishes messages. That is, the publication time m.P BTti of each message m published by a member peer pti is randomly decided from time 0 to time maxT [time unit (tu)]. In the evaluation, maxT is 300 [tu]. At each time unit, a member peer pti is randomly taken in all the peers. Then, the member peer pti publishes one message m in the member subgroup Gt . The receiving time m.RV Ttj of a message m of each member peer ptj is m.P BTti + δtij where δtij is delay time between the peers pti and ptj in the member subgroup Gt . The delay time δtij is randomly taken from 1 to 10 [tu]. The delay time δt0 ,u0 between a pair of gateway peers pt0 and pu0 in the gateway subgroup G0 is also randomly taken from 1 to 10 [tu]. If the gateway peer pt0 receives a message m from a member peer pti in the member subgroup Gt , the gateway peer pt0 forwards the message m to every gateway peer pu0 where pu0 .S ∩ m.P = φ. If the gateway peer pu0 receives the message m from another gateway peer pt0 , the gateway peer pu0 publishes the message m to all the member peers pu1 , ..., pu,lu in the member subgroup Gu . In the evaluation, publication and subscription topics of each member peer pti are randomly taken from the topic set T . Each member peer pti manipulates the LV variable pti .LV . Each message m carries the LV field m.LV and VT field m.V T for evaluation. Each member peer pti also manipulates the linear time (LT ) variable pti .LT and the V T variable pti .V T . In each member peer pti , messages received are stored into the buffer RBti in the receipt order, i.e. m1 precedes m2 in the buffer (m1 ≺≺ti m2 ) if m1 .RV Tti < m2 .RV Tti , i.e. a member peer pti receives a message m2 before another message m1 . In the TLCO protocol, if m2 .LV < m1 .LV , the messages m1 and m2 are exchanged, i.e. ordered by a peer pti in the receipt buffer RBti . Here, unless the message m1 causally precedes the message m2 (m1 → m2 ) and the messages m1 and m2 are related, there is no need to exchange the messages in the receipt buffer, i.e. m1 and m2 are unnecessarily ordered in the TLCO protocol. We measure the numbers of pairs of messages unnecessarily ordered

480

T. Saito et al.

in the TLCO protocol and the LT protocol. In the VT protocol, a message m1 causally precedes a message m2 (m1 → m2 ) if and only if (iff) m1 .V T < m2 .V T [6]. Hence, no message is unnecessarily ordered. We show the number of pairs of unnecessarily ordered messages in the TLCO protocol and the LT protocol. OM T is a set {m1 , m2  | m1 ≺≺ti m2 for a peer pti but m2 .LV < m1 .LV } of messages ordered in the TLCO protocol. OM L is a set {m1 , m2  | m1 ≺≺ti m2 for a pti but m2 .LT < m1 .LT } of messages ordered in the LT protocol. OM V is a set {m1 , m2  | m1 ≺≺ti m2 for a pti but m2 .V T < m1 .V T }. It is noted that ≺≺ti is the same in every protocol. The numbers N OM T , N OM L, and N OM V of unnecessarily ordered pairs of messages in the TLCO, LT, and VT protocols, respectively, are | OM T − OM V |, | OM L − OM V |, and 0, respectively.

Fig. 3. Number of pairs of messages ordered in each protocol.

Figure 3 shows the numbers N OM L, N OM V , and N OM T of pairs of messages ordered in the LT, VT, and the TLCO protocols, respectively, for number g of subgroups. Here, there are forty member peers (n = 40), and thirty topics (tpn = 30). We changed the number of subgroups to 1, 2, 4, 5, 8, 10 and 20. Subscription topics pti .S of each member peer pti are randomly selected from the topic set T . The number of subscription topics of each member peer pti is five. The difference between the LT and TLCO protocols and VT protocol shows how many number of pairs of messages are unnecessarily ordered in the LT protocol and in the TLCO protocol. The number of pairs of ordered messages in the LT and TLCO protocols is same in one subgroup because the behaviors of the LT and TLCO protocols are same in one subgroup. Hence, the numbers

The Group-Based Linear Time Causally Ordering Protocol

481

of pairs of unnecessarily ordered messages in the LT and TLCO protocols is same, i.e. | OM L − OM V |=| OM T − OM V | for g = 1. Then, the number of pairs of messages ordered in each protocol in one subgroup is very few because the delay time between a pair of member subgroups is 0 in one subgroup. The number of pairs of messages ordered in the TLCO protocol is more similar to the number of pairs of messages ordered in the VT protocol as the number g of subgroups increases. In addition, the number of elements in the linear time vector is the number g of subgroups while the number of elements in the vector time is number n of peers, n = 40.

5

Concluding Remarks

In the P2PPS system, a pair of messages which have a common publication topic are related. Related messages have to be causally delivered in every target peer. Current information systems are scalable like IoT. Related messages are required to be efficiently causally delivered in a scalable system. In this paper, we newly proposed the TLCO protocol where a system is composed of member subgroups of peers and a gateway subgroup. Here, the linear time vector (LV ) is newly proposed to causally deliver related messages in a system which is composed of subgroups. In each member subgroup, messages published by a member peer are broadcast to every member peer. Member subgroups are interconnected with gateway peers. On receipt of a message m published by a member peer in a member subgroup, a gateway peer forwards the message m to other gateway peers by unicast communication. The target gateway peer broadcasts the message in the member subgroup. If a message m1 causally precedes a message m2 (m1 → m2 ), the linear time vector (LV ) field m1 .LV of the message m1 is smaller than the LV field m2 .LV of the message m2 (m1 .LV < m2 .LV ). In the linear time (LT) protocol [5], the field m.LT shows the linear time of a message m. Here, we showed m1 .LT < m2 .LT if m1 .LV < m2 .LV . In the evaluation, we showed the number of pairs of messages unnecessarily ordered can be reduced in the TLCO protocol compared with the linear time (LT) protocol. We are now evaluating the TLCO protocol in a scalable system.

References 1. Google alert. http://www.google.com/alerts 2. Blanco, R., Alencar, P.: Event models in distributed event based systems. In: Principles and Applications of Distributed Event-Based Systems, pp. 19–42 (2010) 3. Comer, D.E.: Internetworking with TCP/IP (VOLUME I). Prentice Hall, Upper Saddle River (1991) 4. Hinze, A., Buchmann, A.: Principles and Applications of Distributed Event-Based Systems. IGI Global, Pennsylvania (2010) 5. Lamport, L.: Time, clocks, and the ordering of event in a distributed systems. Commun. ACM 21(7), 558–565 (1978) 6. Mattern, F.: Virtual time and global states of distributed systems. In: Parallel and Distributed Algorithms, pp. 215–226 (1988)

482

T. Saito et al.

7. Nakayama, H., Duolikun, D., Enokido, T., Takizawa, M.: Selective delivery of event messages in peer-to-peer topic-based publish/subscribe systems. In: Proceedings of the 18th International Conference on Network-Based Information Systems, NBiS2015, pp. 379–386 (2015) 8. Nakayama, H., Duolikun, D., Enokido, T., Takizawa, M.: Reduction of unnecessarily ordered event messages in peer-to-peer model of topic-based publish/subscribe systems. In: Proceedings of IEEE the 30th International Conference on Advanced Information Networking and Applications, AINA-2016, pp. 1160–1167 (2016) 9. Nakayama, H., Ogawa, E., Nakamura, S., Enokido, T., Takizawa, M.: Topic-based selective delivery of event messages in peer-to-peer model of publish/subscribe systems in heterogeneous networks. In: Proceedings of the 18th International Conference on Network-Based Information Systems, WAINA-2017, pp. 1162–1168 (2017) 10. Saito, T., Nakamura, S., Duolikun, D., Enokido, T., Takizawa, M.: A hierarchical group of peers in publish/subscribe systems. In: Proceedings of the 11th Intelligent Networking and Collaborative Systems, INCoS-2019 (2019, accepted) 11. Saito, T., Nakamura, S., Duolikun, D., Enokido, T., Takizawa, M.: Object-based selective delivery of event messages in topic-based publish/subscribe systems. In: Proceedings of the 13th International Conference on Broadband and Wireless Computing, Communication and Applications, BWCCA-2018, pp. 444–455 (2018) 12. Saito, T., Nakamura, S., Duolikun, D., Enokido, T., Takizawa, M.: Evaluation of TBC and OBC precedent relations among messages in P2P type of topicbased publish/subscribe system. In: Proceedings of the Workshops of the 33rd International Conference on Advanced Information Networking and Applications, WAINA-2019, pp. 570–581 (2019) 13. Tarkoma, S.: Publish/Subscribe System: Design and Principles, 1st edn. Wiley, Hoboken (2012)

Algorithm for Detecting Implicitly Faulty Replicas Based on the Power Consumption Model Hazuki Ishii(B) , Ryuji Oma, Shigenari Nakamura, Tomoya Enokido, and Makoto Takizawa Hosei University, Tokyo, Japan {hazuki.ishii.5h, ryuji.oma.6r}@stu.hosei.ac.jp, [email protected], [email protected], [email protected]

Abstract. A system can be fault-tolerant by replicating each application process on multiple servers. Each replica is performed on a host server. According to the advances of hardware and architecture technologies of servers, each server can be considered to be free of fault, i.e. always proper. On the other hand, replicas of application processes easily suffer from faults, e.g. infected with virus. A faulty replica may send a proper reply, e.g. wiretapped reply. A replica which sends a proper reply but does faulty computation is implicitly faulty. Implicitly faulty replicas cannot be detected by checking the replies. It takes a longer or shorter time and a server supporting a faulty replica consumes more or smaller electric energy since the faulty replica does computation different from a proper replica. In this paper, we propose an algorithm to detect implicitly faulty replicas of a process by using the power consumption and computation models of a server in addition to checking replies in a cluster. Keywords: Process faults · Implicitly faulty replica · Power consumption model · Computation model · Fault detection

1

Introduction

Information systems are composed of servers where application processes issued by clients are performed. Information systems have to be tolerant of faults. Systems can be more reliable and available by taking checkpoints [4,8] and replicating system components [4,5,7]. In this paper, we discuss the active replication of application processes [4,5,7]. Servers are recently getting more reliable and available according to the advances of hardware and architecture technologies like CPUs [1]. Hence, each server can be assumed to be always proper, i.e. does not suffer from any fault. On the other hand, application processes performed on servers easily suffer from faults due to hacking and virus attacks [6]. In order to reliably perform an c Springer Nature Switzerland AG 2020  L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 483–493, 2020. https://doi.org/10.1007/978-3-030-33506-9_43

484

H. Ishii et al.

application process, multiple replicas of the application process are performed on servers. A faulty replica may return an improper reply or no reply due to Byzantine fault [21]. A majority one of replies from replicas can be taken as a proper reply if fewer than the half of the replicas are assumed to be faulty [21]. A faulty replica may send a proper reply by wiretapping a reply from another replica. Here, a faulty replica may not properly behave, e.g. does computation different from a proper replica even if the replica sends a proper reply. Such a replica is referred to as implicitly f aulty [17]. Implicitly faulty replicas cannot be detected by checking replies from replicas in traditional fault-tolerant technologies [4,6,7]. In order to detect implicitly faulty replicas, further information is needed in addition to checking the replies from replicas. In this paper, we take advantage of information on energy consumption of each server and execution time of each replica on each server in order to detect implicitly faulty replicas. The macro-level power consumption models [10,11,19] are proposed to show how much electric power [W] a server consumes to perform application processes without considering the power consumption of each hardware component like CPU [9–14,19]. In the MLPC (Multi-level Power Consumption) model [10,19], the power consumed by a server depends on the numbers of active CPUs, cores, and threads. If no application process is performed, a server consumes the minimum power. According to the computation models of a server [10,13,19], the execution time of each application process on the server can be estimated under the assumption that the minimum execution time of each application process on a server is apriori known like on-line transaction processing applications [19]. In addition, the energy [J] to be consumed by a server to perform application processes can be estimated by taking advantage of the power consumption and computation models. If a replica of an application process is properly performed on a server, the replica returns a proper reply. In addition, the actual execution time of the replica is the same as the estimated time. Furthermore, the actual energy consumed by a server until a replica terminates since the replica starts is the same as the estimated energy consumption model. On the other hand, if a replica is faulty, the execution time of the replica may be shorter or longer and the server may consume more or less energy than expected even if every replica may send a proper reply. In this paper, we discuss how to detect implicitly faulty replicas on reliable servers by measuring the energy consumption of each server and execution time of each replica in addition to checking replies. In Sect. 2, we present a system model and implicitly faulty replicas. In Sect. 3, we discuss how to estimate the execution time of each replica and the power consumption of each host server. In Sect. 4, we discuss how to detect implicitly faulty replicas.

Algorithm for Detecting Implicitly Faulty Replicas

2 2.1

485

System Model Implicitly Faulty Replicas

A cluster is composed of servers s1 , ..., sn (n ≥ 1) and a load balancer L which are interconnected in networks [Fig. 1]. Application processes p1 , ..., pm (m ≥ 1) are issued to the servers s1 , ..., sn . In this paper, we assume each application process is deterministic [15]. Let P be a set {p1 , ..., pm } of application processes in a cluster. An application process pi is actively replicated on the servers [4,15]. Let pit be a replica of the application process pi on a server st . Here, the server st is a host server of each replica pit . Let Pt be a set of replicas pt1 , ..., ptm of application processes p1 , ..., pm , respectively, on a server st . Here, some pair of servers st and su are heterogeneous in terms of execution time and power consumption. A client issues a request qi to the load balancer L. On receipt of the request qi , the load balancer L creates pi1 , ..., pin replicas of an application process pi to handle the request on servers s1 , ..., sn , respectively. Each replica pit of an application process pi is performed on a server st (i = 1, ..., m). The replica pit on the server st sends a reply rit to the load balancer L on termination of the replica pit . The load balancer L collects the replies ri1 , ..., rin from the replicas pi1 , ..., pin , respectively. Then, the load balancer L obtains a reply ri from a collection of the replies ri1 , ..., rin , e.g. majority one of the replies. The load balancer L forwards a reply ri to the client. In this paper, we assume each server is always proper and the load balancer L is also proper. This means, the load balancer L can anytime get the electric power consumption [W] of each server and the execution time [sec] of each replica performed on the each server. A faulty replica pit may send an improper reply or no reply to the load balancer L [4]. Furthermore, a faulty replica pit may do improper computation different from the specification of the application process pi even if the replica pit sends a proper reply. For example, a replica pit does improper computation if the replica pit is infected with virus. The faulty replica pit wiretraps a reply riu of another replica piu (u = t) which might be proper and just forwards the wiretapped reply riu to the load balancer L. Thus, a replica which sends a proper reply but does different computation is referred to as implicitlyf aulty [17]. Implicitly faulty replicas cannot be found by checking replies in the traditional fault detection algorithms [4,16]. In order to detect implicitly faulty replicas, we need further information in addition to replies from replicas. In this paper, we discuss how to detect an implicitly faulty replica pit on each server st by monitoring the energy consumption of the host server st and the execution time of the replica pit . 2.2

Power Consumption Model

A server st is composed of npt (≥ 1) homogeneous CPUs cpt0 , . . . , cpt,npt −1 (t = 1, ..., n). Each CPU cptk [1] is composed of nct (≥ 1) homogeneous cores ctk0 , . . . , ctk,nct −1 . Each core ctkh supports the same number ctt of threads. Here,

486

H. Ishii et al.

Fig. 1. Replicas of an application process.

ctt is one or two. Totally ntt (= npt ·nct ·ctt ) threads are supported to application processes by a server st . An active thread is a thread where at least one replica is performed. Let CPt (τ ) be a set of replicas performed on a server st at time τ . Replicas are fairly allocated with the threads on a server st in the roundrobin (RR) algorithm [2,3]. Here, the electric power N Et (m) [W] consumed by a server st to concurrently perform m (≥ 0) replicas is given as follows [18–20]: [Power consumption for n processes] [Fig. 2] ⎧ ⎪ minEt if n = 0. ⎪ ⎪ ⎪ ⎪ ⎪ ⎨minEt + m · (bEt + cEt + tEt ) if 1 ≤ m ≤ npt . N Et (m) = minEt + npt · bEt + m · (cEt + tEt ) if npt < m ≤ nct · npt . ⎪ ⎪ ⎪ minEt + npt · (bEt + nct · cEt ) + ntt · tEt if nct · npt < m < ntt . ⎪ ⎪ ⎪ ⎩maxE if m ≥ nt . t t (1) The electric power consumption Et (τ ) [W] of a server st at time τ is assumed to be N Et (m) in this paper, where m is the number |CPt (τ )| of active replicas. 2.3

Computation Model

Each replica pit is at a time performed on a thread of a server st . It takes Tit time units [tu] to perform a replica pit on a thread of a server st . If only a replica pit is performed on a server st without any other replica, the execution time Tit of the replica pit is minimum, i.e. Tit = minTit . Let minTi show a minimum one of minTi1 , . . . , minTin , i.e. minTi = minTf i on the fastest thread which is on a server sf . A server sf with the fastest thread is f astest in a cluster S. Here, one virtual computation step [vs] is assumed to be performed on a thread of a fastest server sf for one time unit [tu].

Algorithm for Detecting Implicitly Faulty Replicas

487

Fig. 2. MLPCM model.

This means, the maximum thread computation rate maxT CRt of a server st is defined to be (minTi /minTit ) · maxT CRf = minTi /minTit [vs/tu]. The maximum computation rate maxSCRt (≤ ntt ) of a server st is ntt · maxCRTt where ntt is the total number of threads. The total number V Ci = minTi [tu] ·maxT CRf [vs/tu] = minTi [vs] of virtual computation steps are performed in a replica pit . The maximum computation rate maxP CRit of a replica pit on a server st is V Ci /minTit = minTi /minTit (≤ 1). For every pair of replicas pit and pjt on a server st , maxP CRit = maxP CRjt = maxT CRt . The computation rate N P Rit (m) (≤ maxSCRt ) [vs/tu] of a replica pit on a server st where m replicas are concurrently performed at time τ is defined [8,10,19] as follows: [MLCM (Multi-Level Computation with Multiple CPUs) model]  ntt · maxT CRt / m if m > ntt . (2) N P Rit (m) = maxT CRt if m ≤ ntt . In this paper, the server computation rate N P Rit (m) of a server st to perform m replicas is given as nt ·maxT CRt (= maxSCRt ) for m > ntt and m·maxT CRt for m ≤ ntt . etSuppose a replica pit on a server st starts at time st and ends at time et. Here, τ =st N P R(|CPt (τ )|) = V Ci [vs] = minTi . Thus, minTit shows the amount of computation of a replica pit . The execution of a replica pit is modeled as follows: [Computation model of a replica pit ] 1. At time τ a replica pit starts, the computation residue Rit of a replica pit is V Ci , i.e. Rit = V Ci (= minTi );

488

H. Ishii et al.

2. At each time τ , the computation residue Rit (τ ) is decremented by the computation rate N P Rit (|CPt (τ )|), i.e. Rit = Rit − N P Rit (|CPt (τ )|); 3. If Rit (≤ 0), the replica pit terminates at time τ .

3

Estimation Model

An application process pi issued by a client is replicated on multiple servers s1 , ..., sn in a cluster. P is a set of processes p1 , ..., pm (m ≥ 1) which clients issue to the cluster. One replica pit of each application process pi is performed on each server st (t = 1, ..., n). Pt is a set of replicas pt1 , ..., ptm of the application processes p1 , ..., pm in the process set P , respectively, on a server st . At time τ , a variable Rit shows the computation residue of a replica pit , i.e. how many number of virtual computation steps of a replica pit are still to be performed. At time stimeit a replica pit starts on a server st , Rit = minTi . At each time τ , the computation residue Rit is decremented by the computation rate N P Rit (|CPt (τ )|) where CPt (τ ) is a set of replicas performed on a server st as presented in the preceding section. A replica pit terminates at time etimeit if the computation residue Rit gets zero. In the load balancer L, an estimation (EST ) algorithm is performed to simulate the execution of each replica pit on each server st while issuing a replica pit to a server st . Each replica pit starts on a server st at time stimeit . By the EST algorithm, the termination time etimeit of each replica pit is estimated. ETit stands for the execution time [sec] of each replica pit on a server st . ETit = etimeit − stimeit + 1. Et shows the energy [J] to be consumed by replicas in the replica set Pt from time stimeit to time a server st to perform  etimeit N Et (|CPt (τ )|). The execution time ETit and etimeit . That is, Eit = τ =stime it energy consumption Eit are estimated for a server st by the algorithm EST (st ) which is shown in Algorithm 1. For each server st , a variable Pt denotes a set of replicas performed on st . If the load balancer L issues the replicas pi1 , ..., pin to the servers, every replica pit starts on a server st at time stimeit . For each replica pit , the starting time stimeit is given to the estimation algorithm EST (st ). In the estimation algorithm EST (st ), the replica pit starts at time τ (= stimeit ) and is added to the set Pt , i.e. Pt = Pt ∪ {pit }. m = |Pt |, i.e. number | CPt (τ ) | of active replicas on the server st at current time τ . Eit = 0 and Rit = minTi . At each time τ , for every replica pit in the set Pt , the variable Eit is incremented by the power consumption N Et (m) [W]. For each replica pit , the computation residue Rit is decremented by the computation rate N P Rit (m). Then, if Rit ≤ 0, the replica pit terminates at time τ and is removed from the replica set Pt , i.e. Pt = Pt − {pit }. Here, the termination time etimeit is τ . The execute time ETit of the replica pit is etimeit − stimeit + 1. Eit shows the energy consumed by the server st during the execution of a replica pit from time stimeit to time etimeit . Time τ is incremented by one, i.e. one time unit [tu] and the procedure is interacted. Thus, in the algorithm EST (st ), the estimated execution time ETit and estimated energy consumption Eit are obtained for each replica pit on a server st .

Algorithm for Detecting Implicitly Faulty Replicas

489

Algorithm 1. Simulation algorithm EST (st ) 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

input : st = server; Pt = set of replicas performed on st ; output : Et = total energy consumption of st ; stimeit = stating time of each replica pit on st ; etimeit = terminate time of each replica pit ; Eit = energy to be consumed by st from stimeit to etimeit for each replica pit ; Pt = φ; τ = 0; while do for each replica pit which starts at time τ , i.e. τ = stimeit do Rit = minTi ; Eit = 0; Pt = Pt ∪ {pit }; for end; m = |Pt |; /* number of active replicas performed at time τ */ if m > 0 then Et = Et + N Et (m); /* energy */ for each replica pit in Pt do Rit = Rit − N P Rit (m); Eit = Eit + N Et (m); if Rit ≤ 0 then /* Pit terminates */ Pt = Pt − {pit }; etimeit = τ ; ETit = etimeit − stimeit + 1; for end; else /* no replica on st , i.e. st is idle. */ Et = Et + minEt ; τ = τ + 1; while end;

On the other hand, each replica pit is started on a server st at time stimeit . Then, the replica pit is actually performed on the server st and terminates at time aetimeit . In addition to the actual termination time aetimeit for each replica pit , each server st monitors the power consumption [W] for each time unit. The energy AEit [J] actually consumed by a server st from time stimeit to time detimeit is sent to the load balancer L [Fig. 3]. Thus, the load balancer L obtains a pair ETit , Eit  of estimated execute time ETit and estimated energy consumption Eit by the EST algorithm. The load balancer L also obtains a pair AETit , AEit  of actual execution time AETit and actual energy computation AEit for each replica pit from a server st . By comparing the actual data AETit ,

490

H. Ishii et al.

Fig. 3. Estimation algorithm.

AEit  with the estimated data ETit , Eit , the load balancer L checks if a replica pit is implicitly faulty.

4

Detection of Implicitly Faulty Replicas

By using the estimation EST (st ) algorithm, the estimated execution time ETit (= etimeit − stimeit + 1) [sec] of each replica pit and the estimated energy consumption Eit [J] of a server st from time stimeit to the time etimeit can be obtained. In addition, the load balancer L obtains the actual execute time AEit (= aetimeit − stimeit + 1) of each replica pit and the actual energy AEit consumed by host server st from time stimeit to time aetimeit . There are totally m replicas pi1 , ..., pim of an application process pi in a cluster. The load balancer L also collects a reply rit of a replica pit from every server st . Let Ri be a set {rit , ..., rim } of replies of replicas pit , ..., pim . The load balancer L decides on a proper reply by collecting a majority reply ri , for Ri , i.e. |{rit ∈ Rit |rit = ri }| > n/2. If there is no majority reply, the load balancer L returns a reply ⊥ to the cluster. The reply ⊥ means no proper reply can be obtained. Then, suppose a proper reply ri is obtained since more than half of the replicas send the proper reply ri to the load balancer L. Here, some replica pit which sends a proper reply ri may be faulty. For each host server st of a replica pit which sends a proper reply ri , it is checked each replica pit is properly performed on the server st in terms of the execution time of the replica pit and the energy consumption of the server st . The load balancer L obtains the actual execution time AEit (= aetimeit − stimeit + 1) for each replica pit and the energy consumption of a server st for each replica pit . The load balancer L also obtains the

Algorithm for Detecting Implicitly Faulty Replicas

491

estimated execution time ETit = etimeit − stimeit + 1 and the estimated energy consumption ETit in the EST algorithm as shown in [Fig. 3]. The load balancer L make a following decision on each server st : [Detection of an Implicitly Faulty Replica] 1. If AETit = ETit and AEit = Eit , a replica pit is decided to be proper. 2. Otherwise, the replica pit is implicitly faulty. For example, even if AETit = ETit , if AEit < Eit , a fewer number of computation steps are performed on a replica pit than a proper replica. Here, some computation steps might be omitted. The load balancer L behaves as follows: [Behavior of a load balancer L] 1. On receipt of an application process pi from a client, the load balancer L creates a replica pit of the application process pi on each server st . 2. The EST (st ) algorithm is performed while replicas are actually performed on each server st . 3. The load balancer L receives a table AETit , AEit  a reply rit , actual execute time AETit , and actual energy consumption AEit of a replica pit of the application process pi from each server st . 4. Let Ri be a set {rit , ..., rin } of replies from the replicas pi1 , ..., pin . Here, if the load balancer L does not receive a reply from a replica pit , rit is ⊥. 5. If the load balancer L gets a majority reply r from the reply set Ri , i.e. |{rit ∈ Ri | rit = r}| > n/2, the load balancer L compares the real execution time AETit with the estimation execute time ETit and the actual energy consumption AEit with the estimated energy consumption Eit of each replica pit which sends the reply rit = r. A replica piu which does not send a proper reply r is faulty. If a replica pit sends a proper reply rit (= r) but AETit = ETit or AEit = Eit , replicas pit is implicitly faulty. A replica pit which sends a proper reply and AETit = ETit and AEit = Eit is proper. 6. Let P Ri be a subset of the proper replicas. If |P Ri | > n/2, a reply r from replicas in the set P Ri is proper and the load balancer L sends the reply r to the client. Otherwise, the load balancer L sends a reply ⊥ to the client. Thus, the load balancer L detects two types of faulty replicas. One is a faulty replica which just sends an improper reply. This type of faulty replica can be detected by the traditional detection algorithms which take a majority reply from the replicas. The second type is an implicitly faulty replica which sends a proper reply but does improper computation. In the estimation EST (st ) algorithm, we can detect implicit faulty replicas by taking usage of execution time of a replica and energy consumption of a server st .

492

5

H. Ishii et al.

Concluding Remarks

In order to reliably perform application processes, multiple replicas of each application process are performed on servers. Servers can be now considered to be reliable according to the advances of hardware and architecture technologies. On the other hand, replicas of an application process still suffer from faults due to security attacks like virus and hacking. A faulty replica may send an improper reply or no reply. In addition, even if a replica sends a proper reply, the replica may do different computation from proper replicas. The replica is implicitly faulty, i.e. sends a proper reply but does improper computation. Implicitly faulty replicas cannot be detected by checking replies from the replicas in the traditional fault detection algorithm. In this paper, we discussed how to detect implicitly faulty replicas in a cluster by taking advantage of the power consumption and computation models. In this paper, we proposed the EST algorithm to detect implicitly faulty replicas by comparing the actual execution time of each replica and energy consumption of a host server with ones estimated by the power consumption and computation models. We are now evaluating the EST algorithm.

References 1. Intel xeon processor 5600 series: The next generation of intelligent server processors, white paper (2010). http://www.intel.com/content/www/us/en/processors/ xeon/xeon-5600-brief.html 2. Job scheduling algorithms in Linux virtual server (2010). http://www. linuxvirtualserver.org/docs/scheduling.html 3. Linux Operating Systems. https://ja.wikipedia.org/wiki/Linux 4. Bernstein, P.A., Goodman, N.: The failure and recovery problem for replicated databases. In: Proceedings of the 2nd ACM Symposium on Principles of Distributed Computing, pp. 114–122 (1998) 5. Defago, X., Schiper, A., Sergent, N.: Semi-passive replication. In: Proceedings of IEEE the 17th Symposium on Reliable Distributed Systems, pp. 43–50 (1998) 6. Denning, D.E.R.: Cryptography and Data Security. Addison Wesley, Boston (1982) 7. Deplanche, A.M., Theaudiere, P.Y., Trinquet, Y.: Implementing a semi-active replication strategy in CHORUS/ClassiX, a distributed real-time executive. In: Proceedings of IEEE the 18th Symposium on Reliable Distributed Systems, pp. 90–101 (1999) 8. Duolikun, D., Aikebaier, A., Enokido, T., Takizawa, M.: Energy-aware passive replication of processes. Int. J. Mob. Multimed. 9(1,2), 53–65 (2013) 9. Duolikun, D., Enokido, T., Takizawa, M.: Dynamic migration of virtual machines to reduce energy consumption in a cluster international journal of grid and utility computing. Int. J. Grid Utility Comput. 9(4), 357–366 (2018) 10. Duolikun, D., Kataoka, H., Enokido, T., Takizawa, M.: Simple algorithms for selecting an energy-efficient server in a cluster of servers. Int. J. Commun. Netw. Distrib. Syst. 21(1), 1–25 (2018) 11. Enokido, T., Aikebaier, A., Takizawa, M.: A model for reducing power consumption in peer-to-peer systems. IEEE Syst. J. 4(2), 221–229 (2010)

Algorithm for Detecting Implicitly Faulty Replicas

493

12. Enokido, T., Aikebaier, A., Takizawa, M.: Process allocation algorithms for saving power consumption in peer-to-peer systems. IEEE Trans. Ind. Electron. 58(6), 2097–2105 (2011) 13. Enokido, T., Ailixier, A., Takizawa, M.: An extended simple power consumption model for selecting a server to perform computation type processes in digital ecosystems. IEEE Trans. Ind. Inform. 10(2), 1627–1636 (2014) 14. Enokido, T., Takizawa, M.: An integrated power consumption model for distributed systems. IEEE Trans. Ind. Electron. 60(2), 824–836 (2013) 15. Fischer, M.J., Lynch, N.A., Paterson, M.S.: Impossibility of distributed consensus with one faulty process. In: Proceedings of the Second ACM SIGACT-SIGMOD Symposium on Principles of Database Systems, 21–23 March 1983, Colony Square Hotel, Atlanta, Georgia, USA, pp. 1–7 (1983) 16. Hayashibara, N., Takizawa, M.: Design of the notification system for failure detectors. Int. J. High Perform. Comput. Netw. 6(1), 25–34 (2009) 17. Ishii, H., Oma, R., Nakamura, S., Enokido, T., Takizawa, M.: Fault detection of process replicas on reliable servers. In: Proceedings of the 22nd International Conference on Network-Based Information System, NBiS-2019 (2019) 18. Kataoka, H., Duolikun, D., Enokido, T., Takizawa, M.: Energy-efficient virtualisation of threads in a server cluster. In: Proceedings of the 10th International Conference on Broadband and Wireless Computing, Communication and Applications, BWCCA-2015, pp. 288–295 (2015) 19. Kataoka, H., Nakamura, S., Duolikun, D., Enokido, T., Takizawa, M.: Multi-level power consumption model and energy-aware server selection algorithm. Int. J. Grid Utility Comput. (IJGUC) 8(3), 201–210 (2017) 20. Kataoka, H., Sawada, A., Duolikun, D., Enokido, T., Takizawa, M.: Energy-aware server selection algorithm in a scalable cluster. In: Proceedings of IEEE the 30th International Conference on Advanced Information Networking and Applications, AINA-2016, pp. 565–572 (2016) 21. Lamport, L., Shostak, R., Pease, M.: The Byzantine generals problems. ACM Trans. Program. Lang. Syst. 4(3), 382–401 (1992)

Parallel Data Transmission Protocols in the Mobile Fog Computing Model Kosuke Gima1(B) , Ryuji Oma1 , Shigenari Nakamura1 , Tomoya Enokido2 , and Makoto Takizawa1 1

Hosei University, Tokyo, Japan {kosuke.gima.3r,ryuji.oma.6r}@stu.hosei.ac.jp, [email protected], [email protected] 2 Rissho University, Tokyo, Japan [email protected]

Abstract. The fog computing (FC) model is proposed to efficiently realize the IoT (Internet of Things). In this paper, we consider the mobile FC (MFC) model including mobile fog nodes which communicate with other fog nodes in wireless networks. Here, each fog node is equipped with some process by which output data obtained by processing the input data from other nodes and devices is sent to other fog nodes in the opportunistic way. In the opportunistic protocols, each fog node exchanges only data with other fog nodes in the communication range. In this paper, we newly discuss the MFC model by which fog nodes exchange not only data but also processes with one another in the communication range. A node sends a process to another node which holds data to be handled by the process so that the electric energy to be consumed by the nodes can be reduced. Here, even if there are multiple target nodes in the communication range, a source node sends data to one of the target nodes. In this paper, we newly propose a parallel data transmission (PDT) algorithm where a fog node in parallel sends segments of data to multiple target nodes. Keywords: IoT · Fog computing model · Mobile Fog Computing (MFC) model · Parallel data transmission (PDT) model · PTD algorithm

1

Introduction

The IoT (Internet of Things) includes millions of sensors and actuators which are interconnected with clouds of servers in networks [9]. Here, networks are congested to forward sensor data to servers and servers are heavily loaded to process the sensor data. Fog computing (FC) models [10,15] are proposed to efficiently realize the IoT by distributing the processing load of sensor data to fog nodes and reducing the network traffic by transmitting processed data. The FC model is composed of fog nodes in addition to sensor and actuator devices c Springer Nature Switzerland AG 2020  L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 494–503, 2020. https://doi.org/10.1007/978-3-030-33506-9_44

Parallel Data Transmission Protocols in the Mobile Fog Computing Model

495

and servers. Each fog node supports a process to handle sensor data in the FC model while every process of an application is performed on a server in the cloud computing model. Sensor data is processed by a process of a fog node pi and is sent to another fog node pj to further process the data. The node pj is a target node of the source node pi . In the most FC models, each fog node is fixed at a location, i.e. does not move. The TBFC (Tree-Based FC) model of the IoT is proposed in our previous studies [11,14], where fog nodes are hierarchically structured in a tree to reduce the energy consumption of the fog nodes. In addition to improving the performance and reducing the energy consumption of the IoT, the FC model is required to be tolerant of faults. In order to make the TBFC model tolerant of faults, the FTBFC (Fault-tolerant TBFC) model is proposed [13]. Mobile nodes like vehicles have to be considered in addition to fixed nodes like servers. In the MFC (Mobile FC) model [6], mobile fog nodes like vehicles are moving and communicate with one another in wireless communication. Here, not only data obtained by processing data from other nodes but also processes to process data are exchanged among fog nodes. Since a node is moving, there may not be another node in the wireless communication range of the node ever if the node would like to deliver output data to a target fog node. Thus, mobile nodes have to take advantage of opportunistic communication protocols [2,16]. A node waits for opportunity that another node comes in the communication range. Once another node comes in the communication range of a source node, the source node can forward the data to the node to deliver to servers. In the MFC protocol [6], each fog node communicates with one fog node exchanging by data and process in the communication range. In this paper, we newly propose a parallel data transmission (PDT) model, where each fog node communicates with one or more than one target node in the communication range. Here, a source fog node divides data to multiple segments and sends each segment to one of multiple target nodes in the communication range. Thus, data segments are in parallel processed by multiple target fog nodes in order to reduce the total energy consumption and execution time of the target nodes. In this paper, each segment of data is the same size. We propose a PDT algorithm to select target fog nodes in the communication range to which a source node sends segments of data so that the total energy consumption of the source and target fog nodes can be reduced. In Sect. 2, we present a system model. In Sect. 3, we present the power consumption and computation models of a fog node. In Sect. 4, we propose the PDT model. In Sect. 5, we propose the PDT algorithm to select multiple target nodes to which segments of the output data are transmitted.

2

System Model

A fog computing (FC) model is composed of clouds of servers, fog nodes, and devices with sensors and actuators [15]. In this paper, we consider the MFC (Mobile FC) model [6] which is composed of mobile fog nodes interconnected in

496

K. Gima et al.

wireless networks. Mobile fog nodes can communicate with one another only if the fog nodes are in the communication range of each other. Let F Fi be a set of fog nodes which are in the communication range of a fog node fi . F Ni (⊆ F Fi ) is a set of target fog nodes of a fog node fi , i.e. F Ni = {fj |fj ∈ F Fi and fi → fj }. In the MFC model, each source node fi selects one target node fj in the set F Ni . In this paper, we consider a way where one or more than one target node is selected in the set F Ni if F Ni includes multiple target nodes. A fog node fi supports a process Fi . A fog node fi receives data from sensors and issues actions to actuators. On receipt of input data from sensors or fog nodes, a fog node fi processes the input data and forwards the output data obtained by processing the input data to a target fog node which supports a process Fj which can process the output data. Servers in clouds finally receive data processed by fog nodes. Then, the servers receive input data processed by fog nodes and make a decision on actions to be done by actuators by processing the input data. The servers deliver the actions to target actuators via fog nodes. The output data of a fog node is smaller than the input data. Processes are performed on fog nodes. Here, network and server traffic can be reduced in the MFC model. In the TBFC model [7,8,12,14], a fog node fi is equipped with a process Fi and memory storage Mi . Here, the node fi is a host node of the process Fi . Let p(fi ) show a process Fi supported by a node fi . A message m is first created by a sensor to carry sensor data. The message m is received and processed by a fog node fi with a process Fi . The output data is obtained by the process Fi through processing the input data. The output data has to be sent to a fog node which supports a process Fj to process the output data. A process Fi is referred to as precede a process Fj (or Fi f ollows Fj )(Fi → Fj ) if the output message of the process Fi is processed by the process Fj . Fj is a target subprocess of Fi (Fig. 1). A message with processed data is sent to a fog node fj with a process Fj such that Fi → Fj . Thus, messages are processed by a sequence of processes. A node fi precedes a node fj (fi → fj ) if Fi → Fj . Here, the node fj is a target node of the node fi . On receipt of a message m, a fog node fi stores the message m in the memory Mi . Let maxMi be the maximum size of the memory storage Mi of a fog node fi , i.e. maximum number of messages which the node fi can receive. Messages with input data received and messages with output data are stoned in the memory storage Mi . A fog node fi receives a message mk from a fog node fk such that fk → fi . Let RMi be a collection of input messages stored in the memory Mi . Then, the node fi processes data in the input messages RMi and obtains the output data by a process Fi . The fog node fi creates a message mi with the processed output data in the memory Mi . Then, the fog node fi has to deliver the message mi to a target fog node fj with a process Fj following the process Fi of the fog node fi (Fi → Fj ). Mij shows a set of output messages in a fog node fi , which are required to be sent to a target fog node with a process Fj such that Fi → Fj . A fog node fi has to deliver the messages Mij to a target node.

Parallel Data Transmission Protocols in the Mobile Fog Computing Model

497

Fig. 1. Target node.

3

Power Consumption and Computation Models of a Fog Node

A fog node fi receives input data, processes the input data, and sends the output data obtained by processing the input data. Let T Ti (x) and RTi (x) be a pair of time [sec] for a node fi to transmit and receive data of size x, respectively. The transmission time T Ti (x) and receiving time RTi (x) are proportional to the size x of the data. T Ti (x) = tci + tti · x.

(1)

RTi (x) = rci + rti · x.

(2)

Here tci , tti , rci , and rti are constants. Suppose there are multiple target nodes fi1 , · · · , fi,si in the communication range of a node fi . The output data Di of a node fi is divided into segments di1 , · · · , di,si (si ≥ 1). Then, the node fi sends each segment dij to a target node fij (j = 1, · · · , si ). Let xj be the size |dij | of each segment dij . Here, x1 + · · · + xsi = x (= |Di |). It takes time T T Ti (x) [sec] for the node fi to send the segments di1 , · · · , di,si to target nodes fi1 , · · · , fi,si , respectively: T T Ti (x) = T Ti (x1 ) + · · · + T Ti (xsi ). = tci · si + tti · x.

(3)

It takes time CTi (x) [sec] for a fog node fi to process input data of size x. The execution time CTi (x) [sec] depends on the computation complexity of the process Fi of each node fi . In this paper, we consider two types of processes in terms of computation complexity, i.e. O(x) and O(x2 ). The execution time CTi (x) is given as follows: CTi (x) = cti · Ci (x).

(4)

498

K. Gima et al.

Here, Ci (x) = x or x2 and cti is a constant. In a Raspberry Pi 3 Model B node [1], tti = 0.00001, rti = 0.00009, tci = 0.00007, rci = 0, and cti = 0.00009. It takes a longer time to receive data than send the data. A fog node fi is assumed to follow the SPC (Simple Power Consumption) model [3–5]. If a process Fi is performed on a node fi , the node fi consumes the maximum electric power maxEi [W]. Otherwise, the node fi consumes the minimum power minEi [W]. In a Raspberry Pi 3 Model B node fi , maxEi = 3.7 [W] and minEi = 2.1 [W]. rei /sei = 5 according to our experiment [1]. In a PC fi with an Intel Core i7-6700K CPU, maxEi is 89.5 [W]. A fog node fi consumes the electric power T Pi and RPi [W] to transmit and receive a message. T Pi and RPi are given as follows (Fig. 2): T Pi = tei · maxEi .

(5)

RPi = rei · maxEi .

(6)

Fig. 2. Power consumption.

Here, tei and rei are constants. tei ≤ 1 and rei ≤ 1. In a Raspberry Pi 3 Model B node fi , tei = 0.676 and rei = 0.729. A fog node fi consumes the maximum power maxEi to perform a process Fi as shown in Fig. 2. The fog node fi consumes the power T Pi (x) = tei · maxEi [W] for RTi (x) time units [tu]. Hence, the energy REi (x) = RPi [W] ·RTi (x) [tu] is consumed by the node fi . Thus, the node fi consumes the energy T Ei (x) = T Pi [W] ·T Ti (x) [tu]. A fog node fi consumes electric energy T Ei (x) and REi (x) [J] to transmit and receive data of size x: REi (x) = RPi [W ] · RTi (x)[sec] = rei · (rci + rti · x) · maxEi .

(7)

T Ei (x) = T Pi [W ] · T Ti (x)[sec] = tei · (tci + tti · x) · maxEi .

(8)

If a node fi sends the segments di1 , · · · , di,si to target nodes fi1 , · · · , fi,si , respectively, the node fi consumes the electric energy T T Ei (x) [J] where x = x1 + · · · xsi : T T Ei (x) = T Pi · T T Ti (x) = tei · (tci · si + tti · x) · maxEi .

(9)

Parallel Data Transmission Protocols in the Mobile Fog Computing Model

499

A fog node fi consumes power maxEi [W] to process data. Hence, the energy CEi (x) = maxEi [W] ·CTi (x) [tu] is consumed by the fog node fi to process data of size x: CEi (x) = CTi (x) · maxEi = cti · Ci (x) · maxEi .

(10)

If a fog node fi receives data Di of size x and sends segments di1 , · · · , di,si to si (≥ 1) target nodes, the fog node fi totally consumes the energy EEi (x) to receive and process the data and send the processed data to target fog nodes: EEi (x) = REi (x) + CEi (x) + T T Ei (ρ · x). = REi (x) + CEi (x) + T E1 (y1 ) + · · · + T Esi (ysi ).

(11)

Here, ρ · x = y1 + · · · + ysi . If yj = yk for every pair of segments dij and dik , yj = ρ · x/si . The node fi consumes the energy EEi (x): (12) EEi (x) = REi (x) + CEi (x) + si · T Ei (ρ · x/si ). = [rei · (rci + rti · x) + cti · Ci (x) + (si · tci + tti · ρ · x)] · maxEi .

4

Parallel Data Transmission (PDT) Model

Each fog node fi moves and communicates with other fog nodes in a wireless network. A fog node fi sends output messages Mij in the memory Mi to a process Fj of another node following the process Fi of the node fi (Fi → Fj ). We discuss how fog nodes communicate with one another to process data and exchange data and processes. In our previous studies [6], the data transmission (DT), process transmission (PT), and data exchange (DE) algorithms are proposed for a fog node to communicate with another node in the communication range. In the DT algorithm, a fog node fi sends the output data Mij to a target node fj . In the PT algorithm, a target node fj sends a process Fj to a node fi . Then, the node fi processes the output data Mij by using the process Fj . In the DE algorithm, a node fi communicates with a node fj in the communication range but the node fj is not a target node. Here, the nodes fi and fj exchange the output messages so that the usage ratios of the memories get the same. In this paper, we newly propose a parallel data transmission (PDT) model to more efficiently process data. Suppose there are multiple target nodes fi1 , · · · , fiti of a node fi are in the communication range of fi . In the DT way, one target node fij is selected and the node fi sends the output data Mij to the target node fij . Let x be the size |Mij | of the output data Mij . Suppose the execution time CTj (x) of the process Fj is O(x2 ), i.e. ctj · x2 for the size x of the data Mij . If the node fi sends the half of the data Mij to a target node fij and the other half to another target node fik , the total execution time of the nodes fij and fik is (ctij + ctik ) · (x/2)2 . If ctij = ctik = c, the total execution time is c · (x/2)2 + c · (x/2)2 = c · x2 /2. If the node fi sends the data Mij to only the node fij , the execution time of fij is c · x2 . Thus, c · x2 /2 is smaller than c · x2 . Thus, the total execution time and total energy consumption of fog nodes can be reduced by dividing output data to segments and in parallel sending the segments to different target nodes as shown in Fig. 3.

500

K. Gima et al.

Fig. 3. Parallel data transmission (PDT).

5

PDT Selection Algorithm

We discuss how each fog node fi behaves to deliver and process data Mij . In this paper, we assume every fog node is homogeneous. Let F Ni be a set of fog nodes which are in the communication range of a fog node fi . Let F Fi be a subset of fog nodes in the set F Ni , which follow a node fi , i.e. F Fi = {fj ∈ F Ni |fi → fj }(⊆ F Ni ). In the set F Fi , fog nodes are sorted with respected to available size of memory. Let AMi be the size of available memory in a node fi . F Fi is an ordered set of target nodes fi1 , · · · , fi,ki (ki ≥ 0). Here, AMij > AMik for nodes fij and fik (j < k) in F Fi . Let x be the size |Mij | of data in a fog node fij and xij be the size of data in a fog node fij . First, a first node fi1 is selected in the ordered set F Fi . If the data Mij of size x is sent to the selected fog node fi1 , the node fi consumes the energy T Ei (x) to send data of size x to the node fi1 . The fog node fij processes not only the data of size x sent by the fog node fi but also its own data which is size xij . Hence, the target node fij consumes the energy REij (x) + CEij (xij + x) to receive and process data of size x. Totally, EE1 = T Ei (x) + REi1 (x) + CEi1 (xi1 + x). Next, two fog nodes fi1 and fi2 are selected in the set F Fi . Here, the fog node fi sends data of x/2 to each of the target fog nodes fi1 and fi2 . Here, the fog node fi and the selected target nodes fi1 and fi2 totally consume the energy EE2 = 2 · T Ei (x/2) + (REi1 (x/2) + CEi1 (xi1 + x/2)) + (REi2 (x/2) + CEi2 (xi2 + x/2)). Thus, the k nodes fi1 , · · · , fik (1 ≤ k ≤ ki ) are taken in the ordered set F Fi (Fig. 4). The total energy consumption EEk of the node fi and target nodes fi1 , · · · , fik is given as follows:

Parallel Data Transmission Protocols in the Mobile Fog Computing Model

EEk = k · T Ei (x/k) +

k 

[REij (x/k) + CEij (xij + x/k)].

501

(13)

j=1

Here, the size of each segment dij is the same x/k(j = 1, · · · , k). We select the number k(≤ ki ) so that the total energy consumption EEk is minimum in the PDT algorithm (Algorithm 1). The node fi divides the data into k segments di1 , · · · , dik , where the size of each segment dij is x/k. The node fi sends a segment dij to each node fij (j = 1, · · · , k). Algorithm 1. PDT selection algorithms 1 2 3 4 5 6 7 8 9 10 11 12

input : F Ni = fi1 , · · · , fiki  = ordered set of target nodes in the communication range; x = size of data; output : k = number of fog nodes; Ei = ∞; for l = 1, · · · , ki do if |AMij | > x/l for every node fij (j ≤ l), then  E = l · T Ei (x/l) + lj=1 [REij (x/l) + CEij (xij + x/l)]; if E < EE, then EE = E; k = li ; if end; if end; for end;

Fig. 4. PDT algorithm.

6

Concluding Remarks

The MFC (Mobile Fog Computing) model is composed of mobile fog nodes which communicate with one another in wireless networks. A fog node receives

502

K. Gima et al.

data from other nodes and processes the data by a process supported by the node. In this paper, we newly proposed the parallel data transmission (PDT) algorithm in the MFC model to efficiently realize the IoT. Here, a fog node divides data to segments and sends in parallel segments to target nodes. In addition, we are considering an algorithm where the size of each segment dij is different depending on the size of available memory of a target node fij . We are evaluating the algorithm.

References 1. Raspberry PI 3 model B. https://www.raspberrypi.org/products/raspberry-pi-3model-b/ 2. Dhurandher, S.K., Sharma, D.K., Woungang, I., Saini, A.: An energy-efficient history-based routing scheme for opportunistic networks. Int. J. Commun. Syst. 30(7), e2989 (2015) 3. Enokido, T., Ailixier, A., Takizawa, M.: A model for reducing power consumption in peer-to-peer systems. IEEE Syst. J. 4(2), 221–229 (2010) 4. Enokido, T., Ailixier, A., Takizawa, M.: Process allocation algorithms for saving power consumption in peer-to-peer systems. IEEE Trans. Ind. Electron. 58(6), 2097–2105 (2011) 5. Enokido, T., Ailixier, A., Takizawa, M.: An extended simple power consumption model for selecting a server to perform computation type processes in digital ecosystems. IEEE Trans. Ind. Inform. 10(2), 1627–1636 (2014) 6. Gima, K., Oma, R., Nakamura, S., Enokido, T., Takizawa, M.: A model for mobile fog computing in the IoT. In: Proceedings of the 22nd International Conference on Network-Based Information Systems, NBiS-2019 (2019, accepted) 7. Guo, Y., Oma, R., Nakamura, S., Duolikun, D., Enokido, T., Takizawa, M.: Data and subprocess transmission on the edge node of TWTBFC model. In: Proceedings of the 11th International Conference on Intelligent Networking and Collaborative Systems, INCoS-2019, pp. 80–90 (2019) 8. Guo, Y., Oma, R., Nakamura, S., Duolikun, D., Enokido, T., Takizawa, M.: Evaluation of a two-way tree-based fog computing (TWTBFC) model. In: Proceedings of the 13th International Conference on Innovative Mobile and Internet Services in Ubiquitous Computing, IMIS-2019, pp. 72–81 (2019) 9. Hanes, D., Salgueiro, G., Grossetete, P., Barton, R., Henry, J.: IoT Fundamentals: Networking Technologies, Protocols, and Use Cases for the Internet of Things. Cisco Press, Indianapolis (2018) 10. Oma, R., Nakamura, S., Duolikun, D., Enokido, T., Takizawa, M.: An energyefficient model for fog computing in the internet of things (IoT). Internet Things 1–2, 14–26 (2018) 11. Oma, R., Nakamura, S., Duolikun, D., Enokido, T., Takizawa, M.: Evaluation of an energy-efficient tree-based model of fog computing. In: Proceedings of the 21st International Conference on Network-Based Information Systems, NBiS-2018, pp. 99–109 (2018) 12. Oma, R., Nakamura, S., Duolikun, D., Enokido, T., Takizawa, M.: Energy-efficient recovery algorithm in the fault-tolerant tree-based fog computing (FTBFC) model. In: Proceedings of the 33rd International Conference on Advanced Information Networking and Applications, AINA-2019, pp. 132–143 (2019)

Parallel Data Transmission Protocols in the Mobile Fog Computing Model

503

13. Oma, R., Nakamura, S., Duolikun, D., Enokido, T., Takizawa, M.: A fault-tolerant tree-based fog computing model. Int. J. Web Grid Serv. (IJWGS) (2019, accepted) 14. Oma, R., Nakamura, S., Enokido, T., Takizawa, M.: A tree-based model of energyefficient fog computing systems in IoT. In: Proceedings of the 12th International Conference on Complex, Intelligent, and Software Intensive Systems, CISIS-2018, pp. 991–1001 (2018) 15. Rahmani, A.M., Liljeberg, P., Preden, J.S., Jantsch, A.: Fog Computing in the Internet of Things. Springer, Heidelberg (2018) 16. Spaho, E., Barolli, L., Kolici, V., Lala, A.: Evaluation of single-copy and multiplecopy routing protocols in a realistic VDTN scenario. In: Proceedings of the 10th International Conference on Complex, Intelligent, and Software Intensive Systems, CISIS-2016, pp. 285–289 (2016)

Recovery of Fiber Networks C/M-Plane via an IoT-Based Narrow-Band Links System Based on LoRa Mesh Platform Goshi Sato1(&), Yoshitaka Shibata2, and Noriki Uchida3 1

National Institute of Information and Communications Technology, 4-2-1 Norikibe-cho, Koganei, Tokyo 184-8795, Japan [email protected] 2 Faculty of Software and Information Science, Iwate Prefectural University, 152-52 Sugo, Takizawa, Iwate 020-0693, Japan [email protected] 3 Department of Information and Communication Engineering, Fukuoka Institute of Technology (FIT), 3-30-1 Wajiro-Higashi, Higashi-Ku, Fukuoka 811-0295, Japan [email protected]

Abstract. Although the LoRa private network has started to be used in various applications due to its characteristics, there are only arrangements for carrier sense time, transmittable time, and post-transmission standby time on the LoRa standard and Radio Law. It is necessary for the user side to independently deal with behaviors related to retransmission of data lost at the time of collision and error correction function. In addition, since it is a relatively new wireless communication method, it can be said that it is almost in a disorderly state without unified guidelines and tacit agreements that have been accumulated in the usual way. This situation is particularly problematic when various services start utilizing LoRa with their own protocols, so a uniformly available LoRa platform is necessary. We demonstrate a field-trial experiment of a lowspeed/latency/loss tolerable SDN control/management-plane which can take advantage of the widely available IoT resources and easy-to-create wireless mesh for the timely recovery of the C/M-plane after disaster.

1 Introduction As fundamental telecommunication infrastructure, optical transport networks are critical in modern society. In the cases of the large-scale disasters, the optical transport networks would be fatally damaged or destroyed (including not only the data-plane (D-plane) but also control and management-plane (C/M-plane) networks). Besides the D-plane recovery, in modern software-defined networking (SDN)-based fiber optical networks, quick recovery of the C/M-plane network is essential not only for emergency control of the surviving optical network resources, but also for the quick collection of information related to network damage/survivability such that the optimal recovery plan can be decided as early as possible [1]. Meanwhile, with the conventional approach, the manually performed information collection and the repair of the original © Springer Nature Switzerland AG 2020 L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 504–511, 2020. https://doi.org/10.1007/978-3-030-33506-9_45

Recovery of Fiber Networks C/M-Plane

505

C/M-plane networks in the optical networks would consume a lot of time. For the quick recovery of C/M-plane network, other complementary approaches are expected, e.g., employing the wired/wireless resources which are outside of the original optical networks such as the surviving LTE/Internet access and satellite links [2]. With the R&D of the internet of things (IoT) technologies, low-energy consumption and low-cost IoT devices are emerging. The corresponding long-distance networking technologies, e.g., low-power wide-area (LPWA) and LPWA-based mesh networks (LPWA-mesh) are promising to offer the wide coverage sensing and environment-data collection capabilities. Meanwhile, due to the bandwidth restriction and the trade-off between the data-rate and range, for the long-distance communication, LPWA-mesh would be the low-data rate, long-delay, lossy environment. Given these constraints, we are motivated to investigate the feasibility of the fiber network C/M-plane recovery via an IoT-based extremely narrow-band and lossy link system (FRENLL). In particular, we introduce a low-speed/latency/loss tolerable SDN control/management system design taking advantage of the widely available IoT resources for the timely recovery of the C/M-plane of optical networks. In this paper, we firstly present the primary study results with a field-trial experiment verifying the feasibility of the IoT-base disaster recovery of optical transport networks within the metropolitan area.

2 Purpose The purpose of this research is to construct a unified LoRa self-employed network platform that can be used in various situations using LoRa mesh network technology. Specifically, by applying the LoRa mesh network technology to design the optimum setting and data structure in the following three communication environments, it is possible to realize communication in which the example application is sufficiently operable in various communication environments with different situations It is aimed to evaluate that performance can be demonstrated through experiments. 1. Mobile phone network Sharing road conditions by inter-vehicle/road-vehicle communication on roads crossing insensitive areas 2. Temporary observation point data transfer and collection in mountainous areas. Such a platform that meets various communication environments and application requirements is one that could not be realized in the LoRa private network system so far and is realized by utilizing the LoRa mesh network technology of our organization.

3 Methods In order to exchange and share information with neighboring shoulder system in real time, we will integrate N different wavelengths of different wavelengths and develop N wavelength cognitive radio. Then, the radio link that maximizes the evaluation function based on the necessary communication quality (throughput, delay, packet loss rate, RSSI) is determined according to the change of the communication environment (communication distance, noise ratio, interference, etc.). By maximizing communication distance

506

G. Sato et al.

and total data transfer amount by realizing inter-vehicle communication. For this reason, we develop a function that can select an optimal wireless link from N wavelength radio links by SDN (Software Defined Network) protocol function. As the N wavelength radio, a radio for IoT such as a wireless LAN or LoRa radio corresponding to the IEEE802.11n (2.4 GHz)/ac (5.6 GHz)/ad/60 GHz) standard that is available for license generation and is the next generation wireless network We develop an N wavelength cognitive radio unit by suitable combination from standards etc. In particular, LoRa wireless communication uses broadcast communication, if collision of wireless signals occurs due to simultaneous transmission, retransmission is not performed. Also, in consideration of various communication environments, nodes that perform communication have hidden terminal problems because radio wave reach ranges are extremely different. In order to avoid simultaneous transmission, LoRa mesh network technology implements time-division access control method after time synchronization by GPS. It is assumed that all the radio stations operated in the system are numbered by ID and the total number is also grasped. In this premise, based on the ID dynamically allocate time slots that it can transmit. For example, assuming that the transmission time per wireless device is 10 s (considering the time required for backoff to avoid collisions with other systems), when using five wireless stations, these five devices do not have any problems A time longer than 50 s that can be transmitted is assigned as a transmission cycle. In this transmission cycle, as shown in Fig. 1, transmission time slots are assigned based on respective IDs. As a result, occurrence of collision in communication in the own system is avoided. The amount of data that can be transmitted per unit time of the wireless device and the communicable distance vary depending on factors such as the spreading factor of LoRa, the used channel bandwidth, the surrounding communication environment, and the like. Also, since the transmission cycle varies according to the number of radio devices used in the system, slot design of the entire system is required in advance according to the requirements.

Fig. 1. Relationship between transmission cycle time and transmission time per wireless device

Therefore, in this research, by appropriately designing the spreading factor of the LoRa, the used channel bandwidth, and the transmission cycle according to the following three kinds of application systems on the basis of the LoRa mesh network technology, Verify feasibility. By the way, data relaying is done by using flooding. However, in addition to the mechanism to avoid transmitting the same information multiple times, in order to reduce the number of flooding transmission, the following implementation is carried out.

Recovery of Fiber Networks C/M-Plane

507

First, by sending information listing the sender IDs of the signals received by each node in the past as control signals at regular intervals, it is possible to grasp the range within which their neighboring nodes can communicate. When a newly received information is received, if there is a node that advertised the information earlier before its own transmission timeslot comes around, if the neighboring node that it covers is adjacent to If it is all covered, it will not retransmit. As a result, retransmission of unnecessary information is suppressed. 3.1

Road Situation Sharing by Road-to-Vehicle Communication on a Road Crossing a Mobile Phone Network Insensitive Area

The proposer has been researching the speeding up of WiFi communication at the time of mistake by pre-sharing of WiFi connection information using LoRa. In order to further develop this, we will develop an in-vehicle type road-vehicle communication system using LoRa mesh network technology that enables bidirectional data storage type communication by LoRa. In advance, share meta information such as the type of application information held by each vehicle, the type of application data requested, the direction of travel of the vehicle, the traveling history, etc., and automatically determine the information to be transmitted. It is possible to more efficiently allocate the communication time by WiFi at the time of limited mismatch to data transfer. Unlike communication between fixed stations, communication time at mismatch is limited also in LoRa communication in road-to-vehicle communication, so design the transmission cycle time as short as possible as shown in Fig. 2 below So, share meta information efficiently.

Fig. 2. Transmission cycle time design for inter-vehicle/road-vehicle communication

In this research, in order to limit the maximum number of terminals simultaneously communicating, the target vehicle is assumed to be a called route bus, and the loadside server is installed at the bus stop or the like. Evaluate the amount of data transfer at the time of passing. 3.2

Transfer and Collect Data Between Temporary Observation Points in the Mountain Area

We develop an IoT sensor data collection network system using LoRa mesh network technology as a system for instantly constructing an information communication network enabling planar and real-time measurement by sensors in a limited mountainous area. Among the collected sensor data, there is a relatively large data amount. In order to cope with such a difference in size of data to be collected, it is possible to variably

508

G. Sato et al.

Fig. 3. Variable slot overview for collecting sensor data with different data amount

secure one slot time on the IoT sensor data collection network system side as a transmission path as shown in Fig. 3. 3.3

FRENLL Architecture in Disaster Recovery

Figure 4 illustrates the FRENLL architecture. In the disaster area, due to the fiber cuts and node (e.g., reconfigurable optical add/drop multiplexer (ROADM)) failures, the original D-plane and C/M-plane networks are isolated, with surviving fibers and nodes unreachable from the SDN controller (Ctrl) or network management system (NMS). To collect the survivability information first, instead of sending technical staffs and manually checking individual nodes, we can employ the wireless resources outside of the original optical network, e.g., with LPWA-mesh for message exchanging. Note that this LPWAmesh can be created or enabled with the LPWA-enabled IoT devices after disasters, or it may have already existed in the field which was established for offering various types of daily IoT services by operators before disaster, e.g., environment sensing and data collection. The details of the concrete services and business models are out of the scope of this paper and omitted herein. To employ this LPWA-mesh as a temporary emergency C/M-plane network, we introduce two types of agents, namely, delegator and delegatee into the SDN-based C/M-plane for information collection and commands’ delegation. Both of them have an interface to access the LPWA-mesh for message exchanging.

Fig. 4. System architecture of the LPWA-mesh-based C/M-plane recovery.

Recovery of Fiber Networks C/M-Plane

509

4 Preparation of Experimental Environment Figure 5 illustrates the setup of the field-trial experiment which included four nodes, from A to D in the optical network. It was assumed that these nodes were originally connected with fibers as a part of a metropolitan optical ring network. In the case of a disaster, it was assumed that in D-plane, node B was damaged, and the fiber between nodes A and B was broken. In the C/M-plane, the SDN Ctrl/NMS which was closed to node A was unreachable from nodes B, C and D. As mentioned above, to perform quick C/M-plane recovery, delegator E, delegatees F and G were placed one after another at node A, B and C, respectively, as shown in Fig. 5. It is noteworthy that, during the placement of delegatee F, it was assumed that the damaged node B was also replaced by emergency optical systems (EOS) which are lightweight portable disaggregated subsystems (e.g., Add/Drop unit, transceiver unit, and optical amplifier etc.) to quickly recreate the necessary functions (Add/Drop, transceiver, and optical signal amplification) [3]. Hence, node B and the surviving nodes C and D and the surviving fibers can be employed to recreate the desired lightpaths e.g., between B and D to carry the important traffic, e.g., for victim relief. However, since the vendor-specific optical supervisory channels (OSC) were broken between (A, B) and (B, C) due to the fiber cut and the original node B failure (not recoverable at once), the restoration of the original C/M-plane network will still take time.

Fig. 5. Setup of the field-trial experiment of FRENLL in disaster recovery.

In addition, prototype N-wavelength wireless communication prototype system (Figs. 5 and 6) implementing this protocol is prototyped using NerveNet [3] developed by NICT and a laboratory test is being conducted. Basic protocol operation can be confirmed by expected operation, basic data collection and system design of N

510

G. Sato et al.

wavelength cognitive radio, which is the target of this fiscal year, system design, implementation per module are expected to be completed in the current fiscal year. Implementation to combine IEEE802.11 bgn and LoRa will be implemented in the future, and information transfer experiments using various data and roadside-vehicle communication will be demonstrated outdoors (Fig. 7).

Fig. 6. System appearance (on vehicle)

Fig. 7. System appearance (loadside)

5 Summary In this paper, we propose a unified LoRa platform that does not rely on existing communication infrastructure and designed to adapt to time-division access control method necessary for realizing it. In addition, we developed a system for sharing vehicle position information using a self-operated network using a basic time division access control method, actually installed a radio at the hospital rooftop of the city center, and then sent out-of-line communication We verified whether or not desired communication can be done even in environment. As a result, it was confirmed that even in the downtown area, it is sufficiently possible to communicate about the sharing of the position information of the vehicle within a radius of about 2 km. In addition, it is shown that time sharing access control communication using basic technology of LoRa platform enables simultaneous sharing of information among multiple sites. And We investigate the feasibility of the fiber network C/M-plane recovery via an IoT-based extremely narrow-band and lossy link system (FRENLL) which takes advantage of the broadly distributed IoT resources and the LPWA-based wireless mesh networks. We present the primary study results with a field-trial experiment successfully verifying the feasibility of the IoT-base disaster recovery of optical transport networks within the metropolitan area. Acknowledgments. This research was supported by Strategic Information and Communications R&D Promotion Program (SCOPE) No. 181502003, Ministry of Internal Affairs and Communications, Japan.

Recovery of Fiber Networks C/M-Plane

511

References 1. Xu, S., et al.: Ann. Telecommun. no. 73 (2018) 2. Monitoring systems for outside plant facilities, ITU-T Recommendations, no. L.81, pp. 1–10 (2009) 3. Shiraiwa, M., et al.: Experimental demonstration of disaggregated emergency optical system for quick disaster recovery. IEEE JLT 36(15), 3083–3096 (2018) 4. Owada, Y., et al.: IEICE Technical report, ASN2018-21, July 2018

Clustering Analysis and Visualization of TCM Patents Based on Deep Learning Na Deng1(&), Xu Chen2, and Caiquan Xiong1 1

School of Computer Science, Hubei University of Technology, Wuhan, China [email protected] 2 School of Information and Safety Engineering, Zhongnan University of Economics and Law, Wuhan, China [email protected]

Abstract. In the process of medicine innovation, pharmaceutical enterprises tend to seize the intellectual property highland actively. They engage in research and development independently, apply for patents for core technologies, or take the initiative to acquire patents from others. Before applying for patents by their own efforts or purchasing patents from others, pharmaceutical companies need to search for related patents in the patent pool and make a comparative analysis of them, in order to find technology blank areas as R&D objectives, or find valuable patents as potential acquisition targets. In this paper, we use deep learning technology and propose a semantic-based clustering algorithm for Traditional Chinese Medicine (TCM) patents, discarding the traditional literal– based text clustering method. We also give a visualization method for TCM patents, so as to facilitate pharmaceutical enterprises to intuitively understand the relevant patents.

1 Introduction In the process of medicine innovation, pharmaceutical enterprises tend to seize the intellectual property highland actively. They engage in research and development independently, apply for patents for core technologies, or take the initiative to acquire patents from others. Before applying for patents by their own efforts or purchasing patents from others, pharmaceutical companies need to search for related patents in the patent pool and make a comparative analysis of them, in order to find technology blank areas as R&D objectives, or find valuable patents as potential acquisition targets. Patent retrieval is an indispensable task for pharmaceutical enterprises before R&D or patent acquisition. If this work is lacking, the biggest consequence will be repeated R&D or patent infringement. In 1994, the National Committee for the Review of New Traditional Chinese Medicine pointed out that 90% of the research and development of new traditional Chinese medicines in China are repetitive studies, and this conclusion is still valid until now. In the millions of patents issued annually by the State Intellectual Property Office, perhaps half of which will be overturned by missed patents or other facts.

© Springer Nature Switzerland AG 2020 L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 512–520, 2020. https://doi.org/10.1007/978-3-030-33506-9_46

Clustering Analysis and Visualization of TCM Patents

513

Because of the characteristics of high investment, high risk and long period in the R&D of traditional Chinese medicine (TCM), the missed retrieval of TCM patents will result in more waste of R&D resources than that of many other technical fields. And patent infringement litigation will bring a fatal blow to pharmaceutical enterprises, which will not only make them fall into a long litigation period, but also may face huge infringement fees. For example, Tianshili sued Wancheng Company for infringement of patent of Chinese medicine “YangXueQingNao Granule” for 19 months [1]. The infringement case between Shanxi HanWang Pharmaceutical Industry and Jiangxi Yintao Pharmaceutical Limited Corporation on patent “a kind of Chinese medicine composition with antihypertensive, lipid-lowering, dizziness-fixing and wind-fixing effects and its preparation method and use” involved 6 million Yuan [2]. In order to facilitate pharmaceutical enterprises to have an overall understanding of similar TCM patents in patent retrieval, different from the traditional text clustering method based on literals, this paper proposes a semantic-based clustering algorithm for TCM patents by using deep learning technology, and presents a visualization method for TCM patents of each cluster in the form of word cloud.

2 Related Work Because of the importance of intellectual property in knowledge economy, A lot of scholars at home and abroad devote themselves to the research of patent retrieval and analysis. [3] proposes a multi-lingual patent retrieval system. [4] emphases that patent retrieval is a recall-oriented information retrieval. In order to improve the efficiency of patent information retrieval and the integrity of retrieval results, [5] applies ontology into patent retrieval. [6] proposes a method to combine text-based and citation-based retrieval methods in the invalidity patent search. In terms of patent analysis, [7] constructs a feature vector space model by mapping patent documents to feature vectors extracted by convolutional neural networks. In addition, the paper also discusses the applications of FVSM for three typical patent analysis tasks, i.e., patents similarity comparison, patent clustering, and patent map generation. [8] describes a new approach to corporate technological performance assessment, based on patent citation analysis. [9] proposes a hybrid-patentclassification approach that combines a novel patent-network-based classification method with three conventional classification methods to analyze query patents and predict their classes. [10] uses SVM for patent classification. [11] proposes a combinatorial model of machine learning to patent quality classification and forecasting in the biomedical industry. In patent clustering, [12] proposes a new Bayesian model for patent clustering. [13] uses adaptive k-means algorithm to cluster patents. [14] applies technology and effect matrix for patent clustering.

514

N. Deng et al.

3 Clustering of TCM Patents 3.1

Representation of TCM Patents

In traditional text analysis, clustering and classification, text is represented as bag-ofwords, which can only reflect whether a word appears in a text, but can not reflect the semantic relationship between words. When we calculate text similarity, using the traditional bag-of-words based vector space method, text is represented as a vector, and the similarity between texts is converted into the distance calculation between vectors. The disadvantage of this kind of method is that the vector dimension is too large and the vector space is sparse, and it can not reflect the semantic similarity between texts. In the era of big data, deep learning develops rapidly, and its excellent feature learning ability makes it perform well in image processing, image recognition and other fields. Word2Vec build a bridge between deep learning and natural language processing. It can express words as vectors and transform texts into the input of neural networks. Different from traditional bag-of-words, Word2Vec represents words as lowdimensional vectors (150–350 or so), which greatly reduces the workload of vector computation. Meanwhile, these low-dimensional vectors also contain semantic information between words. Doc2Vec technology is the derivation of Word2Vec, which can transform texts into low-dimensional vectors. At the same time, these vectors also contain the semantic information of texts. In order to avoid the shortcomings of traditional bag-of-words vector representation, which is computationally heavy and does not contain semantics, this paper will use Doc2Vec to represent TCM patent texts. Based on the text vectors generated by Doc2Vec, K-means algorithm is used to cluster patent texts. 3.2

Flow Chart of Clustering Algorithm

The flow chart of clustering TCM patent texts is shown in Fig. 1. Preprocessing TCM Patents

Chinese Word Segmentation

Clustering Results

Execute K-means Algorithm

Punctuation Removal

Stop Words Removal

Digits Removal

Document ID + Word Sequences

Vectors of Documents

Train Doc2Vec Neural Network

Fig. 1. The flow chart of clustering TCM patent texts.

Clustering Analysis and Visualization of TCM Patents

515

The flow chart is described as follows: 1. Prepare some TCM patent texts to be clustered. 2. Preprocess each TCM patent text, including Chinese word segmentation, punctuation removal, stop words removal and digits removal. 3. For each TCM patent text, the document ID and the word sequence of the document are obtained. 4. Document IDs and word sequences are the input of Doc2Vec neural network. After training of neural network, the vector representations of documents are obtained. 5. Documents are represented as vectors and clustering analysis is carried out by Kmeans algorithm. Finally, the clustering results of TCM patent texts are obtained. 3.3

Preprocessing

Because the abstract part of TCM patents contains its main important information, in this paper, the abstract texts of TCM patents are used as the data source of patent clustering. A screenshot of abstract texts is shown in Fig. 2.

Fig. 2. The flow chart of clustering TCM patent texts.

We need to preprocess on the original data source, including Chinese word segmentation, punctuation removal, stop words removal and digits removal, to get word sequences. The code in Python is shown in Fig. 3.

Fig. 3. Python code of preprocessing on the original data source

516

N. Deng et al.

In the above code, we show a process of patent preprocessing for TCM patents. Jieba is used for Chinese word segmentation. Stop words are removed by using stop words list obtained by word frequency statistics and manual screening. Digits are removed by regular expressions. Part of word sequences is shown in Fig. 4.

Fig. 4. Word sequence after preprocessing.

3.4

Generation of Data Set

After preprocessing on the original data source, the acceptable input to Doc2Vec needs to be generated. Figure 5 shows the Python code for generating the data set. Unlike Word2Vec, Doc2Vec requires not only the word sequence of the text, but also the serial number of the text, which is encapsulated by gensim. models. doc2vec. TaggedDocument. The list of Tagged Documents of all texts constitutes the input data set train_input required by the Doc2Vec neural network.

Fig. 5. Python code of generating data set for Doc2Vec neutral network.

3.5

Training of Doc2Vec Neural Network

The data set obtained in the previous section serves as the input of the Doc2Vec neural network. Figure 6 is the Python code for training Doc2Vec neural network. After setting the operating parameters of Doc2Vec, train_input is used as the input of Doc2Vec, and the training model results are saved after five iterations.

Clustering Analysis and Visualization of TCM Patents

517

Fig. 6. Python code of training Doc2Vec neutral network.

3.6

K-means Clustering

After the training of Doc2Vec neural network, each patent abstract text is expressed as a low-dimensional vector. Based on these vectors, we use classical K-means algorithm to cluster patents. The Python code for clustering is shown in Fig. 7. Each patent abstract text is transformed into a vector accepted by K-means model through function infer_vector of Doc2Vec. n_clusters sets the number of clusters generated by clustering algorithm, and we set it to 15 here. In this paper, we need to cluster 7000 patent abstract texts; thus, labels stores the cluster number of 7000 patent abstract texts (i.e. 0–14).

Fig. 7. Python code of clustering.

3.7

The Storage of Clustering Results

In the previous section, labels has stored the cluster number of each TCM patent abstract text, but in order to facilitate the visualization of each cluster of patents, we will store the word sequence of patents belonging to each cluster in a separate file. The code is shown in Fig. 8. For example, the word sequences of all patent abstracts of cluster i is stored in file named classify_i.txt.

518

N. Deng et al.

Fig. 8. Python code of storing clustering results.

4 Visualization of TCM Patents In order to enable pharmaceutical companies to have an intuitive and visual understanding of TCM patents in each cluster, we intend to use Word Cloud technology to generate word clouds for each cluster. The code is shown in Fig. 9. After setting the parameters such as font, background color, height and width of the word cloud, the word cloud of each cluster can be obtained by using the word sequences of each cluster.

Fig. 9. Python code of generating word cloud.

Clustering Analysis and Visualization of TCM Patents

519

Table 1 shows some word clouds of different clusters. Table 1. Some word clouds of different clusters.

5 Conclusion In order to make pharmaceutical enterprises have an intuitive understanding of TCM patents, this paper proposes a clustering algorithm of TCM patents based on Doc2Vec technology in deep learning, and visualizes the clustering results using word cloud technology. The results show the effectiveness of the method. Acknowledgments. This work was supported by National Key Research and Development Program of China under Grant 2017YFC1405403; National Natural Science Foundation of China under Grant 61075059; Philosophical and Social Sciences Research Project of Hubei Education Department under Grant 19Q054; Green Industry Technology Leading Project (product development category) of Hubei University of Technology under Grant CPYF2017008; Research Foundation for Advanced Talents of Hubei University of Technology under Grant BSQD12131; Natural Science Foundation of Anhui Province under Grant 1708085MF161; and Key Project of Natural Science Research of Universities in Anhui under Grant KJ2015A236.

520

N. Deng et al.

References 1. https://www.zhihuiya.com/articles/view/151/ 2. http://ipr.court.gov.cn/zgrmfy/zlq/201304/t20130410_153707.html 3. Higuchi, S., Fukui, M., Fujii, A., et al.: PRIME: a system for multi-lingual patent retrieval. In: Proceedings of Mt Summit VIII (2002) 4. Jones, G.: Toward higher effectiveness for recall-oriented information retrieval: a patent retrieval case study. Machine Translating (2012) 5. Chen, J.X., Gu, X.J., Chen, G.H., et al.: Ontology-based patent retrieval technologies. J. Zhejiang Univ. 43(12), 2213–2217+2224 (2009) 6. Fujii, A.: Enhancing patent retrieval by citation analysis. In: International ACM SIGIR Conference on Research & Development in Information Retrieval. ACM (2007) 7. Lei, L., Qi, J., Zheng, K.: Patent analytics based on feature vector space model: a case of IoT. IEEE Access 7, 45705–45715 (2019) 8. Narin, F., Carpenter, M.P., Woolf, P.: Technological performance assessments based on patents and patent citations. IEEE Trans. Eng. Manag. 31(4), 172–183 (2017) 9. Liu, D.R., Shih, M.J.: Hybrid-patent classification based on patent-network analysis. J. Am. Soc. Inform. Sci. Technol. 62(2), 246–256 (2011) 10. Chu, X.L., Ma, C., Li, J., et al.: Large-Scale patent classification with min-max modular support vector machines. In: Proceedings of the International Joint Conference on Neural Networks. IEEE, Piscataway (2008) 11. Liu, B., Lai, M., Wu, J.L., et al.: Patent analysis and classification prediction of biomedicine industry: SOM-KPCA-SVM model. Multimed. Tools Appl. 2019, 1–21 (2019) 12. Choi, S., Jun, S.: Vacant technology forecasting using new bayesian patent clustering. Technol. Anal. Strateg. Manag. 26(3), 241–251 (2014) 13. Shanie, T., Suprijadi, J., Zulhanif: Text grouping in patent analysis using adaptive k-means clustering algorithm. In: American Institute of Physics Conference Series (2017) 14. Xu, C., Peng, Z.Y., Liu, B.: Technology and effect matrix for patent clustering. In: Web Information System & Application Conference (2014)

Efficient Resource Utilization Using Blockchain Network for IoT Devices in Smart City Muhammad Zohaib Iftikhar, Muhammad Sohaib Iftikhar, Muhammad Jawad, Annas Chand, Zain Khan, Abdul Basit Majeed Khan, and Nadeem Javaid(B) Department of Computer Science, COMSATS University Islamabad, Islamabad, Pakistan [email protected]

Abstract. With rapid increase in the use of technology, the world is now moving towards smart cities which require the communication and collaboration of Internet-of-Things (IoT) devices. The smart city enhances the use of technology to share information and data among devices. These devices are producing a huge volume of data that needs to be tackled carefully. Different works have already been proposed to provide a communication in a network for the IoT devices; however, nothing has been founded more effective in terms of resource utilization. Hybrid network architecture is the combination of a centralized and distributed network architectures. The centralized network is used for the communication of IoT devices with edge nodes and distributed network for communicating miner nodes with edge nodes. In this way, the network utilize a lot of resources. In this paper, we are proposing a single network which is the combination of both edge nodes and miner nodes. Blockchain is also implemented in this network to provide secure communication between the devices. The evaluation of the proposed model is done using different performance parameters such as time and cost against the number of devices. Limited number of devices are used to perform this evaluation. Furthermore, the results are obtained by utilizing Proof-of-Work consensus mechanism. Keywords: IoT · Smart city Resources · Network

1

· Communication · Blockchain ·

Introduction

The smart city is the collection of different components such as the hospitals, the smart home, the vehicular transportation, etc. Each component further consists of IoT devices that need to communicate with one another to do real-time data processing and to share data using different sensors. In this way, it increases the standard of living, the government services and the environment as well [1]. The applications of a smart city, include a biased medical data sharing between cloud c Springer Nature Switzerland AG 2020  L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 521–534, 2020. https://doi.org/10.1007/978-3-030-33506-9_47

522

M. Z. Iftikhar et al.

servers through blockchain which is proposed in [2]. The devices do not trust each other in order to communicate with each other. A secured mechanism based on blockchain is proposed by [3], where hybrid network architecture is proposed to communicate the devices with each other. The model proposed in this paper consists of two networks. One is edge network which is connected with IoT devices forming a centralized network while the other is a miner network which is connected with the edge nodes forming a distributed network. Miner nodes have more computational power and storage resources than edge nodes. Here, the edge nodes are responsible to provide communication between devices and miner nodes are responsible to perform high computational tasks. The hybrid network uses more resources. The security is achieved using blockchain, which is considered as the most secure technology until now. Trust is achieved by using a trust management scheme proposed in [4]. Every node/block in the blockchain stores the record of all the transactions that take place in the network which is called ledger. Due to this, it’s really hard for the attacker to tamper data in the blockchain because every block in the blockchain maintains a ledger. Using blockchain, the devices do not need to depend on the third party for any type of authentication and verification instead, they can communicate directly. A smart contract is also used to handle transactions that are stored on the blockchain. A consensus mechanism Proof-of-Work (PoW) is also implemented in the block-chain to protect each node from the attacks which is proposed by Nakamoto [5]. Each transaction is verified at each node by performing mining. In the mining process, a complex mathematical puzzle is assigned to every miner and they compete to solve the puzzle. A reward is given to the miner who solves the puzzle first. After verification, the new transaction is then broadcasted in the network and stores in the blockchain. By analyzing the previously proposed model, we come to know that the model is still lacking in the efficient placement of the nodes in a network. Also, there is not any need for the two networks in the model as it will use additional computational power and resources. The IoT devices are linked to the edge nodes forming a centralized network which is a single point of failure. So, a solution to this is discussed in this paper. We have used a single network and replace some edge nodes with the miner nodes in the same network. In this way, we do not need to maintain a separate network for different nodes.

2

Motivation

A smart city is based on intelligent information sharing between different IoT devices. We are taking motivation from [3], which proposed hybrid network architecture. One network is called the core network while the other is called an edge network. Edge nodes are connected directly to the smart devices in the network while miner nodes are responsible to perform high computational tasks. Also, we are getting motivation from [6], which proposed a crowd sensing network to communicate with IoT devices. It further deals with a huge amount of data. Furthermore, [9] proposed blockchain architecture for an intelligent vehicle.

Efficient Resource Utilization Using Blockchain Network for IoT Devices

3

523

Problem Statement

The hybrid network architecture for a smart city network proposed in [3], is not efficiently utilizing the resources. The edge nodes have low storage and they need to process raw data, uploaded by end devices. The placement of edge nodes is not handled, which is also stated as its limitation. Moreover, this architecture utilizes more time and cost. To implement the two networks, we require extra computational resources and cost. Also, it takes more time for the devices to communicate between different networks. So, we are more concerned to reduce cost and time of communication between devices by using single network instead of multiple networks.

4

Related Work

In this section, we are presenting some techniques that are using blockchain in different networks model. Existing work is also included along with the relevant research trend. 4.1

Wireless Sensor Networks

Resource-constrained IoT devices that are of limited resources are discussed in [6]. The devices run services beyond their capabilities which increases the security risks usually due to untrustworthiness. A mechanism is suggested to keep the validity status of the edge servers. The proposed model is working on the on-chain and off-chain communication modes, in a way that service business is at off-chain and the security business is on-chain. A consortium blockchain is deployed with the Proof of Authority (PoA) consensus mechanism which ensures the high throughput and low latency in the system. An other blockchain based node recovery scheme is proposed in [10]. There are three ways of data storage in Wireless Sensor Networks (WSNs) available until now [11]. However, there is not any incentive mechanism available for the data storage in WSN. An incentive mechanism is proposed, based on the blockchain through which the node who will store data for another node will get a reward. Provable Data Possession (PDP) is used which causes the waste of resources in terms of electricity, etc. Smart city network architecture [3] and in the cloud based service provider [12], a number of IoT devices are connected to each other. It arises the issues like high latency rate and more bandwidth usage. Moreover, it also affects the efficiency and scalability of the trust management system. The proposed network is divided into two groups, one is called the miner nodes network and the other one is called the edge nodes network. Edge nodes offer real-time processing and are deployed at the edge network. The IoT devices in the smart city components are connected to these edge nodes forming a centralized network whereas the edge nodes are connected to the miner nodes forming a distributed network. Miner nodes are responsible for generating new blocks and applying consensus on them. [7] have worked on the smart networks using blockchain and [7] has worked on the underwater WSNs.

524

M. Z. Iftikhar et al.

4.2

Vehicular Network

Vehicular communication protocols based on cellular networks are not very updated and these are even not suitable for Intelligent Transport System (ITS) [9]. Also, the issues related to ITS communication are trust, privacy, and security. A central point of trust in the communication network is proposed which is called Intelligent Vehicle Trust Point (IVTP) using blockchain without interference from other vehicles. It also assigns a unique crypto ID to every vehicle that is the part of the network. In the Vehicular Network (VN), the amount of traffic is huge which cannot be managed by a centralized network. It raises the issue of interoperability and compatibility [13]. Moreover, the security between vehicles is also very important. A blockchain based decentralized system is applied to handle the security issues, high cost, and efficiency. Five different blockchains are used, each handle different types of data. The first one tackles car monitoring data such as speed. The second one handles sensory data like temperature. The next one is used to store streaming data like audio, video. The Matlab tool is used to perform simulations for the proposed blockchain based vehicular network. The parameters that are used for performance evaluation include throughput and average delay time. In [14], secure communication between vehicles is the core requirement for ITS because vehicles have to share data with each other. Moreover, the proposed system tackles the challenges of flexibility and robustness. Block-VN network is proposed to provide communication between vehicles in a distributed network. The miner nodes in the network are the vehicles that handle requests. The remaining vehicles are termed as ordinary nodes. A new vehicle/node can send a request to either the controller or the ordinary node. In this way, the system is achieving high trust between the vehicles. The controller node is responsible for the computation of data. Furthermore, no mechanism is proposed for the verification of the client and the authenticity of messages. 4.3

Data Sharing

Data sharing in smart grid is discussed in [15]. Another data sharing scenario between multiple devices that cause long time latency is discussed in [16]. The distrust issues between devices are further highlighted. A blockchain is proposed to handle the sharing of data at the edge nodes. This blockchain framework is using a consensus called Proof of Collaboration (PoC). Every edge node compete to create new blocks by showing their credits instead of showing computation power which utilizes large resources. Moreover, express transactions and hollow blocks are also proposed to increase the overall network efficiency. Different Mobile Network Operators (MNOs) communicate with each other to share data due to trust issues [17]. The term Artificial Intelligence (AI) is used in a sense to make this communication process between networks automatically. A trust-based data sharing framework is proposed using permissioned blockchain which is based on the smart contract. Restrictions are applied to the data accessed, and this permission is only granted by the owner of the data

Efficient Resource Utilization Using Blockchain Network for IoT Devices

525

through a smart contract. Furthermore, the data access is controlled and monitored using datachain and behavior chain. Datachain is used to control the access of the data while the behavior chain is used to keep the record of the data access. To achieve security better, permission levels are divided into four levels. L0, L1, L2, and L3. In level L0, the data is shared with the user only. In level L1, only the computation data is visible and in level L2, access is given to the authorized persons. In level L3, the data is available publicly to everyone. The model of system is implemented in Hyperledger fabric. A smart contract is used to execute all transactions between the user layer and system management layer. The economic benefits of the proposed system are not discussed properly. 4.4

Data Access Management

The validity of Channel State Information (CSI) is important because resources are assigned to the users on this basis [18]. It discusses the issues of CSI and uses a statistical-based algorithm to solve these; however, no one addressed the issues of access control in network for the authenticity of CSI. The main contributions include to find out the efficiency of the mobile users and validation of CSI of each user using blockchain consensus mechanisms. User’s information is gathered including the CSI of mobile users using devices like signal meters and Global Positioning Systems (GPS). A network representing transmittingreceiving pairs of users is considered for simulation especially the Erd¨ os-R´enyi network. Moreover, the proposed algorithm will work best and provide the optimal results for highly concentrated transmitting and receiving pair. [19] have also worked on data access rights using blockchain and [20] contains work on the monetization of data. The already existing centralized system is dependent on the third party for the electronic payments which raises the issue of trust [21]. Every node in the blockchain stores the record of each transaction which requires a huge storage space which in the future will become difficult to manage. A storage mechanism called Network Coded Distributed Storage (NC-DS) is proposed to store the blockchain. The lack of storage and bandwidth problem collectively called bloating problem which is tackled by the proposed framework without losing any data. A block is divided into 3 sub-blocks which are further divided into 6. NC-DS is implemented in two ways. Deterministic Rate (DR) and Rate Less (RL). NC-DRDS is used when there is a finite number of encoded packets while NC-RLDS is implemented when encoded packets are unknown. Moreover, when the system wants to get data from these encoded blocks, then it has to perform decoding first. Also, every packet contains different data so it becomes easier for the attacker to manipulate the data. [22] have worked on trustfulness in complex networks and [23] implemented blockchain in smart grids. 4.5

Smart Cities

No single protocol exists for the communication of different IoT devices [24]. Also, security issues related to these devices are further addressed. Moreover, the

526

M. Z. Iftikhar et al.

standards that exist already are not scalable enough. The blockchain contains a distributed ledger that is maintained through a consensus at each node. IoT chain, IoTex, and NeuroMesh have used to overcome storage issues but they all have some drawbacks. Their solution provides communication for IoT based on integrity. A huge data is shared between different devices [25]. Also, a low-latency trusted data exchange platform is required. Moreover, a blockchain based trusted data exchange system is proposed which is called SpeedyChain. It further consists of smart vehicles, Roadside Infrastructure Units (RSIs) and Service Providers (SPs). Furthermore, it provides more secure communication between vehicles to infrastructure and maintains privacy by continuously changeable keys. Smart city environment challenges are addressed in [27], like storage and communication. Moreover, a blockchain based scheme is proposed for big data auditing. Also, a Data Auditing Blockchain (DAB) that collects proofs instead of a transaction is proposed. They are performing both the batch auditing and dynamic auditing. In this way, the security, reliability and efficiency have been improved. Furthermore, the scheme is validated using Java JDK and all algorithms are implemented using Java Pairing Base Cryptography library (JPBC). [26] have worked on light weight clients based on blockchain.

5

System Model

Different components combine together to form a smart city network. The components/systems that are connected to this network are called smart components/systems because they are using different IoT devices for communication. So in this system model, each system is named as a smart application as clearly seen from Fig. 1. Secure data sharing among different IoT devices was a problem as stated in [28]. According to this, a device will not share any information unless it makes sure that communication is secure. This paper proposes a more efficient solution. This model is categorized into three layers. At the bottom, there are some applications of a smart city. These applications can be a smart home, hospital, building, roads, smart grid or industry. A building can be said smart, if it contains a large number of smart devices that communicate with each other. At the middle layer, a whole smart city network is illustrated which is communicating with the layer above it by using different kinds of arrows. Different arrows show that devices are communicating using different mediums. These mediums can be different kind of networks through which the devices communicate with each other. Moreover, the devices are using the above layer which contains blockchain network. This network contains multiple blockchains that are communicating with each other for the proper functioning of the devices. The proposed system model is motivated from [3,4] and [16]. At the top level, two blockchains are used, in which a miner node is surrounded by few edge nodes. Edge nodes are connected to a single miner node in a blockchain which means that one edge node is connected to multiple IoT devices. Miner

Efficient Resource Utilization Using Blockchain Network for IoT Devices

Fig. 1. Proposed system model

527

528

M. Z. Iftikhar et al.

node is responsible for creating new blocks and verifying through PoW consensus mechanism. Miner nodes have high computational power and storage space as compared to the edge nodes. This is basically a mesh network of different IoT devices. These devices communicate with each other through edge nodes and generate an enormous amount of data which require huge storage space. Moreover, when a smart application tries to interact with the system, it first send request to the edge nodes which further passes the request to the core nodes for validation and approval. The work flow of the proposed model can be clearly understandable from Fig. 1. It can be clearly seen that the smart city is surrounded by blockchain clouds. The previously proposed model uses multiple networks that will utilize extra resources and more computational power to run the network. We have eliminated two networks that were proposed in the base model and have merged them to one. Miners are responsible to verify the new devices that wanted to enter into the network. As all miners are performing the same task, so instead of using multiple nodes, we can assign this task to any of the edge nodes. We can do this by using high storage capacity and powerful edge nodes or placing a single miner node to each edge node network. Smart city is the collection of multiple components of a city like smart hospitals, smart parking, smart organization and smart home, etc. As it can be seen in Fig. 2, that the interaction of each single component of a smart city; such as a smart home. The smart home contains multiple IoT devices that will communicate with each other using the proposed blockchain model. A device may need to communicate with other devices for it’s proper functioning. But sometimes they do not trust each other because of their privacy and security.

Smart Appliances

Smart Appliance

Smart Home

Blockchain

Components

Devices Name

Fig. 2. Smart home with IoT devices

Efficient Resource Utilization Using Blockchain Network for IoT Devices

529

The security of the model is tackled in a way that edge nodes manage all IoT devices because all the devices are directly connected to the edge nodes. Data is passed by the end devices (IoT) to the edge node which provides real-time processing of the data. The edge nodes process the data that is uploaded by the end devices. They perform the scrutinization process to get useful information/data. The edge node then transfers this data to the core node (if required in terms as needed more resources and power). The miner node will validate the data by applying PoW to find out any illegalities in the data. The Miner node is responsible to perform the high computational tasks. The Miner nodes compete each other to solve the task to get the reward. The reward may be in terms of bitcoins or ratings, etc. It is a kind of game, in which a puzzle is given to the nodes to solve.

6

Results and Discussion

We have used the solidity language to implement the proposed idea. A blockchain is used to store records of multiple transactions. A smart contract is written in .sol file to handle the transactions in the blockchain. Furthermore, we have used an array list to store the transactions. All the transactions are then added in the array of blockchain manually using the code. The results are obtained by deploying the contract in an online tool called Remix using Ganache and metamask. A remix is an online tool used to implement the contract directly without making its javascript file. Ganache provides some fake accounts with real money

Fig. 3. Calculated cost against transactions

530

M. Z. Iftikhar et al.

to perform transactions. The transaction id is linked with the online contract deployment tool account and the money is provided in the form of ether. Ether is the currency used for the ethereum platform to exchange transactions. For every transaction, a block is created in Ganache. The PoW consensus mechanism is applied in the edge node network to process and to validate new transactions in the blockchain. The total cost is calculated against the values of gas consumed in Rinkeby and Ropsten network and the transaction rate as shown in Fig. 3. Gas consumed is calculated using Ganache and the transaction rate is provided by the metamask. At the x-axis, the number of transactions are presented ranging from 1 to 5. On the y-axis, the estimated cost is given. The figure shows that by increasing the number of transactions, the cost price remains almost stable except for the first transaction as clearly seen from the figure above. This cost is calculated against each gas consumption. The gas which is consumed by each transaction to mine the block and the transaction rate that it takes using metamask. From this, we can conclude that there are not any major changes in cost. As it remains almost the same in almost every transaction. This is also represented in the tabular form in the following Table 1. Table 1. Number of transactions vs cost Number of Transactions 1

2

3

4

5

Gas consumed

127059

97059

97315

97699

97315

Transactions rate

0.002541 0.001941 0.001946 0.001954 0.001946

Cost (gas/sec)

322.85

188.39

189.37

190.90

189.37

These both values are then multiplied in order to get the value of the cost. As stated above, Ropsten test network is also used to run the same contract. This network uses PoW consensus mechanism by default. Figure 4 looks almost same as the previous one but the results are different. It also shows that by increasing the number of transactions, the cost remains almost constant except for the second transaction. Furthermore, the total gas is calculated by multiplying the transaction rate and gas consumed as calculated in the previous network.

Efficient Resource Utilization Using Blockchain Network for IoT Devices

531

This tells us that the cost is totally dependent on the type of network, used to run the transactions. As Ropsten uses PoW consensus that is why it costs less than the other network which was a Rinkeby network in this case.

Fig. 4. Calculated cost against transactions using Ropsten Network

The above data is also represented in the tabular form in the following Table 2. Table 2. Number of transactions vs cost in Ropsten Network Number of transactions 1

2

3

4

5

Gas consumed

646494

127059

97059

97379

97315

Transactions rate

0.000646 0.000127 0.000097 0.000097 0.000097

Cost (gas/sec)

9.44

16.13

9.41

9.44

9.43

The computational resources are considered as the number of devices. As by increasing the number of devices, we are assuming that they will increase the computational resources, as shown in Fig. 5.

532

M. Z. Iftikhar et al.

Fig. 5. No. of devices vs time/sec

In the above chart, the time is calculated against the number of devices. The number of devices are increased one by one manually to see the changes in the block time and the changes in the results as shown in Fig. 5. As it can be clearly seen that every device time is different from each other except the device number 3 and 4. This may be due to several reasons; same type of devices or devices are added at the same time. Furthermore, it means that as the number of devices increased, the time also fluctuates just because every device is different from others. The parameters at both the axes are also shown in Table 3 below. Table 3. No. of devices vs time No. of devices 1 Time/sec

2

3

4

5

0.24 0.17 0.12 0.12 0.08

Trade-off exists when we are achieving the value of something by compromising/degrading the value of another thing. In the proposed system, the trade-off between two parameters exist such as between the number of transactions and the cost. It means that we are controlling/minimizing the cost by paying the price in terms of increasing the number of transactions. The cost is calculated on two different networks and we are concluding that the cost is reducing by using PoW consensus mechanism network.

7

Conclusion

Huge amount of data is generated by different IoT devices in the smart city network. Our proposed model provides a single network that reduces the time

Efficient Resource Utilization Using Blockchain Network for IoT Devices

533

and cost for communication between devices. Also, the consensus mechanism is implemented to make the network secure and to achieve privacy among the devices. The model is deployed on two different networks to find out the difference in the transaction cost. The results show that the network using the PoW consensus mechanism obtained better cost than the other network. The results are limited to a specific network and are not applied against other different consensus mechanisms.

8

Limitations and Future Work

Our work still have some limitations regarding the number of blockchains used by the network. We only use single blockchain in the implementation. More then one blockchain, can be used to see the impact on results. The scope of proposed model is small so that the results obtained are also single sided. Also, we have not tested the scalability of the proposed model. We are aiming to continue this work in the future.

References 1. Vanus, J., Belesova, J., Martinek, R., Nedoma, J., Fajkus, M., Bilik, P., Zidek, J.: Monitoring of the daily living activities in smart home care. Hum.-Centric Comput. Inf. Sci. 7(1), 30 (2017) 2. Li, Z., Kang, J., Yu, R., Ye, D., Deng, Q., Zhang, Y.: Consortium blockchain for secure energy trading in industrial internet of things. IEEE Trans. Ind. Inform. 14(8), 3690–3700 (2017) 3. Sharma, P.K., Park, J.H.: Blockchain based hybrid network architecture for the smart city. Future Gener. Comput. Syst. 86, 650–655 (2018) 4. Kim, H., Lee, E.A.: Authentication and authorization for the Internet of Things. IT Prof. 19(5), 27–33 (2017) 5. Nakamoto, S.: Bitcoin: a peer-to-peer electronic cash system (2008). https:// bitcoin.org/bitcoin.pdf. Accessed 5 Sept 2019 6. Xu, Y., Wang, G., Yang, J., Ren, J., Zhang, Y., Zhang, C.: Towards secure network computing services for lightweight clients using blockchain. Wirel. Commun. Mob. Comput. 2018, 12 (2018) 7. Noshad, Z., Javaid, N., Imran, M.: Analyzing and securing data using data science and blockchain in smart networks. MS thesis, COMSATS University Islamabad (CUI), Islamabad (2019) 8. Mateen, A., Javaid, N., Iqbal, S.: Towards energy efficient routing in blockchain based underwater WSNs via recovering the void holes. MS thesis, COMSATS University Islamabad (CUI), Islamabad (2019) 9. Singh, M., Kim, S.: Branch based blockchain technology in intelligent vehicle. Comput. Netw. 145, 219–231 (2018) 10. Khan, R.J.u.H., Javaid, N., Iqbal, S.: Blockchain based node recovery scheme for wireless sensor networks. MS thesis, COMSATS University Islamabad (CUI), Islamabad (2019) 11. Ren, Y., Liu, Y., Ji, S., Sangaiah, A.K., Wang, J.: Incentive mechanism of data storage based on blockchain for wireless sensor networks. Mob. Inf. Syst. (2018, accepted)

534

M. Z. Iftikhar et al.

12. Rehman, M., Javaid, N., Awais, M., Imran, M., Naseer, N.: Cloud based secure service providing for IoTs using blockchain. In: Global Communications Conference, GLOBCOM 2019. IEEE (2019, accepted) 13. Jiang, T., Fang, H., Wang, H.: Blockchain-based internet of vehicles: distributed network architecture and performance analysis. IEEE Internet Things J. (2018, accepted) 14. Sharma, P.K., Moon, S.Y., Park, J.H.: Block-VN: a distributed blockchain based vehicular network architecture in smart City. JIPS 13(1), 184–195 (2017) 15. Omaji, S., Javaid, N., Awais, M., Ahmed, Z., Imran, M., Guizani, M.: A blockchain model for fair data sharing in deregulated smart grids. In: Global Communications Conference. IEEE (2019, accepted) 16. Xu, C., Wang, K., Li, P., Guo, S., Luo, J., Ye, B., Guo, M.: Making big data open in edges: a resource-efficient blockchain-based approach. IEEE Trans. Parallell Distrib. Syst. 30(4), 870–882 (2018) 17. Zhang, G., Li, T., Li, Y., Hui, P., Jin, D.: Blockchain-based data sharing system for AI-powered network operations. J. Commun. Inf. Netw. 3(3), 1–8 (2018) 18. Lin, D., Tang, Y.: Blockchain consensus based user access strategies in D2D networks for data-intensive applications. IEEE Access 6, 72683–72690 (2018) 19. Naz, M., Javaid, N., Iqbal, S.: Research based data rights management using blockchain over ethereum network. MS thesis, COMSATS University Islamabad (CUI), Islamabad (2019) 20. Javaid, A., Javaid, N., Imran, M.: Ensuring analyzing and monetization of data using data science and blockchain in loT devices. MS thesis, COMSATS University Islamabad (CUI), Islamabad (2019) 21. Dai, M., Zhang, S., Wang, H., Jin, S.: A low storage room requirement framework for distributed ledger in blockchain. IEEE Access 6, 22970–22975 (2018) 22. Kazmi, H.S.Z., Javaid, N., Imran, M.: Towards energy efficiency and trustfulness in complex networks using data science techniques and blockchain. MS thesis, COMSATS University Islamabad (CUI), Islamabad (2019) 23. Zahid, M., Javaid, N., Rasheed, M.B.: Balancing electricity demand and supply in smart grids using blockchain. MS thesis, COMSATS University Islamabad (CUI), Islamabad (2019) 24. Reilly, E., Maloney, M., Siegel, M., Falco, G.: A smart city IoT integrity-first communication protocol via an ethereum blockchain light client. In: Proceedings of the International Workshop on Software Engineering Research and Practices for the Internet of Things, SERP4IoT 2019, Marrakech, Morocco, pp. 15–19, April 2019 25. Michelin, R.A., Dorri, A., Steger, M., Lunardi, R.C., Kanhere, S.S., Jurdak, R., Zorzo, A.F.: SpeedyChain: a framework for decoupling data from blockchain for smart cities. In: Proceedings of the 15th EAI International Conference on Mobile and Ubiquitous Systems: Computing, Networking and Services, pp. 145–154. ACM, November 2018 26. Ali, I., Javaid, N., Iqbal, S.: An incentive mechanism for secure service provisioning for lightweight clients based on blockchain. MS thesis, COMSATS University Islamabad (CUI), Islamabad (2019) 27. Zhang, Y., Sun, W., Xie, C.: Blockchain in smart city development—The knowledge governance framework in dynamic alliance. In: International Conference on Smart City and Intelligent Building, pp. 137–152. Springer, Singapore, September 2018 28. Bahga, A., Madisetti, V.K.: Blockchain platform for industrial internet of things. J. Softw. Eng. Appl. 9(10), 533 (2016)

Recommendation System Based on Deep Learning Tianhan Gao1(B) , Lei Jiang2(B) , and Xibao Wang3 1

Engineering Research Center of Security Technology of Complex Network System, Ministry of Education, Liaoning Research Center of Safety Engineering Technology in Industrial Control, Northeastern University, Shenyang, China [email protected] 2 Northeastern University, Shenyang, China [email protected] 3 Dalian University of Technology, Dalian, China [email protected] Abstract. With the exponential growth of digital resource from Internet, search engines and recommendation systems have become the effective way to find relevant information in a short period of time. In recent years, advances in deep learning have received great attention in the fields of speech recognition, image processing, and natural language processing. The recommendation system is an important technology to alleviate information overload. How to integrate deep learning into the recommendation system, use the advantages of deep learning to learn the inherent essential characteristics of users and items from various complex multi-dimensional data, and build a model that more closely matches the user’s interest needs has become a hotpot in the research field. This paper reviews the research and application status of recommendation algorithms based on deep learning, and tries to discusses and forecasts the research trends of deep learning approaches applied to recommendation systems. proceedings.

1

Introduction

Our daily needs, including shopping, books, news articles, songs, movies and research documents have flooded a large number of data warehouses and databases. To this end, intelligent recommendation systems and powerful search engines provide users with very useful help. The popularity and practicability of such systems are attributed to their ability to extract information of interest from large amounts of data. Therefore, Amazon [1], Netflix [2] and other recommendation systems will actively understand the user’s interests and inform users of the items they are interested in. Although these systems differ from each other depending on the application they are using, the core mechanism for finding items of interest to the user is the user’s interest in matching the items. In the recommended literature, these are divided into three main categories: collaborative filtering [3] (using only user project interaction information for recommendations), content-based [4] (using user preferences, project preferences, or c Springer Nature Switzerland AG 2020  L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 535–543, 2020. https://doi.org/10.1007/978-3-030-33506-9_48

536

T. Gao et al.

both) and hybrid recommendation models [5] (using interactive information and user and Project metadata). The models under each category have their own limitations, such as data sparsity, cold start of users and projects. A general view of the implications of incorporating deep learning into the recommendation system [6] is a significant improvement over traditional models. In this review, a systematic review of the various tasks related to the integration of deep learning and recommendation systems is conducted.

2 2.1

Traditional Recommendation System Content-Based Recommendation System

The content-based recommendation algorithm is based on the user’s behavioral historical behaviors, such as evaluation, sharing, etc., and integrates these behaviors to calculate the user’s preferences, and then calculates the similarity between the recommended items and the user’s preferences, and recommends the most similar items to the user. The basic idea of the content-based recommendation algorithm is to calculate the similarity between the items that the user has not purchased and the items selected in the current user’s historical behavior. Firstly, the historical behavior data of the user is obtained, mainly the items that the user interacts with, and secondly, the user’s preferences are extracted from the attribute features of the items that the user interacts with, and expressed as features, the similarity between the user and the items to be recommended can be calculated. Finally, all the items to be recommended are sorted according to the similarity, so that the target users are provided with items similar to their past behavioral preferences. 2.2

Collaborative Filtering Recommendation System

Collaborative filtering recommendation: Compared with the recommendation based on association rules, it is a static recommendation, which is based on the analysis of the user’s existing historical behavior. It is usually divided into three categories, domain-based methods, implicit semantic model based methods and graph-based random walk algorithms. Among these methods, the most famous algorithm that is the most widely used in the industry is a domain-based approach. The domain-based approach mainly includes user-based collaborative filtering and item-based collaborative filtering. 2.2.1 User-Based Collaborative Filtering The user-based collaborative filtering algorithm is the earliest recommendation algorithm. The algorithm first calculates a user set similar to the target user interest, and then recommends the user of the similar user set that the user likes and has not touched. According to this basic principle, user-based collaborative filtering can be divided into two steps: 1. Find a user collection similar to the target user’s interest. 2. Find the user that the user likes in the collection, and

Recommendation System Based on Deep Learning

537

the target user has not heard of the item recommendation to the target user. In order to calculate the user similarity, we first need to convert the data purchased by the user into the index data that the item has been purchased by the user, that is, the inverted index of the item. After the inverted index of the item is established, the similarity between users can be calculated according to the similarity formula: |N (a) ∩ N (b)| (1) Wa,b =  |N (a)| ∗ |N (b)| Where N (a) represents the number of items purchased by the user a, and N (b) represents the number of items purchased by the user b, indicating the number of the same items purchased by the user a and the user b. With the similar data of the user, the K most similar users are selected for the user U, and the U-unsold items are recommended to the user U in the items they have purchased. 2.2.2 Item-Based Collaborative Filtering The item-based collaborative filtering algorithm is currently the most widely used algorithm in the industry, which recommends users to items that are similar to the items they liked before. The Item-CF algorithm does not use the content attribute of the item to calculate the similarity between items. It mainly calculates the similarity between items by analyzing the user’s behavior record. The algorithm believes that item A and item B have a large similarity because users who like item A mostly like item B. There is a hypothesis that each user’s interests are limited to a few aspects, so if two items belong to the same user’s interest list, then the two items may belong to a limited number of areas. And if two items appear in many users’ interest lists at the same time, they may belong to the same field and thus have great similarities. 2.2.3

Differences Between User-Based and Item-Based Collaborative Filtering The User-CF recommends to the user those items that the user has a common interest with, and the Item-CF recommends the item to the user that is similar to the item he liked before. It can be seen from the principle of the two algorithms that User-CF is suitable for occasions with few users. If there are many users, calculating the user similarity matrix is very expensive, and Item-CF is suitable for occasions where the number of items is significantly smaller than the number of users. The item similarity matrix is very costly. So User-CF is generally used in news websites, while Item-CF is used in other non-news websites. 2.3

Hybrid Recommendation System

Since the single recommendation algorithm has its shortcomings, it is considered to combine the different recommendation algorithms for the mixed recommendation. It is found through experiments that this can often improve the recommendation quality. Hybrid recommendation algorithms generally use the advantages

538

T. Gao et al.

of one algorithm to compensate for the shortcomings of another algorithm by combining collaborative filtering and content-based recommendation. The most important thing about the hybrid recommendation algorithm is to use deep learning to extract the features of the project content and the user’s historical behavior, and to calculate the similarity between users or projects through the extracted features for collaborative filtering recommendation. At present, deep learning is used for the recommended hybrid recommendation algorithm, which is already a trend in related research fields.

3 3.1

Deep Learning Approaches Autoencoder

The AutoEncoder (AE) is the simplest artificial neural network model that contains only one hidden layer and is widely used for sample feature extraction [7]. The sample data x of the AE is encoded by the encoder function f to obtain the coding feature y, and x and y satisfy the following formula: y = fθ (x) = s(W x + b)

(2)

where: s is a neural network excitation function, generally using a nonlinear function such as sigmoid Function; θ = {W, b} is a set of parameters. Then pass the following formula: (3) x ˆ = gθ (y) = s(W  y + b ) Converting the coding feature y into a reconstructed representation of the original input x, Eq. (3) is the optimization goal of the AE L = x − x ˆ

2

(4)

By continuously correcting the parameters θ and θ , the average reconstruction error L is minimized, and the obtained y can be considered to retain most of the information of the original sample, the equivalent feature of the sample x. 3.2

Denoising AutoEncoder

When only the information of the original input data is retained, the AE cannot guarantee to learn a valid feature representation. For example, in an extreme case, the AE can easily learn an identity function, so it is necessary to give the AE a certain constraint to learn the input ratio. A better feature representation of the data. The Denoising AutoEncoder (DAE) is based on the AE [8–10]. In order to prevent the over-fitting problem, random noise is added to the input data, and the noise is added to the data to reproduce the input.

Recommendation System Based on Deep Learning

3.3

539

Stacked Denoising Autoencoder

SDAE is a network structure that is cascaded by multiple DAEs. It can process larger data sets and mine deeper features of the samples [11–13]. The SDAE network model is the output of the previous layer as the input to the next layer, and the output of the last layer of DAE is used as the reconstructed data of the original input. SDAE network training uses an unsupervised layer-by-layer greedy training algorithm, which is proposed by Hitton et al. [14–16] based on Deep Belief Net (DBN) and is an effective method for training deep networks. When the SDAE network training is completed, the n/2th layer of the encoding function output is used as a valid feature of the original input sample.

4 4.1

Recommendation System Based on Deep Learning Recommendation System Based on Autoencoder

Literature [17] proposed a collaborative filtering method based on AutoEncoder to solve the problem of scoring prediction. Combining the traditional recommendation method collaborative filtering with the deep learning model AutoEncoder, in the collaborative filtering algorithm, assuming that there are m users and n items, and the user’s scoring matrix R for the item, the task is to guess the user as accurately as possible. The rating value of the graded item. This paper [17] proposes to use AutoEncoder to extract the compressed representation of users and projects in the scoring matrix. As a deep feature of users and projects, the extracted features are used for scoring prediction. Experiments prove that the number of RMSE indicators is better than traditional models such as collaborative filtering. On the other hand, the literature [17] uses an automatic encoder that does not extract the deep features of the user. It can be considered to use a stack-type noise reduction encoder, so that deep feature vectors can be obtained and the recommendation quality can be improved. 4.2

Dynamic Hybrid Recommendation System Based on Denoising Autoencoder

In the real recommendation system, the top-N recommended task has many scenarios. Based on the Denoising Autoencoder, the author proposes a system filtering recommendation model CDAE for completing top-N recommendation tasks based on user preferences [18]. The CDAE model takes the row of the user-item evaluation matrix as input, obtains the hidden representation of the user through a layer of neural network coding, and restores the user’s interaction behavior through a layer of neural network. Unlike the simplest Autorec model, the CDAE model incorporates user-specific considerations when coding for hidden representations, with more semantics. In order to make the model more robust, the CDAE model performs noise processing on the input features, either by dropout or by adding Gaussian noise. The CDAE model can be generalized for a variety of classical collaborative filtering models under certain circumstances, and has strong interpretability.

540

4.3

T. Gao et al.

Improved Hybrid Recommendation System Based on Deep Learning

With the deep learning model in the application of the recommendation system has achieved good results, it proves that the deep neural network has the ability to refine complex interactive features. However, most deep learning models rely on feature extraction. There are often problems with more noise and are limited by feature extraction methods. It is difficult to effectively model the complex relationship between content information and user-item interaction. Based on the contractive auto-encoder (CAE) generalization matrix decomposition, this paper [19] proposes a composite recommendation model AutoSVD++ with high computational efficiency and scalability. The model can effectively use content information to enrich the hidden representation of the item and provide users with high quality item recommendations by using implicit feedback. Some researchers have proposed [20] this paper proposes a Bayesian generation model CVAE for the composite recommendation task in multimedia scenes. The model can be mainly divided into two parts: a content information processing part based on the self-encoder and a Bayesian prediction part based on the probability model. Taking the user-item pair as an example, the variational self-encoder encodes the content information of the item to obtain the probability of the item content information; the item rating information is similar to the probability matrix decomposition, firstly hiding each user and item by a normal distribution. It is indicated that the probability concealment of the item indicates that the probability information hidden representation of the rating information and the content information is comprehensively considered, and finally the implicit feedback is predicted by the item hidden representation and the user hidden representation. Later, some researchers applied the variational self-encoder to the collaborative filtering recommendation task based on implicit feedback, and hoped to overcome the limitations of the linear factor model through the nonlinear probability model. In this paper [21], the generation model VAE-CF based on Variational Autoencoder is proposed, and the regular parameters and probability model selection of the variational self-encoder are adjusted to achieve the best results in the current recommended tasks. The model extracts the K-dimensional hidden factor according to the standard Gaussian distribution, then generates the probability distribution of the user clicking all the items according to the nonlinear function, and finally reconstructs the user click history according to the multivariate distribution. 4.4

MLP-Based YouTube Video Recommendation System

Literature [22] proposed a deep neural network model for YouTube video recommendation by using multi-source heterogeneous data such as user information, context information, historical behavior data and project feature information. YouTube video recommendations face three main challenges: scalability, freshness, and data noise issues. To overcome these three challenges, the study applied

Recommendation System Based on Deep Learning

541

the deep neural network model to two key processes of the video recommendation system: candidate set generation and sorting. The purpose of candidate set generation is to screen hundreds of videos related to users from the massive video library, and mainly use the historical behavior data, user characteristics and context information of users on YouTube to model the user’s personalized preferences for video. The method is to transform the recommendation problem into a classification problem based on deep neural network, and find N videos that are closest to the user vector (the neural network transformed feature vector). The sorting process is to further score each candidate video by using neural network and logistic regression model by further considering more video features, and sorting the video according to the scoring value. 4.5

Session-Based Recommendation System with Graph

Taobao’s platform recommendation mainly faces three problems: large amount of data, sparseness of user data and cold start of items. Literature [23] proposed three models for these problems: BES, GES, EGES. Taobao’s homepage recommendation is based on the user’s past behavior. The problem that the paper focuses on is the matching of the recommendation system, that is, the stage of recalling the candidate products from the product pool. The core task is to calculate the pairwise similarity of all items. The paper proposes to build an item graph based on the user’s behavior history, and then use the state-of-the-art graph embedding method to learn the embedding of each item. This is called Base Graph Embedding (BGE). In this way, the pairwise similarity can be calculated based on the embeddings vector of items. BGE is better than CF, but for items with little or no interaction, it is still difficult to get accurate embedding. In order to alleviate this problem, the paper proposes to use side information to enhance the embedding process, and proposes Graph Embedding with Side information (GES).

5

Conclusion

Future research directions focus on the combination of deep learning and traditional recommendation methods, with emphasis on the extension of the selfencoder and its combination with collaborative filtering and implicit semantic models. At the same time, we should study more and improve the traditional recommendation methods, such as the Semantics Enhanced Latent Factor Model (SELFM), and the combination of these methods with the self-encoder and other deep learning methods. There are still many papers in this area that propose a new deep learning recommendation system model, but there are still many areas for improvement in the existing models, and further optimization is needed. But the problem to be solved is data sparseness, cold start issues for new users and new projects. Sparseness and cold-start problems are problems that have long plagued the recommendation system, including classic collaborative filtering: algorithms and emerging network-based recommendation algorithms have this

542

T. Gao et al.

problem. Many researchers have studied this problem and proposed solutions. But the problem still exists and needs to be studied. In addition, the research on the performance evaluation index of the recommended system, the sensitivity of the user to the accuracy of the algorithm, and the generalized quality evaluation method of the algorithm to different fields are all future research goals. Acknowledgement. This paper is supported by China Fundamental Research Funds for the Central Universities under Grant No. N180716019 and Grant No. N182808003.

References 1. Linden, G., Smith, B., York, J.: Amazon.com recommendations item-to-item collaborative filtering. IEEE Internet Comput. 1, 76–80 (2003) 2. Koren, Y., Bell, R., Volinsky, C.: Matrix factorization techniques for recommeder system. Computer 8, 30–37 (2009) 3. Kadie, C., Breese, J.S., Heckerman, D.: Empirical analysis of predictive algorithms for collaborative filtering, pp. 43–52 (1998) 4. Roy, L., Mooney R.J.: Content-based book recommending using learning for text categorization, pp. 195–204 (2000) 5. Shohom, Y., Balabanovic, M.: Content-based, collaborative recommendation. Commun. ACM 40(3), 66–72 (1997) 6. Huang, L., Jiang, B., Lu, S., Liu, Y., Li, D.: Dynamic hybrid recommendation algorithm based on stack noise reduction encoder. Chin. J. Comput. 41(7), 1619– 1647 (2018) 7. Bengio, Y.: Learning deep architectures for AI. Found. Trends Mach. Learn. 2(1), 1–127 (2009) 8. Wang, H., Shi, X., Yeung, Y.: Relational stacked denoising autoencoder for tag recommendation. In: Proceedings of the Twenty-Ninth AAAI Conference on Artificial Intelligence, pp. 3052–3058. AAAI Press, Menlo Park (2015) 9. Hinton, G., Osindero, S., Teh, Y.: A fast learning algorithm for deep belief nets. Neural Comput. 18(7), 1527–1554 (2006) 10. Vincent, P., Larochelle, H., Lajoie, I.: Stacked denoising autoencoders: learning useful representations in a deep network with a local denoising criterion. J. Mach. Learn. Res. 11(12), 3371–3408 (2010) 11. Van den Oord, A., Dieleman, S., Schrauwen, B.: Deep content-based music recommendation. In: Proceedings of the Advances in Neural Information Processing Systems, Lake Tahoe, USA, pp. 2643–2651 (2013) 12. Krizhevsky, A., Sutskever, I., Hinton, G.E.: ImageNet classification with deep convolutional neural networks. In: Proceedings of the Advances in Neural Information Processing Systems, Lake Tahoe, USA, pp. 1097–1105 (2012) 13. Zhang, F., Yuan, N.J., Lian, D.: Collaborative knowledge base embedding for recommender systems. In: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, San Francisco, USA, pp. 353– 362 (2016) 14. Williams, D., Hinton, G.: Learning representations by back-propagating errors. Nature 323(6088), 533–538 (1986) 15. Geng, X., Zhang, H., Bian, J.: Learning image and user feature recommendation in social networks. In: Proceedings of the IEEE International Conference on Computer Vision, Santiago, Chile, pp. 4274–4282 (2015)

Recommendation System Based on Deep Learning

543

16. Wu, C.Y., Ahmed, A., Beutel, A.: Recurrent recommender networks. In: Proceedings of the 10th ACM International Conference on Web Search and Data Mining, Cambridge, UK, pp. 495–503 (2017) 17. Sedhain, S., Menon, A.K., Sanner, S., Xie, L.: AutoRec: autoencoders meet collaborative filtering. In: WWW 2015 Companion Proceedings of the 24th International Conference on World Wide Web, pp. 111–112 (2015) 18. Wu, Y., DuBois, C., Zheng, A.X., Ester, M.: Collaborative denoising auto-encoders for top-N recommender systems. In: WSDM 2016 Proceedings of the Ninth ACM International Conference on Web Search and Data Mining (2016) 19. Zhang, S., Yao, L., Xu, X.: AutoSVD++: an efficient hybrid collaborative filtering model via contractive auto-encoders. In: SIGIR 2017 Proceedings of the 40th International ACM SIGIR Conference on Research and Development in Information (2017) 20. Liang, D., Krishnan, R.G., Hoffman, M.D., Jebara, T.: Variational autoencoders for collaborative filtering. In: WWW 2018 Proceedings of the 2018 World Wide Web Conference (2018) 21. Li, X., She, J.: Collaborative variational autoencoder for recommender systems. In: KDD 2017 Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (2007) 22. Covington, P., Adams, J., Sargin, E.: Deep neural networks for Youtube recommendations. In: RecSys 2016-Proceedings of the 10th ACM Conference on Recommender Systems, pp. 191–198 (2016) 23. Wang, J., Zhao, H., Zhang, Z., Zhao, B., Lee, D.L., Huang, P.: Billion-scale commodity embedding for e-commerce recommendation in Alibaba. In: Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pp. 839-848 (2018)

Routing Method Based on Data Transfer Path in DTN Environments Kazuma Ikenoue(B) and Kazunori Ueda Kochi University of Technology, 185 Miyanokuchi, Tosayamada Town, Kami-shi, Kochi 782-8502, Japan [email protected], [email protected]

Abstract. In recent years, temporary network construction using Delay Tolerant Networking (DTN) has attracted attention as a preparation for communication infrastructure failures that may occur due to disasters. DTN is a technology that allows data packets to be delivered late due to temporarily failure of networking. We have proposed an aggregated message ferry method to improve performance of data transfer with the message ferry method which is the routing method of DTN. In our proposed method, each data are aggregated in advance to a node with high probability to communicate with the ferry node. However, our proposed method has a problem that data cannot be transferred to a destination node if the number of hops on a route is large when data are aggregated. In this paper, we propose a new message ferry method enables each node to establish a route in order to increase the number of data arrivals by shortening paths. In the proposed method, each node attempt to transfer data to a node that is included in the communication route to a ferry node. To show the effectiveness of our proposed method, we compare our proposed method with a conventional method in terms of hop count and data arrivals. Simulation results showed that our proposed method achieved higher data arrival number and hop count reduction than the conventional method.

1

Introduction

In recent years, communication infrastructure and the like have been developed due to improvements in information and communication technology. However, there is a risk that an area where communication cannot be performed due to a disaster or a failure that causes noise or delay in communication may occur. In such a situation, it is necessary to continue communications that are not affected by network failures in order for the disaster response headquarters to grasp the information in the disaster area. Therefore, the construction of alternative networks using a technology called Delay Tolerant Networking (DTN), which is a network independent of communication infrastructure, has been studied [1]. There are several DTN route selection methods [2–4]. Among them, there is a message ferry method that is expected to secure a transfer route to the destination node [5]. In this method, a node called a ferry node circulates and collects data from each node and transports it to the destination node. c Springer Nature Switzerland AG 2020  L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 544–552, 2020. https://doi.org/10.1007/978-3-030-33506-9_49

Routing Method Based on Data Transfer Path in DTN Environments

545

However, with the message ferry method, there is a problem that it is impossible to acquire data of nodes that do not exist near the ferry node. Therefore, in order to solve this problem, aggregated message ferry method has been proposed as a prior study. In this method, data is pre-aggregated to nodes that have a high probability of communicating with the ferry node. However, this increases the number of hops and makes it impossible to transfer data to the destination node. In this study, we propose a message ferry method that considers transfer routes in order to reduce the number of hops and improve the number of data arrivals by changing the communication route determination algorithm for data aggregation. In addition, the usefulness of the proposed method is confirmed by comparing the number of hops and the number of data arrivals with the aggregated message ferry method in the previous study.

2

Related

In this chapter, we explain the aggregated message ferry method that is the basis of the proposed method. This method is proposed based on the message ferry method, which is a kind of DTN routing method. DTN is a network that is resistant to failures due to delays that occur between nodes during communication. Each node moves to a range where it can communicate with other nodes when there is no data transfer destination nearby. Therefore, data may be transferred even when there is no transfer path to the destination node when the data is generated. In the message ferry method, when each node is transfers data to the destination node, it uses a node called a ferry node to transfer the data to the destination instead of relaying nearby nodes. The operation of the ferry node is shown in Fig. 1. The ferry node circulates between clusters that are a collection of nodes, collects and accumulates data from each node in the cluster, and carries the accumulated data to the destination node. Since the ferry node is likely to reach the destination node, this method can construct a route with stable data transfer. However, the message ferry method has a problem that data of a node outside the communicable range of the ferry node cannot be acquired when the ferry node is reaches the cluster. In order to solve this problem, a aggregated message ferry method has been proposed. The aggregated message ferry method is a method in which each node aggregates data to a small number of nodes in advance in order to receive data held by nodes outside the communication range when the ferry node is arrives. Figure 2 shows the communication flow. Each node aggregates data in advance to the node with a high ferry score that indicates the possibility of communication when the ferry node is reaches the cluster. The ferry score increases the score of each node when it is communicates with the ferry node. In addition, the aggregation node transfers the accumulated data when node reaches the ferry node Therefore, even if there is a node outside the communication range when the

546

K. Ikenoue and K. Ueda

Transfer

The ferry node goes around each cluster

Ferry Node Wireless Node Cluster Fig. 1. Ferry node operation

Ferry Score 2

Transfer

Ferry Node

Transfer

Ferry Score 1 Transfer

Transfer data to a node with a high ferry score.

Ferry Score 0 Fig. 2. Overview of the aggregated message ferry method

ferry node is arrives, the data is likely to be transferred to the destination node because it is transferred by the aggregation node. However, if high score node is moves, there is no suitable node for the transfer destination, and the nearby nodes cannot transfer data. In addition, the com-

Routing Method Based on Data Transfer Path in DTN Environments

547

munication path longer and the number of times of communication increases, which may increase the amount of battery consumed by each node.

3

Message Ferry Method Considering Forwarding Routes

This chapter describes the proposal method. In the proposal method, the ferry score addition method was changed in order to easily construct an alternative route that can transfer data to a ferry node even if a high score node was moves. The ferry score is the value calculated from the transfer route when data is transferred to the ferry node is added to each node existing in the transfer route. In addition, in order to reduce the number of trials for determining the communication destination, each node creates a list that lists the ferry scores of adjacent nodes, and refers to the list when performing of communication. 3.1

Assumed Environment and Required System

In this paper, based on the assumed environment of previous research, we assumed a situation where a communication infrastructure failure occurred after a major disaster. In the event of a disaster, it is necessary to collect information on evacuees and carry out prompt rescue activities. Therefore, in this study, we define the shelter where the evacuees gather in the cluster, which is a set of nodes, and the disaster response headquarters that requests rescue activities based on the information of the disaster area. The nodes are assumed to be smartphones of disaster victims, and the ferry nodes use portable communication devices. 3.2

Proposed Method

In the proposed method, the ferry score addition method set in the conventional method is changed in order to shorten the communication route by concentrating the data transfer route to the route that is likely to reach the ferry node. In addition, in order to reduce the number of trials when determining the communication destination, a list describing the ferry scores of the nodes adjacent to each node is created to determine the communication destination. 3.2.1 Ferry Score Calculation Method In the proposed method, the score addition method is changed so that the ferry score is a value that represents the possibility that each node is included in the data transfer route to the ferry node, taking into account the data transfer history. In the aggregated message ferry method, the score assigned to each node is addition only to the node that sent the data to the ferry node. In the proposed method, when a node is transfers data to a ferry node, all nodes on the data transfer path are targeted. The value to be added to the nodes existing in the transfer route is set so that the value added for each node close to the ferry node on the transfer route when the data arrives increases. By adding the scores to all

548

K. Ikenoue and K. Ueda

the nodes existing on the route, it is considered easier to construct an alternative route that can transfer data to the ferry node even if the node moves. Describes how the ferry score is calculated. First, each data creates a transfer route to reach the ferry node as shown in Eq. 1.

(1)

The data generation node is i1 , and the node that transferred the data to the ferry node is iN . The value to be added to each node is determined by the position in the route so that the ferry score is a value that represents the possibility of being included in the route that can transfer data to the ferry node. That is, iN is added with N which is the number of hops, and other nodes in are added with n which is the order in its route. Figure 3 shows the flow for determining the added value for each node. Data is generated at node A and transferred to node C through node B. In this case, the maximum number of hops 3 is added to the score of node C that transferred the data to the ferry node. Then, add 2 to node B, which is the node shown in the transfer path next to node C, and finally add 1 to node A as well.

Data arrives from node A to node C via node B. Transfer to the ferry node. The number of hops for this communication is 3.

Node A Ferry Score 1 0

Node B Ferry Score 0 2

Node C Ferry Score 0

Ferry Node

Fig. 3. How to determine the value to add to the ferry score

3.2.2 Node Communication Destination Determination Method In the proposed method, to reduce the number of trials when a node is determines the communication destination and to easily determine the communication

Routing Method Based on Data Transfer Path in DTN Environments

549

destination, a list of ferry scores of adjacent nodes that can communicate with the node is created. In the list, nodes that can communicate with each node are listed in order, and nodes that cannot communicate are deleted. This list is updated when communication with the node is becomes possible or impossible. Figure 4 shows communication using lists. When the node has accumulated data, the nodes in the list are sorted in descending order using the ferry score. Then, transfer the data to the node at the top of the list. After communicating, return to the process of adding a communicable node. The data is sent to the node below it if data cannot be sent to the node at the top of the list. In previous studies, when data is transferring, the ferry scores of neighboring nodes were confirmed one by one, but using a list makes it possible to find a node with the maximum ferry score with a small number of trials.

Transfer data to node C, which is the highest in the list.

Source Node List Node C : 8 Node D : 4 Node A : 2 Node B : 1

Node A Ferry Score 2 Node B Ferry Score 1 Node C Ferry Score 8

Ferry Node

Node D Ferry Score 4

Fig. 4. Communication flow using lists

4

Evaluation of Hop Count and Data Arrival Count by Reducing the Number of Communication Paths

This chapter describes the simulation for evaluating the proposal method. In the proposed method, the method for adding the ferry score and the process for determining the communication destination are changed, so it is expected to reduce the number of communication and improve the data arrival number when each data is transferring to the ferry node. Therefore, the average number of hops and the number of data arrivals are compared with the aggregate message ferry method to confirm the usefulness of the proposed method. The average hop

550

K. Ikenoue and K. Ueda

count is the average number of communications when data is transferred to the ferry node, and the data arrival count is the total number of data that reaches the ferry node. 4.1

Purpose of Evaluation

In the simulation, the number of hops and the number of data arrivals are compared with the aggregated message ferry method, which was the previous research. In addition, the average number of hops is compared to confirm that the communication path is shortened. The average number of hops is the total number of hops divided by the number of data arrivals. 4.2

Scenario and Simulation Environment

Each parameter of this simulation was determined based on the experiment conducted in the conventional method because the assumed environment of the proposed method is equal to that of the previous research. In this evaluation experiment, the simulation time is 10500 s, and two patterns with different simulation area sizes are used to confirm that the proposed method is effective regardless of the area size. Two patterns with different areas are simulated, and the effectiveness of the proposed method is confirmed when the density of wireless nodes in the area is different. The scenario 1 area is 200 m2 , and the scenario 2 area is 300 m2 . The nodes of evacuees who generate and relay data are assumed to be those who move within the shelters in the event of a disaster and those who remain in the place due to the effects of injury. Therefore, 30 nodes move in a random direction, and 30 nodes to stay on the spot without moving. Also, the data to be transferred is generated at random nodes every 60 s, and there are sufficient buffers for evacuees and ferry node. The ferry node does not exist at the start of the simulation, and stays in the area after 1500 s, which is the cyclic period. The dwell time is 600 s, during which data is collected from the nodes. After 600 s, move out of the time area for the cyclic period again and repeat the standby. The communication speed and communication radius of the ferry node are the same as those of the evacuee node. In this simulation, the proposed method was implemented on a simulator called The ONE (The Opportunistic Network Environment simulator) [6]. 4.3

Comparison Result

A simulation was performed to compare the conventional method with the proposed method. The comparison results of the number of hops and the number of data arrivals and the average number of hops are shown below. The number of hops and the number of data arrivals in each scenario are shown in Table 1.

Routing Method Based on Data Transfer Path in DTN Environments

551

Table 1. Result of hop count and data arrival count 200 m2 300 m2 Conventional Proposal Conventional Proposal Number of data generation 175 Number of data arrivals

172

175

145

175

Hop count

359

345

519

426

It was confirmed that the number of hops in the proposed method was lower and the number of data arrivals was improved in both scenarios compared to previous studies. In Scenario 1, there is no significant difference in the number of hops from the previous survey, but in Scenario 2, the number of hops and the number of data arrivals are decreasing. Table 2 shows the average number of hops in each scenario. Table 2. Average hop count results 200 m2 300 m2 Conventional Proposal Conventional Proposal Average number of hops 2.1

2.0

3.6

2.4

It was confirmed that the average hop count of the proposed method was short in both scenarios, and the communication path for transferring data to the ferry node was short. Therefore, using the proposed method, it is possible to suppress the increase in the number of routes, reduce the number of hops, and improve the data arrival rate in an environment where the communication route tends to be long. 4.4

Consideration

In this study, we changed the algorithm used when aggregating data to a small number of nodes in order to shorten the communication path so that the data can be transferred to the destination. However, aggregation nodes accumulate a large amount of data, which can lead to faster battery consumption and shorter operating time. Therefore, in the future, it will be necessary to verify the proposed method when it is closer to the actual environment by incorporating factors such as the ferry node movement algorithm and the battery of each node that were not considered this time.

5

Conclusion

In this paper, we focused on DTN that is tolerant to situations where delays occur frequently even in ad-hoc networks that do not depend on communication

552

K. Ikenoue and K. Ueda

infrastructure. And we described the message ferry method that is the routing method. In the message ferry method, a node that is called gferry node h is responsible for carrying data, however cannot acquire data from nodes that could not communicate with the ferry node. Therefore, the aggregated message ferry method had been proposed to aggregate data of each node to a small number of nodes in advance. However, the aggregated message ferry method increases the number of hops and causes the case where it is impossible to transfer data to the destination node because the communication path for data aggregation becomes longer. In this paper, we proposed a new message ferry method enables each node to transfer data on a shorter route in order to improve the number of data arrivals by changing scoring algorithm based on past routes that were used for data transfer. Then, in order to verify effectiveness of our proposed method, we compared our proposed method with conventional method by measuring the number of hop count and the number of data arrivals at the destination node. Simulation results showed that our proposed method achieved shorter path than conventional method and reduction of the number of hop count, furthermore, the number of data arrivals was also improved. In the future, in order to reveal the usefulness of our proposed method on an actual environment such as serious disasters, it will be necessary to examine the evaluation when considering the node battery and network communication time.

References 1. Fall, K.: A delay-tolerant network architecture for challenged internets. In: Proceedings of the SIGCOMM 2003, pp. 27–34 (2003) 2. Grossglauser, M., Tse, D.: Mobility increases the capacity of ad hoc wireless networks. IEEE/ACM Trans. Network. 10, 477–486 (2002) 3. Spyropoulos, T., Psounius, K., Raghavendra, C.: Spray and wait an efficient routing scheme for intermittently connected mobile networks. In: Proceedings of the WDN 2005, pp. 252–259 (2005) 4. Vahdat, A., Becker, D.: Epidemic routing for partially-connected ad hoc networks. Technical report, CS-200006 (2000) 5. Zhao, W., Ammar, M., Zegura, E.: A message ferrying approach for data delivery in sparse mobile ad hoc networks. In: Proceedings of the MobiHoc 2004, pp. 187–198 (2004) 6. Keranen, A., Ott, J., Karkkainen, T.: The ONE simulator for DTN protocol evaluation. In: Proceedings of the SIMUTools 2009 (2018)

The 12th International Workshop on Next Generation of Wireless and Mobile Networks (NGWMN-2019)

A Hybrid Intelligent Simulation System for Node Placement in WMNs Considering Load Balancing: A Comparison Study for Exponential and Normal Distribution of Mesh Clients Seiji Ohara1(B) , Heidi Durresi4 , Admir Barolli2 , Shinji Sakamoto3 , and Leonard Barolli4 1

2

Graduate School of Engineering, Fukuoka Institute of Technology, 3-30-1 Wajiro-Higashi, Higashi-Ku, Fukuoka 811-0295, Japan [email protected] Department of Information Technology, Aleksander Moisiu University of Durres, L.1, Rruga e Currilave, Durres, Albania [email protected] 3 Department of Computer and Information Science, Seikei University, 3-3-1 Kichijoji-Kitamachi, Musashino-shi, Tokyo 180-8633, Japan [email protected] 4 Department of Information and Communication Engineering, Fukuoka Institute of Technology, 3-30-1 Wajiro-Higashi, Higashi-Ku, Fukuoka 811-0295, Japan [email protected], [email protected]

Abstract. Wireless Mesh Networks (WMNs) are becoming an important networking infrastructure because it has many advantages such as low cost and increased high-speed wireless Internet connectivity. In our previous work, we implemented a Particle Swarm Optimization (PSO) based simulation system, called WMN-PSO, and a simulation system based on Genetic Algorithm (GA), called WMN-GA, for solving node placement problem in WMNs. Then, we implemented a hybrid simulation system based on PSO and distributed GA (DGA), called WMNPSODGA. Moreover, we added in the fitness function a new parameter for the load balancing of the mesh routers called NCMCpR (Number of Covered Mesh Clients per Router). In this paper, we consider Exponential and Normal distributions of mesh clients and carry out a comparison study. The simulation results show that the performance of the Exponential and Normal distribution was improved by considering load balancing when using WMN-PSODGA. Moreover, for the same number of mesh clients, the Normal distribution has better behavior than the Exponential distribution, because all mesh clients are covered by a smaller number of mesh routers.

c Springer Nature Switzerland AG 2020  L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 555–569, 2020. https://doi.org/10.1007/978-3-030-33506-9_50

556

1

S. Ohara et al.

Introduction

The wireless networks and devices are becoming increasingly popular and they provide users access to information and communication anytime and anywhere [3,8–11,14,20,26,27,29,33]. Wireless Mesh Networks (WMNs) are gaining a lot of attention because of its low-cost nature that makes it attractive for providing wireless Internet connectivity. A WMN is dynamically self-organized and self-configured, with the nodes in the network automatically establishing and maintaining mesh connectivity among itself (creating, in effect, an ad hoc network). This feature brings many advantages to WMN such as low up-front cost, easy network maintenance, robustness and reliable service coverage [1]. Moreover, such infrastructure can be used to deploy community networks, metropolitan area networks, municipal and corporative networks, and to support applications for urban areas, medical, transport and surveillance systems. Mesh node placement in WMNs can be seen as a family of problems, which is shown (through graph theoretic approaches or placement problems, e.g. [6,15]) to be computationally hard to solve for most of the formulations [37]. In fact, the node placement problem considered here is even more challenging due to two additional characteristics: (a) locations of mesh router nodes are not pre-determined, in other wards, any available position in the considered area can be used for deploying the mesh routers. (b) routers are assumed to have their own radio coverage area. We consider the version of the mesh router nodes placement problem in which we are given a grid area where to deploy a number of mesh router nodes and a number of mesh client nodes of fixed positions (of an arbitrary distribution) in the grid area. The objective is to find a location assignment for the mesh routers to the cells of the grid area that maximizes the network connectivity and client coverage. Node placement problems are known to be computationally hard to solve [12, 13,38]. In some previous works, intelligent algorithms have been recently investigated [4,7,16,18,21–23,31,32]. In [24], we implemented a Particle Swarm Optimization (PSO) based simulation system, called WMN-PSO. Also, we implemented another simulation system based on Genetic Algorithm (GA), called WMN-GA [19], for solving node placement problem in WMNs. Then, we designed and implemented a hybrid simulation system based on PSO and distributed GA (DGA). We call this system WMN-PSODGA. We considered also the load balancing problem. We added in the fitness function a new parameter called NCMCpR (Number of Covered Mesh Clients per Router). In this paper, we present the performance analysis of WMNs by WMNPSODGA system considering Exponential and Normal distribution of mesh clients. The rest of the paper is organized as follows. We present our designed and implemented hybrid simulation system in Sect. 2. We present WMN-PSODGA

A Hybrid Intelligent Simulation System for Node Placement in WMNs

557

Web GUI tool and pseudo code in Sect. 3. The simulation results are given in Sect. 4. Finally, we give conclusions and future work in Sect. 5.

2 2.1

Proposed and Implemented Simulation System: Algorithms and Methods Particle Swarm Optimization

In PSO a number of simple entities (the particles) are placed in the search space of some problem or function and each evaluates the objective function at its current location. The objective function is often minimized and the exploration of the search space is not through evolution [17]. Each particle then determines its movement through the search space by combining some aspect of the history of its own current and best (best-fitness) locations with those of one or more members of the swarm, with some random perturbations. The next iteration takes place after all particles have been moved. Eventually the swarm as a whole, like a flock of birds collectively foraging for food, is likely to move close to an optimum of the fitness function. Each individual in the particle swarm is composed of three D-dimensional vectors, where D is the dimensionality of the search space. These are the current position xi , the previous best position pi and the velocity vi . The particle swarm is more than just a collection of particles. A particle by itself has almost no power to solve any problem; progress occurs only when the particles interact. Problem solving is a population-wide phenomenon, emerging from the individual behaviors of the particles through their interactions. In any case, populations are organized according to some sort of communication structure or topology, often thought of as a social network. The topology typically consists of bidirectional edges connecting pairs of particles, so that if j is in i’s neighborhood, i is also in j’s. Each particle communicates with some other particles and is affected by the best point found by any member of its topological neighborhood. This is just the vector pi for that best neighbor, which we will denote with pg . The potential kinds of population “social networks” are hugely varied, but in practice certain types have been used more frequently. We show the pseudo code of PSO in Algorithm 1. In the PSO process, the velocity of each particle is iteratively adjusted so that the particle stochastically oscillates around pi and pg locations. 2.2

Distributed Genetic Algorithm

Distributed Genetic Algorithm (DGA) has been used in various fields of science. DGA has shown their usefulness for the resolution of many computationally hard combinatorial optimization problems. We show the pseudo code of DGA in Algorithm 2.

558

S. Ohara et al.

Algorithm 1. Pseudo code of PSO. /* Initialize all parameters for PSO */ Computation maxtime:= T pmax , t := 0; Number of particle-patterns:= m, 2 ≤ m ∈ N 1 ; Particle-patterns initial solution:= P 0i ; Particle-patterns initial position:= x0ij ; Particles initial velocity:= v 0ij ; PSO parameter:= ω, 0 < ω ∈ R1 ; PSO parameter:= C1 , 0 < C1 ∈ R1 ; PSO parameter:= C2 , 0 < C2 ∈ R1 ; /* Start PSO */ Evaluate(G0 , P 0 ); while t < T pmax do /* Update velocities and positions */ = ω · v tij v t+1 ij +C1 · rand() · (best(Pijt ) − xtij ) +C2 · rand() · (best(Gt ) − xtij ); t+1 xij = xtij + v t+1 ij ; /* if fitness value is increased, a new solution will be accepted. */ Update Solutions(Gt , P t ); t = t + 1; end while Update Solutions(Gt , P t ); return Best found pattern of particles as solution;

Population of individuals: Unlike local search techniques that construct a path in the solution space jumping from one solution to another one through local perturbations, DGA use a population of individuals giving thus the search a larger scope and chances to find better solutions. This feature is also known as “exploration” process in difference to “exploitation” process of local search methods. Fitness: The determination of an appropriate fitness function, together with the chromosome encoding are crucial to the performance of DGA. Ideally we would construct objective functions with “certain regularities”, i.e. objective functions that verify that for any two individuals which are close in the search space, their respective values in the objective functions are similar. Selection: The selection of individuals to be crossed is another important aspect in DGA as it impacts on the convergence of the algorithm. Several selection schemes have been proposed in the literature for selection operators trying to cope with premature convergence of DGA. There are many selection methods in GA. In our system, we implement 2 selection methods: Random method and Roulette wheel method. Crossover operators: Use of crossover operators is one of the most important characteristics. Crossover operator is the means of DGA to transmit best genetic features of parents to offsprings during generations of the evolution process. Many methods for crossover operators have been proposed such as Blend

A Hybrid Intelligent Simulation System for Node Placement in WMNs

559

Crossover (BLX-α), Unimodal Normal Distribution Crossover (UNDX), Simplex Crossover (SPX). Mutation operators: These operators intend to improve the individuals of a population by small local perturbations. They aim to provide a component of randomness in the neighborhood of the individuals of the population. In our system, we implemented two mutation methods: uniformly random mutation and boundary mutation. Escaping from local optima: GA itself has the ability to avoid falling prematurely into local optima and can eventually escape from them during the search process. DGA has one more mechanism to escape from local optima by considering some islands. Each island computes GA for optimizing and they migrate its gene to provide the ability to avoid from local optima (See Fig. 1). Convergence: The convergence of the algorithm is the mechanism of DGA to reach to good solutions. A premature convergence of the algorithm would cause that all individuals of the population be similar in their genetic features and thus the search would result ineffective and the algorithm getting stuck into local optima. Maintaining the diversity of the population is therefore very important to this family of evolutionary algorithms. Algorithm 2. Pseudo code of DGA. /* Initialize all parameters for DGA */ Computation maxtime:= T gmax , t := 0; Number of islands:= n, 1 ≤ n ∈ N 1 ; initial solution:= P 0i ; /* Start DGA */ Evaluate(G0 , P 0 ); while t < T gmax do for all islands do Selection(); Crossover(); Mutation(); end for t = t + 1; end while Update Solutions(Gt , P t ); return Best found pattern of particles as solution;

Fig. 1. Model of Migration in DGA.

560

S. Ohara et al.

Fig. 2. Model of WMN-PSODGA migration.

2.3

WMN-PSODGA Hybrid Simulation System

In this subsection, we present the initialization, particle-pattern, fitness function and client distributions. Also, our implemented simulation system uses Migration function as shown in Fig. 2. The Migration function swaps solutions among lands included in PSO part. Initialization We decide the velocity of particles by a random process considering the area size. For√instance, when √ the area size is W × H, the velocity is decided randomly from − W 2 + H 2 to W 2 + H 2 . Particle-Pattern A particle is a mesh router. A fitness value of a particle-pattern is computed by combination of mesh routers and mesh clients positions. In other words, each particle-pattern is a solution as shown is Fig. 3.

Fig. 3. Relationship among global solution, particle-patterns and mesh routers in PSO part.

Gene Coding A gene describes a WMN. Each individual has its own combination of mesh nodes. In other words, each individual has a fitness value. Therefore, the combination of mesh nodes is a solution.

A Hybrid Intelligent Simulation System for Node Placement in WMNs

561

Fitness Function WMN-PSODGA has the fitness function to evaluate the temporary solution of the router’s placements. The fitness function is defined as: F itness = α × SGC(xij , y ij ) + β × N CM C(xij , y ij ) + γ × N CM CpR(xij , y ij ). This function uses the following indicators. • SGC (Size of Giant Component) The SGC is the maximum number of the routers constructing in the same network. The SGC indicator means the connectivity of the routers. • NCMC (Number of Covered Mesh Clients) The NCMC is the number of the clients belong to the network constructed by the SGC’s routers. The NCMC indicator means the covering rate of the clients. • NCMCpR (Number of Covered Mesh Clients per Router) The NCMCpR is the number of clients covered by each router. The NCMCpR indicator means load balancing. WMN-PSODGA aims to maximize the value of the fitness function in order to optimize the placements of the routers using the above three indicators. The fitness function has weight-coefficients α, β, and γ for SGC, NCMC, and NCMCpR. Moreover, the weight-coefficients are implemented as α + β + γ = 1. Router Replacement Methods A mesh router has x, y positions and velocity. Mesh routers are moved based on velocities. There are many router replacement methods, such as: Constriction Method (CM) CM is a method which PSO parameters are set to a week stable region (ω = 0.729, C1 = C2 = 1.4955) based on analysis of PSO by Clerc et al. [2,5,35]. Random Inertia Weight Method (RIWM) In RIWM, the ω parameter is changing randomly from 0.5 to 1.0. The C1 and C2 are kept 2.0. The ω can be estimated by the week stable region. The average of ω is 0.75 [28,35]. Linearly Decreasing Inertia Weight Method (LDIWM) In LDIWM, C1 and C2 are set to 2.0, constantly. On the other hand, the ω parameter is changed linearly from unstable region (ω = 0.9) to stable region (ω = 0.4) with increasing of iterations of computations [35,36]. Linearly Decreasing Vmax Method (LDVM) In LDVM, PSO parameters are set to unstable region (ω = 0.9, C1 = C2 = 2.0). A value of Vmax which is maximum velocity of particles is considered. With increasing of iteration of computations, the Vmax is kept decreasing linearly [30,34]. Rational Decrement of Vmax Method (RDVM) In RDVM, PSO parameters are set to unstable region (ω = 0.9, C1 = C2 = 2.0). The Vmax is kept decreasing with the increasing of iterations as  T −x . Vmax (x) = W 2 + H 2 × x

562

S. Ohara et al.

Where, W and H are the width and the height of the considered area, respectively. Also, T and x are the total number of iterations and a current number of iteration, respectively [25].

3

WMN-PSODGA Web GUI Tool and Pseudo Code

The Web application follows a standard Client-Server architecture and is implemented using LAMP (Linux + Apache + MySQL + PHP) technology (see Fig. 4). Remote users (clients) submit their requests by completing first the parameter setting. The parameter values to be provided by the user are classified into three groups, as follows. • Parameters related to the problem instance: These include parameter values that determine a problem instance to be solved and consist of number of router nodes, number of mesh client nodes, client mesh distribution, radio coverage interval and size of the deployment area. • Parameters of the resolution method: Each method has its own parameters. • Execution parameters: These parameters are used for stopping condition of the resolution methods and include number of iterations and number of independent runs. The former is provided as a total number of iterations and depending on the method is also divided per phase (e.g., number of iterations in a exploration). The later is used to run the same configuration for the same problem instance and parameter configuration a certain number of times. We show WMN-PSODGA Web GUI tool in Fig. 5. The pseudo code of our implemented system is shown in Algorithm 3.

Fig. 4. System structure for web interface.

A Hybrid Intelligent Simulation System for Node Placement in WMNs

563

Fig. 5. WMN-PSODGA Web GUI Tool.

Algorithm 3. Pseudo code of WMN-PSODGA system. Computation maxtime:= Tmax , t := 0; Initial solutions: P . Initial global solutions: G. /* Start PSODGA */ while t < Tmax do Subprocess(PSO); Subprocess(DGA); WaitSubprocesses(); Evaluate(Gt , P t ) /* Migration() swaps solutions (see Fig. 2). */ Migration(); t = t + 1; end while Update Solutions(Gt , P t ); return Best found pattern of particles as solution;

4

Simulation Results

In this section, we present the simulation results for Exponential and Normal distribution of mesh clients. Table 1 shows the common parameters for simulations. Besides, the number of mesh routers is 16 for the Exponential distribution. On the other hand, the number of mesh routers is 8 for the Normal distribution because 16 is too many for the Normal distribution. Moreover, there are two cases for each distribution to evaluate the effect of load balancing. In the first case are shown the simulation results of WMNPSODGA system when the weight-coefficients are α = 0.3, β = 0.7, γ = 0.

564

S. Ohara et al. Table 1. The common parameters of the simulations. Parameters

Values

Number of mesh clients 48 Radius of a mesh router 2.0–3.5 Number of GA Islands

16

Number of migrations

200

Evolution steps

9

Selection method

Roulette wheel method

Crossover method

SPX

Mutation method

Uniform mutation

Crossover rate

0.8

Mutation rate

0.2

Replacement method

LDVM

Area size

32.0 × 32.0

In the second case are shown the simulation results of WMN-PSODGA system when the weight-coefficients are α = 0.3, β = 0.6, γ = 0.1. 4.1

Exponential Distribution

Figure 6 shows the visualization results after the optimization for Exponential distribution. Figure 7 shows the number of covered clients by each router. Figure 8 shows the transition of the standard deviations. The value of r means the correlation coefficient. In Fig. 8(a), when the load balancing is not considered, there is no correlation between the standard deviation and the number of

Fig. 6. Visualization results after the optimization for Exponential distribution.

A Hybrid Intelligent Simulation System for Node Placement in WMNs

565

updates. On the other hand, in Fig. 8(b), when the load balancing is considered, the standard deviation is decreased with the increase of the number of updates. These show that better optimization is achieved by considering load balancing.

Fig. 7. Number of covered clients by each router after the optimization for Exponential distribution.

Fig. 8. Transition of the standard deviations for Exponential distribution.

4.2

Normal Distribution

Figure 9 shows the visualization results after the optimization for Normal distribution. Figure 10 shows the number of covered clients by each router. Figure 11 shows the transition of the standard deviations. In Fig. 11(a), when the load balancing is not considered, there is almost no correlation between the standard deviation and the number of updates. On the other hand, in Fig. 11(b), when the load balancing is considered, the standard deviation is decreased with the increase of the number of updates. These show that better optimization is achieved by considering load balancing.

566

S. Ohara et al.

Fig. 9. Visualization results after the optimization for Normal distribution.

Fig. 10. Number of covered clients by each router after the optimization for Normal distribution.

Fig. 11. Transition of the standard deviations for Normal distribution.

A Hybrid Intelligent Simulation System for Node Placement in WMNs

5

567

Conclusions

In this work, we evaluated the performance of WMNs using a hybrid simulation system based on PSO and DGA (called WMN-PSODGA). We considered the Exponential and Normal distribution of mesh clients. From the simulation results, we found that the performance of the Exponential and Normal distribution was improved by considering load balancing when using WMN-PSODGA. Moreover, for the same number of mesh clients, the Normal distribution has better behavior than the Exponential distribution, because all mesh clients are covered by a smaller number of mesh routers. In future work, we will consider other distributions of mesh clients.

References 1. Akyildiz, I.F., Wang, X., Wang, W.: Wireless mesh networks: a survey. Comput. Netw. 47(4), 445–487 (2005) 2. Barolli, A., Sakamoto, S., Ozera, K., Ikeda, M., Barolli, L., Takizawa, M.: Performance evaluation of WMNs by WMN-PSOSA simulation system considering constriction and linearly decreasing Vmax methods. In: International Conference on P2P, Parallel, Grid, Cloud and Internet Computing, pp. 111–121. Springer (2017) 3. Barolli, A., Sakamoto, S., Barolli, L., Takizawa, M.: Performance analysis of simulation system based on particle swarm optimization and distributed genetic algorithm for WMNs considering different distributions of mesh clients. In: International Conference on Innovative Mobile and Internet Services in Ubiquitous Computing, pp. 32–45. Springer (2018) 4. Barolli, A., Sakamoto, S., Ozera, K., Barolli, L., Kulla, E., Takizawa, M.: Design and implementation of a hybrid intelligent system based on particle swarm optimization and distributed genetic algorithm. In: International Conference on Emerging Internetworking, Data & Web Technologies, pp. 79–93. Springer (2018) 5. Clerc, M., Kennedy, J.: The particle swarm-explosion, stability, and convergence in a multidimensional complex space. IEEE Trans. Evol. Comput. 6(1), 58–73 (2002) 6. Franklin, A.A., Murthy, C.S.R.: Node placement algorithm for deployment of twotier wireless mesh networks. In: Proceedings of Global Telecommunications Conference, pp. 4823–4827 (2007) 7. Girgis, M.R., Mahmoud, T.M., Abdullatif, B.A., Rabie, A.M.: Solving the wireless mesh network design problem using genetic algorithm and simulated annealing optimization methods. Int. J. Comput. Appl. 96(11), 1–10 (2014) 8. Goto, K., Sasaki, Y., Hara, T., Nishio, S.: Data gathering using mobile agents for reducing traffic in dense mobile wireless sensor networks. Mob. Inf. Syst. 9(4), 295–314 (2013) 9. Inaba, T., Elmazi, D., Sakamoto, S., Oda, T., Ikeda, M., Barolli, L.: A secure-aware call admission control scheme for wireless cellular networks using fuzzy logic and its performance evaluation. J. Mob. Multimed. 11(3&4), 213–222 (2015) 10. Inaba, T., Obukata, R., Sakamoto, S., Oda, T., Ikeda, M., Barolli, L.: Performance evaluation of a QoS-aware fuzzy-based CAC for LAN access. Int. J. Space-Based Situated Comput. 6(4), 228–238 (2016) 11. Inaba, T., Sakamoto, S., Oda, T., Ikeda, M., Barolli, L.: A testbed for admission control in WLAN: a fuzzy approach and its performance evaluation. In: International Conference on Broadband and Wireless Computing, Communication and Applications, pp. 559–571. Springer (2016)

568

S. Ohara et al.

12. Lim, A., Rodrigues, B., Wang, F., Xu, Z.: k-center problems with minimum coverage. Theoret. Comput. Sci. 332(1–3), 1–17 (2005) 13. Maolin, T., et al.: Gateways placement in backbone wireless mesh networks. Int. J. Commun. Netw. Syst. Sci. 2(1), 44–50 (2009) 14. Matsuo, K., Sakamoto, S., Oda, T., Barolli, A., Ikeda, M., Barolli, L.: Performance analysis of WMNs by WMN-GA simulation system for two WMN architectures and different TCP congestion-avoidance algorithms and client distributions. Int. J. Commun. Netw. Distrib. Syst. 20(3), 335–351 (2018) 15. Muthaiah, S.N., Rosenberg, C.P.: Single gateway placement in wireless mesh networks. In: Proceedings of 8th International IEEE Symposium on Computer Networks, pp. 4754–4759 (2008) 16. Naka, S., Genji, T., Yura, T., Fukuyama, Y.: A hybrid particle swarm optimization for distribution state estimation. IEEE Trans. Power Syst. 18(1), 60–68 (2003) 17. Poli, R., Kennedy, J., Blackwell, T.: Particle swarm optimization. Swarm Intell. 1(1), 33–57 (2007) 18. Sakamoto, S., Kulla, E., Oda, T., Ikeda, M., Barolli, L., Xhafa, F.: A comparison study of simulated annealing and genetic algorithm for node placement problem in wireless mesh networks. J. Mob. Multimed. 9(1–2), 101–110 (2013) 19. Sakamoto, S., Kulla, E., Oda, T., Ikeda, M., Barolli, L., Xhafa, F.: A comparison study of hill climbing, simulated annealing and genetic algorithm for node placement problem in WMNs. J. High Speed Netw. 20(1), 55–66 (2014) 20. Sakamoto, S., Kulla, E., Oda, T., Ikeda, M., Barolli, L., Xhafa, F.: A simulation system for WMN based on SA: performance evaluation for different instances and starting temperature values. Int. J. Space-Based Situated Comput. 4(3–4), 209–216 (2014) 21. Sakamoto, S., Kulla, E., Oda, T., Ikeda, M., Barolli, L., Xhafa, F.: Performance evaluation considering iterations per phase and SA temperature in WMN-SA system. Mob. Inf. Syst. 10(3), 321–330 (2014) 22. Sakamoto, S., Lala, A., Oda, T., Kolici, V., Barolli, L., Xhafa, F.: Application of WMN-SA simulation system for node placement in wireless mesh networks: a case study for a realistic scenario. Int. J. Mob. Comput. Multimed. Commun. (IJMCMC) 6(2), 13–21 (2014) 23. Sakamoto, S., Oda, T., Ikeda, M., Barolli, L., Xhafa, F.: An integrated simulation system considering WMN-PSO simulation system and network simulator 3. In: International Conference on Broadband and Wireless Computing, Communication and Applications, pp. 187–198. Springer (2016) 24. Sakamoto, S., Oda, T., Ikeda, M., Barolli, L., Xhafa, F.: Implementation and evaluation of a simulation system based on particle swarm optimisation for node placement problem in wireless mesh networks. Int. J. Commun. Netw. Distrib. Syst. 17(1), 1–13 (2016) 25. Sakamoto, S., Oda, T., Ikeda, M., Barolli, L., Xhafa, F.: Implementation of a new replacement method in WMN-PSO simulation system and its performance evaluation. In: The 30th IEEE International Conference on Advanced Information Networking and Applications (AINA 2016), pp. 206–211 (2016) 26. Sakamoto, S., Obukata, R., Oda, T., Barolli, L., Ikeda, M., Barolli, A.: Performance analysis of two wireless mesh network architectures by WMN-SA and WMN-TS simulation systems. J. High Speed Netw. 23(4), 311–322 (2017) 27. Sakamoto, S., Ozera, K., Barolli, A., Ikeda, M., Barolli, L., Takizawa, M.: Implementation of an intelligent hybrid simulation systems for WMNs based on particle swarm optimization and simulated annealing: performance evaluation for different replacement methods. Soft Comput. 23(9), 3029–3035 (2017)

A Hybrid Intelligent Simulation System for Node Placement in WMNs

569

28. Sakamoto, S., Ozera, K., Barolli, A., Ikeda, M., Barolli, L., Takizawa, M.: Performance evaluation of WMNs by WMN-PSOSA simulation system considering random inertia weight method and linearly decreasing Vmax method. In: International Conference on Broadband and Wireless Computing, Communication and Applications, pp. 114–124. Springer (2017) 29. Sakamoto, S., Ozera, K., Ikeda, M., Barolli, L.: Implementation of intelligent hybrid systems for node placement problem in WMNs considering particle swarm optimization, hill climbing and simulated annealing. Mob. Netw. Appl. 23(1), 27–33 (2017) 30. Sakamoto, S., Ozera, K., Ikeda, M., Barolli, L.: Performance evaluation of WMNs by WMN-PSOSA simulation system considering constriction and linearly decreasing inertia weight methods. In: International Conference on Network-Based Information Systems, pp. 3–13. Springer (2017) 31. Sakamoto, S., Ozera, K., Oda, T., Ikeda, M., Barolli, L.: Performance evaluation of intelligent hybrid systems for node placement in wireless mesh networks: a comparison study of WMN-PSOHC and WMN-PSOSA. In: International Conference on Innovative Mobile and Internet Services in Ubiquitous Computing, pp. 16–26. Springer (2017) 32. Sakamoto, S., Ozera, K., Oda, T., Ikeda, M., Barolli, L.: Performance evaluation of WMN-PSOHC and WMN-PSO simulation systems for node placement in wireless mesh networks: a comparison study. In: International Conference on Emerging Internetworking, Data & Web Technologies, pp. 64–74. Springer (2017) 33. Sakamoto, S., Ozera, K., Barolli, A., Barolli, L., Kolici, V., Takizawa, M.: Performance evaluation of WMN-PSOSA considering four different replacement methods. In: International Conference on Emerging Internetworking, Data & Web Technologies, pp. 51–64. Springer (2018) 34. Schutte, J.F., Groenwold, A.A.: A study of global optimization using particle swarms. J. Global Optim. 31(1), 93–108 (2005) 35. Shi, Y.: Particle swarm optimization. IEEE Connect. 2(1), 8–13 (2004) 36. Shi, Y., Eberhart, R.C.: Parameter selection in particle swarm optimization. In: Evolutionary Programming VII, pp. 591–600 (1998) 37. Vanhatupa, T., Hannikainen, M., Hamalainen, T.: Genetic algorithm to optimize node placement and configuration for WLAN planning. In: Proceedings of The 4th IEEE International Symposium on Wireless Communication Systems, pp. 612–616 (2007) 38. Wang, J., Xie, B., Cai, K., Agrawal, D.P.: Efficient mesh router placement in wireless mesh networks. In: Proceedings of IEEE International Conference on Mobile Adhoc and Sensor Systems (MASS 2007), pp. 1–9 (2007)

Multi-dimensional Contract Incentive Design for Mobile Crowdsourcing Networks Nan Zhao(&), Menglin Fan, Chao Tian, Pengfei Fan, and Xiao He Hubei University of Technology, Wuhan 430068, China [email protected], [email protected], [email protected], [email protected], [email protected]

Abstract. Through utilizing sensing and computing capabilities of mobile devices, mobile crowdsourcing network (MCN) can collect and analyze data in a cost-effective way. However, due to the selfishness of mobile devices, they may be reluctant to participate in crowdsourcing without additional incentives. In this paper, the incentive mechanism to encourage mobile devices’ participation in the multi-tasks of crowdsourcing is designed. By modelling MCN as a labour market, the crowdsourcing incentive mechanism is regarded as a moral hazard model under contract-based asymmetric information scenarios. Moreover, considering the interaction among crowdsourcing tasks, a multidimensional contract model is proposed. By evaluating mobile users’ performance, the service provider will reward and stimulate them to participate in crowdsourcing and work harder. Results demonstrate that the proposed contract has excellent performance in crowdsourcing incentives.

1 Introduction Mobile crowdsourcing network (MCN) has broad application prospects in many fields such as public utilities, public health, environmental protection, energy conservation and emission reduction, which has a positive role in promoting the realization of national sustainable development strategic goals [1, 2]. In the process of participating in the crowdsourcing tasks, the mobile users (MUs) may consume various resources, such as data processing and transmission costs, bandwidth required for data transmission, energy consumption of the mobile device battery, and possible discomfort caused by manual operation of submitting data [3, 4]. As a result, MUs will be reluctant to participate in crowdsourcing tasks unless they receive a satisfactory reward to compensate for their resource consumption and potential security threats [5]. Moreover, when MU participates in the crowdsourcing task, the MU may hide its true information to avoid revealing its sensitive information. Thus, the SP cannot obtain the MU’s actual effort in crowdsourcing tasks, which leads to the problem of asymmetric information [6]. We focus on the incentive mechanism to encourage MUs in participating in crowdsourcing tasks. Many researches have noted that there is an urgent need to design a reasonable incentive mechanism to motivate the MU participating in crowdsourcing tasks. In [7], an insurance-based incentive framework was designed to motivate the users to upgrade to a higher security level. Zhao et al. proposed an incentive mechanism to encourage © Springer Nature Switzerland AG 2020 L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 570–578, 2020. https://doi.org/10.1007/978-3-030-33506-9_51

Multi-dimensional Contract Incentive Design for MCNs

571

multi-user collaboration to complete one crowdsourcing task [8]. Wang et al. studied a novel online incentive mechanism for constructing a fine-grained REM with crowdsourcing in a realistic scenario [9]. However, most of the existing researches may just define incentive mechanism as one-dimensional [10, 11]. Practically, considering that the MU needs to participate in multiple crowdsourcing tasks simultaneously in most cases, a multi-dimensional incentive mechanism should be designed to replace onedimensional incentive scheme. Therefore, a multi-dimensional contract incentive mechanism is designed for MCN to solve these problems. Firstly, we model the incentive scenario as multi-dimensional, which can attract the MU to participate in multiple crowdsourcing tasks. Then, we propose the contract-based incentive model that leads to the MU’s bonus ratio being proportional to its performance. The contract maximizes the utility of the service provider (SP) and encourages the MU to perform well in crowdsourcing tasks under asymmetric information scenarios. Moreover, the optimization incentive problem is to maximize the expected utility of the SP under conditions acceptable to the MU. Besides, this paper makes the incentive mechanism more reasonable by analyzing the risk appetite of crowdsourcing participants. Simulation verifies the performance of the multi-dimensional contract-based incentive mechanism.

2 System Model In this section, the concept of MCN will be introduced at first. Then, the optimal contract will be set by considering both utility of the MU and the SP. We assume that the incentive mechanism to encourage mobile devices’ participation in the multi-task of crowdsourcing. 2.1

Model of Mobile Crowdsourcing Network

In Fig. 1, a mobile crowdsourcing network includes three parts: a service provider (SP), a lot of MUs and end users. First, end users seek help from the SP according to their requests. Then, the SP distributes crowdsourcing tasks on the service platform according to end users’ request. The SP hires MUs to complete crowdsourcing tasks. After the MU accomplished tasks, the SP feeds back results to end users.

Service Provider Computer Base Station Smartphone

Smart vehicular

Fig. 1. Mobile crowdsourcing network (MCN)

Mobile Users

572

2.2

N. Zhao et al.

Utility of Mobile Users

When MU participates in the multi-task of crowdsourcing, which offers crowdsourcing effort e ¼ ðe1 ; e2 ; . . .; en ÞT to acquire its reward from the SP. The crowdsourcing effort e results in the SP reaping the profits. Since a large number of measurement errors exist, the achieved benefit of the SP may not be consistent with the MU’s actual efforts. Then, the SP’s actual profit p ¼ ðp1 ; p2 ; . . .; pn ÞT is supposed to be a noisy signal as: p ¼ he þ f; 0 B where h ¼ @

h1

..

0 .

ð1Þ

1 C A means that the MU’s profit in paying for per unit

hn crowdsourcing effort, and f ¼ ðf1 ; f2 ; . . .; fn ÞT is the random variable with f  N ð0; r2 Þ. Accordingly, the SP’s total profit R is given by: 0

R¼

n X

pi :

ð2Þ

i¼1

When the MU accomplish the crowdsourcing tasks for the SP, the MU has an operating cost. Moreover, considering that operational costs are related to the level of MU’s effort, we assumed that the crowdsourcing cost wðeÞ increases much faster with greater effort than that with lesser effort. That is w0 ðeÞ [ 0, w00 ðeÞ [ 0. For simplicity, the MU’s crowdsourcing cost wðeÞ is defined as: 1 wðeÞ ¼ eT Ce; 2

ð3Þ

0

1 c11 . . . c1n B . .. . C where C ¼ @ .. . .. A is a symmetric matrix of crowdsourcing cost, and the cn1    cnn main diagonal element cii represents the operational cost of completing the task, and the remaining corresponding positions represent the links between task i and task j. Moreover, assume that the payment which the SP offers to the MU is WM , which can be defined as: WM ¼ a þ bT he;

ð4Þ

where a is the basic salary of the MU, and b ¼ ðb1 ; b2 ; . . .; bn ÞT is the bonus coefficient associated with the MU’s crowdsourcing performance. The MU has a different bi when it chooses different crowdsourcing actions. Furthermore, we introduce risk aversion coefficients gM to indicate how much the MU is willing to participate in crowdsourcing tasks. Because of the MU’s conservative properties, we need to define the utility function of the MU as a concave function. In

Multi-dimensional Contract Incentive Design for MCNs

573

addition, assume that the utility function must be an exponential form because of the symmetric matrix form of measurement errors and cost coefficients. Then, we take the form of negative exponential utility to express the MU’s utility, which is written as: U ðWM Þ ¼ egM ðWM wðeÞÞ ;

ð5Þ

where gM [ 0 is the absolute risk aversion coefficient of the MU (gM ¼ ðu00 ðWM  wðeÞÞ=  u0 ðWM  wðeÞÞÞ). Next, the MU’s expected utility E½U ðWM Þ is given by: h i E ½U ðWM Þ ¼ E egM ðWM wðeÞÞ :

ð6Þ

Let fM ¼ WM  wðeÞ ¼ a þ bT he  12 eT Ce. Next, the MU’s expected utility E ½U ðWM Þ is simplified as: E ½U ðWM Þ ¼ egM fM ¼ ¼  egM ½a þ b

T

he12eT Ce12gM bT Rb

;

ð7Þ

0

1 . . . d1n .. C .. . A is a symmetric matrix of the covariance, and the main . dn1    d2n diagonal element d2i represents the variance of fi , and the remaining corresponding positions represent the covariance between fi and fj . The variance indicates the difficulty in ensuring the accuracy of the measurements. As the variance d2i increases, the SP is increasingly able to observe the performance of the MU. d21 B . where R ¼ @ ..

2.3

Utility of Service Provider

Considering the efforts of MU, the SP’s total utility is defined as: Ws ¼ R  WM :

ð8Þ

Then, the expected utility profit of the SP is given by: E ½Ws  ¼

n X

pi  bT he  a:

ð9Þ

i¼1

Similarly to the MU, we also let fS ¼ utility E ½U ðWs Þ can be defined as:

n P

pi  bT he  a, then, the SP’s expected

i¼1

E ½U ðWs Þ ¼ eðfS Þ ¼ e



 n P i¼1

 pi bT hea

:

ð10Þ

574

N. Zhao et al.

From (1) and (2), we can simplify the fS as: fS ¼ ¼

n X i¼1 n X

pi  bT he  a hi

i¼1

¼ Q1 b1

n X

Qij hj bj  bT he  a

ð11Þ

j¼1 n X

hi Qi1 i¼1 T T

þ Q2 b2

n X

hi Qi2 þ . . . þ Qn bn

i¼1

n X

hi Qin  bT he  a

i¼1

 ¼ diagðhÞ Q hb  bT he  a; where C1 ¼ Q. 2.4

Problem Formulation

Due to the selfishness and limited resources of MU, it may make a little effort in crowdsourcing tasks. Therefore, in order to encourage the MU in participating in crowdsourcing tasks, the SP needs to design a reliable contract-based incentive mechanism. Service Provider

Mobile Users

Offers a task with contract

Accepts or refuses the contract

Exerts an effort or not

Task completed or not

The contract is excuted

Fig. 2. Contract-based incentive mechanism for mobile crowdsourcing

As it shows in Fig. 2, after the SP designs the optimal contract, it broadcasts the contract items to the potential MU. Note that contract items include the optimal basic salary a and the optimal bonus coefficient b . Once the MU decides whether it is willing to sign the contract and informs the SP of its choice. Then, the SP notifies the crowdsourcing task of the adopted MU, and MU helps to accomplish crowdsourcing tasks. Finally, the SP will receive results from the MU, and the SP decide whether to pay the MU based on the information received.

Multi-dimensional Contract Incentive Design for MCNs

575

3 Contract-Based Crowdsourcing Incentive Mechanism Although the SP has signed a contract with the MU, it still cannot get the exact crowdsourcing effort of the MU because of information asymmetry. Therefore, the contract should ensure that the MU offers the optimal effort e to get its maximum utility. The following incentive compatibility (IC) constraint is given by: ðICÞ Because

@E ½uðWM Þ @fM

max

e  0

E ½U ðWM Þ:

ð12Þ

¼ gM egM fM [ 0, the IC constraint (14) can be rewritten as: 1 1 max fM ¼ a þ bT he  eT Ce  gM bT Rb: ei  0 2 2

ðICÞ

ð13Þ

Moreover, to ensure that the utility received by the MU is no less than its reserved utility U, the following individual rational (IR) constraint is given by: 1 1 ðIRÞ a þ bT he  eT Ce  gM bT Rb  U: 2 2

ð14Þ

Based on the IC and IR constraints, we can obtain the optimal contract, which can maximize the SP’s expected utility. The optimal contract design problem is written as:

max

ffai ;bi  0gg

s:t: ðICÞ ðIRÞ

E ½U ðWs Þ ¼ eðfS Þ ¼ e



 n P

 pi bT hea

i¼1

T T T ¼ ef½diagðhÞ Q hbb heag ;

1 1 max fM ¼ a þ bT he  eT Ce  gM bT Rb; ei  0 2 2 1 T 1 T T a þ b he  e Ce  gM b Rb  U: 2 2

ð15Þ

Then, by the IC constraint (15), the optimal effort e is given by, e ¼ C1 hT b:

ð16Þ

Since @E½u@fðSWS Þ ¼ efS [ 0, by combining (18), the SP’s optimization problem can be simplified as: max

  1 fS ¼ diagðhÞT QT hb  a  bT he þ ðe ÞT Ce ; 2 1 1 a þ bT he  ðe ÞT Ce  gM bT Rb  U: 2 2

ffai ;bi  0gg

s:t:

ðIRÞ

ð17Þ

576

N. Zhao et al.

Then, from the IR constraint (19), it can be observed that fS increases as the basic salary a decreases. The SP can obtain its maximum utility while the basic salary a takes the minimum value, the optimal basic salary a can be written as: 1 1 a ¼ U  bT he þ ðe ÞT Ce þ gM bT Rb: 2 2

ð18Þ

Then, the SP’s optimization problem can be simplified as: max

fbi  0g

  1 fS ¼ diag(h)T QT hb  a  bT he þ ðe ÞT Ce : 2

ð19Þ

Accordingly, by combining the equivalent of optimal effort e (16) and the equivalent of optimal basic salary a (18), we can obtain the optimal bonus coefficient b from (19)  1 b ¼ hT C1 h þ gM R hT C1 ½diagðhÞ:

ð20Þ

4 Numerical Results and Analysis Simulation result is used to verify the performance of the multi-dimensional contractbased incentive mechanism. We study how optimal effort e and the payment WM changes by changing the variance d2i with n ¼ 2.

(a) δ 12 increases with δ 22 =3.5

(b) δ 22 increases with δ 12 =3.5

Fig. 3. MU’s optimal effort with the different d2i

We first investigate the influence of variance d21 and d22 on the optimal efforts ei , when one variance d2i changes, the other variance d2i remains the same. As we shown in Fig. 3, as the variance d21 increases, the MU can hide its efforts in crowdsourcing task 1, the optimal effort e1 is reduced. While variance d21 increases, the SP is increasingly able

Multi-dimensional Contract Incentive Design for MCNs

577

to observe the performance of the MU in crowdsourcing task 1. It means that the MU can reduce its efforts on crowdsourcing task 1 to get more utility to deceive the SP. However, the increase of optimal effort e2 makes the SP obtain more utility from crowdsourcing task 2, which insures the SP’s actual utility in this scenario. When the variance d22 changes, we can achieve the same results. Results verify the excellent performance of our method under asymmetric information scenario.

(a) δ 12 increases with δ 22 =3

(b) δ 22 increases with δ12 =3

Fig. 4. The payment WM with different d2i

In Fig. 4, we investigate the influence of variance d2i on the optimal basic salary a and the payment WM . Similar to the above, we also let one variance d2i remain the same, while the other variance d2i changes. As the variance d21 increases, the optimal basic salary a , the optimal bonus coefficient b1 and the payment WM decreases while the optimal bonus coefficient b2 increases. In the above experiment, as the variance d21 increases, the optimal effort e1 of the MU decreases, and the optimal effort e2 increases. The reduction of the optimal effort e1 causes the bonus coefficient b1 to decrease, while the increase in the optimal effort e2 causes the optimal bonus coefficient b2 to increase. According to this experiment, we verify that the MU’s bonus coefficient is directly proportional to its efforts. Thus, the changes in the bonus coefficient bi can motivate the MU to work hard to complete crowdsourcing tasks while participating in them. The changes in the payment WM ensure that the SP’s utility from the crowdsourcing tasks.

5 Conclusions In this article, we investigate a contract-based crowdsourcing incentive mechanism in the MCN. In the MCN, the SP rewards MU with a multi-dimensional evaluation to encourage it in participating in crowdsourcing. Due to the selfish nature of the SP and the MU, the proposed incentive mechanism can maximize utility for both parties by considering the both requirements. In addition, we consider the risk preference of the

578

N. Zhao et al.

MU and study the optimal under the situation of asymmetric information. Finally, we validate our proposed contract by changing the parameters to analyze the numerical results of the rewards and optimal efforts. Results demonstrate that the contract-based incentive mechanism has excellent performance in mobile crowdsourcing.

References 1. Jarrett, J., Saleh, I., Blake, M., Thorpe, S., Grandison T., Malcolm, R.: Mobile services for enhancing human crowdsourcing with computing elements. In: IEEE International Conference on Mobile Services, pp. 149–152 (2014) 2. Duan, L., Kubo, T., Sugiyama, K., Huang, J., Hasegawa, T., Walrand, J.: Incentive mechanisms for smartphone collaboration in data acquisition and distributed computing. In: Proceedings of the IEEE International Conference on Computer Communications, pp. 1701– 1709 (2012) 3. Zhao, N., Chen, Y., Liu, R., Wu, M., Xiong, W.: Monitoring strategy for relay incentive mechanism in cooperative communications networks. Comput. Electr. Eng. 60, 14–29 (2017) 4. Zhao, N., Wu, M., Xiong, W., Liu, C.: Cooperative communication in cognitive radio networks under asymmetric information: a contract-theory based approach. Int. J. Distrib. Sens. Netw. 11, 1–11 (2015) 5. Qiu, T., Chen, B., Sangaiah, K.: A survey of mobile social networks: applications, social characteristics, and challenges. IEEE Syst. J. 12, 1–16 (2017) 6. Lu, J., Xin, Y., Zhang, Z., Liu, X., Li, K.: Game-theoretic design of optimal two-sided rating protocols for service exchange dilemma in crowdsourcing. IEEE Trans. Inf. Forensics Secur. 13, 2801–2815 (2018) 7. Gogo, A., Cybenko, G., Garmire, E.: A crowd sourced pharmacovigilance approach using SMS-based asymmetric encryption. In: International Multi-conference on Computing in the Global Information Technology, pp. 226–231 (2010) 8. Zhao, N., Fan, M., Tian, C., Fan, P.: Contract-based incentive mechanism for mobile crowdsourcing networks. Algorithms 10, 1–13 (2017) 9. Wang, X., Umehira, M., Han, B., Li, P., Gu, Y., Wu, C.: Online incentive mechanism for crowdsourced radio environment map construction. In: IEEE International Conference on Communications, pp. 1–6 (2019) 10. Gao, L., Iosifidis, G., Huang, J., Tassiulas, L.: Hybrid data pricing for network-assisted userprovided connectivity. In: IEEE International Conference on Computer Communications, pp. 682–690 (2014) 11. Ota, K., Dong, M., Gui, J., Liu, A.: QUOIN: incentive mechanisms for crowd sensing networks. IEEE Netw. 32, 1–6 (2018)

Evaluation and Comparison of CO2 and Fuel Consumption for Different Car Following Models Ningling Jiang1 and Elis Kulla2(B) 1

Graduate School of Engineering, Okayama University of Science, 1-1 Ridai-cho, Kita-ku, Okayama 700-0005, Japan [email protected] 2 Department of Information and Computer Engineering, Okayama University of Science, 1-1 Ridai-cho, Kita-ku, Okayama 700-0005, Japan [email protected]

Abstract. Vehicular communication systems are computer networks in which vehicles and RoadSide Units (RSUs) are the communicating nodes, providing each other with information, such as safety warnings and traffic information. Despite the advanced development in self-driving technologies, intervehicular communication is still an emerging field, with different applications, such as, platooning, road safety and so on. Platooning is an application where vehicles create a formation and travel together and other vehicles follow a physical leader (front car) or a virtual leader, depending on the implementation. Energy saving, labor saving and improvement of safety are expected outcomes of this technology. In this work, a comprehensive simulation called VENTOS is used to evaluate and compare ACC, Krauss and CACC car following models. The results show that, the following cars that use CACC have a better efficiency regarding CO2 emission and fuel consumption. Keywords: Vehicular communications · Platooning model · CACC · ACC · KRAUSS · Simulation

1

· Car following

Introduction

Intelligent Transportation Systems (ITS) can send and receive information between wearables, road infrastructure, vehicles and other devices present nearby, in order to solve problems such as accidents avoidance, traffic congestion and environmental measures. Vehicular communications are a promising area of Intelligent Transportation Systems (ITS). Platooning and the car-following methods pose some interesting research issues in the field of vehicular communications, such as, fuel efficiency, environment pollution, and so on. Car-following models have been studied for nearly a half century [1]. They describe the process of following vehicles in the traffic flow, as shown in Fig. 1. They simulate vehicle behavior by setting various rules and thresholds to avoid c Springer Nature Switzerland AG 2020  L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 579–588, 2020. https://doi.org/10.1007/978-3-030-33506-9_52

580

N. Jiang and E. Kulla

hitting the leading vehicle. Vehicles accelerate and decelerate in accordance to the distance and acceleration of the leading vehicle. Some experiments in vehicular communications are costly, and resource consuming [2,3]. So, researchers often use simulation tools instead of outdoor experiments. In this work simulations were carried out on VENTOS simulator [4]. Mobility model and network simulation are two aspects of the simulation of VENTOS. OMNET++ is used for the simulation of network, while SUMO is for simulation of vehicular mobility. The paper is organized as follows. In Sect. 2, we briefly introduce different car-following models. In Sect. 3.1, We describe the simulation environment. Simulation results are presented and analyzed in Sect. 4. Finally, in Sect. 5, conclusions and future work are summarized.

2 2.1

Car-Following Models Krauss Model

The model developed by Kraußin 1997 is a microscopic, space-continuous, carfollowing model based on the safe speed rule. Safe speed refers to the law of driving at a speed limit to ensure safe driving according to circumstances, road conditions and traffic conditions. The speed of vehicles must be controlled within safe speed to ensure driving safety on the road. This safe speed is computed as follows: g(t) − vl (t)tr (1) vsaf e = vl (t) + v (t)+v (t) l f + tr 2b Where vl (t) represents the speed of the leading vehicle in time t, g(t) is the gap to the leading vehicle in time t, tr is the driver’s reaction time (about 1 s) and b is the maximum deceleration of the vehicle (m/s2 ). Because vsaf e may be larger than the maximum speed allowed on the road or larger than the vehicle is capable to reach until the next step due to its acceleration capabilities [5], the minimum of these values is computed as resulting speed which is called the “desired” or “wished” speed. vdes = min[vsaf e , v + at, vsaf e ]

(2)

Where t represents step duration of the simulation. 2.2

ACC Model

The ACC car-following model has multiple operation modes, and the system transits between these modes in order to generate the desired acceleration. In the following we will describe Speed Control Mode and Gap Control Mode.

Evaluation and Comparison of CO2 and Fuel Consumption

581

Speed Control Mode. For the ACC car-following model, the feedback control law in speed mode is activated when there are no preceding vehicles in the range covered by the sensors or preceding vehicles exist in a spacing larger of 120 m [6]. This mode aims to eliminate the deviation between the vehicle speed and the desired speed and is given as: αi,k+1 = k1 (vd − vi,k ), k1 > 0

(3)

where, αi,k+1 presents the acceleration recommended by speed control mode of the i-th consecutive (subject) vehicle for the next time step k + 1; v(d) and vi−1,k indicate the desired cruising speed and the speed of the i-th vehicle at the current time step k, respectively; and k1 is the control gain determining the rate of speed deviation for acceleration. Typical values for this gain are 0.4 s−1 . Gap Control Mode. In gap control mode, the acceleration in the next time step k + 1 is modelled as a second-order transfer function based on the gap and speed deviations with respect to the preceding vehicle, and is defined as: αi,k+1 = k2 ei,k + k3 (vi−1,k − vi,k ), k2 , k3 > 0

(4)

in which ei,k is the gap deviation of the i-th consecutive vehicle at the current time step k, and vi−1,k is the current speed of the preceding vehicle (index i − 1 refers to the leader of vehicle i; k2 and k3 are the control gains on both the positioning and speed deviations, respectively. The proposed optimal values for the gains are k2 = 0.23 s−2 and k3 = 0.07 s−1 [7]. The gap control mode is activated when the gap and speed deviations are concurrently smaller than 0.2 m and 0.1 m/s respectively [7]. 2.3

CACC Model

The CACC car-following model is based on the ACC model and allows vehicles to drive with even closer space gap. This is achievable by parameter sharing using the V2V wireless communication. Speed Control Mode. The speed controller for CACC vehicles is the same with the ACC ones since the additional information exchange (either V2V or V2I) does not influence the vehicle cruising mode. This control mode is activated when the time-gap is larger than 2 s [7] and is given as: αi,k+1 = k4 (vd − −vi,k ), k4 > 0 where the control gain k4 is equal to 0.4 s−1 .

(5)

582

N. Jiang and E. Kulla

Fig. 1. Car-following model at constant headway-time

Gap Control Mode. For the CACC car-following model, the speed of the equipped vehicles in the next time step k + 1 is represented by a first-order transfer function, according to: vi,k+1 = vi,k + k5 ei,k + k6 ei,k , k5 , k6 > 0

(6)

where (ei,k ) is the derivative of the gap deviation ei,k , and is defined as: ei,k = vi−1,k − vi,k − td αi,k

(7)

with td being the desired time gap of the CACC controller. The values of the control gains k5 and k6 of Equation are set as 0.45 s−2 and 0.25 s−1 , respectively [7]. The gap control mode for CACC vehicles is activated when the gap and speed deviations are concurrently smaller than 0.2 m and 0.1 m/s respectively [7].

3 3.1

Simulation Envornment The VENTOS Simulation Environment

VENTOS is an integrated simulator based on two well-known simulators: SUMO and OMNET++. It uses the mobility information of cars, bikes and pedestrians to perform realistic simulations. – SUMO (Simulator of Urban MObility) is an open-source, high-portable, and fast in run-time simulator created by the German Aerospace Centre in 2001, which enables users to develop and integrate new algorithms; despite its name, SUMO can also simulate highway traffic networks [8,9]. The simulator also includes the TraCI (Traffic Control Interface) tool, which is a Python API offering users the ability to interact with the running simulation in order to control the vehicle parameters. – OMNET++ is an open-source, extensible, modular, component-based C++ simulation package and captures the wireless communication simulation. IEEE 802.11p physical layer modeling and IEEE 1609.4 are implemented in the Veins (Vehicles in Network Simulation) framework and is used for wireless V2X communication between different modules. Many well-known TCP/IP protocols can be added from INET framework [10,11].

Evaluation and Comparison of CO2 and Fuel Consumption

3.2

583

Simulation Scenarios

Scenario 1. In scenario 1, the CO2 and fuel consumption of the ACC, KRUASS and CACC car following models were compared. As shown in Fig. 2, the scenario is performed on the expressway. On a three-lane highway, five vehicles (V1, V2, V3, V4, V5) travel in the direction of the arrow from the starting point (green) to the destination (red).

Fig. 2. Scenario 1: The highway map

Table 1 shows the parameters set in for Scenario 1 simulation. As a result of the simulation, as described in HBEFA (Handbook Emission Factors for Road Transport), the fuel consumption and CO2 of each vehicle are collected and compared and evaluated. The tool collects data. HBEFA is a database application that provides emission factors and specific emissions for all current road vehicles (passenger cars, light vehicles, large vehicles, buses, motorcycles). Table 1. Simulation parameters: Scenario 1 Parameters

Value

MAP

Highway

Vehicle number

5

Vehicle length (m)

5

Vehicle max speed (km/h)

80

Vehicle max acceleration (m/s2 ) 3 Vehicle max deceleration (m/s2 ) 5 Time (s)

80

Car following model

KRAUSS, ACC, CACC

Scenario 2. In scenario 2, We use a real street map of our city of Okayama (Fig. 3). The map is imported from the OpenStreetMap (Fig. 4) which is a free editable map of the whole world [12]. As shown in Fig. 5, five vehicles (V1, V2, V3, V4, V5) travel in the direction of the arrow from the starting point (green) to the destination (red). We use CACC car following model, but vehicles apply different timegap values, namely 0.03 s and 0.5 s. Table 2 shows the parameters used in Scenario 2.

584

N. Jiang and E. Kulla

Fig. 3. Map of Okayama city

Fig. 4. Map of Okayama city from SUMO

Fig. 5. Scenario 2: The city map

Table 2. Simulation parameters: Scenario 2 Parameters

Value

MAP

Okayama

Vehicle number

5

Maximum acceleration (m/s2 ) 3.0 Maximum deceleration (m/s2 ) 5.0 Time-gap (s)

0.3, 0.05

Speed control gain (1/s)

0.4

Speed gain (1/s)

1.0

Vehicle max speed (km/h)

80

Time (s)

80

Evaluation and Comparison of CO2 and Fuel Consumption

4 4.1

585

Simulation Results Scenario 1

The simulation results for Scenario 1 are shown in Figs. 6 and 7 for the CO2 emission and fuel consumption of each moving vehicle. The horizontal axis shows the number of each vehicle, where v1 is the leading vehicle and v5 is the last vehicle tail. The vertical axis shows the total CO2 emissions and total consumed fuel for each vehicle. We should note that, the leading vehicle moves identically the same in each simulation settings, so we use that as a comparison basis. In general the vehicles which use the CACC model emit less CO2 and consumes less fuel than when they use the other models. When using CACC, the following vehicles are able to follow the preceding vehicles at a closer distance gap, thus optimizing air resistance. In addition, CACC-enabled vehicles break quicker than ACC-enabled ones.

Fig. 6. CO2 emission for each car-following model

4.2

Scenario 2

The simulation results for Scenario 2 are shown in Figs. 8 and 9 for the CO2 emission and fuel consumption of each moving vehicle. The horizontal axis shows the number of each vehicle, where v1 is the leading vehicle and v5 is the last vehicle tail. The vertical axis shows the total CO2 emissions and total consumed fuel for each vehicle. We should note that, the leading vehicle moves identically the same in each simulation settings, so we use that as a comparison basis. In general the CACC-enabled vehicles with gaptime 0.3 s emit less CO2 and consume less fuel than vehicles using the CACC model with gaptime 0.05 s).

586

N. Jiang and E. Kulla

Fig. 7. Fuel consumption for each car-following model

Fig. 8. CO2 emission for each car-following model

Fig. 9. Fuel consumption for each car-following model

Evaluation and Comparison of CO2 and Fuel Consumption

5

587

Conclusions and Future Work

In this paper we conducted simulations in two different scenarios, in order to evaluate the efficiency of three different car-following models: ACC, CACC and KRAUSS. We also compared the performance of CACC model for different gaptime: 0.05 s and 0.3 s. From the simulation results we found the following: • When using CACC model, the following vehicles emit less CO2 and consume less fuel than when using ACC and KRUASS model. • In order to optimize CO2 emissions and fuel consumption in CACC model, we can control the timegap settings of the model. We should note, that the abovementioned findings are limited to our simulation scenarios. Furthermore, for Scenario 2, decreasing the timegap of CACC, might also cause crashes in certain extreme cases, so a more thorough study is required to analyze different behaviour of the leading vehicle, such as lane changes, traffic light crossings, immediate breaking and so on. We plan to work towards solving those problems, in our future work. Our final goal is to implement a real laboratory environment, and evaluate different car-following models.

References 1. Khodayari, A., Ghaffari, A., Kazemi, R., Alimardani, F., Braunstingl, R.: Improved adaptive neuro fuzzy inference system car-following behaviour model based on the driver–vehicle delay. IET Intell. Transp. Syst. 8(4), 323–332 (2013) 2. Coll-Perales, B., Gruteser, M., Gozalvez, J.: Evaluation of IEEE 802.11ad for mmWave V2V communications. In: 2018 IEEE Wireless Communications and Networking Conference Workshops (WCNCW), pp. 290–295, April 2018 3. Coll-Perales, B., Gozalvez, J., Gruteser, M.: Sub-6GHz assisted MAC for millimeter wave vehicular communications. IEEE Commun. Mag. 57(3), 125–131 (2019) 4. VENTOS open simulator. https://maniam.github.io/VENTOS/. Accessed 30 Aug 2019 5. Krauß, S., Wagner, P., Gawron, C.: Metastable states in a microscopic model of traffic flow. Phys. Rev. E 55(5), 5597 (1997) 6. Liu, H., Xiao, L., Kan, X.D., Shladover, S.E., Lu, X.-Y., Wang, M., Schakel, W., van Arem, B.: Using cooperative adaptive cruise control (CACC) to form highperformance vehicle streams. Final report (2018) 7. Xiao, L., Wang, M., van Arem, B.: Realistic car-following models for microscopic simulation of adaptive and cooperative adaptive cruise control vehicles. Transp. Res. Rec. 2623(1), 1–9 (2017) 8. Behrisch, M., Bieker, L., Erdmann, J., Krajzewicz, D.: SUMO–simulation of urban mobility: an overview. In: Proceedings of SIMUL 2011, The Third International Conference on Advances in System Simulation. ThinkMind (2011) 9. Song, J., Wu, Y., Xu, Z., Lin, X.: Research on car-following model based on SUMO. In: The 7th IEEE/International Conference on Advanced Infocomm Technology, pp. 47–55. IEEE (2014)

588

N. Jiang and E. Kulla

10. Varga, A., Hornig, R.: An overview of the OMNeT++ simulation environment. In: Proceedings of the 1st International Conference on Simulation Tools and Techniques for Communications, Networks and Systems & Workshops, p. 60. ICST (Institute for Computer Sciences, Social-Informatics and Telecommunications Engineering) (2008) 11. Amoozadeh, M., Deng, H., Chuah, C.-N., Zhang, H.M., Ghosal, D.: Platoon management with cooperative adaptive cruise control enabled by VANET. Veh. Commun. 2(2), 110–123 (2015) 12. Haklay, M., Weber, P.: OpenStreetMap: user-generated street maps. IEEE Pervasive Comput. 7(4), 12–18 (2008)

Individually Separated Wireless Access Point to Protect User’s Private Information Myoungsu Kim(&) and Kangbin Yim Department of Information Security Engineering, Soonchunhyang University, Asan, South Korea {brightprice,yim}@sch.ac.kr

Abstract. Due to the miniaturization, diversification, and massification of devices, the amount of installation of wireless AP (Access Point)s has increased, and the users can utilize the Internet by connecting their devices such as laptops, mobile phones, and tablet PCs to wireless APs when desired. However, there has been a big concern about the threat to steal user’s personal information from the devices connected to the wireless APs. People think this can be prevented by using the encrypted channel of the wireless AP, but the passwords of the wireless APs used in coffee shops, restaurants, or public places are null or a shared-single, so that the malicious user can also obtain such a password and the user’s personal information may be easily exposed. In this paper, we propose a protection scheme for user’s personal information delivered within the domain of a wireless AP and we introduce a method using a POS (Point of Sale) system for the scheme. A random SSID and password are generated through the POS system, and a user’s dedicated AP is generated based on this. Users can register their SSID and password with the device to protect them from the threat of personal information leakage by malicious users.

1 Introduction A wireless AP (Access Point) is a device that wirelessly connects other devices that requires network connections to the Internet. Recently, due to the miniaturization, diversification, and massification of devices [1, 2], as the demand for APs to connect to the Internet wirelessly no matter of locations or places, in response, various types of APs have emerged [2, 3] and installed in coffee shops and restaurants or public places as the spaces for providing the wireless Internet [2, 4]. In addition, as the development of devices has increased the number of service providers supporting the device platform [5, 6], when user authentication is performed through the device, the service can be simply used by the device without the need for complicated offline and complicated authentication procedures. If a user connects to the unencrypted channel of an AP to use a service, there will be a risk that personal information may be stolen by a malicious user [7–10]. When a malicious user connects to the AP, the user’s personal information can be collected and abused without restriction, resulting in secondary damage such as unauthorized service subscription, payment, or phishing [11, 12]. In order to prevent this, most places where the AP is recently operated have a password by default, so it is not easy for a malicious © Springer Nature Switzerland AG 2020 L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 589–596, 2020. https://doi.org/10.1007/978-3-030-33506-9_53

590

M. Kim and K. Yim

user to bypass the password and steal the data. However, a malicious user can also obtain the password of a wireless AP in usual environments through physical means (for example, a password written with an SSID is posted in a specific space), it eventually becomes like an unencrypted AP, and the user’s personal data is inevitably exposed. Due to the encryption technology and the authentication process, it is not easy for a malicious user to access the internal data protected by the secure domain through a general method without any other assistance [13–15]. There have been several researches discussing the vulnerability in the encryption technology or the authentication process and suggesting improved or new methods of encryption and authentication as the countermeasure [13–15]. However, as described above, when a malicious user physically obtains the password, the AP’s internal data is exposed as it is, no matter how good encryption technologies or password processes they have. In order to prevent this problem, it is necessary not only to defend the AP from an external attack but also to respond to a case where a malicious user obtains the password and accesses it internally, as in the normal cases for public secure wireless domains. In this paper, we discuss a way that even if a malicious user obtains a password, other users’ personal data cannot be delivered. This creates unique personal information based on a POS (Point of Sale) data, which is easily taken from a payment in a coffee shop or a restaurant, and transmits the data to an AP connected to the POS to create a user’s own SSID and a password as a secure virtual private AP. The user inputs the SSID and the password in the printed receipt from the POS into the wireless network and connects to the Internet so that the malicious user cannot access the user’s virtual AP unless the receipt is obtained.

2 Related Works 2.1

Background

The malicious user connects to the AP in order to collect personal information data of each user, easily connects to the AP without a password, collects the user’s data, even in APs with weak passwords and authentication methods, user data is collected by decrypting and connecting through packet data analysis [7–9]. However, as many studies have been conducted on encryption and authentication methods related to wireless APs, it has become difficult for a malicious user to connect to a wireless AP in a general manner and steal user data [13–15]. If this situation makes it difficult to collect personal information, a malicious user may think of a way to collect personal information after obtaining a password using a social engineering hacking method, rather than externally analyzing and accessing it. For example, in coffee shops and restaurants where people gather frequently and frequently, even if you set a password for the wireless AP, if the password is posted for everyone to see inside such as the SSID, a malicious user can easily obtain the password.

Individually Separated Wireless AP to Protect User’s Private Information

591

Although the password of the wireless AP is given one password per SSID, and the password is transmitted through a conversation rather than a posting in a coffee shop or a restaurant, it is inevitably forced to reveal the password, this can only be an environment where malicious users can easily obtain a password. Therefore, the individual SSID and password should be given to each user, so that even if a malicious user obtains a password and accesses the internal network, the personal information data of the user accessing the AP through the individual SSID should not be confirmed (Fig. 1).

Fig. 1. Problem of the current wireless network sharing the same password.

2.2

The Concept of Encryption of Internal Communication Data

The concept of encryption of internal communication data of wireless AP is that even if a user connects to AP using multiple SSIDs and passwords in one AP, each user cannot check each other’s data. This means that even if it is connected to the network, other users’ packets cannot be checked and collected [16–18]. It calculates the value given by the input device that is a specific subject, generates the SSID and password, and sends it to the user and the AP, the AP receives the SSID and password to create a user-only AP, a user enters a given SSID and password into an AP to access a user-only AP (Fig. 2).

592

M. Kim and K. Yim

Fig. 2. Wireless AP internal data encryption concept.

The concept is summarized as follows: (1) Create an SSID and password for creating a user-only AP in the input device serving as a specific subject. (2) Separate SSID and password are given for each user or group using wireless AP. (3) Even if a malicious user obtains a password and is connected to the AP, the personal information data of the user connected to the AP cannot be verified with a separate SSID and password.

3 Proposed User Data Protection Scheme in AP The internal user data protection scheme proposed in this paper is to create a virtual AP (user-only AP) in the AP and connect it to a given virtual AP for each device. It generates the SSID and password through the manager’s specific input device and delivers it to the AP, and the AP creates a virtual AP based on the given value, and the SSID and password information is output from the manager’s input device and delivered to the user, and the user inputs the output information to the device and is configured to connect to the virtual AP. 3.1

Environment and Scope

The environmental element of the proposed solution is to provide the wireless network only to users who have just paid for the sale of seats, objects, food, and drinks at their place, and detailed scope of these elements include coffee shops selling drinks such as coffee or restaurants serving food.

Individually Separated Wireless AP to Protect User’s Private Information

3.2

593

Subject Specific Input Device

The elements for generating the SSID and the password used when creating the virtual AP are specific input devices for calculating and paying for products to be sold, which are POS (Point of Sale) system used in restaurants and coffee shops. When the manager inputs the payment related information into the POS and confirms the payment progress and payment, the SSID and password are generated by calculating the payment related contents (item, quantity, price, payment time, etc.) inside the POS. 3.3

Method of Creating an SSID and the Password

In the POS system, the value is calculated using the payment-related information, and then the final calculated value is generated by using a hash function to generate a code (hash value), and the range of values to be used as the SSID and password in the code is specified. For example, in the generated hash code value, 1 to 25 digits are truncated, and then 1 to 20 digits are defined by the password, and 21 to 25 digits are defined by the SSID (Fig. 3).

Fig. 3. Flow chart about generating an SSID and the password.

3.4

Method of Creating a Virtual AP

When the code of the SSID and password generated in the POS system is transmitted to the AP, the AP creates a virtual AP based on the value and delivers the generation result to the POS.

594

3.5

M. Kim and K. Yim

Method of Delivery to User

When the POS system confirms that the virtual AP is generated, the generated SSID and password are added to the payment receipt and output. The printed receipt is sent to the user. 3.6

Method of Connecting a Virtual AP to User Device

The user checks the SSID and password of the receipt output from the POS system and enters the contents into the user’s device to connect to the virtual AP (Fig. 4).

Fig. 4. Virtual AP (user-dedicated AP) creation and connection.

3.7

Virtual AP Connectivity Scenarios for User Devices

This section implements and describes the scenario of connecting to the virtual AP from the user device based on the above description. This scenario consists of a user’s device, a manager’s POS system, a receipt printer, and a wireless AP. In the detailed configuration of the environment, the receipt machine is connected to the POS system, and the POS system and AP are connected to the wired network. (1) The user performs the payment process and payment through the order such as coffee or meal to the manager. (2) The manager enters the order details of the user at the POS system and proceeds with the payment. (3) The POS system calculates payment-related information entered by the manager, derives a value, and executes a hash function to generate a code (hash value).

Individually Separated Wireless AP to Protect User’s Private Information

595

(4) The POS system generates SSID and password values by separating the digits of the generated code and delivers them to the AP. (5) The AP creates a virtual AP based on the given SSID and password and sends the generated result to the POS system. (6) After confirming the virtual AP creation, the POS system outputs the generated SSID and password in the receipt. (7) The user receives the receipt, checks the SSID and password, and enters the corresponding information on the device to connect to the virtual AP.

4 Conclusion and Future Works In order to cope with external attacks, the wireless AP is continuously researching encryption technologies and authentication methods, as a result, a security environment is now established that cannot be easily attacked from the outside. However, due to the single password characteristic of the wireless AP, there is still a threat of internal information data leakage when a malicious user acquires and connects to the password of the wireless AP. In this paper, we propose a method to protect user privacy data through POS (Point of Sales) used in coffee shops or restaurants. When the SSID and password are generated through the POS system, a user’s individual AP is created based on this, and the user can access User-only AP and protect the user’s personal data from malicious users. In future research, we will implement the proposed method in this paper and test whether it can be executed in various environments. Acknowledgments. This research was supported by the MSIT (Ministry of Science and ICT), Korea, under the ITRC (Information Technology Research Center) support program (IITP-20192015-0-00403) supervised by the IITP (Institute for Information & communications Technology Planning & Evaluation) and the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIT) (NRF-2018R1A4A1025632).

References 1. Qualcomm: The Wi-Fi Evolution an integral part of the wireless landscape. https://www. qualcomm.com/media/documents/files/the-wi-fi-evolution-an-integral-part-of-the-wirelesslandscape.pdf 2. Telecom Advisory Services: The Economic Value of Wi-Fi: a Global View (2018 and 2023), October 2018. https://www.wifi.org/download.php?file=/sites/default/files/private/Economic %20Value%20of%20Wi-Fi%202018.pdf 3. Rojone: Wireless CPE, Access Points, Adapters Wireless Networking Tutorial. https://www. rojone.com/assets/files/L-Com-03-Wireless-CPE-Access-Points-Adapters.pdf 4. ABI Research: Global Wi-Fi Hotspots will Grow to 7.1 Million in 2015 as a Method to Offload Traffic. https://www.abiresearch.com/press/global-wi-fi-hotspots-will-grow-to-71million-in-2/. Accessed July 2019 5. Tomić, V., Stojanović, D.: Trends and innovations in mobile banking (2018). https://www. researchgate.net/publication/330468035_Trends_and_innovations_in_mobile_banking

596

M. Kim and K. Yim

6. Ganapati, S.: Using Mobile Apps in Government, Washington DC: IBM Center for the Business of Government (2015). http://www.businessofgovernment.org/sites/default/files/ Using%20Mobile%20Apps%20in%20Government.pdf 7. Maimon, D., Becker, M., Patil, S., Katz, J.: Self-protective behaviors over public WiFi networks. In: The Workshop: Learning from Authoritative Security Experiment Results, pp. 69–76 (2017). https://www.usenix.org/system/files/conference/laser2017/laser2017_ maimon.pdf 8. Public WiFi: The Hidden Dangers of Public WiFi (2014). https://www.privatewifi.com/wpcontent/uploads/2015/01/PWF_whitepaper_v6.pdf 9. Muppavarapu, R.: Open Wi-Fi hotspots-Threats and Mitigations (2014). https://dl. packetstormsecurity.net/papers/wireless/openwifimitigations.pdf 10. Norton: The risks of public Wi-Fi. https://us.norton.com/internetsecurity-privacy-risks-ofpublic-wi-fi.html. Accessed July 2019 11. Shahin, E.: Is WiFi worth it: the hidden dangers of public WiFi. Catholic Univ. J. Law Technol. 25(1), 7 (2017). https://scholarship.law.edu/cgi/viewcontent.cgi?article=1023&con text=jlt 12. Sombatruang, N., Sasse, M.A., Baddeley, M.: Why do people use unsecure public Wi-Fi?: an investigation of behaviour and factors driving decisions. In: Proceedings of the 6th Workshop on Socio-Technical Aspects in Security and Trust. ACM (2016). https://www. researchgate.net/publication/319694418_Why_do_people_use_unsecure_public_Wi-Fi_ An_investigation_of_behaviour_and_factors_driving_decisions 13. Kalniņš, R., Puriņš, J., Alksnis, G.: Security evaluation of wireless network access points. Appl. Comput. Syst. 21(1), 38–45 (2017). https://www.degruyter.com/downloadpdf/j/acss. 2017.21.issue-1/acss-2017-0005/acss-2017-0005.pdf 14. Zou, Y., Zhu, J., Wang, X., Hanzo, L.: A survey on wireless security: technical challenges, recent advances, and future trends. Proc. IEEE 104(9), 1727–1765 (2016). https://ieeexplore. ieee.org/stamp/stamp.jsp?arnumber=7467419 15. Alblwi, S., Shujaee, K.: A survey on wireless security protocol WPA2. In: Proceedings of the International Conference on Security and Management (SAM). The Steering Committee of The World Congress in Computer Science, Computer Engineering and Applied Computing (WorldComp) (2017). https://csce.ucmss.com/cr/books/2017/LFS/CSREA2017/ SAM3445.pdf 16. Xia, L., Kumar, S., Yang, X., Gopalakrishnan, P., Liu, Y., Schoenberg, S., Guo, X.: Virtual WiFi: bring virtualization from wired to wireless. In: ACM SIGPLAN Notices, vol. 46, no. 7, pp. 181–192. ACM, March 2011. http://citeseerx.ist.psu.edu/viewdoc/download?doi=10. 1.1.362.2789&rep=rep1&type=pdf 17. Lee, N.S., Lee, J.H., Jeong, C.K.: PS-Net: personalized secure Wi-Fi networks. J. Korean Inst. Commun. Inf. Sci. 40(3), 497–505 (2015). http://www.koreascience.or.kr/article/ JAKO201510665813311.page 18. Chandra, R., Bahl, P.: MultiNet: connecting to multiple IEEE 802.11 networks using a single wireless card. In: IEEE INFOCOM 2004, vol. 2, pp. 882–893. IEEE, March 2004. https:// ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=1356976

Long-Term Care (LTC) Monitoring System for Caregivers Based on Wireless Sensing Technology Hsing-Chung Chen1,2(&), Mei-He Jiang1, and Tzu-Ya Chen1 1

2

Department of Computer Science and Information Engineering, Asia University, Taichung City, Taiwan [email protected], [email protected] Department of Medical Research, China Medical University Hospital, China Medical University, Taichung City, Taiwan

Abstract. Continued decline in fertility rates and increased life expectancy cause serious population ageing in the world. As the older population increases, the demand of caregivers increases also, long-term care (LTC) is now to become a serious topic for each country who is facing this problem undoubtedly. In Taiwan, caregivers who need to work 16–24 h a day are maybe in the majority. Under the influence of prolonged over-exertion, high repetition or bad posture probably make caregivers in the risk of waist injuries. In this article, a LTC monitoring system for caregivers based on wireless sensing is designed. Through the change of waist bending rate, the system can remind caregivers to keep in a correct posture. In addition to this, caregivers can also track their own personal visualization chart via a mobile app at any time. Keywords: ThingSpeak IoT sensor


Long-term care (LTC)
Arduino UNO
Flex

1 Introduction Both the continued decline in fertility rates and the increased life expectancy cause serious population ageing in the world [1]. According to the World Population Ageing 2017: Highlights by United Nations, the population aged 60 or over in the world are 962 million in 2017, and that had been already more than twice as large as in 1980 [2]. In 2030, older people are expected to be more than children under age 10. In 2050, they whose age more than 60 years old will even be more than adolescents whose ages are from 10 to 24 [2]. As the older population increases, the demand of caregivers increases, too. Hence, the long-term care (LTC) is now to become a serious topic for each country who is facing this undoubted problem [2]. In Taiwan, the majority part within the total caregivers who need to work more than 16 h a day [3]. However, the caregivers have the problems of not only excessive working hours, but also heavy working contents. For example, their daily care works are preparing meals, feeding, and helping with displacement and bathing, etc. These over-exertion and high repetition works probably make the caregivers in the risk of © Springer Nature Switzerland AG 2020 L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 597–605, 2020. https://doi.org/10.1007/978-3-030-33506-9_54

598

H.-C. Chen et al.

waist injuries some time later, especially, for the older caregivers [4]. Under the high demand of caregivers, their physical conditions also needs to be evaluated to avoid losing health, due to the kind of works mentioned above, and make the caregivers become more and more shortage in serious. In this paper, a LTC monitoring system consisting of Arduino UNO, flex sensor and ESP8266-01s (Wi-Fi module) is designed and implemented, which could collect the bending value data of the monitored caregivers’ waist. The collected data generated by flex sensor implemented on Arduino UNO as well as a Wi-Fi module will be sent to ThingSpeak IoT for every two seconds. According to ThingSpeak IoT, all collected bending value data could be organized in personal chart, or download as a sheet in CSV format for the further analysis. The remainder of this paper is described below. Section 2 is related works, Sect. 3 talks about design of LTC Monitoring System Structure, Sect. 4 is the implementation results and analysis, and finally is our conclusions.

2 Related Works Musculoskeletal injuries have been identified as one of the major occupational injuries for caregivers in many industrialized countries, while in the musculoskeletal injury, the following back injuries are most common, followed by the hands, neck and shoulders [4]. In recent years, it has been found that the rate of musculoskeletal injuries in Taiwan has increased. According to the number of notified occupational diseases in the 104 national occupational injuries and diseases in Taiwan, 2,242, the second highest rate is 610 occupational musculoskeletal diseases, accounting for 27.2% [5]. In Taiwan’s 105 national occupational injury diagnosis and treatment network occupational diseases, the number of notifications was 2,242, the second highest rate was 534 occupational musculoskeletal diseases, accounting for 20.7% [6]. In 2018, there were 2,242 notifications of occupational diseases in the national occupational injury and diagnosis network. The highest rate was 813 occupational musculoskeletal diseases, accounting for 37.7% [7]. In addition, ThingSpeak was originally launched by ioBridge in 2010 as a service in support of IoT applications [19]. ThingSpeak is an open-source Internet of Things (IoT) application and API to store and retrieve data from things using the HTTP protocol over the Internet or via a Local Area Network [20]. It could enable the creation of sensor logging applications, location tracking applications, and a social network of things with status updates [20].

3 LTC Monitoring System Structure First, the tools we use to implement are drew up in Subsect. 3.1. Second, the system setup procedure is formulated in Subsect. 3.2. The LTC monitoring system structure in this paper is shown in Fig. 1.

Long-Term Care (LTC) Monitoring System for Caregivers

599

Fig. 1. Structure

3.1

Tools for System

a. Arduino UNO [8] Arduino UNO is an open-source microcontroller board developed by Arduino.cc, and it can be powered by a USB cable or by an external 9 V battery. For the purpose of convenient and safe use with caregivers, the Arduino UNO is powered by battery in this study. b. ESP8266-01 s [9] Considering that caregivers need to walk around often during the work, so the sensor value data should be sent by wireless. ESP8266-01s Wi-Fi module can reach this goal, and it can both sense and upload the data simultaneously. c. Flex Sensor [10] Flex sensor is known also bend sensor. The principle behind flex sensor is that when the sensing metal is bent outward, the resistance of it will change. According to the change resistance, the bend rate of sensing metal can be computed. d. ThingSpeak (Database) [11] ThingSpeak is a platform specially designed for IoT and developed by MathWorks. It provide user to upload real-time data to cloud and create the visualization chart or raw data sheet base on all collected data. e. App Inventor2 (Mobile Device) [12] App Inventor is an open-source web application with graphical interface, it provide user to create an android application by drag and drop objects. In this study, the android application designed by App Inventor 2 can allow caregivers to track their own waist bending rate anytime and remind them to take a break duly. For the selfdata protection, caregivers need to login with account and password to read their own data. 3.2

System Setup Procedure

The system setup procedure is formulated below.

600

H.-C. Chen et al.

System Setup Procedure Step:1. Choose tools in Subsect. 3.1. Use Arduino UNO as a hardware device and ThingSpeak for data receiving and storage. ThingSpeak can analyse the received data by C, C++ and Java, and then generates the results [13, 14] Step:2. Test whether the flex sensor can work in normal. Use Arduino UNO and flex sensor (see Fig. 2). Test whether the flex sensor can generate different values base on the increased or decreased resistance on Arduino IDE.

Fig. 2. Circuit diagram of Arduino and flex sensor

Step: 3. Test and change the baud of ESP8266-01s. If the initially baud of ESP8266-01s is set to 115200, change it to 9600. If the initially baud is 9600, nothing to do. ESP8266-01s needs to connect with FT232 and computer first, and the VCC of it is 3.3 V, not 5 V (see Fig. 3a). To change the baud of ESP8266-01s need to use RealTerm (Fig. 3b) [15, 16]. (a)

Fig. 3. a: ESP8266-01s connect with FT232; b: Realterm

(b)

Long-Term Care (LTC) Monitoring System for Caregivers

601

Step: 4. Setup ThingSpeak and channel. Go to the official website of ThingSpeak (Fig. 4a), create a new account and setup the channel to receive data (Fig. 4b).

(a)

(b)

Fig. 4. a: ThingSpeak Website; b: Setup Channel

Step: 5. Combine flex sensor, ESP8266-01s and ThingSpeak. After finishing circuit connection (Fig. 5a), unplug the TX and RX when program is burning and re-insert them when burning is end. Then test whether channel of ThingSpeak receive the data and show them on the chart when the flex sensor is bend or no bend (Fig. 5b) [17].

(a)

(b)

Fig. 5. a: Circuit diagram of ESP8266-01s and flex sensor; b: Chart of ThingSpeak

602

H.-C. Chen et al.

Step: 6. Design the app which could show the personal bending chart after user (caregiver) login. Check the personal chart of caregiver’s waist bending rate at any time after user login by the mobile app (Fig. 6a and b) [18].

(a)

(b)

Fig. 6. a: Login interface; b: Personal chart of caregiver’s waist bending rate

▓

4 Results and Analysis The result of caregiver’s waist bending rate can be shown in the visualization chart (Fig. 7) after ESP8266-01s receiving the data from Arduino [13]. The personal chart can let user know their waist bending situation. Use the sheet in ThingSpeak to calculate and check if any bending rate is over the normal reference. The normal waist bending rate is around 76*83. In the system, it will warn user when the value is lower than 75 (Table 1). When the value is lower than 60, it means that the user’s waist is bent at 90°.

Long-Term Care (LTC) Monitoring System for Caregivers

Fig. 7. Chart of ThingSpeak Table 1. The collected data

created_at 2018-12-19 14:58:03 2018-12-19 14:58:24 2018-12-19 14:58:45 2018-12-19 14:59:07 2018-12-19 14:59:28 2018-12-19 14:59:49 2018-12-19 15:00:10 2018-12-19 15:00:31 2018-12-19 15:00:59 2018-12-19 15:01:20 2018-12-19 15:01:41 2018-12-19 15:02:02 2018-12-19 15:02:23 2018-12-19 15:02:44 2018-12-19 15:03:05 2018-12-19 15:03:26 2018-12-19 15:03:47 2018-12-19 15:04:07 2018-12-19 15:04:29 2018-12-19 15:04:50 … 2018-12-19 15:15:18 2018-12-19 15:15:39

entry_id 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 … 50 51

field1 82 82 82 72 81 81 81 82 82 82 82 82 82 82 71 65 68 67 82 82 … 82 82

Value 82 82 82 72 81 81 81 82 82 82 82 82 82 82 71 65 68 67 82 82 … 82 82

Value < 75 None None None Warning None None None None None None None None None None Warning Warning Warning Warning None None … None None

603

604

H.-C. Chen et al.

5 Conclusion In this paper, a LTC monitoring system for caregivers based on wireless sensing is designed. By the system, caregivers can check their own waist bending rate in any time, even they are too busy to do that, the system can also remind them to keep in correct posture when needed. Caregivers in addition to take care of others, also should pay attention to their physical conditions. If the professional injury of the caregiver can be reduced, the quality of care will also increase. Hope there will be more useful function combine with this system in the future. For example: Remind caregivers which subject is now need to do something? Become one of the integrated system in hospital, and so on. Then, through the systematic management in hospital, the problem of resource shortage can decrease and the quality of medical can be upgraded.

References 1. W. H. Organization. Population ageing. https://www.who.int/features/qa/72/en/ 2. United Nations and Department of Economic and Social Affairs: World population ageing 2017: highlights, ed: United Nations, New York (2017) 3. Chen, S.-H., Liu, H.-E., Li, C.-L., Kao, C.-Y.: An exploration of quality of life and related factors in foreign nurse aides. J. Health Sci. 14(1), 57–68 (2012) 4. Chen, M.: Caring for the musculoskeletal injury of the waiter. Labor Saf. Health Newslett. 83, 22–23 (2007). 200706 5. Department of Occupational Safety and Health, Ministry of Labor. National occupational injury diagnosis and treatment network occupational disease notification statistics, 25 December 2018. https://www.osha.gov.tw/1106/1113/1114/16681/ 6. Department of Occupational Safety and Health of the Ministry of Labor. Statistics on the National Occupational Diseases and Diseases Network Occupational Diseases, 25 December 2018. https://www.osha.gov.tw/1106/1113/1114/18497/ 7. Department of Occupational Safety and Health of the Ministry of Labor. Statistics on the National Occupational Diseases Diagnosis and Treatment Network Occupational Diseases, 25 December 2018. https://www.osha.gov.tw/1106/1113/1114/24256/ 8. ARDUINO. Arduino Products, 25 December 2018. https://www.arduino.cc/en/Main/ Products 9. Robot, P.: Arduino ESP8266 Serial Wi-Fi module (ESP-01) (CGGs), 25 December 2018. https://www.playrobot.com/arduino/711-arduino-esp8266-wifi-esp01.html 10. TAIWANNIOT. Flex Sensor SpectraSymbol 4.5 Bending sensor SPARKFUN imported, 25 December 2018. https://www.taiwaniot.com.tw/product/flex-sensor-4-5-sparkfun/ 11. T. MathWorks. ThingSpeak, 25 December 2018. https://thingspeak.com/ 12. MIT App Inventor 2. App Inventor 2, 25 December 2018. http://ai2.appinventor.mit.edu/? locale=en 13. A. Wikipedia. Arduino. https://zh.wikipedia.org/wiki/Arduino 14. Yen, T.: [Arduino Starter Kit 02] Arduino Uno Board 的各個部位 – Parts of Arduino Uno Board, 25 December 2013/2018. http://thats-worth.blogspot.com/2013/12/arduino-unoboard-parts-of-arduino-uno.html 15. Two small pigs. *Teaching*Arduino Wi-Fi (ESP8266) First time application, 25 December 2015/2018. http://lolwarden.pixnet.net/blog/post/82031214-%2A%E6%95%99%E5%AD%

Long-Term Care (LTC) Monitoring System for Caregivers

16. 17. 18. 19. 20.

605

B8%2Aarduino-wifi%28esp8266%29-%E5%88%9D%E6%AC%A1%E4%B8%8A%E6% 89%8B%E6%87%89%E7%94%A8 DK. [Arduino] The confusion of Serial port, 25 December 2015/2018. http://eva54185418. blogspot.com/2013/01/arduino-serial-port.html P. C. (Painting the rain on the banana leaf). Talking about IoT and ThingSpeak.com, 25 December 2015/2018. http://pizgchen.blogspot.com/2015/07/iot-thingspeakcom.html CAVEDU Education Team. About App Inventor, 25 December 2018. http://www. appinventor.tw/whatis/ Llawlor: ThingSpeak, 29 October 2014. https://github.com/iobridge/thingspeak/blob/master/ README.textile Hans: ThingSpeak Plus Third-party Hardware – an Alternative to ioBridge Hardware, ioBridge Blog Post, 19 December 2014

The 10th International Workshop on Methods, Analysis and Protocols for Wireless Communication (MAPWC2019)

Concatenated Path Domain for Dijkstra’s Algorithm Based Ray Tracing to Enhance Computational Areas Kazunori Uchida(B) and Leonard Barolli Department of Information and Communication Engineering, Fukuoka Institute of Technology (FIT), 3-30-1 Wajiro-Higashi, Higashi-Ku, Fukuoka 811-0295, Japan {k-uchida,barolli}@fit.ac.jp

Abstract. This paper provides a simulation method which can apply the Dijkstra’s algorithm (DA) based ray tracing to a large size of random rough surface (RRS). Since the RRS and path concatenations are performed, we can deal with this difficult problem even with a small size of personal computer (PC). By using the convolution method to generate 3D RRSs, concatenation of two adjacent RRSs is possible by keeping the random variables with respect to the 2D conjunction area between them. Concatenation of traced rays can also be executed by keeping the path data at the 2D conjunction area. First we start ray tracing for the first RRS with a source node, and next, keeping the 2D path data at the conjunction area, we move to the second RRS to execute ray tracing. We repeat this procedure until ray tracing for the last RRS is finished. All paths computed by the present method constitute shortest paths. However, the shortest paths thus obtained are different from the optical rays, and consequently three path modifications, path-linearization, path-selection and line of sight (LOS)-check, are required. Numerical examples reveal that the proposed concatenation method is an effective tool for a small size of PC to execute ray tracing along a large scale of RRS. Keywords: Concatenated computational domain · Random rough surface · Dijkstra’s algorithm · Ray tracing · Wave prop

1 Introduction Dijkstra algorithm (DA) is an effective graph search algorithm to solve the shortest paths from a source node to other destination nodes of a graph with proper link costs between connected nodes [1, 2]. This algorithm has often been used not only in graph theory but also in many other applications such as operations research (OR), management science, transportation, and geographic information system (GIS) [3]. As is well known, wavefront movements are governed by the Fermat’s principle [4] which tells us that an optical ray from source to destination takes the path that can be traversed in the least time. Considering ray traveling time as link cost, we can simulate various types of wave propagation by DA-based ray tracing methods [5–7]. c Springer Nature Switzerland AG 2020  L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 609–620, 2020. https://doi.org/10.1007/978-3-030-33506-9_55

610

K. Uchida and L. Barolli

In relation to radio communications for sensor networks, we have applied the DAbased ray tracing method to simulations of the wave propagation in complicated structures such as random rough surface (RRS) and others [8, 9]. The main purpose of these investigations is to clarify the statistical relationship between radio wave coverages and propagation environments by computing ensemble average of propagation characteristics. In numerical simulations, however, it is obvious that the larger RRS size is, the more reliable statistical results may be obtained. In line with this context, we should enlarge the computational areas by introducing concatenated domains for RRSs and shortest paths. First, we discuss the concatenation of RRSs. In this paper, we use the convolution method to generate the Gaussian type of RRS in Cartesian coordinate systems (x, y) [11]. Due to its analytical property, we can easily concatenate RRSs with arbitrary number of sections in x-direction. Assuming the node numbers which can be processed by a PC are (Nx , Ny , Nz ) in the Cartesian coordinate systems, the resultant concatenated node numbers are (M × Nx , Ny , Nz ) where M is a concatenation number. Since M is arbitrary, we can deal with a fairly large size of RRS even by using a small size of PC, if we choose large numbers for Ny and Nz together with M(> 1). Second, we consider concatenation of shortest paths along RRS, since DA based ray-tracing is a useful tool to simulate electromagnetic (EM) propagation in complicated structures. In order to increase the accuracy of the method, three procedures, path-selection, path-linearization and line of sight (LOS) check, are necessary to convert the original shortest paths to optical rays so that we can simulate wave propagation in high frequency regions [8, 9]. In this paper, we describe an essence of the DA based ray tracing together with the three procedures to modify the original DA based shortest paths to the realistic optical rays. We also discuss the concatenation of the original shortest paths as well as the modified shortest path between the adjacent two concatenated RRSs. Thus, the present DA-based ray tracing can be applied to a large size of RRS even with a small size of PC. Third, we consider propagation order of distance β of EM waves traveling along RRS. This parameter was fist introduced by Hata to estimate path loss of VHF and UHF waves in urban ares [12]. The essence of this parameter is described such that EM waves decay as |E| ∼ r−1 in the free space, but they decay as |E| ∼ r−β with β > 1 in the urban areas, and the parameter β is given by the simple empirical equations [12]. These results were derived from many experimental data [13]. We have also discussed the same problem from a simulation view point by investigating wave propagation along RRS. In this paper, concatenation of β is also discussed for concatenated RRS in order to enlarge the computational areas. The first section of this paper is an introduction. The second section discusses concatenation of RRS. The third section describes the essential parts of DA procedures. The fourth section shows concatenation of paths concerning original and modified shortest paths. The fifth section is a conclusion of this paper.

2 RRS Concatenation Starting from the direct RRS generation based on discrete Fourier transformation (DFT) [10], we have devised the analytically expressed convolution method for generating

Concatenated Path Domain for Dijkstra’s Algorithm

611

inhomogeneous RRS [11]. In this section, we consider only the 2D RRS with Gaussian type of spectrum. The amplitude weight function h(rr ) of the Gaussian type of spectrum is given as follows: √     r 2 4 π dv exp −2 (1) h(rr ) = cl cl where r = |rr | is the amplitude of position vector r = (x, y). Moreover, dv is the height standard deviation of RRS and cl is its correlation length. Since the amplitude weighting function is given analytically in Eq. (1), we can easily discretize it as follows:

Δ h(μΔ , νΔ ) (2) h˜ μν = 2π where Δ is a discretized interval in x and y directions. Moreover, RRS is truncated by truncation number L such as μ , ν = 0, ±1, ±2, · · · , ±L. It is worth noting that numerical condition Δ ≤ c/5 is enough for the discretized interval to obtain a smooth RRS profile. Consequently, we just choose the truncation number as L ≥ 5 depending on the correlation length c. Based on the convolution method, expression for 2D RRS generation is summarized as follows: fi j =

L

L

∑ ∑

μ =−L ν =−L

h˜ μν gi+μ

j+ν

(3)

where fi j denotes the discrete RRS profile at (x, y) = (iΔ , jΔ ) for i = 0, 1, 2, · · · , Nx − 1 and j = 0, 1, 2, · · · , Ny − 1. Moreover, gi+μ j+ν are the 2D series of the Gaussian random number with zero mean and unit deviation, that is gi+μ

j+ν

∈ N(0, 1)

(4)

where N(0, 1) indicates a set of random variables of Gaussian or normal distribution. Equation (3) indicates that the number of points of discrete RRS profile is given by Nx × Ny and the total number of generated random variables is given by (Nx + 2L + 1) × (Ny + 2L + 1). Now, we keep the random variables corresponding to x = (Nx − 1)Δ , whose total number amounts to (2L + 1) × (Ny + 2L + 1), in order to concatenate RRS. Then, using the stored random variables, we start to generate another RRS based on Eq. (3). It is evident from analytic nature of the convolution method that the end surface at x = (Nx − 1)Δ of the first RRS is smoothly matched to the beginning surface at x = 0 of the next RRS. Thus, we can concatenate multiple elements of RRS in x direction successively, and thus we can generate large size of discrete RRS with points M × Nx × Ny where M ≥ 1 is a concatenation number. In Fig. 1 we show an example of concatenated RRS where the concatenation number is chosen as M = 3. In these numerical computations we have made height shift by fi j → fi j + 3dv in order to obtain positive values for the height of generated RRS. Since size of each figure is (Nx , Ny ) = (120, 40), total size of the concatenated RRS amounts to (Nx , Ny ) = (120, 120). In this way, concatenation enables us to deal with large size of RRS even by a small size of PC.

612

K. Uchida and L. Barolli

Fig. 1. Concatenated RRSs with M = 3 for dv = 3 and cl = 3.

3 Dijkstra’s Algorithm As discussed in the preceding section, (x, y) plane of RRS has been discretized into Nx × Ny points or nodes. Similarly, the RRS height in Eq. (3) can be discretized into Nz points. As a result, nodes of the graph corresponding to a 3D RRS are composed of eight apexes of cubes as follows: (xi , y j , zk ) = (Δ i, Δ j, Δ k)

(5)

where i = 0, 1, 2, · · · , Nx − 1, j = 0, 1, 2, · · · , Ny − 1 and k = 0, 1, 2, · · · , Nz − 1. Assume Δ = 1, we may express the node at (xi , y j , zk ) by N(i, j, k). It should be noted that each node constitutes the cube of which diagonal line is from the node N(i, j, k) to node N(i + 1, j + 1, k + 1). 3.1

Node Type and Link Type

We consider node categorizations from geometrical point of view. Since the node N(i, j, k) is surrounded by eight cubes generated by eight nodes N(i − 1 + ic , j − 1 + jc , k − 1 + kc ) where ic , jc , kc = 0 or 1, and thus we can assign each cube a unique number  = ic + 2 jc + 4kc . Since the geometry of a node is characterized by the combinations of eight cubes that are filled with free space or obstacle, the number of node types characterizing node geometries amounts to 28 = 256. Thus, we can categorize any node by the node-type number m defined by m = b0 + 2b1 + 4b2 + 8b3 + 16b4 + 32b5 + 64b6 + 128b7

(6)

where  = 0, 1, ..., 7 and b = 1 or 0 depending on whether the -th cube is occupied by obstacle or free space. In this paper, we denote a set of nodes by placing a tilde over the symbol. The set of all nodes, for example, is expressed as N˜ nod and the number of elements of this set

Concatenated Path Domain for Dijkstra’s Algorithm

613

is given by Nnod = Nx × Ny × Nz and it is classified into three types from a geometrical viewpoint. One is the free-space node set N˜ f re of which adjacent eight cubes are all freespace with node-type number m = 0. Second is the inner node set N˜ inn of which adjacent eight cubes are all obstacle with node-type number m = 255. Third is the surface node set N˜ sur of which adjacent eight cubes are combinations of free-space and obstacle nodes with node-type number 0 < m < 255. Now we consider links between adjacent two nodes N(i, j, k) and N(i , j , k ) where  i = i + i0 , j = j + j0 and k = k + k0 for i0 , j0 , k0 = −1, 0 or 1. Consequently, there exist twenty-seven types of links from one node, and each link is assigned by the link-type number defined by (7) n = (i0 + 1) + 3( j0 + 1) + 9(k0 + 1). For example, the link corresponding to i0 = j0 = k0 = 0 is a self-loop with zero link-cost and link-type number n = 13. Next, we assign link-cost to each link between the adjacent two nodes in terms of their distance as follows:  i20 + j02 + k02 (connected)    (8) C(i, j, k; i , j , k ) = ∞ (disconnected). √ It is √ evident that there are three types of link-costs for the connected links, that is, 1, 2 and 3. 3.2 Initialization ˜ j, k) as In this paper, we denote the connected-node set of a node N(i, j, k) by D(i, follows: ˜ j, k) = {N(i , j , k )|∞ > C(i, j, k; i , j , k ) > 0}. D(i, (9) ˜ ˜ From a computational viewpoint, we classify the node set Nnod into fixed-node set N f ix , candidate-node set N˜ can and another one. The candidate-node set is defined by the nodes which are connected to nodes in the fixed-node set. At the initial stage, the fixed-node set is the source node N(i0 , j0 , k0 ) as follows: N˜ f ix = {N(i0 , j0 , k0 )}

(10)

and the candidate node set is the connected node set of the source node as follows: ˜ 0 , j0 , k0 ). N˜ can = D(i

(11)

Now, we express the total link-cost of a destination node N(i, j, k) by the scaler matrix Ct (i, j, k) which indicates the total sum of link-costs of a path from the source node to the destination node. We also denote the proximity node of the destination node ¯ j, k) to which the destination node is linked backwardly. At the by the dyadic matrix P(i, initial stage, the total link-cost matrix and proximity node matrix are apparently given by Ct (i0 , j0 , k0 ) = 0 (12) ¯ 0 , j0 , k0 ) = N(i0 , j0 , k0 ). P(i It should be noted that the total link-cost of a source node is zero and its proximity node is itself.

614

3.3

K. Uchida and L. Barolli

Repetition

We can determine fixed-nodes step by step in the following way. First, out of the candidate-node set, we find the node N(, m, n) which has the minimum total link-cost as follows: min Ct (i, j, k). Ct (, m, n) = (13) N(i, j,k)∈N˜ can

We select this node as a new fixed-node, and we renew the fixed-node set as follows: N˜ f ix = N˜ f ix ∪ {N(, m, n)}.

(14)

We also renew the candidate-node set by adding the nodes connected to the newly selected fixed-node as follows: ˜ m, n) \ N˜ f ix ]. N˜ can = [N˜ can \ {N(, m, n)}] ∪ [D(,

(15)

Moreover, we modify the total link-cost and proximity-node matrices of the nodes connected to the newly selected fixed node in the following way. If Ct (, m, n) + C(, m, n; i, j, k) < Ct (i, j, k) holds, we surely renew the total link-cost matrix and proximity-node matrix as follows: Ct (i, j, k) = Ct (, m, n) +C(, m, n; i, j, k) ¯ j, k) = N(, m, n). P(i,

(16)

If Ct (i, j, k) = Ct (, m, n) + C(, m, n; i, j, k) holds, we may renew only the proximitynode matrix as follows: ¯ j, k) = N(, m, n). P(i, (17) We call this process as proximity-node selection which is performed among the proximity nodes with an equal total link-cost. The former renewals in Eq. (16) are mandatory in order to obtain the unique solution for the connected nodes having the minimum total link-costs. However, the latter renewal in Eq. (17) is optional, because it has no influence on distribution of the minimum total link-costs. In particular, this renewal is useful when we want to select a specific path out of multiple shortest paths having the same total link-cost. We can thus select one desirable shortest path by controlling choice of proximity node in Eq. (17). 3.4

Ending

Repeat the above mentioned repetition process until there exist no candidate nodes, that is, N˜ can = 0. / Then, computations are completed with the repetition times Nrep < Nnod , and we obtain the total link-cost matrix Ct (i, j, k) as well as the proximity-node ¯ j, k) for the fixed-node set. Thus, we can construct shortest paths from any matrix P(i, destination nodes to the source node by tracing the proximity-node matrix backwardly.

Concatenated Path Domain for Dijkstra’s Algorithm

615

3.5 Shortest Path Now we define the position vector of a node constructing a shortest path as follows: r n = (in , jn , kn )

(18)

where n = 0, 1, · · · , N. The source node is at r 0 and the destination node is at r N . Then, the shortest path can be traced from the proximity matrix by the following recurrence relations: ¯ r n ). r n−1 = P(r (19) As a result, total link-cost or path-length of the shortest path is computed as follows: R=

N

∑ |rr n − r n−1 |.

(20)

n=1

4 Path Concatenation In this section, we consider path concatenation in relation to the RRS concatenation discussed in the preceding section. Numerical examples to depict path concatenation are given for the RRS examples shown in Fig. 1. Computational size of each RRS is chosen as (Nx , Ny , Nz ) = (40, 120, 30), and thus the concatenated RRS size amounts to (Nx , Ny , Nz ) = (120, 120, 30). The source node is at (i0 , j0 , k0 ) = (0, 60, 21) in the first RRS with a height of unit length above its surface node. Destination nodes are also unit length above the surface nodes of first, second or third RRS. 4.1 Original Shortest Path As described in the preceding section, DA data enable us to obtain the shortest paths from any destination nodes to the source node by tracing the proximity-node matrix recurrently. This fact implies that we can concatenate shortest paths with respect to the first RRS to the next one, only if we match the cost-matrices and proximity-matrices at the connecting nodes between the two concatenated RRSs. It is worth noting that DA data to be stored for the path concatenation are 2D ones corresponding to the aperture nodes which connect the two concatenated RRSs. Figure 2 shows the concatenated distributions of shortest path-length computed by Eq. (20) based on the original DA data. It is found from the numerical results that the distributions of path-length are as smoothly concatenated as the case of concatenated RRSs. In order to check the accuracy of the concatenation process, we show distributions of path-length in a different 2D way at the cross section of (i, j0 , k) as shown in Fig. 3. It is well demonstrated that path-lengths of the first RRS are smoothly continued to the next RRS. Thus, the path concatenation seems to be complete as RRS concatenation does. However, it is shown that the original data for shortest paths do not accurately simulate wave propagation where contours of shortest path constitute spherical wave fronts in the free space.

616

K. Uchida and L. Barolli

Fig. 2. Concatenated distributions of original path-lengths at unit length above RRS with with source node at i = i0 = 0 and, j = j0 = 60.

Fig. 3. Concatenated distributions of original path-lengths in the cross sections of j = j0 = 60 with source node at i = i0 = 0 and, k = k0 = 21.

As a result, the paths obtained straightforwardly from the proximity-node matrix exhibit piecewise-linear lines which are far from the optical rays. In order to construct optical rays from the original shortest paths, we should execute three types of numerical procedures, path-selection, path-linearization and LOS-check [8, 9]. 4.2

Modified Shortest Path

We discuss modified shortest path which means that an original shortest path has been modified by the path-selection, path-linearization and LOS-check. The first two procedures are executed in the frame of DA-computations described in the preceding section. Path-selection is processed by selecting diffraction nodes, such as apex or wedge nodes, as priorities for the proximity-node selection in Eq. (17). In this way, we can select exclusively a path passing through apex or wedge nodes. This process can be executed at each DA repetition. Path-linearization is to linearize a path in the free space region by modifying the proximity-node matrix at free-space nodes so that the proximity-node matrix of a free-space node is replaced by the backwardly next proximity node of the path. This process can also be executed at each DA-repetition. The final procedure LOS-check, however, cannot be executed completely within the one frame of DA-computations which are closely related to each RRS. Consequently, we should pay a careful attention to the concatenation of modified paths which were

Concatenated Path Domain for Dijkstra’s Algorithm

617

Fig. 4. Distributions of concatenated modified path-lengths at unit length above RRS with source node at i = i0 = 0 and j = j0 = 60.

Fig. 5. Concatenated distributions of modified path-lengths in the cross sections of j = j0 = 60 with source node at i = i0 = 0 and k = k0 = 21.

computed with respect to each RRS. We propose a method which can execute path concatenation effectively by using the path data of the former RRS. It should be noted that these paths must pass through the aperture nodes which connect the two adjacent RRSs. Figure 4 shows concatenated distributions of modified shortest path-lengths from the source node in the first RRS to the destination nodes in the concatenated RRSs. The destination nodes are unit length above the surface nodes of the concatenated RRSs shown in Fig. 1. It is found from the numerical results that distributions of path-length are concatenated as smoothly as the RRS concatenation. However, there exist few differences between the original data in Fig. 2 and the modified data in Fig. 4. We also show distributions of modified path-length in a 2D way for the nodes at (i, j0 , k) as shown in Fig. 5. It is well demonstrated that modified path-lengths are smoothly continued to the next RRS and thus path-concatenation is complete. However, there exist a distinct difference between the original data in Fig. 3 and the modified data in Fig. 5. The main reason is the wave front movements, that is, in a polyhedron way for the original data and in a spherical way for the modified data. Anyway, we can conclude that the concept of concatenation is an effective idea for attacking the problem of ray-tracing with respect to a large size of RRS by use of a small size of PC.

618

4.3

K. Uchida and L. Barolli

Propagation Order of Distance

So far, we have reported that the propagation oder of distance β is an important quantity for EM wave propagation along RRS. The parameter β was fist introduced by Hata to estimate path losses of VHF and UHF EM waves in urban ares for mobile communications. Essence is described such that EM waves decay as |E| ∼ r−1 in the free space, but they decay as |E| ∼ r−β with β > 1 in the urban areas. Moreover, the parameter β are given in a simple form of empirical equations [12, 13]. In this paper, from a simulation point of view, we discuss a simulation method for β estimation and show some numerical examples. In the preceding subsections, we have numerically shown that path concatenation is possible as well as RRS concatenation. It is evident that concatenation of the other physical quantities derived from shortest paths are also possible by utilizing the data of nodes which connect two adjacent RRSs. Now we consider the relation between the propagation oder of distance β and the optical rays along RRS. Taking into account of diffraction effects by introducing a new diffraction coefficient, we define the equivalent distance of an optical ray as follows [14]: Re = |rr 1 − r 0 | +

∑N−1 n=1 |rr n+1 − r n | ∏N−1 n=1 |G(Xn )|

(21)

where the approximate diffraction coefficient is defined by π

e− 4 j G(Xn ) = √ π 2 π Xn + e− 4 j  Xn = κ |rr n+1 − r n |(1 − cos θn ) (rr n+1 − r n ) · (rr n − r n−1 ) cos θn = |rr n+1 − r n ||rr n − r n−1 |

(22)

where κ is the wave number in the free space. It should be noted that above relations have been obtained asymptotically from the rigorous diffraction coefficient given by complex type of Fresnel function [15]. Consequently, we can estimate propagation order of distance β as follows:

β = log Re / log R

(23)

where R is the path-length given by Eq. (20) or the direct distance from source node to destination given by |rr N − r 0 |. Figure 6 shows concatenated distributions of the propagation order of distance β at the nodes which are unit length above the concatenated RRSs shown in Fig. 1. It is found that the parameter β can also be smoothly concatenated as in the case of RRS concatenation. We also show β distributions in a 2D way with respect to the nodes at (i, j0 , k) as shown in Fig. 7. It is well demonstrated that order of propagation is also smoothly continued to the next RRS and thus β concatenation seems to be complete. As a result, we can conclude that the concept of concatenation is an effective idea for computing physical quantities related to the shortest paths along RRS.

Concatenated Path Domain for Dijkstra’s Algorithm

619

Fig. 6. Distributions of concatenated modified path-lengths at unit length above RRS with source node at i = i0 = 0 and j = j0 = 53.

Fig. 7. Concatenated distributions of modified path-lengths in the cross sections of j = j0 = 60 with source node at i = i0 = 0 and k = k0 = 5.

5 Conclusion We have proposed some concatenation methods in order to enlarge application areas of DA based ray tracing for solving propagation characteristics along RRS by using a small size of PC. The essence of the proposed concatenation is that the 2D DA data with respect to the nodes connecting two adjacent RRSs are stored and handed to the next DA computations. Consequently, memories and computation time required for a PC are considerably small. In the numerical simulations, it has been demonstrated that the present concatenation is effective to computations of the original shortest paths, modified shortest paths and propagation order of distances. Applications of the proposed method to the investigation of propagation characteristics along a big scale of RRS by using a small size of PC deserves as a future problem.

References 1. Dijkstra, E.W.: A note on two problems in connexion with graphs. Numerische Mathematlk 1, 269–271 (1959) 2. Deo, N.: Graph Theory with Applications to Engineering and Computer Science, pp. 268– 327. Prentice-Hall Inc, Englewood Cliffs (1974). ISBN 0-13-363473-6 3. Zhan, F.B., Noon, C.E.: Shortest path algorithms: an evaluation using real road networks. Transp. Sci. 32(1), 65–73 (1998). https://doi.org/10.1287/trsc.32.1.65

620

K. Uchida and L. Barolli

4. Schuster, S.A.: An Introduction to the Theory of Optics. E. Arnold, London (1904). Chapter III 5. Uchida, K., Nogami, S., Takematsu, M., Honda, J.: Tsunami simulation based on dijkstra algorithm. In: 2014 International Conference on Network-Based Information Systems, pp. 114–119. Salerno, Italy (2014) 6. Uchida, K.: Discrete ray tracing based on Dijkstra Algorithm for diffraction and reflection. Int. J. Microwave Opt. Technol. 10(6), 377–384 (2015) 7. Wu, Y., Li, X.: Numerical simulation of the propagation of hydraulic and natural fracture using Dijkstra’s Algorithm. Energies 9, 519 (2016). https://doi.org/10.3390/en9070519 8. Uchida, K., Barolli, L.: Dijkstra-Algorithm based ray-tracing by controlling proximity node mapping. In: IEEE 31st International Conference on Advanced Information Networking and Applications Workshops (IEEE WAINA 2017), Taipei, Taiwan, pp.189-194, March 2017 9. Uchida, K., Leonard, B.: Dijkstra Algorithm based ray tracing: a case study for tunnel structures. In: The 32nd IEEE International Conference on Advanced Information Networking and Applications, AINA 2018, Cracow, Poland, May 2018 10. Thoros, E.I.: The validity of the Kirchhoff approximation for rough surface scattering using a Gaussian roughness spectrum. J. Acoust. Soc. Am. 83(1), 78–92 (1988) 11. Uchida, K., Takematsu, M., Lee, J.H., Shigetomi, K., Honda, J.: An analytic procedure to generate inhomogeneous random rough surface. In: The 16th International Conference on Network-Based Information Systems, NBiS 2013, pp. 494–501, Gwangju, Korea (2013) 12. Hata, M.: Empirical formula for propagation loss in land mobile radio services. IEEE Trans. Veh. Technol. VT–29(3), 317–325 (1980) 13. Okumura, Y., et al.: Field strength and its variability in UHF and VHF land-mobile radio service. Rev. Elec. Commun. Lab. 16, 825–873 (1986) 14. Uchida, K.: Path loss along random rough surface in terms of propagation order of distance. In: 8th Asia-Pacific Conference of Antenna and Propagation (APCAP2019), Incheon, Korea (2019) 15. Noble, B.: Methods Based on the Wiener-Hopf Technique. Pergamon Press, London (1958)

Routing of Optical Baseband Signal Depending on Wavelength in Periodic Structure Naoki Higashinaka1 and Hiroshi Maeda2(&) 1

Graduate School of Communication and Information Networking, Fukuoka Institute of Technology, Fukuoka, Japan [email protected] 2 Department of Information and Communication Engineering, Fukuoka Institute of Technology, Fukuoka, Japan [email protected]

Abstract. Square lattice two-dimensional photonic crystal waveguide with nonlinearity and linear dispersion is numerically analyzed for optical wavelength division multiplexing (WDM). In steady state for Gaussian beam incidence, the distribution ratio is calculated from the electric field profile of the three output ports. As a result, it is possible to switch the output port by changing the wavelength of the input signal. This result shows that passive alloptical switching is possible.

1 Introduction Since the 2000s, the traffic of the Internet has been increased by the spread of optical fiber network and of applications. Along with that, information processing has been increased in electronic integrated circuits of network routers, and there are problems such as heat generation and increase in power consumption. There is a possibility that these problems can be avoided by using optical technology instead of electronic signal processing in integrated circuit [1]. Photonic crystal has attracted attention for the realization of optical integrated circuits. The photonic crystal is artificial periodic structures matched to the wavelength of light. By modifying the periodic lattice and the thickness of the pillar etc., it is possible to create a wavelength region that cannot propagate in the structure. This is called photonic band gap (PBG) [2]. In this paper, nonlinear dielectric material is installed as a resonator in the photonic crystal waveguide. In addition, switching of the output port depending on the wavelength of the input signal is considered. In previous study, the optical signal output to each port did not reach to sufficiently steady state due to the shorter computation time steps [3]. Therefore, to set the optical signal to steady state, it is necessary to allow enough time to pass after the optical signal is input. FDTD (Finite Difference Time Domain) method is used to calculate the electromagnetic field in the waveguide. The FDTD method can be applied to nonlinear dielectric materials and distributed linear and nonlinear dielectric materials [4]. In this paper, the analysis of optical switching in photonic crystal waveguides is reported. © Springer Nature Switzerland AG 2020 L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 621–629, 2020. https://doi.org/10.1007/978-3-030-33506-9_56

622

N. Higashinaka and H. Maeda

2 FDTD Method
According to Yee’s FDTD scheme [5], the two dimensional i:e:

@ @y

 ¼ 0 , discretized

Maxwell’s equation for transverse electric (TE) mode, which propagates to x and z axis, are described as follows;   1 iþ ;k 2      Dt 1 1 1 1 n n þ Ey i þ ; k þ  Ey i þ ; k  l0 Dz 2 2 2 2     1 1 n þ 12 n12 Hz i; k þ i; k þ ¼ Hz 2 2      Dt 1 1 1 1 n n  Ey i þ ; k þ  Ey i  ; k  l0 Dx 2 2 2 2     1 1 1 1 nþ1 Dy 2 i þ ; k þ ¼ Dny i þ ; k þ 2 2 2 2      1 Dt 1 1 nþ nþ1 Hx 2 i þ ; k þ 1  Hx 2 i þ ; k þ Dz 2 2      1 1 Dt 1 1 nþ nþ  Hz 2 i þ 1; k þ  Hz 2 i; k þ Dx 2 2 nþ1 Hx 2



1 iþ ;k 2



n1 ¼ Hx 2

ð1Þ

ð2Þ

ð3Þ

where E and H are electric and magnetic field and D is electric flux intensity. Space and time discretization are Dx; Dz; Dt. Number of corresponding grids are i; k; n. Giving a set of suitable initial conditions and a constitutive equation between D and E to above equations, the latest field is calculated successively as the increase of time step number. Let us consider composite space of free space, linear dielectric material and dispersive linear and nonlinear dielectric material. Eyn ¼

Dny e0

ð4Þ

where e0 is permittivity of free space. The electric field in linear dielectric material is described as follows; Eyn ¼

Dny e0 er

where er is relative permittivity of linear dielectric material.

ð5Þ

Routing of Optical Baseband Signal Depending on Wavelength

623

The electric field in dispersive linear and nonlinear dielectric material is described as follows [4];
3 ð3Þ n1 Dny  Sn1 þ 2v a E L y 0 Eyn ¼
2 ð3Þ ð3Þ n1 e1 þ v0 ð1  aÞSn1 R þ 3v0 a Ey 1 e0

SL ðzÞ ¼ 2eaL Dt cosðbL DtÞ  z1 SL ðzÞ  e2aL Dt  z2 SL ðzÞ þ cL DteaL Dt sinðbL DtÞ  E ðzÞ aL ¼ 2pfL  dL bL ¼ 2pfL cL ¼

qffiffiffiffiffiffiffiffiffiffiffiffiffi 1  d2L

2pfL  ðes  e1 Þ qffiffiffiffiffiffiffiffiffiffiffiffiffi 1  d2L

SnR ¼ 2eaR Dt cosðbR DtÞ  Sn1  e2aR Dt  Sn2 R R þ cR DteaR Dt sinðbR DtÞ  ðEn Þ2 aR ¼ 2pfNL  dNL bR ¼ 2pfNL

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 1  d2NL

ð6Þ

ð7Þ ð8Þ ð9Þ ð10Þ

ð11Þ ð12Þ ð13Þ

ð3Þ

cR ¼

2pfNL  v0  ð1  aÞ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 1  d2NL

ð14Þ

where e1 and es are relative permittivity of dispersive linear and nonlinear dielectric material, fL is linear relaxation frequency, fNL is nonlinear relaxation frequency, dL is linear decaying factor and dNL is nonlinear decaying factor. Following equation is developed from the above equation.
3 ð3Þ n1 Dny  Sn1 þ 2v a E L y 0 Eyn ¼
2 ð3Þ ð3Þ n1 e1 þ v0 ð1  aÞSn1 þ 3v a E R y 0 1 e0

ð15Þ

The electric field in linearly dispersive and nonlinear dielectrics is calculated from this equation.

624

N. Higashinaka and H. Maeda

3 Simulation Method Optical signal propagating in two-dimensional photonic crystal waveguide with a duplexer with three ports is analyzed. The photonic crystal waveguide used in the simulation is shown in Fig. 1. In Fig. 1, background of the waveguide is free space and the refractive index n0 ¼ 1:0. Black circles are linear dielectric material and the refractive index n1 ¼ 3:6 as is in Ref. [6]. In addition, blue circles are dispersive linear and nonlinear dielectric material. The refractive index n2 ¼ 1:5, relative permittivity of dispersive linear and nonlinear dielectric material e1 ¼ 2:25, es ¼ 5:25, linear relaxation frequency fL ¼ 63:7½THz, linear decaying factor dL ¼ 2:5  104 , nonlinear relaxation frequency fNL ¼ 14:8½THz, nonlinear decaying factor dNL ¼ 3:36  101 , ð3Þ nonlinear susceptibility v0 ¼ 0:07, and weight factor for Kerr effect a ¼ 0:7 as is in Ref. [4]. The lattice period L ¼ 551:8½nm, and the radius of dielectric material R ¼ 0:2L are used. The FDTD discretization are Dx ¼ Dz ¼ 9:85½nm and Dt ¼ 0:0209½fs, respectively. The analysis area is 14.105[lm] in the propagation axis and 10.244[lm] in the transverse axis. The wavelength k = 1.350 – 1.600[lm]. Berenger’s perfectly matched layer [7] is installed as absorbing boundary condition.

Fig. 1. Top of view 2-D photonicrystal waveguide (duplexer is composed of linear dispersive and nonlinear media at the center of the waveguide)

The electric field input to port 1 is Gaussian beam and given by

Routing of Optical Baseband Signal Depending on Wavelength

(  )   x 2 2pct Ey ðx; tÞ ¼ E0 exp  sin w0 k

625

ð16Þ

where the amplitude E0 ¼ 1:0, the beam spot w0 ¼ 28½nm, optical speed in vacuum c ¼ 2:998  108 ½m=s. For steady state, 10; 000Dt are sufficient for the electric field to reach port 2 and port 3 from port 1. However, computation up to 100; 000Dt is done to get convergence of the electric field in the waveguide. The light intensity Pi of port 1, port 2 and port 3 is obtained from the electric field of 90; 000  100; 000Dt where the electric field converges. The evaluation is described as follows; P100;000 Pi ¼

Eyi ðt ¼ nDtÞ2 ði ¼ 1; 2; 3Þ 10; 000

n¼90;000

ð17Þ

Definition of distribution ratio Si is described as follows; R P dw R PORTi i R Si ¼ R PORT1 P1 dw þ PORT2 P2 dw þ PORT3 P3 dw

ði ¼ 1; 2; 3Þ

ð18Þ

where w is the direction traversing the waveguide.

4 Simulation Results Distribution ratio is calculated by converged electric field. At wavelength k ¼ 1:386½lm, The electric fields that propagate the center of the waveguide and output to port 1, port 2, and port 3 are shown in Figs. 2, 3 and 4. It is confirmed that the electric field is converged at 90; 000Dt  100; 000Dt by these figures. Figure 5 shows distribution ratio of port 1, port 2 and port 3. It was confirmed that the distribution ratio to the ports 2 and port 3 is changed along with change of the input signal wavelength. The optical signal does not reach port 1 at any wavelength because it is closed by linear dielectric pillars. The distribution ratio of port 2 and that of port 3 were switched twice at wavelength of 1:350  1:600½lm. At the wavelength k ¼ 1:369  1:437½lm, the distribution ratio of port 2 is higher than that of port 3. At the wavelength k ¼ 1:350  1:368½lm and k ¼ 1:438  1:600½lm, the distribution rate of port 3 is higher than that of port 2. At the wavelength k ¼ 1:386½lm, the distribution ratio of the optical intensity of port 1 is S1 ¼ 0:179%, that of port 2 is S2 ¼ 99:149% and that of port 3 is S3 ¼ 0:672%. Most of optical intensity is obtained from port 2. The electric field distribution is shown in Fig. 6. At the wavelength k ¼ 1:504½lm, the distribution ratio of the optical intensity of port 1 is S1 ¼ 1:380%, that of port 2 is S2 ¼ 3:962% and that of port 3 is S3 ¼ 94:657%. Most of optical intensity is obtained from port 3. The electric field distribution is shown in Fig. 7.

626

N. Higashinaka and H. Maeda

Fig. 2. Electric field output to port 1 (Left side is 0Dt  100; 000Dt and Right side is 90; 000Dt  100; 000Dt)

Fig. 3. Electric field output to port 2 (Left side is 0Dt  100; 000Dt and Right side is 90; 000Dt  100; 000Dt)

Fig. 4. Electric field output to port 3 (Left side is 0Dt  100; 000Dt and Right side is 90; 000Dt  100; 000Dt)

Routing of Optical Baseband Signal Depending on Wavelength

627

100

Distribution ratio[%]

90 80 70 60

50 40 30 20 10

0 1.350E-06

1.400E-06

1.450E-06

1.500E-06

1.550E-06

1.600E-06

Wavelength[m] port1

port2

port3

Fig. 5. Distribution ratio of port 1, port 2 and port 3 ðk ¼ 1:350  1:600½lmÞ

Fig. 6. Electric field distribution ðk ¼ 1:386½lmÞ

628

N. Higashinaka and H. Maeda

Fig. 7. Electric field distribution ðk ¼ 1:504½lmÞ.

5 Conclusion It was found that the output port of the photonic crystal waveguide can be switched depending on the wavelength by using a linear dispersive and nonlinear medium duplexer. It is easy to use as optical integrated circuit because the waveguide size is order of square of 10½lm. As future work, we would like to consider narrowing the wavelength interval and providing broadband filter function.

References 1. NTT Technical Journal 2010, vol. 22, no. 5. The Telecommunications Association 2. Noda, S., Baba, T. (ed.): Roadmap on Photonic Crystals. The Netherlands, Kluwer Academic Publishers (2003) 3. Haari, K., Maeda, H., Meng, X., Higashinaka, N.: Numerical analysis of optical duplexer composed of photonic crystal with square lattice. In: Proceedings of BWCCA 2018, The 13th International Conference on Broadband, Wireless Computing, Communication and Applications. Lecture Notes on Data Engineering and Communications Technologies 25, pp. 548– 558 (2018) 4. Sullivan, D.M.: Nonlinear FDTD formulations using Z transforms. IEEE Trans. Microw. Theory Tech. 43(3), 676–682 (1995) 5. Yee, K.S.: Numerical solution of initial boundary value problems involving Maxwell’s equation. IEEE Trans. Antennas Propaga. 14(3), 302–307 (1996). IEEE Trans. EMC, 32(3), 222–227 (1990)

Routing of Optical Baseband Signal Depending on Wavelength

629

6. Li, X., Maeda, H.: Numerical analysis of photonic crystal directional coupler with kerr-type nonlinearity. In: Proceedings of 5th Asia-Pacific Engineering Research Forum on Microwaves and Electromagnetic Theory, pp. 165–168 (2004) 7. Berenger, J.: A perfectly matched layer for the absorption of electromagnetic waves. J. Comput. Phys. 114(2), 185–200 (1994)

Two-Stage Dynamic Contract Design Under Asymmetric Information in Cooperative Communication Nan Zhao(&), Pengfei Fan, Xiao He, Menglin Fan, and Chao Tian Hubei University of Technology, Wuhan 430068, China [email protected], [email protected], [email protected], [email protected], [email protected]

Abstract. Cooperative communication is considered as a prospective technique to increase spectral efficiency. Because of the selfishness of relay nodes, it is necessary to offer relay nodes long-term incentives in the dynamic network. Two-stage dynamic contract with dynamic asymmetric information scenarios is studied in this paper. The type of each relay node is considered to be independent of the two stages with the same continuous probability distribution. By deriving the enough and essential conditions, a contract model is developed to identify the type of information on relay nodes. Moreover, to achieve the source’s maximum expected utility, this paper proposes an optimal sequential method to acquire the optimal relay-reward policy. Experimental results demonstrate that the system performance of cooperative communication is improved by the design of the optimal two-stage dynamic contract.

1 Introduction Cooperative communication is considered as a promising method for spectrum efficiency improvement [1–3]. Nevertheless, because of the shortage of wireless resources and the selfishness of wireless nodes, the potential relay nodes (RNs) may be reluctant to participate in cooperative without incentives [4, 5]. Besides, the various selection methods require complete network information during the potential RNs selection [6]. However, the fad effect of wireless channel and the mobility of wireless users lead to the problem of asymmetric network information between the source node (SN) and the RNs [7, 8]. Therefore, it is a challenging topic to encourage RNs to participate in cooperative communication efficiently. At present, the incentive problem of cooperative communication under asymmetric information conditions is attracting the attention of researchers. Duan et al. studied the wireless resource exchanges strategy based on contract theory [9]. Li et al. proposed a contract incentive mechanism about traffic balancing [10]. Sheng et al. apply contract theory to cooperative communication between primary and secondary users [11]. The existing cooperative communication incentive method mainly studies static contract design [12, 13]. However, in actual cooperative communication, we are faced with a flexible and dynamic network environment. To avoid the additional transaction costs © Springer Nature Switzerland AG 2020 L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 630–637, 2020. https://doi.org/10.1007/978-3-030-33506-9_57

Two-Stage Dynamic Contract Design Under Asymmetric Information

631

caused by the frequently selecting RNs is important. Therefore, it is necessary to design a two-stage dynamic contract to encourage RNs to participate in cooperative communication. This paper concentrates on solving the problem of network information asymmetry in cooperative communication. By introducing the market-driven contract model into cooperative communication, a dynamic incentive mechanism based on contract theory is proposed. The dynamic characteristic of the network environment is considered, and a two-stage dynamic contract model is designed. For the purpose of encouraging the RNs to participate in cooperative communication, the combination of incentive compatibility and personal rationality is implemented to realize the identification of RNs’ private information. The performance of the proposed two-stage dynamic contract incentive mechanism is verified by simulation.

2 System Model and Problem Formulation A general cooperative network consists of an SN, a destination node, and N RNs. The SN wants to transmit the data onto the destination node. Because of the poor channel condition, the SN needs the help of the RNs. However, due to the selfishness of both parties, the SN expects RNs to transmit data with high power, which will increase the cost of RNs. The RN hopes to get the maximum reward with less effort. Therefore, this paper will design a dynamic contract incentive mechanism to resolve the contradiction between the two parties. 2.1

Relay Node Model

Assume hRTi ;D is the channel gain between the transmitter RTi and the destination D. Since pi is the received power of the destination from the ith RN, the transmit power of the ith RN is hRTpi ;D . Therefore, the cost of the ith RN is i

Ci ðpi Þ ¼

pi c ¼ hi pi ; i 2 ½1; N], hRTi ;D i

ð1Þ

where ci is the cost of unit transmission power, here, we define hi ¼ hRTci ;D as the cost i

coefficient of unit transmission power. When hi increases, it means that the ith RN has higher transmission cost. Suppose each RN knows hi , which is a random variable and distributed on a strictly positive interval H 2 ½hL ; hH  with the distribution function Fi ðhÞ. Then, the utility of the ith RN can be expressed as the obtained reward wi minus the transmission cost Ci ðpi Þ, that is, URNi ðpi ; wi Þ ¼ wi  hi pi ;

i 2 ½1; N]:

ð2Þ

632

2.2

N. Zhao et al.

Source Node Model

The utility of SN is the total achievable cooperative rate minus the reward wi paid to the ith RN, that is Z US ¼

hH

hL

½q logð1 þ

pi Þ  wi dFi ðhÞ; n0

i 2 ½1; N;

ð3Þ

where q is the equivalent yield coefficient per unit channel capacity, n0 is the noise power. To simplify the analysis, usually set the normalized value of n0 to 1. 2.3

Contract Formulation

In the asymmetric network environment, the ability of each ith RN to transmit data is not known to the SN. The private ability information of the ith RN is different in each stage. Therefore, in order to increase the cooperative enthusiasm of the ith RN, this paper proposes a two-stage dynamic contract, in Fig. beginning of the  1 1shown  1. At the 2 1 2 th contract, the SN provides the contract wi ðpi Þ; wi ðpi ; pi Þ to the i RN, where p1i is the cooperative power of the first stage, w1i is the reward received in the first stage of the ith RN, p2i is the cooperative power of the second stage, and w2i is the reward received in the second stage of the ith RN. When the ith RN chooses to accept a contract, it will notify the SN of its choice. If the collaboration is successful, SN will pay the ith RN according to the contract. Otherwise, the ith RN will not be paid.

t=0

t=0.25

t=0.5

t=1

t=1.5

t=2 time

θi1 is realized Only RNs learn

The source RNs accept First period offer power offers a longor refuse and reward pi1 term contract the {wi1 ( pi1 ); wi2 ( pi1, pi2 )} contract take place wi1

Second period θi2 is realized offer power only RNs and reward pi2 learn take place wi2

Fig. 1. Two-stage dynamic contract design.

3 Dynamic Incentive Design Under Asymmetric Information Considering the two-stage contract design, the total expected utility of SN can be written as Us ¼ U1s þ dU2s ;

ð4Þ

Rh where U1s ¼ hLH ½q logð1 þ p1i Þ  w1i dFi1 ðh1i Þ is the utility of the first stage of the SN, Rh and U2s ¼ hLH ½q logð1 þ p2i Þ  w2i dFi2 ðh2i Þ is the utility of the second stage of the SN,

Two-Stage Dynamic Contract Design Under Asymmetric Information

633

d is the discount factor. When d is greater than 1, it means that the second stage is longer than the first stage. In the case where the ith RN and the SN have the same discount factor, the twostage utility of the ith RN is URNi ¼ ½w1i  h1i p1i  þ d½w2i  h2i p2i ;

3.1

i 2 ½1; N]:

ð5Þ

Contracting Design in Stage 2

2 In stage 2, every type-h2k RN utility is URN ð~hi Þ ¼ w2k ð~ hi Þ  h2k p2k ð~ hi Þ; where ~ hi is the k 2 th first-stage announcement about its type of the i RN, hk is the second stage of private information, and p2k ð~hi Þ is the second stage of cooperative power, w2k ð~ hi Þ is the reward 2 obtained by the type-hk RN of the second stage. To ensure that the type-h2k RN achieves a non-negative value by choosing contract ðw2k ð~hi Þ; p2k ð~hi ÞÞ, the individual rationality (IR) constraint must be satisfied, that is, 2 ð~hi Þ ¼ w2k ð~hi Þ  h2k p2k ð~hi Þ  0: i; k 2 ½1; N: URN k

ð6Þ

Then, considering that the type-h2k RN should choose the appropriate contract to achieve the maximum utility, the following incentive compatibility (IC) constraints should be met, w2k ð~hi Þ  h2k p2k ð~hi Þ  w2j ð~hi Þ  h2k p2j ð~ hi Þ;

i; k; j 2 ½1; N:

ð7Þ

Then, since 2 dURN k

dh2k

¼ p2k ðh2k Þ  0:

k; 2 ½1; N;

ð8Þ

2 the type-h2k RN’s utility URN is decreasing in h2k . k 2 2 2 From (8), we have URN ðhH Þ ¼ minURN ðh2k Þ ¼ 0. Thus, URN can be rewritten as k k k

Z 2 2 URN ¼ URN ðhH Þ  k k

hH h2k

Z P2k ðsk Þdsk ¼

hH

h2k

p2k ðsk Þdsk :

k; 2 ½1; N:

ð9Þ

Then, by combining (6) and (9), we have w2k ¼ h2k p2k ð~hi Þ þ

Z

hH h2k

p2k ðsk Þdsk ;

i; k; 2 ½1; N:

ð10Þ

634

N. Zhao et al.

3.2

Contracting Design in Stage 1

1 Assume that the type-h1i RN’s utility in stage 1 can be written as URN ¼ w1i  h1i p1i , i then, the intertemporal IC constraint of the two-stage is written as

Z ½w1i  h1i p1i  þ d

hH hL

Z 2 URN dFk2 ðh2k Þ  ½w1j  h1i p1j  þ d k

hH

hL

2 URN dFk2 ðh2j Þ: i; k; j 2 ½1; N: k

ð11Þ where

R hH hL

2 URN dFk2 ðh2k Þ is the expected continuing utility of the type-h1i RN in stage 2. k

2 Considering that the type-h1i RNs’ utility URN in stage 2 is independent of h1i , that k R hH 2 R h 2 is hL URNk dFk2 ðh2k Þ ¼ hLH URN dFk2 ðh2j Þ, the above IC constraint can be simplified as k

w1i  h1i p1i  w1j  h1i p1j

i; j 2 ½1; N:

ð12Þ

Then, considering the expected continuation utility of stage 2, the ith RN’s intertemporal IR constraint is written as Z w1i



h1i p1i

þd

hH hL

2 URN dFk2 ðh2k Þ  0: k

i; k; 2 ½1; N:

ð13Þ

Similarly to the analysis of stage 2, we can get Z w1i ¼ h1i p1i þ where Hk2 ðh2k Þ ¼

hH h1i

Z p1i ðsi Þdsi  d

hH

hL

Hk2 ðh2k Þp2k ðh2k ÞdFk2 ðh2k Þ;

i; k; 2 ½1; N;

ð14Þ

Fk2 ðh2k Þ . fk2 ðh2k Þ

Then, the intertemporal contract design optimization problem can be defined as, max

fp1i ;p2k ;w1i ;w2k g

s:t: where dFk2 ðh2k Þ.

U1s ¼

R hH hL

U1s þ dUs2 ;

ð15Þ

ð6Þ; (7), (12), and (13),

½q logð1 þ p1i Þ  w1i dFi1 ðh1i Þ

and

Us2 ¼

R hH hL

½q logð1 þ p2k Þ  w2k 

Then, by combining (10) and (14), the above maximum optimization problem can be simplified as max

fp1i ;p2k g

U1s þ dUs2 ;

 Rh  Us1 ¼ hLH q logð1 þ p1i Þ  h1i p1i  Hi1 ðh1i Þp1i þ dHk2 ðh2k Þp2k ðh2k Þ dFi1 ðh1i Þ,  Rh  Us2 ¼ hLH q logð1 þ p2k ðh2k ÞÞ  h2k p2k ðh2k Þ  Hk2 ðh2k Þp2k ðh2k Þ dFk2 ðh2k Þ.

where

ð16Þ and

Two-Stage Dynamic Contract Design Under Asymmetric Information

635

Therefore, the optimal cooperative power can be given by (

q 1 p1 i ðhi Þ ¼ h1 þ H 1 ðh1 Þ  1 i

i

q 2 p2 k ðhk Þ ¼ h2  1

i

i; k 2 ½1; N:

ð17Þ

k

4 Numerical Results and Discussion In this section, the feasibility of a two-stage dynamic contract will be verified by MATLAB simulation experiments. The experiment parameters are set as: a ¼ 1, b ¼ 3, d ¼ 0:3, q ¼ 15, N ¼ 21, and h ¼ ½1 : 0:1 : 3.  Figure 2 describes the optimal expected utility URN of RN for different types of RNs with the same q. As hi increases, the cooperation cost of RN increases, and the utility obtained by the relative RN decreases. Moreover, we can see the utility of the type-3, type-6 and type-9 RNs obtained when selecting all types of contracts provided by the SN. The RN can only get the most utility when choosing the type of contract suitable for its ability. Then, the SN can know the information related to the nature of RN. Therefore, the problem of asymmetric network information can be solved through this dynamic contract.

 Fig. 2. RN’s optimal expected utility URN with the different type for effort incentive design.

Next, we study the system performance of the type-hi RN under the different distributions. In case A, the probability of all type-hi RNs is uniformly distributed in the [1, 3] with fi ðhi Þ ¼ 1=4. In case B, the probability of low cost for all type-hi RNs is

636

N. Zhao et al.

greater than the probability of high cost with fi ðhi Þ ¼ ð6  hi Þ=12. In case C, all type-hi RNs are low cost, and the probability is less than that of high cost with fi ðhi Þ ¼ hi =12. Figure 3 depicts the optimal utility us of the SN under the three different distributions of the type-hi RN with the same distributions q in all cases. In each case, when the value of q is low, the utility of SN is insufficient, resulting in no utility in the whole communication process. Then, as q increases, the utility acquired by the SN increase, and the SN gained a higher non-negative profit. Besides, with the greater the probability of the type-hi RN low-cost coefficient, the RN may have the greater cooperative power, which leads to the greater optimal utility of the SN.

Fig. 3. Source’s optimal utilities us with different distributions of RNs’ type.

5 Conclusion This paper proposes a contract-based dynamic incentive mechanism to solve the problem of network information asymmetry in cooperative communication. Considering the selfishness of the two parties and the time-varying characteristics of the network environment, a two-stage dynamic contract with the same continuous probability distribution is designed. Moreover, the incentive compatibility and participation constraints are combined to encourage the relay nodes to participate in cooperation actively. A sequence optimization method is designed to acquire the optimal relay cooperative power. The experimental results show that the system performance of cooperative communication is improved by the design on a two-stage dynamic contract.

Two-Stage Dynamic Contract Design Under Asymmetric Information

637

References 1. Nosratinia, A., Hunter, T.E., Hedayat, A.: Cooperative communication in wireless networks. IEEE Commun. Mag. 42, 74–80 (2004) 2. Dhanasekaran, S., Reshma, T.: Full-rate cooperative sharing scheme for cognitive radio communications. IEEE Commun. Lett. 22, 97–100 (2018) 3. Zhao, N., Liang, C., Pei, Y.: Dynamic contract incentive mechanism for cooperative wireless networks. IEEE Trans. Veh. Technol. 67, 10970–10982 (2018) 4. Tang, R., Cheng, L., Cao, X.: Contract-based incentive mechanism for cooperative NOMA systems. IEEE Commun. Lett. 23, 172–175 (2019) 5. Zhang, M., Huang, W.: Efficient network sharing with asymmetric constraint information. IEEE J. Sel. Areas Commun. 37, 1898–1910 (2019) 6. Hong, X., Wang, J., Wang, C., Shi, J.: Cognitive radio in 5G: a perspective on energyspectral efficiency trade-off. IEEE Commun. Mag. 52, 46–53 (2014) 7. Zhao, N., Liang, Y., Pei, Y.: Dynamic contract design for cooperative wireless networks. In: IEEE Globe Communication Conference, pp. 1–6 (2017) 8. Zhang, Y., Song, L., Saad, W., Han, Z.: Contract-based incentive mechanisms for device-todevice communications in cellular networks. IEEE J. Sel. Areas Commun. 33, 2144–2155 (2015) 9. Duan, L., Gao, L., Huang, J.: Cooperative spectrum sharing: a contract-based approach. IEEE Trans. Mob. Comput. 13, 174–187 (2014) 10. Li, Y., Zhang, J., Gan, X.: Contract-based incentive mechanism for delayed traffic offloading in cellular networks. IEEE Trans. Wirel. Commun. 15, 5314–5327 (2016) 11. Sheng, P., Liu, F.: Profit incentive in trading nonexclusive access on a secondary spectrum market through contract design. Trans. Netw. 22, 1190–1203 (2014) 12. Ju, P., Song, W.: Repeated game analysis for cooperative MAC with incentive design for wireless networks. IEEE Trans. Veh. Technol. 65, 5045–5059 (2016) 13. Zhao, N., Wu, M., Xiong, W., Liu, C.: Cooperative communication in cognitive radio networks under asymmetric information: a contract-theory based approach. Int. J. Distrib. Sens. Netw. 2015, 1–11 (2015)

Minimizing Control Overhead of Routing Protocols in Wireless Multihop Networks: Simulation Evaluation Soushi Morita1 and Elis Kulla2(B) 1

Department of Information and Computer Engineering, Okayama University of Science, 1-1 Ridai-cho, Kita-ku, Okayama 700-0005, Japan [email protected] 2 Graduate School of Engineering, Okayama University of Science, 1-1 Ridai-cho, Kita-ku, Okayama 700-0005, Japan [email protected]

Abstract. In recently developed unmaned vehicles technologies, a lot of interest from the research community is focused on the decision making of autopilot agents, which are programmed to safely drive or pilot an unmaned vehicle through road traffic or aerial traffic. A lot of work has to be done in connecting those agent-driven vehicles, in order to create possibilities for new applications, business models and communication paradigms. A lot of research has been done on the communication channel (i.e. mm wave, tdma techniques). However, vehicle communication applications and infrastructure require more attention and development. In this paper we investigate the performance of AODV routing protocol, and simulate different scenarios of environmental settings, vehicle movements, AODV operation and so on. We then discuss the findings based on the simulation results.

1

Introduction

Vehicular Ad-hoc Networks (VANETs) are a special type of Ad-hoc networks and are an important component of the future Intelligent Transportation Systems (ITS). They can been utilized to guarantee road safety, to avoid potential accidents by creating new forms of intervehicle communications and applications. In recent years, drone demand, such as drone delivery, has greatly increased and is approaching the practical stage. However, communication between drones is still under development. Therefore, in this work we evaluate the performance of AODV (Adhoc On-demand Distance Vector) [1,5] for different ring search diameters in mobile drone environment. Our objective is to help the research community to identify improvements in AODV [2,3]. The contents of the following sections is as follows. In Sect. 2, we describe AODV protocol, its route lifetime field and ring search procedure. Then, in Sect. 3, we describe the simulation environment and settings. Simulation results are presented in Sect. 4, and finally we show our conclusions and future considerations. c Springer Nature Switzerland AG 2020  L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 638–645, 2020. https://doi.org/10.1007/978-3-030-33506-9_58

Minimizing Control Overhead of Routing Protocols

2

639

Ad-Hoc On-Demand Distance Vector (AODV) Routing Protocol

AODV is an ad-hoc on-demand routing protocol. It performs Route Discovery using control messages: Route Request (RREQ) and Route Reply (RREP). In AODV, routes are set up by flooding the network with RREQ packets. As a RREQ traverses the network, the traversed nodes store information about the source, the destination, and the node from which they received the RREQ. The later information is used to set up the reverse path back to the source. When the RREQ reaches a node, which knows a route to the destination or it is the destination itself, the node responds to the source with a RREP packet which is routed through the reverse path set up by the RREQ. This sets the forward route from the source to the destination. To avoid overburdening the nodes with information about routes which are no longer (if ever) used, nodes discard this information after a timeout. When either destination or intermediate node moves, a Route Error (RERR) is sent to the affected source nodes. When source node receives the RERR, it can reinitiate route discovery if the route is still needed. Neighborhood information is obtained by periodically broadcasting Hello packets. For the maintenance of the routes, two methods can be used: – ACK messages in MAC level or – HELLO messages in network layer The main advantage of this protocol is that routes are established on demand and destination sequence numbers are used to find the latest route to the destination. The connection setup delay is lower. One of the disadvantages of this protocol is that intermediate nodes can lead to inconsistent routes if the source sequence number is very old and the intermediate nodes have a higher but not the latest destination sequence number, thereby having stale entries. Also multiple RREP packets in response to a single RREQ packet can lead to heavy control overhead. Another disadvantage of AODV is that the periodic beaconing leads to unnecessary bandwidth consumption. An example of “route discovery” procedure is described below in reference to Fig. 1. A route from node S to node D is established through flooding RREQ packets in the network, until node D is found. Every node forwards RREQ packets, while increasing its sequence number. A node which has already received a RREQ with a lower sequence number, refuses the reception of all RREQ packets with higher sequence number. This decreases the overhead and avoids loops in AODV. When node D is found, a RREP packet is sent back to node S and the route to node D is registered in S’s routing table. On the way back node 7 and node 3 also register the route to node D for future reference (another RREQ packet). 2.1

Link Lifetime and Route Lifetime

An AODV node keeps track of all known routes to other nodes in the network, with a route lifetime field on each of them. The Route Lifetime field of

640

S. Morita and E. Kulla

Fig. 1. AODV route discovery using RREQ & RREP packets.

Fig. 2. Ring search (T T L < 7).

the routing table entry is initially set to a constant value (ACTIVE-ROUTETIMEOUT). Then, every time a new control packet is received, it is updated accordingly. However, ACTIVE-ROUTE-TIMEOUT does not take into account any information about the actual network conditions. 2.2

Ring Search

Extended ring search is an AODV function that prevents the spread of unnecessary RREQ throughout the network [6–10]. In the extended ring search, the originating node first sets the initial value of TTL to T T L in the RREQ packet IP header, and sets the timeout for receiving RREP to Ring T raversal T ime milliseconds. If the RREQ times out, the caller broadcasts the RREQ again with a TTL incremented by the “incremented TTL” value. This continues until the TTL set by RREQ reaches the TTL threshold, after which NetDiameter is set to TTL. In standard AODV, as shown in Fig. 2, since TTL Start is 1, TTL starts

Minimizing Control Overhead of Routing Protocols

641

Fig. 3. Ring search (T T L ≥ 7).

from 1 and increases by 2 each time RREQ times out. If TTL Threshold 7 is exceeded, TTl is set to Net Diameter (35) is set to TTL as shown in Fig. 3. Refer to Table 1 for TTL values, that we used in our simulations.

3

Simulation Environment

We use a 500 m × 500 m area to conduct our simulations (as shown in Fig. 4). The detailed parameters are as shown in Table. 1. If the sender node and the receiver node are moved randomly, the difference in the communication result is mainly based on the randomness of the mobility model, rather than the routing protocol mechanism. Thus, the sender node is set to fixed coordinates (0, 0) and the receiver node is set to fixed coordinates (500, 500). 100 nodes move around the area and they are used to relay packets from sender to receiver. We modify the TTL settings in order to alter the ring search diameter, and evaluate the performance based on number of RREQ packets and total PDR of the 60 s communication. First we set TTL Start to 1. Then, we increase the TTL Threshold from 1 to 9, and examine the effect of the change, by measuring the number of RREQ and the value of PDR.

4

Simulation Result

Simulation results for PDR are shown in Fig. 5. The graph show the relationship between TTL threshold value and PDR, for different communication distance values. When communication distance increases, the number of hops required to reach the destination is smaller (2 hops), so the PDR value is higher. Also, when the TTL threshold value increases, the PDR value increases in general.

642

S. Morita and E. Kulla

Fig. 4. 500 m × 500 m simulation area

Fig. 5. PDR vs TTL threshold for PDR.

Simulation results for number of RREQ packets are shown in Fig. 6. The graphs show the relationship between the communication distance and the number of unique transmitted RREQ packets (RREQ u), the total number of forwarded RREQ packets (RREQ t), the total/unique rate (RREQ t/u), which shows the overhead of control packets, when transmitting RREq packets. We found out that while the communication distance increases the number of RREQ packets decreases (both RREQu and RREQt), while the RREQ t/u rate increases. Increasing the communication distance makes the communication easier but it increases the power consumption, so it will be necessary to find the optimal communication distance based on each application.

Minimizing Control Overhead of Routing Protocols

(a) Packet Rate=0.04s

(b) Packet Rate=0.25s

(c) Packet Rate=2.00s

Fig. 6. RREQ number vs communication distance.

643

644

S. Morita and E. Kulla Table 1. Simulation parameters Parameter

Value

Map size Static node number Moving node number Communication distance

500 m × 500 m 2 100 150 m–350 m

Packet size Packet interval

256 Bytes 0.04 s, 0.25 s, 2 s

TTL start TTL increment TTL threshold Net diameter

1 2 1–9 35

Simulator

NS-3

In general, the packet rate of transmitted data has little effect on the number of RREQ, but we see a decrease in the total number of RREQ, when the packet rate is lower.

5

Conclusion

In this paper, we investigated the effect of TTL settings, in AODV performance by simulations. From simulation results, we conclude the following. – When communication distance or the TTL threshold value increases, PDR value is higher. – When communication distance increases the number of RREQ packets decreases, while the overhead increases. – It is necessary to find the optimal communication distance based on each application, considering transmission power and control overhead. – For lower packet rates, the number of RREQ packets decreases. In future research, we would like to investigate the relationship between the communication distance and power consumption, and improve AODV to dynamically set the optimal communication distance.

References 1. Perkins, C.E., Royer, E.M.: Ad-hoc on-demand distance vector routing. In: Proceedings of the Second IEEE Workshop on Mobile Computing Systems and Applications, WMCSA 1999, New Orleans, LA, USA, pp. 90–100 (1999) 2. Mai, Y., Rodriguez, F.M., Wang, N.: CC-ADOV: an effective multiple paths congestion control AODV. In: 2018 IEEE 8th Annual Computing and Communication Workshop and Conference (CCWC), Las Vegas, NV, pp. 1000–1004 (2018)

Minimizing Control Overhead of Routing Protocols

645

3. Ashwini, H.K, Vyshali Rao, K.P., Ginimav, I.: CM-AODV: an efficient usage of network bandwidth in AODV protocol. In: 2018 International Conference on Design Innovations for 3Cs Compute Communicate Control (ICDI3C), Bangalore, pp. 111–114 (2018) 4. Bhagyalakshmi, Dogra, A.K.: Q-AODV: a flood control ad-hoc on demand distance vector routing protocol. In: 2018 First International Conference on Secure Cyber Computing and Communication (ICSCCC), Jalandhar, India, pp. 294–299 (2018) 5. Perkins, C., Belding-Royer, E., Das, S.: Ad hoc on-demand distance vector (AODV) routing. Internet Request for Comments (RFC), pp. 1–37, July 2013. https://www. ietf.org/rfc/rfc3561.txt. Accessed 30 Aug 2019 6. Jayakody, A., Jawadul, A.S.: Modified expanding ring search in common node scenario for AODV. In: 2017 6th National Conference on Technology and Management (NCTM), Malabe, pp. 185–188 (2017) 7. Singh, V.K., Gabrani, G., Dubey, S.: Efficient routing algorithm using sectorized antenna for mobile adhoc networks. In: 2016 3rd International Conference on Computing for Sustainable Global Development (INDIACom), New Delhi, pp. 3635– 3639 (2016) 8. Gwak, K., Park, S., Song, J.: Performance evaluation: two flood-cancellation methods of the blocking expanding ring searches on the AODV/WiFi MANET environment. In: 2019 21st International Conference on Advanced Communication Technology (ICACT), PyeongChang Kwangwoon Do, Korea (South), pp. 239–248 (2019) 9. Sharma, H.L., Agrawal, P., Kshirsagar, R.V.: Multipath reliable range node selection distance vector routing for VANET: design approach. In: 2014 International Conference on Electronic Systems, Signal Processing and Computing Technologies, Nagpur, pp. 280–283 (2014) 10. Huang, Y.-F., Wang, H.-W., Hsu, H.-C.: A novel management methodology or adhoc wireless networks to improve the routing efficiency with adaptive broadcasting range under lower transmitting energy. In: 2015 11th International Conference on Heterogeneous Networking for Quality, Reliability, Security and Robustness (QSHINE), Taipei, pp. 381–386 (2015)

Effect of Parasitic Element on Communication Performance of 13.56 MHz RFID System Kiyotaka Fujisaki(B) and Yuki Yoshigai Fukuoka Institute of Technology, 3-30-1 Wajiro-higashi, Higashi-ku, Fukuoka 811-0295, Japan [email protected], [email protected]

Abstract. RFID is a very important automatic recognition tool for management of things. Recently, RFID technique is widespread in various situations for improving user’s convenience. However, RFID is affected by the environment in the vicinity, which effect the communication performance. In the previous work, we have shown the possibility to expand the communication distance when we put a parasitic element on the middle position between the table type RFID reader and the target tag. In this paper, when a parasitic element is put on the fixed position on the table type RFID reader we evaluate the communication performance of RFID system and show its usefulness.

1

Introduction

Radio Frequency Identification (RFID) technique realizes the contactless data exchange by using the induction field or the radio wave between a communication terminal and tags. Because this system can be communicate contactlessly, RFID technique is a useful tool for management of a large amount of goods or for data exchange between the near field devices. In recent years, the rapid spread of the system based on the RFID technique is remarkable. For example, the RFID is applied to goods management, electronic money, train tickets, and so on. A lot of applications using RFID technique have been proposed [7,8,10– 12,14]. Furthermore, the improvement of RFID technique and performance evaluation using RFID system were performed to implement reliable RFID system [1–6,9,13]. In [4], the basic performance of a table type RFID reader was evaluated and the effect of metallic plate to the reading rate of RFID system was investigated when the RFID reader was placed on the metallic plate. In our previous work [5], using 13.56 MHz RFID system, we evaluated the basic performance of table type RFID reader. Furthermore, in order to expand the communication performance, the use of a parasitic element was considered. As a parasitic element, a open loop coil was placed on half position between the reader and the tag. In this paper, in order to improve the communication performance on the table type RFID reader, we made some coils with different number of turns as a c Springer Nature Switzerland AG 2020  L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 646–654, 2020. https://doi.org/10.1007/978-3-030-33506-9_59

Effect of Parasitic Element on Performance of 13.56 MHz RFID System

647

parasitic element and evaluate the communication performance of RFID system when the parasitic element is placed on the fixed height from the reader face parallely. The paper structure is as follows. In next section, we introduce the RFID system. Then, the experimental method is shown and the communication performance is evaluated when the parasitic element is put on the table type RFID reader. Finally, we conclude the paper.

2

RFID System

The RFID system is the major automatic identification tool. The system uses the wireless communication technique in order to get the ID from the tag without touching it. As shown in Fig. 1, an RFID system consists of two components. One is the RFID tag and sticks it on the object which we want to manage. Another is the reader/writer and is controlled by a management system. The RFID tag normally does not have the power supply to work, so the reader/writer not only exchange the data, but also supply the power and clock signal to the RFID tag.

Reader / Writer

Clock & Power Data

RFID Tag

Controller

Fig. 1. Structure of RFID system

The tag has various shapes according to the purpose of use. For example, the card type RFID tag is used for member’s card, and the label type RFID tag is stuck on goods and used for the management of them. On the other hand, as an RFID reader/writer there are two types: fixed one or portable one. In this experiment, we use a card type RFID tag and a table type reader/writer.

3

Experimental Method

Figure 2 shows the photograph of 13.56 MHz table type RFID reader/writer STRW01 produced by SOFEL, a RFID tag, and a sample of the parasitic element. This RFID system is based on the ISO 15693 International Standard. The size of the housing of the table type RFID reader is 25 cm × 35 cm. This reader has an approximately 20 cm square loop antenna in the housing and this system communicates using electromagnetic induction.

648

K. Fujisaki and Y. Yoshigai

Table type RFID reader

Square open loop coil

RFID tag

Fig. 2. Photo of 13.56 MHz table type RFID reader ST-RW01, RFID tag, and square open loop coil

By using these three items, we carried out the evaluation of the reading performance of RFID system. In Fig. 3 is shown the evaluation image. The parasitic element is placed on the height h [cm] from the reader face parallely, and we evaluate the reading area where communication is enabled on the observation point at each communication distance d [cm]. In this experiment, as a parasitic element, a 20 cm square open loop coil is used. We made some open loop coils

y

Clear acrylic board Tag

x d

Parasitic coil

h Loop antenna of RFID reader

Fig. 3. Evaluation image of communication performance in each distance between reader and target tag

Effect of Parasitic Element on Performance of 13.56 MHz RFID System

649

Fig. 4. Distribution of points where reader is able to communicate with tag when there is no parasitic element

by changing the number of turns and evaluated the communication performance using each coil.

4

Evaluation of Communication Performance Between Reader and Target Tag

The distribution of points where the reader was able to communicate with the target tag was evaluated in each communication distance between the reader and a tag. In this experiment, the coil of tag is put parallel to the loop antenna of reader.

650

K. Fujisaki and Y. Yoshigai

Fig. 5. Distribution of points where reader is able to communicate with tag in the case of the number of turns of a coil is 3 times

Figure 4 shows the distribution of points where reader is able to communicate with tag when there is no parasitic element. In this figure, the point with a red circle means that the reader could communicate with a tag. As shown in these figures, communication area decreases and becomes near the center of antenna of reader as the distance h between the tag and the reader increases. In order to expand the communication distance between the table type RFID reader and the tag, we use a open loop coil and changed the number of turns as a parasitic element. The open loop coil is placed on the height h [cm] from the reader face. In this paper, we show the results which a parasitic element is placed on h = 6 cm. Figures 5 and 6 show distribution of the point which can

Effect of Parasitic Element on Performance of 13.56 MHz RFID System

651

Fig. 6. Distribution of points where reader is able to communicate with tag in the case of the number of turns of a coil is 3.5 times

communicate with tag when the number of turns of a coil is made into 3 times and 3.5 times, respectively. As shown in these figures, by using a parasitic element the communication length was expanded. Especially, the improvement of the communication performance is large in the case when the number of turns of a coil is 3.5 times. However, in the case of the distance d = 3 and 6 cm the communication performance is degraded. Furthermore, in the case of the distance d = 15 cm the center of reading area could not communicate. These causes are unknown and are under investigation now. Finally, in Fig. 7 and Table 1, we show the relation between the number of turns of the parasitic coil and the reading area rate in each communication

652

K. Fujisaki and Y. Yoshigai

Reading area rate [%]

100

50

0

0

3

6

9

12

15

18

Communication distance between reader and tag : d [cm] N: The number of turns of parasitic coil no coil N=1 N = 1.5 N=3 N = 3.5 N=4

N=2 N = 4.5

N = 2.5 N=5

Fig. 7. Relation between the number of turns of a coil and reading area rate in each communication distance d between reader and tag

distance d. By using parasitic element, it is clearly shown that the reading area can be expanded and the communication distance can be extended. However, the communication performance does not improve only by increasing the number of Table 1. Relation between the number of turns of a coil and reading area rate in each communication distance d between reader and tag No. of turns of parasitic coil

Distance between reader and tag: d [cm]

0

82.54 65.53 58.73 43.49 14.05

0

3

6

9

12

15 0.00

18 0.00

1

82.54 64.79 58.43 43.49 14.50

0.00

0.00

1.5

82.54 66.42 58.88 44.97 16.12

0.00

0.00

2

82.99 65.53 59.91 46.60 18.20

0.00

0.00

2.5

83.14 66.86 63.02 52.37 25.74

0.00

0.00

3

82.99 67.46 64.50 55.77 36.39

5.03

0.00

3.5

80.62 56.51 12.87 73.22 68.79 43.20 48.96

4

82.99 30.18 17.01

7.10

2.07 39.20 29.14

4.5

81.36

9.76 51.33

7.99

0.00

0.00

0.00

5

81.51 49.70 21.45

0.00

0.00

0.00

0.00

Effect of Parasitic Element on Performance of 13.56 MHz RFID System

653

turns of the open loop coil simply. The number of turns of the open loop coil requires optimization.

5

Conclusion

In this paper, we show the effect of the parasitic element put on the fixed position of the table type RFID reader to the communication performance. By using a open loop coil between the reader and the tag as a parasitic element, the points where communication is enabled with tag increased as the number of turns of coil increased. Furthermore, the parasitic element expanded the communication length. However, the communication performance does not improve only by increasing the number of turns of the open loop coil simply. The number of turns of the open loop coil requires optimization. In the future work, we want to evaluate in details the influence of parasitic element to the communication performance and to suggest a method to improve the communication performance by using parasitic element.

References 1. Basat, S.S., Kyutae, L., Laskar, J., Tentzeris, M.M.: Design and modeling of embedded 13.56 MHz RFID antennas. In: Proceedings of 2005 IEEE International Symposium on Antennas and Propagation Society, pp. 64–67 (2005). https://doi.org/ 10.1109/APS.2005.1552740 2. Cantatore, E., Geuns, T.C.T., Gelinck, G.H., et al.: A 13.56-MHz RFID system based on organic transponders. IEEE J. Solid-State Circuits 42(1), 84–92 (2007). https://doi.org/10.1109/JSSC.2006.886556 3. Fujisaki, K.: Implementation of a RFID-based system for library management. Int. J. Distrib. Syst. Technol. 6(3), 1–10 (2015). https://doi.org/10.4018/IJDST. 2015070101 4. Fujisaki, K.: Evaluation and measurements of main features of a table type RFID reader. J. Mob. Multimed. 11(1–2), 21–33 (2015) 5. Fujisaki, K.: Evaluation of table type reader for 13.56 MHz RFID system considering distance between reader and tag. In: Proceedings of the 21st International Conference on Network-Based Information Systems, vol. 22, pp. 56–64 (2019). https://doi.org/10.1007/978-3-319-98530-5 5 6. Fujisaki, K.: Evaluation of 13.56 MHz RFID system performance considering communication distance between reader and tag. J. High Speed Netw. 25(1), 61–71 (2019). https://doi.org/10.3233/JHS-190603 7. Li, N., Gerber, B.B.: Performance-based evaluation of RFID-based indoor location sensing solutions for the built environment. Adv. Eng. Inf. 25(3), 535–546 (2011). https://doi.org/10.1016/j.aei.2011.02.004 8. Minami, T.: RFID tag based library marketing for improving patron services. In: Advances in Knowledge Acquisition and Management, vol. 2303 (2006). https:// doi.org/10.1007/11961239 5 9. Potyrailo, R.A., Morris, W.G., Sivavec, T., et al.: RFID sensors based on ubiquitous passive 13.56-MHz RFID tags and complex impedance detection. Wirel. Commun. Mob. Comput. 9(10), 1318–1330 (2009). https://doi.org/10.1002/wcm.711

654

K. Fujisaki and Y. Yoshigai

10. Prasad, N.R.K., Rajesh, A.: RFID-based hospital real time patient management system. Int. J. Comput. Trends Technol. 3(3), 509–517 (2012) 11. Symonds, J., Seet, B.C., Xiong, J.: Activity inference for RFID-based assisted living applications. J. Mob. Multimed. 6(1), 15–25 (2010) 12. Tajima, M.: Strategic value of RFID in supply chain management. J. Purch. Supply Manag. 13(4), 261–273 (2007). https://doi.org/10.1016/j.pursup.2007.11.001 13. Yoshigai, Y., Fujisaki, K.: Evaluation of 13.56 MHz RFID system considering tag magnetic field intensity. In: Proceedings of the 21st International Conference on Network-Based Information Systems, vol. 1036, pp. 620–629 (2020). https://doi. org/10.1007/978-3-030-29029-0 61 14. Zhonga, R.Y., Dai, Q.Y., Qu, T., et al.: RFID-enabled real-time manufacturing execution system for mass-customization production. Robot. Comput.-Integr. Manuf. 29(2), 283–292 (2013). https://doi.org/10.1016/j.rcim.2012.08.001

The 10th International Workshop on Cloud, Wireless and e-Commerce Security (CWECS-2019)

Perception Mining of Network Protocol’s Stealth Attack Behaviors Yan-Jing Hu1,2 and Xu An Wang1,2(&) 1

2

Network and Information Security Key Laboratory, Engineering University of the Armed Police Force, Xi’an 710086, China [email protected], [email protected] National Key Laboratory of Integrated Services Networks, Xidian University, Xi’an 710071, China

Abstract. Unknown network protocol’s stealth attack behavior is becoming a new type of attack, which greatly harms the cyber space security. The stealth behaviors are not easy to be detected by existing security measures. Starting with the implementation of the instructions of the protocol programs, the normal behavior instruction sequences are captured by dynamic binary analysis. The algorithm of instruction clustering and feature distance computation is designed to mine the potential stealth attack behavior instruction sequences. The mined stealth attack behavior instruction sequences (for inline assembly) are loaded into the general executing framework. A virtual protocol behavior analysis platform HiddenDisc has been developed, and the Dynamic analysis is implemented on the platform. Then the protocol execution security evaluation scheme is proposed and implemented. Using the stealth transformation method designed by ourselves, the stealth attack behaviors are transformed. We successfully attacked the virtual target machine by using the transformed stealth attack behaviors, but the stealth behaviors were not captured. The experimental results show that the present method can accurately and efficiently perception mining unknown protocol’s stealth attack behaviors, transform and use of stealth attack behavior can also enhance our information offensive and defensive capabilities. Keywords: Protocol reverse


Stealth attack behavior
Instruction clustering

1 Introduction Network protocol is the fundamental cornerstone of cyberspace, it is also one of the most important infrastructures of the information age. The stealth attack behavior of the protocol is based on the experience of traditional attack methods, and incorporates the concealment technology that specifically circumvents the detection and forensics of security defense facilities, further enhancing the concealment, pertinence and deception of the attack behavior, giving personal privacy information and Public infrastructure poses a serious security threat [1–3]. In recent years, some private network protocols secretly steal important information of the target network through carefully designed stealth functions. On the other hand, they use the vulnerabilities of existing protocols to silently embed hidden functions into normal protocols. Such protocols are long-term © Springer Nature Switzerland AG 2020 L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 657–669, 2020. https://doi.org/10.1007/978-3-030-33506-9_60

658

Y.-J. Hu and X. A. Wang

latent, and only trigger execution under special conditions and special circumstances. After completing the task, they continue to crouch and control the target host for a long time without being discovered. Such stealth attack behavior usually hides or pretends to be mixed into the protocol binary code. Therefore, this important information is suitable for reverse retrieval from the binary code that implements the protocol. However, this work is facing great challenges. Because the stealth attack behaviors we face are diverse, and under normal conditions they are difficult to perceive and mine. Some protocols have malicious features, special modules, or critical code snippets that are encrypted, spoofed, or confusing. The malicious behavior of other protocols is embedded in normal behavior and can only be triggered under special conditions. Due to the increasingly complex and diverse application protocols, existing signature scanning techniques, integrity detection techniques, and network behavior anomaly detection techniques are almost ineffective in the face of endless stealth behavior [4]. Traditional network security technologies are almost ineffective for the perception and mining of stealth attack behaviors [5]. More seriously, the intervention of security technology is likely to interfere with normal network communication. Because a hidden malicious behavior does not replicate and spread itself, or even has obvious malicious features. The increasing concealment, robustness and survivability of stealth attack behavior make traditional analysis, tracking and recognition more difficult. Protocol’s stealth attack behavior continues to erode the lines of defense built by traditional security measures. While inheriting traditional technologies, it gradually goes deeper into the bottom of the system, and draws on new technologies and social engineering principles to continuously enhance concealment and aggression. Therefore, how to master a common analysis method, which can quickly and accurately identify and mine the stealth attack behavior of unknown protocols will become a new challenge for cyberspace security. Mining the protocol’s stealth attack behavior is also one of the key issues that the protocol reverse analysis cannot avoid. The protocol’s stealth attack behavior can successfully achieve the attack target through the network protocol, which is difficult to detect by existing security devices and technologies [1, 3, 6]. The protocol reverse analysis technique [7] is an effective method. The current protocol reverse analysis mainly focuses on two major tasks, one is the identification of the message format, and the other is the identification of the protocol behavior pattern [8]. The analysis methods are roughly divided into two categories: network message flow analysis based and instruction sequence analysis based. Network traffic analysis method directly analyzes the network traffic of unknown protocols to obtain message format and internal structure information. In addition, the meaning of each field in the message and the semantic information of the message text can be further inferred. The instruction sequence analysis method mainly analyzes the instruction sequence of the protocol entity in the communication process. This method appears to be later than the former, but it has rapidly emerged and flourished in recent years, becoming the most powerful method for protocol reverse analysis. The theoretical basis of this method is that the protocol entity resolves the protocol message according to the protocol specifications. By monitoring and analyzing the process of protocol protocol parsing of protocol messages and the use of message fragments, the internal structure and semantic information of the protocol messages can be accurately obtained. In order to capture the instruction sequences in

Perception Mining of Network Protocol’s Stealth Attack Behaviors

659

the protocol message processing, the current research mainly focuses on the dynamic stain analysis technology, which is a technique for dynamically tracking and analyzing the message parsing process at the instruction level.

2 Related Work Numerous attempts have been made to reverse the protocol’s behaviors generated by the network applications, which is crucial for cyberspace security [9–12]. Some existing approaches employ Obfuscation techniques many times [13–15]. The same protocol can be changed to unrecognized and inheritance invisible through repeated obfuscation and processing [16]. The implementation of reverse analysis becomes more and more difficult, and it is also more and more difficult to capture and Mining [17]. To implement the protocol’s stealth attack, we need to execute the corresponding instruction sequences, but the execution time is difficult to grasp, and the instruction sequences are difficult to mine. In this paper, we try to make remarkable progress by using the instruction clustering method [18]. Due to the limitation of protocol message information, relatively little useful information can be obtained from protocol message reverse analysis, such as the behavior of protocol is difficult to get only from network traffic. There’s nothing you can do about it. Lim et al. [19] Extract Protocol Message Format and protocol output data by Static Analysis Program Binary Code. Such methods require the user to provide a prototype function for the input to the output buffer, which is often difficult to provide, such as the ability to write data when the program has no output. This kind of static analysis method process is often very complex, because there is no actual implementation of the protocol, so the information obtained is indirect. Since the output message is analyzed, the method cannot extract the format of the received message or infer the semantics of the field, let alone study the behavior of the protocol. Static analysis is done under the premise that the protocol program is not actually running. It can be used as the first step of protocol behavior analysis, so it will not damage the operating system. However, it is precisely because it does not need to run the code that the key information such as the logical structure and system call of the program can be hidden at the code level due to the present technical means such as shell-adding, encryption, compression, polymorphic deformation, etc. It can only be released when the code is actually executing. For a thorough analysis of a protocol with potentially aggressive behavior, static analysis techniques fall far short of what we need. The static analysis technology based on disassembly is difficult to extract the key information from the sample because of the complexity and poor performance of the Assembly Code. The static analysis method [20] is widely used in anti-virus technology based on static feature scanning, but the feature-based detection technology cannot identify the malicious behavior after deformation. Therefore, the focus of the research has shifted to the Syntax and semantic analysis of the protocol, which can detect malicious behavior by semantic analysis and identifying the characteristic codes between different variables. However, these methods do not perform well in analyzing stealth behavior. Then, the anti-detection of protocol program is attempted by using Fuzzy technology,

660

Y.-J. Hu and X. A. Wang

and the method of stack analysis system adjustment and code normalization is proposed [20]. A significant limitation of analyzing protocol messages only is that the information available from protocol messages is very limited and it is difficult to analyze cryptographic protocols. In order to make up for the shortcomings of this method, the new research trend is reverse analysis of the implementation of protocol procedures. The dynamic binary program analysis technology is the way to combine protocol message and Protocol Program. Polyglot [21] is the first automated reverse analysis protocol binary system that uses dynamic binary analysis technology, but it can only extract the received message format. The Dispatcher [22] has improved Polyglot to extract the format of outgoing messages, and it has developed semantic inference capabilities to analyze both outgoing and incoming messages. After the release of Polyglot, reverse protocol analysis using dynamic binary analysis technology has been springing up all over the world. However, the reverse analysis of the protocol is still focused on the extraction of the syntax level of the protocol, and further research is only to infer the protocol state machine. It seems that the main work of protocol reverse analysis is to get the protocol message format, field and infer the protocol state machine. However, our research shows that the protocol reverse analysis is much more than that. In fact, the behavior of protocol has more direct and fundamental influence on network security.

3 Perceptual Mining of Protocol’s Stealth Attack Behavior 3.1

System Framework Design

Through monitoring and tracking the execution process of a large number of unknown network protocols, we find that the protocol program can be used to analyze the protocol message parsing process to explore the stealth attack behaviors of the protocols, and then evaluate the operational credibility of the protocols. The overall framework of the system is shown in Fig. 1. As shown in Fig. 1, the design goal of the system is to automatically analyze the behavior of the unknown protocol, obtain the disclosure behavior instruction sequences of the protocol, and then perform instruction cluster analysis to perceive and mine the message formats, stealth attack behaviors and trigger conditions of the protocols. The behavioral calculation distinguishes the stealth attack behavior from the disclosure behavior, and finally triggers, monitors and analyzes these stealth attack behaviors and evaluates the credibility of the protocols. 3.2

Implementation of Analysis and Evaluation

3.2.1 Acquisition of Protocol Disclosure Behavior Instruction Sequence We integrated the dynamic stain propagation analysis technology into the HiddenDisc analysis platform. Each byte of the protocol message is marked as a stain source, and each instruction of the protocol program process protocol message is recorded. Based on these instruction sequences and observing the propagation of the source of contamination in registers and memory, the disclosure behavior instruction sequences can be captured. It should be noted that the original basic block is generated by a dynamic

Perception Mining of Network Protocol’s Stealth Attack Behaviors

661

Protocols

Dynamic instruction sequence extractor

Disclosure behavior instruction sequences

Static instruction clustering analysis

Stealth behavior trigger condition

Stealth behavior instruction sequences

Sensitive message generation

Dynamic regression testing analysis

Protocol execution security evaluation report

Fig. 1. System architecture of stealth attack behavior perception mining

execution monitor, and each basic block is composed of a disassembled instruction sequence. Since the nature of the protocol behavior is mainly related to the operation code, we remove all the operands (parameters) and reduce the original basic block to a disclosure behavior instruction sequence containing only the operation code, which lays a good and solid foundation for further instruction clustering analysis.

662

Y.-J. Hu and X. A. Wang

3.2.2 Perceptual Mining Algorithm for Protocol’s Stealth Attack Behaviors The protocol’s stealth attack behavior is essentially a deceptive behavior. Because its nature and intention are concealed, it is difficult to disclose it in advance, which increases the security risk of the protocol. This section mainly studies the perceptual mining algorithm of protocol’s stealth attack behaviors. We performed an n-gram analysis of all instruction sequences for the 2063 protocol samples in the protocol sample set, yielding a normal n-gram distribution representing all protocol behaviors. The stealth attack behavior hidden in the protocol program binary code is perceived and mined by the instruction clustering algorithm. Algorithm 1 can divide all the behaviors of all protocols in the sample set, and can perceive and expose the stealth attack behavior instruction sequences. Algorithm 1 is described as follows.

Perception Mining of Network Protocol’s Stealth Attack Behaviors

663

Potential stealth attack behaviors can be perceived by calculating the behavioral distance between all sequences of unknown behavioral instructions and known sequences of public behavioral instructions. For example, behaviors A and B are the two public behaviors we captured, and their functions are known. There is an unknown protocol P, and its behavior is unknown. We can distinguish behaviors A and B by calculating the behavior patterns PA and PB of behaviors A and B. If an unknown behavior Pi is very close to the behavioral pattern PA, i.e. d (Pi, PA) = 0, d is the behavioral distance function, which means that the unknown behavior Pi and the known behavior A are very close, they can be classified into the same behavior cluster. If the known behavior A is credible, then the unknown behavior Pi is also considered to be credible. By the same token, if Pi and the known behavior B are very close, it can form a behavior cluster with behavior B. If the n-gram instruction characteristics of behavior Pi are far from all known trusted behavior patterns, we consider Pi to be a new stealth attack behavior pattern and generate a new cluster separately. Whether its behavior is really untrustworthy, you need to actually trigger it to verification run. 3.2.3 Automatic Generation of Sensitive Messages The instruction clustering algorithm can mine the potential stealth attack behaviors of unknown protocols, but whether these behaviors must be untrustworthy behaviors and the specific content of the stealth behaviors need further analysis and verification. The mined potential stealth attack behavior instruction sequences will run in our selfdeveloped virtual protocol analysis platform HiddenDisc. Since the stealth attack behavior can trigger execution under special conditions, it is necessary to construct a sensitive message to trigger it. The sensitive message generation algorithm is based on all sequences of instructions, trigger conditions, and captured protocol messages mined from the binary code. As shown in Algorithm 2.

664

Y.-J. Hu and X. A. Wang

According to the new message instance generated by Algorithm 2, the parsing process of the new protocol message is dynamically monitored again, and the stealth attack behaviors of the protocols are triggered, captured, and analyzed until all the stealth attack behaviors are mined. In the smudge propagation analysis of the protocol message data, the special control flow graph generated by all conditional jumps in the protocol program parsing message process is called a conditional jump graph, G ¼ hV; Ei. The node V in the figure consists of a branch point slice and a basic block. The comparison constant is the value of the protocol message field or part of it, and may also be the special data contained in the protocol message and the protocol program binary code. Edge E is the control flow relationship between basic blocks or conditional jumps. The basic block we define refers to a series of sequential execution instructions between two conditional jump instructions during protocol message parsing. Conditional jump diagrams are semantic abstractions of protocol behavior that represent control dependencies and data dependencies between protocol behaviors. The new message generated by Algorithm 2 triggers the execution of the stealth attack behaviors, an example is shown in Fig. 2, where the stealth path consists of the basic blocks represented by the dashed boxes.

B1

public block condition jump block public flow

B2

stealth flow stealth block

B3

B4

B5

B6

B13

B7

B8

B9

B10

B14

B15

B16

B11

B17

B12

B18

Fig. 2. Example of triggering stealth attack behavior

As shown in Fig. 2, in the above conditional jump diagram generated by the cooperation of Algorithm 1 and Algorithm 2, two stealth paths are exposed, and two stealth attack behavior instruction sequences can be obtained, respectively: H1 = {B5,

Perception Mining of Network Protocol’s Stealth Attack Behaviors

665

B13, B14}, H2 = {B5, B13, B15, B16, B17, B18,}. Based on the perception and mining of the stealth attack behaviors, the above algorithm generates the stealth attack behavior instruction sequences constructed with basic blocks. Finally, the execution credibility of the protocol is evaluated based on the results of the stealth behavior analysis, no further elaboration.

4 Experiment and Evaluation 4.1

Analysis Platform Construction

Through a large number of long-term protocol instance sample analysis and technology accumulation, we extended the two functions of instruction sequence clustering mining and sensitive message automatic generation based on the open source virtual machine TEMU, and developed the protocol behavior virtual analysis platform-HiddenDisc prototype system. At present, 2063 protocol samples have been analyzed, 32 normal protocols have been mastered, and 271 malicious behavior protocols have been analyzed and confirmed. The other protocols are unknown. The six analyzers are networked PCs and the HiddenDisc prototyping systems are deployed on them. The host machine is installed with Ubuntu linux operating system, one control server is installed with Windows Server 2008 operating system, one analysis server is installed with Ubuntu linux operating system, and the hardware protocol analyzer RADCOM is connected. The rest of the analysis machines are installed with Wireshark software. Since the unknown stealth attack behavior may have security risks, the analysis of all protocol samples is performed on the virtual analysis platform HiddenDisc. Due to the uncertainty of the triggering of the stealth attack behavior, we adopt a combination of online acquisition and offline analysis. While continuously recording the protocol sample flow data, the instruction sequences of the protocols are intelligently monitored, which makes the message flows and the instruction sequences complement each other, and mutual confirmation. 4.2

Experimental Results and Analysis

Analysis of the behavioral instruction sequences of the Android system on HiddenDisc, combined with data traffic analysis, found that the behavior invoked a self-hiding API. Table 1 shows the origin of each self-hiding behavior and the calling API. For example, to detect the self-hiding behavior of a “hidden application”, you need to analyze and check if the start Service().on Receive() accessed at startup in Context.in. Broadcast Receiver is called.

666

Y.-J. Hu and X. A. Wang Table 1. Analysis of self-hiding call behavior

Self-hiding behavior Hide application

Origin

BroadcastReceiver.onReceive (ACTION_BOOT_COMPLETED/ SMS_RECEIVED/ NEW_OUTGOING_CALL/ SCREEN_OFF/ PACKAGE_INSTALL/ PACKAGE_ADDED/SIG_STR) Hide activity Any entry point Delete email BroadcastReceiver.onReceive (SMS_RECEIVED/ ACTION_VIEW) BroadcastReceiver onReceive Delete call history (PHONE_STATE) PhoneStateListener OnCallStateChanged (TelephonyManager. CALL_STATE_RINGING) BroadcastReceiver.onReceive Block message (SMS_RECEIVED) Block calls BroadcastReceiveronReceive (PHONE_STATE) PhoneStateListener OnCallStateChanged (TelephonyManager. CALL_STATE_RINGING) Hide Any entry point reminder Hide BroadcastReceiver.onReceive() notification Electrostatic Any entry point telephone

Delete system log

Any entry point

Self-hiding call API Context. startService()

Window.addFlags (FLAG_NOT_TOUCH_MODAL) ContentResolver.Delete (“content: // sms”) ContentResolver.Delete (“content: // cal_log / calls”)

BroadcastReceiver.abortBroadcast (“content: // sms”) ITelephony.endCall()

Context.sendBroadcast(Intent. ATION_CLOSE_SYSTEM_DIALOGS) NotificationManager.cancel() NotificationManager.cancelAll() Vibrator.Cancel() Audio Manager.set Ringer Mode (RINGER_ MODE_ SILENT) Audio Manager. Set Stream Mute(true) Audio Manager. Adjust Stream Volume (ADJUST_LOWER) Runtime.Exec(“logcat -c”)

Perception Mining of Network Protocol’s Stealth Attack Behaviors

667

For example, to check the “delete message” behavior, first track Broadcast Receiver.On Receive (SMS_RECEIVED/ACTION_VIEW) to see if it accesses Content Resolver.Delete().Next, look in the reverse view of the message flow def-use chain with “content:// sms” as a parameter. Take the malware Droid KungFu1 as an example. It deletes all SMS messages. After removing the irrelevant instructions, only the simplified decompiled code is shown as follows:

The hypothesis analysis determines that the call to Content Resolver.Delete() can be accessed from Broad cast Receiver.on Receive (PHONE_STATE). To check the value of the Delete() parameter, look back at the def-use chain from $r5 (used on line 4 and defined on line 3). When line 3 calls the method parse(), continue with $r4 and then $r3. Finally, on the first line, I saw the definition, which is “content:// sms /”. Analysis of this can determine the potential “delete message” behavior.

5 Conclusion and Outlook In order to deal with the characteristics of the stealth behavior of network protocols, this paper proposes and designs two algorithms of perceptual mining and sensitive message generation of stealth attack behaviors, and develops the HiddenDisc protocol behavior simulation analysis platform. Through the self-developed instruction clustering algorithm, 2063 protocol samples were analyzed and evaluated. For the first time, the behavior of network protocols was clustered successfully at the instruction sequence level, and 138 hidden behaviors were perceived and mined in 173 s. After a long and repeated path coverage test, in addition to confirming these 138 hidden behaviors, there is no new stealth behavior discovery, and both the analysis efficiency and the reliability have reached the expected goal. However, the types of stealth attack behaviors and the means of attack vary widely. At present, there is only a preliminary understanding of the credibility of the protocol execution. The relationship between the stealth behavior and the execution credibility needs to be further studied. Next, we will continue to improve the HiddenDisc virtual operation and analysis platform, improve the accuracy and versatility of Algorithms 1 and 2, improve the ability to perceive complex and hidden behaviors, conduct in-depth research and continuously improve the credibility evaluation scheme of protocol operation, and In contrast, two methods are used to study the stealth attack methods and the reliable techniques of anti-hidden behavior. Acknowledgements. This work is supported by the National Key Research and Development Program of China Under Grants No. 2017YFB0802000, National Cryptography Development Fund of China Under Grants No. MMJJ20170112, the Natural Science Basic Research Plan in Shaanxi Province of china (Grant Nos. 2018JM6028), National Nature Science Foundation of

668

Y.-J. Hu and X. A. Wang

China (Grant Nos. 61772550, 61572521, U1636114, 61402531, 61103178, 61373170, 61402530, 61309022 and 61309008.), Engineering University of PAP’s Funding for Scientific Research Innovation Team (grant no. KYTD201805).

References 1. Harale, A., Tambe, S.: Detection and analysis of network & application layer attacks using honey pot with system security features. Int. J. Adv. Res. Ideas Innov. Technol. 3, 1–4 (2017) 2. Meng, B., et al.: DDOS attack detection system based on analysis of users’ behaviors for application layer. In: 2017 IEEE International Conference on Computational Science and Engineering (CSE) and IEEE International Conference on Embedded and Ubiquitous Computing (EUC) 2017, pp. 596–599 (2017) 3. Wang, Y., Yang, J.: Ethical hacking and network defense: choose your best network vulnerability scanning tool. In: 31st International Conference on Advanced Information Networking and Applications Workshops (WAINA) 2017. IEEE Conference Publications, pp. 110–113 (2017) 4. Bateman, W.M., Amaya, A., Fenstermaker, J.: Securing the grid and your critical utility functions. In: 2017 IEEE Rural Electric Power Conference (REPC) 2017, pp. 29–37 (2017) 5. Dooley, M., Rooney, T.: DNS vulnerabilities. In: DNS Security Management 2017, p. 324. Wiley-IEEE Press (2017) 6. Almubairik, N.A., Wills, G.: Automated penetration testing based on a threat model. In: 11th International Conference for Internet Technology and Secured Transactions (ICITST) 2016, pp. 413–414. IEEE Conference Publications (2016) 7. Narayan, J., Shukla, S.K., Clancy, T.C.: A survey of automatic protocol reverse engineering tools. ACM Comput. Surv. 48(3), 1–26 (2015) 8. Zhang Zhao, W.Q.-Y., Wen, T.: Survey of mining protocol specifications. Comput. Eng. Appl. 49, 1–9 (2013) 9. Luo, X., et al.: A type-aware approach to message clustering for protocol reverse engineering. Sensors 19(3), 716 (2019) 10. Votipka, D., et al.: An observational investigation of reverse engineers’ process and mental models. In: Extended Abstracts of the 2019 CHI Conference on Human Factors in Computing Systems 2019, pp. 1–6. ACM, Glasgow (2019) 11. Li, P., Mao, K.: Knowledge-oriented convolutional neural network for causal relation extraction from natural language texts. Expert Syst. Appl. 115, 512–523 (2019) 12. Bossert, G., Guihéry, F., Hiet, G.: Towards automated protocol reverse engineering using semantic information. In: Proceedings of the 9th ACM Symposium on Information, Computer and Communications Security 2014, pp. 51–62. ACM, Kyoto (2014) 13. Koganti, V.S., Galla, L.K., Nuthalapati, N.: Internet worms and its detection. In: International Conference on Control, Instrumentation, Communication and Computational Technologies (ICCICCT) 2016, pp. 64–73. IEEE Conference Publications (2016) 14. Pawlowski, A., Contag, M., Holz, T.: Probfuscation: an obfuscation approach using probabilistic control flows. In: Caballero, J., Zurutuza, U., Rodríguez, R. (eds.) Detection of Intrusions and Malware, and Vulnerability Assessment: Proceedings of the 13th International Conference, DIMVA 2016, San Sebastián, Spain, 7–8 July 2016, pp. 165–185. Springer, Cham (2016)

Perception Mining of Network Protocol’s Stealth Attack Behaviors

669

15. Xie, X., et al.: Mixed obfuscation of overlapping instruction and self-modify code based on hyper-chaotic opaque predicates. In: Tenth International Conference on Computational Intelligence and Security 2014, pp. 524–528. IEEE Conference Publications (2014) 16. Payer, M.: HexPADS: a platform to detect “stealth” attacks. In: Caballero, J., Bodden, E., Athanasopoulos, E. (eds.) Engineering Secure Software and Systems: Proceedings of the 8th International Symposium, ESSoS 2016, London, UK, 6–8 April 2016, pp. 138–154. Springer, Cham (2016) 17. Karim, A., et al.: Botnet detection techniques: review, future trends, and issues. J. Zhejiang Univ. Sci. C 15(11), 943–983 (2014) 18. Abul Hasan, M.J., Ramakrishnan, S.: A survey: hybrid evolutionary algorithms for cluster analysis. Artif. Intell. Rev. 36(3), 179–204 (2011) 19. Lim, J., Reps, T., Liblit, B.: Extracting output formats from executables. In: Proceedings of the Working Conference on Reverse Engineering, Benevento, Italy (2006) 20. Egele, M., et al.: A survey on automated dynamic malware-analysis techniques and tools. ACM Comput. Surv. 44(2), 1–42 (2012) 21. Caballero, J., Yin, H., Liang, Z., Dawn, S.: Polyglot: automatic extraction of protocol message format using dynamic binary analysis. In: Proceedings of the 14th ACM Conference on Computer and Communications Security, pp. 317–329 (2007) 22. Caballero, J., Poosankam, P., Kreibich, C., Song, D.: Dispatcher: enabling active botnet infiltration using automatic protocol reverse-engineering. In: Proceedings of the 16th ACM Conference on Computer and Communications Security, pp. 621–634 (2009)

Digital Image Anti-counterfeiting Technology Chin-Ling Chen1,2,3, Chin-Feng Lee4, Fang-Wei Hsu5, Yong-Yuan Deng1(&), and Ching-Cheng Liu6

4

1 Department of Computer Science and Information Engineering, Chaoyang University of Technology, 168 Jifeng E. Rd., Wufeng District, Taichung 41349, Taiwan, R.O.C. [email protected], [email protected] 2 School of Information Engineering, Changchun Sci-Tech University, Changchun 130600, Jilin, China 3 School of Computer and Information Engineering, Xiamen University of Technology, Xiamen, People’s Republic of China Department of Information Management, Chaoyang University of Technology, Taichung 41349, Taiwan, R.O.C. [email protected] 5 Department of Management Information Systems, National Chung Hsing University, Taichung, Taiwan [email protected] 6 Department of Information and Communication Engineering, Chaoyang University of Technology, 168 Jifeng E. Rd., Wufeng District, Taichung 41349, Taiwan, R.O.C. [email protected]

Abstract. There are many trademarks and logos around the world. Every trademark or logo represents its own brand or intellectual property. However, some people who are interested will benefit from counterfeiting trademarks or logos. As the number of counterfeit cases has increased, the technology of anticounterfeiting has gradually been taken. Watermark is the protection method of today’s trademarks. Since the naked eye cannot guarantee the recognition rate, this study uses scientific and technical testing methods to distinguish the authenticity of the trademark. Through the method of this study combined with watermarking technology, not only the problem of misjudgment is improved, but also the location and area where the image has been tampered with can be accurately determined. It also enhances image recovery and image quality in tamper areas. Keywords: Trademark protection technology




Anti-counterfeiting




Watermarking

1 Introduction Due to the rapid development of technology, various printing technologies have also been greatly improved, and various trademarks and logos have been designed with many anti-counterfeiting treatments. However, in the face of high-resolution reversal and the current mature printing technology used for counterfeiting, the current © Springer Nature Switzerland AG 2020 L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 670–677, 2020. https://doi.org/10.1007/978-3-030-33506-9_61

Digital Image Anti-counterfeiting Technology

671

trademark anti-counterfeiting is still likely to be copied and abused by malicious people. With the surge in global counterfeiting cases, anti-counterfeiting technology is growing at a rate of 30% per year. The digital watermarking technology has become the new darling of anti-counterfeiting technology because of its ability to prevent unauthorized printing and copying without incurring huge cost. The watermarking technology embeds the digital watermark information into the multimedia information to be protected through some algorithm, in order to hide the digital watermark information. At present, digital watermarking technology has been widely used in images, audio and video. Therefore, this technology has become a new hot spot in the field of multimedia information research. Nowadays, the protection of trademarks is passive visual inspection for logo and the word stream. Nowadays, the protection of trademarks is passive visual inspection for logo and serial number. The purpose of this study is to identify the authenticity of a trademark by means of scientific and technological testing, relying on scientific and technological testing to reduce the factors of human identification and misjudgment. The current image authentication technologies can be divided into two types: encryption (CryptographyBased) [1, 2] and watermarking technology [3–8] (Watermark-Authentication). Encryption technology is the use of hash function to calculate the image authentication code. The comparison of the authentication codes determines whether the image has been tampered with during transmission. However, the disadvantage of encryption technology is that it cannot detect the location of the falsified area, and the digital signature is one of the common encryption type authentication techniques. In contrast, the watermarking technique calculates the authentication information of each image block and self-embeds every watermark code into the associated block. The position and area where the image has been tampered can be detected by comparing the watermark code of each block. The watermarking technology can be divided into robust watermarking [3, 4], fragile watermarking [5, 6] and semi-fragile watermarking [7, 8] by different technical requirements. Robust watermarking is often used for the certification of intellectual property rights and copyrights. Fragile watermarking is used to detect tampering and reply to images. Since the embedded watermark code is very sensitive to subtle changes, the location of the tampering can be accurately detected. The semi-fragile watermarking is a mixture of the robust watermarking and the fragile watermarking. Its ability to resist image processing and tamper detection is worse than the other two methods. The proposed method is a self-embedding watermarking method, which effectively reduces the false positive rate by embedded each block. This study not only can reduce the false positive rate, but also improve the image recovery ability of the tampering area.

2 Related Works QR Code is a matrix two-dimensional code. A specification developed by Denso Corporation of Japan in 1994. QR Code has been used in many fields due to its high reliability, high speed scanning and large data capacity. QR code information can be decoded through a dedicated card reader device or a normal smart phone. The QR code is currently licensed as an international standard published by ISO and has a large

672

C.-L. Chen et al.

number of real-world applications, including: information storage, website redirection, product tracking passenger identification, etc. The QR code not only brings us convenience but also privacy issues. The use of QR code through encryption anti-counterfeiting mechanism has been widely applied to identify the authenticity of various products. In general, the method of verifying the authenticity of a product is to add a laser tag to the packaging and printing, so that the consumer can judge the authenticity of the product from the appearance, but the laser tag counterfeiting is not easy to eliminate. In recent years, the business circles have used information technology to adopt the QR Code anticounterfeiting mechanism. The application of QR code in other areas such as preventing counterfeit drugs is a major task for public health care around the world. The increase in these drugs makes treatment potentially harmful and even fatal. Other application areas such as ID card, passport, visa, etc., including QR code reader, visual secret sharing mechanism and database comparison. As shown in Fig. 1 below, first obtain the QR code printed on the ID card, such as passport and ID card, through the mobile phone. Once the QR Code is obtained, the code in the corresponding QR Code database is obtained by the same device. As shown in Fig. 1(b), the system decodes each QR code to obtain a secret share value for each code. As shown in Fig. 1(c), the mechanism for using the visual secret sharing threshold is applied in this authentication mechanism. Anticounterfeiting work has been widely valued by major companies. If the secret image can be visualized by human vision, then the two obtained secret sharing values are stacked together. Through system verification, we can treat the ID file as real, otherwise it will be considered fake. The mechanism is that it does not require software. It can be implemented in a mobile phone and can be used by most people because the mobile device already uses a QR reader. Therefore, using QR Code is one of the effective ways to get undistorted. Since the system is implemented using secret images, security issues can be guaranteed. Due to the sharing of the database, the secret image values in the ID document cannot be known. The proposed method generates a QR code created using the algorithm shown in Fig. 2 [9]. (a)

Passport

(b)

Mobile

QR Data

(c)

Share

Share

Stacking Result

Fig. 1. QR Code authentication system [9]

Digital Image Anti-counterfeiting Technology

673

Fig. 2. Flowchart of algorithm [9]

3 Methodology For the anti-counterfeiting issue of digital images, this study proposes two methods as follows: Method 1: The scattering method of logo and serial number In this study, the logo and the serial number are first scattered in the form of seed [10]. Then, the “Digital Print Scanning”, “Plain Scanning”, and “Rolling Scanning” are used to print out the scattering logo and serial number. And the seed’s ordering method is used to reply the scattered logo and the serial number, thereby judging the legitimacy and legality of the trademark. This study uses different trademarks and font sizes, and uses different seeds to generate different scattered trademarks as shown in Fig. 3. The scattered logo and the serial number are reordered by taking a photo scan. If the similarity of the logo and the track after the seed is compared with the original image is higher than 90%, the legality of the trademark can be verified. Method 2: Anti-counterfeiting method based on QR code and pattern spectrum The QR code is square, and the largest feature is the upper left, upper right, and lower left three large like “回” words as the QR code to identify the positioning mark. Losing

674

C.-L. Chen et al.

Fig. 3. Scattered trademark and serial number

one of the positioning marks will affect the identification. The QR code is fault-tolerant, and if the QR code graphic is damaged, it can still be read by the machine. The general fault tolerance rate is 7%–30%. The higher the fault tolerance rate, the larger the area of the QR code pattern. The highest to 30% of the damage to the graphic area can still be read by the machine [11]. Method 2 will use singular value decomposition (SVD) to find out the characteristics of each block feature and generate block verification information to ensure image integrity. We embed a QR code in the original QR code or logo, embedding another QR code with the intention of high QR code fault tolerance [11]. In the case where the original QR code is highly corrupted, the general watermark may have been difficult to identify. However, in the case of the high tampering rate of the QR code, if edge detection is used to enhance the contour of the QR code, the extracted QR code has a high probability of being successfully read. Figure 4 is the overall flow of the study. First, the invoice number will be converted to a QR-code format, just like the current electronic invoice. Then convert the logo or related authentication information to be embedded into another QR code as the embedded information in the embedded process. Finally, generate QR code for watermark authentication after entering the embedding process and the authentication code generating. This method can also be used to protect the logo, just replace the step of invoice number conversion with the logo to be protected in Fig. 4.

Digital Image Anti-counterfeiting Technology

675

Convert Generate

Invoice number

Convert

Authentication code generation

Logo, Authentication message

Fig. 4. Flowchart of the study

The authentication code generation and embedding process of method 2 are shown in Fig. 5.

Fig. 5. Verification code generation and QR Code embedding process of this method

First, the QR code is divided into a plurality of 4  4 blocks, and each block is subjected to singular value decomposition to obtain a singular value matrix. Combine the singular values of each singular value matrix into exclusive-OR operations and combine them into a 16-bit authentication message. Next, convert a QR code to be embedded into a binary image for embedding as information. In order to make method 2 more convenient for anti-counterfeiting detection, the pattern spectrum will be added to the watermark image generated by method 2. Special logo or anti-counterfeit patterns will be printed on the watermark image by special transparent ink. Special spectral filters must be used to observe with the naked eye, as shown in Fig. 6.

676

C.-L. Chen et al.

4 Experimental Result The method 1 of this study will take a photo scan of scattered logo and serial number, followed by trademark reordering. And then, the re-ordered trademarks are compared. If the similarity between the logo and the serial number of the seed sorted by the original image is higher than 90%, the legality of the trademark can be verified. Good results can be obtained in the verification results in the electronic. However, it is verified by the photo scan that the image characteristics such as white balance and color temperature of the image are not as good as those of the electronic image, and the discrimination result is also very different. Therefore, the image correction experiment must be performed through the reference device parameters to obtain the photo scan processing mode parameters, which may make the results close to the electronic verification.

Fig. 6. Pattern spectrum printing anti-counterfeiting process

The expected results from the method 2 in electronic verification are: (1) Use block verification codes to accurately locate the area under attack. And through this method, it is decided whether to take out the QR code information in this block. (2) The QR code taken out under low tamper rate can be easily read. (3) In the case of high tampering rate, although the extracted QR code is lost, the extracted QR code can still be successfully read after the edge detection algorithm. The addition of the pattern spectrum enables the logo and various trademarks to be visualized using a specific spectrum to achieve a simple detection effect. The proposed method 2 can effectively possess the anti-counterfeiting function in both the electronic file and the finished product after printing.

Digital Image Anti-counterfeiting Technology

677

5 Conclusions In the method 1 of the proposed scheme, because current mobile devices and photographic technology can’t accurately grasp the position of each pixel in the scattered trademark and the serial number image. If the pixel position of the scattered image after photographing or scanning is slightly displaced, the success rate of sorting it into the original image by seed is low. So, this study is difficult to implement in practice. In the method 2, the uniqueness of the electronic file of the logo and the trademark can be simultaneously prevented and the tampering can be prevented. At the same time, it can protect the anti-counterfeiting by using the pattern spectrum after printing. The embedded technology used in this study can be applied to the general intellectual property protection in the future. Acknowledgments. This research was supported by the Ministry of Science and Technology, Taiwan, R.O.C., under contract number MOST 108-2221-E-324-013 and MOST 108-2221-E324-014.

References 1. Yan, G., Olariu, S., Weigle, M.C.: Providing VANET security through active position detection. Comput. Commun. 31(12), 2883–2897 (2008) 2. Hu, P.Y., Chang, C.: Novel image authentication scheme based on quadtree segmentation. Imaging Sci. J. 53(3), 149–162 (2005) 3. Parah, S.A., Sheikh, J.A., Loan, N.A., Bhat, G.M.: Robust and blind watermarking technique in DCT domain using inter-block coefficient differencing. Digit. Signal Process. 53, 11–24 (2016) 4. Singh, A.K.: Improved hybrid algorithm for robust and imperceptible multiple watermarking using digital images. Multimed. Tools Appl. 76(6), 8881–8900 (2017) 5. Zear, A., Singh, A.K., Kumar, P.: A proposed secure multiple watermarking technique based on DWT, DCT and SVD for application in medicine. Multimed. Tools Appl. 77(4), 4863– 4882 (2018) 6. Ansari, I.A., Pant, M., Ahn, C.W.: SVD based fragile watermarking scheme for tamper localization and self-recovery. Int. J. Mach. Learn. Cybern. 7(6), 1225–1239 (2016) 7. Lee, C.F., Shen, J.J., Chen, Z.R., Agrawal, S.: Self-embedding authentication watermarking with effective tampered location detection and high-quality image recovery. Sensors 19(10), 2267 (2019) 8. Chen, F., He, H., Huo, Y.: Self-embedding watermarking scheme against JPEG compression with superior imperceptibility. Multimed. Tools Appl. 76(7), 9681–9712 (2017) 9. Espejel-Trujillo, A., Castillo-Camacho, I., Nakano-Miyatake, M., Perez-Meana, H.: Identity document authentication based on VSS and QR codes. Procedia Technol. 3, 241–250 (2012) 10. Tiwari, A., Sharma, M., Tamrakar, R.K.: Watermarking based image authentication and tamper detection algorithm using vector quantization approach. AEU-Int. J. Electron. Commun. 78, 114–123 (2017) 11. Park, S.K., Miller, K.W.: Random number generators: good ones are hard to find. Commun. ACM 31(10), 1192–1201 (1988)

System Implementation of AUSF Fault Tolerance Wei-Sheng Chen1, Fang-Yie Leu1(&), and Heru Susanto2,3 1

2

Department of Computer Science, Tunghai University, Taichung City, Taiwan {g06350033,leufy}@thu.edu.tw Research Center for Informatics, The Indonesian Institute of Science, Jakarta, Indonesia [email protected] 3 Information Management Department, Tunghai University, Taichung City, Taiwan

Abstract. In this study, we deal with two topics. The first is that a machine, named Mediator, is added to a 5G system for managing and keeping track of UE’s authentication. The purpose is that when an AUSF fails, other AUSFs can successfully take over its authentication process on UEs. The second is that the proposed mechanism can detect this failure immediately and response properly, aiming to increase the QoS that an UE can receive from 5G networks. Experimental results show that the proposed system is better than those compared systems. Keywords: SDN


AUSF
vEPC
VM
EPS-AKA
Fault tolerance

1 Introduction As we know the functional complexity of IoT devices has been higher day by day. Also, 5G network will soon available to substitute for 4G. Some issues of 5G, for example, data transmission velocity and capacity and system reliability, have to be urgently enhanced. Generally, 5G Core (5GC) consists of control plane and data plane. The former, often implemented by SDN Controller, like ONOS, OpenDaylight, Ryu, etc. is used to control network entities of data plane, like switches and VMs, so as to provide users with data delivery services. In all control-plane functions, VM Machine Management is an important one. Currently, due to some reasons, e.g., business secret and un-developed features, to the best of our knowledge, no VM failure detection and take-over procedure can be found when an AUSF fails. Of course, if no such management mechanism is available, when UE’s authentication fails, UE can access this service by scanning nearby stations as a network reentry. But in this case, the time when UE is switched on to the time when UE successfully connects to the underlying network will be long. Basically, IoT devices often connect to a network intermittently to transmit the data they sense. After © Springer Nature Switzerland AG 2020 L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 678–687, 2020. https://doi.org/10.1007/978-3-030-33506-9_62

System Implementation of AUSF Fault Tolerance

679

delivering the data, they will disconnect from the network. The intermittent connection may result in AUSF’s and SDN controller’s operation burden. Therefore, in this study, we try to effectively manage AUSF pool to mitigate the burden of AUSF and SDN Controller, especially when network is busy. We employ a delicate machine, named Mediator, which records status of UE authentication. If an AUSF does not work properly during an UE’s authentication, the Mediator will assist available AUSFs to successfully take over those UEs authenticated by the failed, aiming to make 5G networks to provide high-quality services, and offer high bandwidth to carry IoT traffic. To ensure network reliability and availability, a system with fault tolerant mechanisms is required. The rest of this paper is organized as follows. Section 2 briefly describes related works and background of this study. Section 3 introduces the Mediator System. Simulation and discussion are presented in Sect. 4. Section 5 concludes this paper and addresses our future studies.

2 Literature Review and Background Schiller et al. [1] discussed related work of 3GPP networks and Mobile Edge Computing (MEC), and introduced architecture and implementation details of MEC platform and SDN controller. MEC could be implemented by using existing management techniques (i.e., excluding Network Function Virtualization (NFV)/Software Defined Networking (SDN)). However, NFV/SDN would greatly improve flexibility of network entity deployment and rapidly build network services at an edge computer. OVS performance in terms of throughputs for delivering smaller packets may not be higher than that of line rate of the interface. To overcome this limitation, OVS has been ported to Data Plane Development Kit (DPDK), namely OVDK. Shanmugalingam et al. [2] presented the experimental results of OVDK performance test when flow and port mirroring were activated. The OVDK achieves impressive line rate throughputs across physical interfaces. vMEC as a new IoT applications framework helps service provider to offer an improved user experience and a network platform that reduces network loading. Hsieh et al. [3] discussed the system model of vMEC and specified the traffic control mechanism. They also designed the architecture of vMEC IoT gateway platform and summarized the development steps of virtualized IoT gateway platform and performance analyses. Arins [4] proposed Firewall as a service for internet service providers (ISP), allowing end users to request and install match-action rules in ISPs edge routers. SDN Controller takes charge of translating high-level logics into low-level rules for OpenFlow switches. Openflow, a level – 2 protocol, is employed to protect users from network attacks. However, this study only supported three APIs for users to develop their high-level services. Additionally, it would be better for authors to survey the situation in which network entities fail.

680

W.-S. Chen et al.

3 Mediator System In our system, there is an AUSF pool, which is managed by SDN Controller. We also develop its APIs with which users can check the pool’s statuses. Any two AUSFs are connected by N14 interface [5]. It means all AUSFs are connected by using a complete graph. Figure 1 shows the 16 Steps [6] of an 5G-AKA authentication process. In 5G networks, since most entities employed are those with lower reliabilities compared with those of dedicated entities used by 4G/3G networks. The key reason is reducing deployment cost. When an AUSF, e.g., AUSFf, fails, we need to deal with the following problems.

Fig. 1. The sequence chart of 5G-AKA [6].

(1) The failure of AUSFf needs to be detected. (2) It would be better if alive AUSFs can handle all UEs currently authenticated by AUSFf to continue the authentication processes. In 4G networks, when dedicate network entities fail, network operators have to prepare a spare entity. Otherwise, no AUSF will take over for the failed, resulting in the fact that the system will be out of service. Users often request high-quality services. Services cannot be continued should be a problem. 3.1

Mediator

Basically, Mediator consists of Receiver and Writer. The former takes charged of receiving packets from other network entities. Writer is responsible for writing packet data into our Common Storage which is a database. Also, when an AUSF, e.g., AUSFi, 1  i  m, fails where m is the number of AUSFs currently in AUSF Pool, Mediator sends a notification message to SDN Controller. SDN Controller fairly allocates those UEs authenticated by AUSFi to other m − 1 AUSFs.

System Implementation of AUSF Fault Tolerance

681

3.1.1 AUSF to Common Storage Through Mediator When AUSFi fails, we divide an authentication step performed by an AUSF into three sub-steps, including receiving a packet, processing a packet and sending a packet. The symbol A-B represents the sub-step B of step A, A = 0, 4, 8 or 16, B = 1, 2, or 3, since the steps related to AUSF as shown in Fig. 1 are Steps 4, 8 and 16. But steps 16-2 and 16-3 are unavailable. The reason will be described later. In the Common Storage, i.e., the database, for each AUSF, we establish a Data Table, denoted by DataTablei, which records authentication data for the UE during UE’s authentication. When AUSFi sends a packet to Mediator, a flag, named Flag, is added. Its value can only be 0 or 1. Its initial value is zero. When Receiver receives a packet delivered by AUSFi, if Flag = 1, indicating that AUSFi is authenticating UE, Writer writes the packet data into DataTablei. The DataTable has a field Status which keeps track of the sub-steps of an authentication step. When an authentication sub-step is finished, the Status is then increased by 1. e.g., when AUSFi receives Auth-Info Response (Authentication Information Response, step 7) from ARPF, it sends the information to Mediator. Writer stores the information with Flag = 1 and Status = 8-1 in DataTablei, 1  i  m where m is the amount of AUSF. After processing this packet, Writer changes Status from 8-1 to 8-2. AUSF then transmits a 5G-AIA to SEAF (step9) and SEAF delivers Auth-Req (Authentication Request, step 11) to UE, Writer changes the UE’s authentication Status to 8-3. But when AUSF receives user 5G-AC (5G Authentication Confirmation) from SEAF (step 15), since Flag = 1 and it is the beginning of Step 16, Writer keeps the packet data, and Status = 16-1. After that it changes Status to 0 and Flag = 0. That is why, steps 16-2 and 16-3 do not exist. Following that, SEAF notifies gNB to provide this UE with network service. It is also the end of UE authentication. 3.1.2 Failure Identification by Mediator The Failure identification procedure is as follows. (1) Mediator periodically delivers a heartbeat to AUSFi to verify whether AUSFi is still alive or not. (2) If Mediator receives no reply from AUSFi for three times, it notifies SDN Controller to allocate those UEs currently authenticated by AUSFi to other alive AUSFs, following load balance policy with which a light load AUSF will be allocated more UEs. If UEj is distributed to AUSFk, SDN Controller authorizes AUSFk to access UEj and its data in DataTablei. (3) After that, AUSFk accesses UEj’s final authentication status, i.e., Status, from the tuple concerning UEj in DataTablei via Mediator, 1  k  n, 1  j  m, where n is the number of alive AUSFs and m is the number of UEs needed to be allocated, and continues UEj’s authentication. (4) However, after sending a heartbeat to AUSFi if Mediator receives an ACK from AUSFi, it will wait for the following packet sent by AUSFi. (5) After the authentication on an UE if Flag = 0, the tuple created for the UE in DataTablei will be removed by Mediator.

682

W.-S. Chen et al.

(6) If DataTablei is now empty, Mediator notifies SDN Controller to recover AUSFi to be VM again. (7) Mediator also removes DataTablei from the system database. 3.2

SDN Controller

In order to reduce the workload of SDN Controller, we add Mediator to assist SDN Controller to complete the mission of UE’s authentication and fault recognition of AUSFs. 3.2.1 Starting up an AUSF-VM Once an UE, e.g., UEj, is turned on, its authentication process will start. Basically, SDN Controller assigns the authentication task of UEj to an AUSF that has lighter load, where 1  j  n, and n is the number of UEs currently demanding authentication. If all AUSFs have reached their full load, SDN Controller will wake up an AUSF-VM, requesting this VM to act as an AUSF, e.g., AUSFi, and then authenticate UEj. 3.2.2 Distributing UEs Authenticated by the Failed AUSF After AUSF, e.g., AUSFf fails, the rest of the authentication steps of those UEs authenticated by AUSFf need to be completed by other AUSFs. Therefore, when receiving the distributing message sent by Mediator, SDN Controller follows load balance policy to distribute these UEs to other AUSFs. Now, we only take the number of UEs into account as the load of an AUSF.

4 Simulation and Discussions Our system simulation topology includes three main subsystems, including E-UTRAN, 5GC and Edge Computers. E-UTRAN consists of UEs and gNBs. 5GC involves AMF, AUSFs, SDN Controller of AUSF Pool and virtual machine (VM) pool. Mediator System contains Mediator, Common Storage and DataTables. 4.1

Topology and Settings

The network topology of the Mediator System is established by using Mininet (see Fig. 2). Host 1 (h1), host 4 (h4), and host 5 (h5) represent the three AUSF Servers deployed in the 5GC and organized a complete graph; Hosts 2 (h2), host 3 (h3), host 6 (h6) and host 7 (h7) are UE, AMF, MySQL and Mediator (Edge Computer), respectively, where MySQL is Common Storage. Switches s1–s7 are used to individually connect h1–h7 hosts.

System Implementation of AUSF Fault Tolerance

683

Fig. 2. Network topology established on using Mininet.

4.2

Experiments

In addition to the aforementioned topology and settings, QoS data is measured by tools such as iperf, NS3, and Wireshark. The compared systems include 4G network, our system (MS), remote firewall which is installed in a remote firewall server, denoted by Remote, and Open vSwitch (OVS) which as a switch is connected to SGW. We installs firewall policies to its flow entries so that OVS can act as a firewall to filter IP packets when SGW passes these IP packets to it. 4.2.1 Different Packet Sizes In Experiment 1, different packet sizes ranging from 1 KB to 16 KB. Data rate is fixed to 100 Mbps. Note that number of packets multiplied by the packet size is roughly a constant. Figure 3a shows throughputs of this experiment which range between 10 and 12 MB/sec. MS’s Throughputs is slightly lower than 4G’ because 4G does not provide a fault-tolerant firewall mechanism. Throughputs of the Remote are lower than those of MS because the probability that a packet is dropped on its way to the Remote server is higher than that when this packet is transmitted inside 5GC due to long-distance transmission. OVS’s throughputs is slightly higher than the Remote’s and the difference between OVS’s and MS’s is small also because both transmit packets inside the 5GC. Figure 3b shows the end-to-end delays. Obviously, the 4G’s end-to-end delays are shorter than the MS’s, since packets do not go through Mediator and edge computers. The Remote’s and OVS’s end-to-end delays are longer than MS’s because the Remote needs to encrypt (decrypt) data, before (after) data pass through VPN, and the time that OVS checks firewall policies installed in its flow entries is longer than an edge computer’s packet processing delays. The difference of the end-to-end delays between MS and 4G is smaller than that between Remote and MS, while the difference between

684

W.-S. Chen et al.

Fig. 3. QoS on different packet sizes ranging from 1 KB to 16 KB.

OVS’s and MS’s end-to-end delays is smaller than that between MS’s and Remote’s and also between MS’s and 4G’s. The latter indicates that the processing speed of MS is not far away from that of OVS. But when OVS is busy, the difference may be larger. Figure 3c shows packet loss rates. The larger a packet, the longer its transmission time. This also increases the probability that the packet is discarded. The packet loss rates of MS is much smaller than those of the Remote. Owing to checking firewall policies by looking up its flowtable, OVS’s performance is not better than the MS’s. The MS’s delays are longer than 4G’s. The reasons are mentioned above. Figure 3d shows the jitters. The unit is in lsec. In fact, the jitters of the four systems are all stable. 4.2.2 Different Numbers of Packet Sent Per Second In Experiment 2, the size of a packet is fixed to 1.5 KB. The number of packets sent per second ranges from 80 to 42K, and bandwidth is set to 200 Mbps (=25 MB/sec). The purpose is observing the trend of the four tested systems before and after bandwidth of the transmission link is saturated. Figure 4a illustrates the throughputs. If the packet size is fixed, the more the packets sent per second, the larger the throughputs. When number of packets sent per second exceeds 16K, the occupied bandwidth is 24 MB/sec (= 192Mbps). Throughputs of the only UE begin to decline because high data rates seriously congest network links. 4G and MS have reached about 21 MB (=168 Mbps) and 18 MB (=144 Mbps), respectively. As shown, when the number of packets sent is less than 16K, throughputs

System Implementation of AUSF Fault Tolerance

685

Fig. 4. QoS on different numbers of packet sent per second which range from 80 to 43K

of MS are not better than those of 4G, but the difference is very small. The Remote’s throughputs are lower than those of the other three. Packets of 4G do not pass through firewalls. So 4G’s throughputs are better than MS’s. Figure 4b shows end-to-end delays. Although between 80 and 12K, the delays slightly increase with the number of packets sent, after reaching 16K, the delays and packet loss rates rise sharply (see Fig. 4c). Jitters of the four systems are somewhat similar on low numbers of packet sent per second, and they change highly otherwise because the data rates exceed the bandwidth, resulting in higher delays (see Fig. 4d). 4.2.3 Different Packet Loss Rates The packet loss rates are changed from 0% to 80%. Figures 5a–c show the throughputs, end-to-end delays, and jitters, respectively. When packet loss rates increase, throughputs decrease linearly. The packets which cannot reach the destination lead to the fact that the number of packets received by the receiving end is reduced. When packet loss rates are 0%, the difference between the four tested systems is obvious. The difference on throughputs between 4G and MS is less than 1 MB/sec. The Remote’s and OVS’s throughputs are lower than those of MS. However, when packet loss rates reach 80%, the throughputs of the four are similar because the number of packets arriving at the receiving end is limited. Figure 5b shows that when packet loss rates are higher, a large number of packets will be blocked on the ways to their destinations, thus causing longer delays. 4G’s and MS’s are less than 2 ms, and 4G’s delays are about 4 to 6 ms shorter than Remote’s. OVS’s delays are about 3 ms to 5 ms shorter than Remote’s. Jitters becomes unstable as packet size is larger.

686

W.-S. Chen et al.

Fig. 5. QoS on different packet loss rates ranging from 0% to 80%

5 Conclusion and Future Studies Generally, this study proposes a fault-tolerant mechanism to solve the fault problem of AUSF. The experimental results also show that our approach is feasible. However, there are still some issues that have to be explored in the near future. For example, if AUSF is down in the process of handover, how other AUSFs take over for it to continue the handover process. This may conduct a trust issue or the authorization problem of accessing data in Common Storage. In the future, we will also propose a more complete handover mechanism, and continue improving this system. We would also like to derive its behavior model and reliability model so that users can comprehend its behaviors and reliabilities before using it. These constitute our future studies.

References 1. Schiller, E., Nikaein, N., Kalogeiton, E., Gasparyan, M., Braun, T.: CDS-MEC: NFV/SDNbased application management for MEC in 5G systems. Comput. Netw. 135(22), 96–107 (2018) 2. Shanmugalingam, S., Ksentini, A., Bertin, P.: DPDK Open vSwitch performance validation with mirroring feature. In: Proceeding of the 23rd IEEE International Conference on Telecommunications (IEEE ICT 2016), Thessaloniki, Greece, pp. 1–6 (2016) 3. Hsieh, H.-C., Chen, J.-L., Benslimane, A.: 5G virtualized multi-access edge computing platform for IoT applications. J. Netw. Comput. Appl. 115(1), 94–102 (2018)

System Implementation of AUSF Fault Tolerance

687

4. Arins, A.: Firewall as a service in SDN OpenFlow network. In: IEEE Workshop on Advances in Information, Electronic and Electrical Engineering, pp. 1–5, November 2015 5. Cisco, Load Balance MME in Pool, 19 June 2015. https://www.cisco.com/c/en/us/support/ docs/wireless/mme-mobility-management-entity/119021-config-mme-00.html 6. Dehnel-Wild, M., Cremers, C.: Security vulnerability in 5G-AKA draft. 3GPP TS 33.501 draft v0.7.0, February 2018

News Collection and Analysis on Public Political Opinions Zhi-Qian Hong1, Fang-Yie Leu1(&), and Heru Susanto2,3 1

2

Department of Computer Science, Tunghai University, Taichung City 40704, Taiwan [email protected], [email protected] Research Center for Informatics, The Indonesian Institute of Sciences, Jakarta, Indonesia [email protected] 3 Information Management Department, Tunghai University, Taichung City 40704, Taiwan

Abstract. With the fast development of news media and freedom of speech in Taiwan, some news is not objectively reported. In fact, in order to attract people’s attention and increase the click rates of news, many journalists did not convey the exact meanings of news, even distorting news meanings or adding some subjective criticisms or opinions. As a result, news confusions come out one after the other. Based on the analysis of political opinion news, this study would like to analyze certain political characters, such as candidates during a certain period of time, for example, the election period. Last year (2018), Kaohsiung-city-mayor election was held in December. We develop a news gathering and analytical scheme, named Focused News Collection and analytical System (FNCaS), which predicts which candidate might be the winner. By analyzing the possible outcomes for readers through big data analysis techniques and deep learning approaches after some amount of news were gathered. The purpose is to reduce the time for readers to absorb news essentials, and to conclude the possible results of the analyses immediately, aiming to improve the efficiency that people access to news contents and understand the implications behind it. Our conclusion is that the FNCaS has capability in collecting news immediately and analyzing some amount of news of focused domains efficiently. Keywords: News
Public political opinion intelligence
LSTM




Deep learning




Artificial

1 Introduction In the era of information explosion, the quantity of news significantly increases. Every day, a large number of news is broadcast. Under this circumstance, it is almost impossible for readers to read all news presented in front of them in a limited time and absorb the news contents immediately. On the other hand, readers are frequently confused by some news and then intercept wrong information. © Springer Nature Switzerland AG 2020 L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 688–697, 2020. https://doi.org/10.1007/978-3-030-33506-9_63

News Collection and Analysis on Public Political Opinions

689

One of the authors of this paper has the experience of news collection during his part time job in an institute of Tunghai University, Taiwan. He found that many news are duplicated on Internet and embedded with specific advertisement. For example, a building designed by a well-known architect was written into news for the purpose of better sales. The former wasted readers’ time to deal with them, while the latter contradicts the news neutrality. To prevent this problem, in this study, we propose a news gathering and analytical scheme, named Focused News Collection and analytical System (FNCaS) which collects a large amount of currently-focused news via web crawlers, and then applies bigdata analytical techniques to find valuable knowledge hidden behind the news of concerned and focused topics. Readers who are interested in these topics or focuses can quickly retrieve the deep knowledge of collected news. The FNCaS is presented based on the Kaohsiung Mayor’s election in 2018, and the candidates were GUO-YU HAN and QI-MAI CHEN. In fact, from the analytical results on the political news obtained in the passing days, readers can realize possible current status of a certain political guide. The FNCaS can also judge whether a certain news is in a neutral state. Furthermore, we can use the proposed system to monitor the freedom level of press and respond for news supervision. The rest of this paper is organized as follows. Section 2 briefly describes background knowledge and related research of this study. Section 3 introduces the architecture of the FNCaS. Experiments and their discussion are presented in Sect. 4. Section 5 concludes this study and addresses our future studies.

2 Background Knowledge and Research Human beings have begun to engage in Natural Language Processing (NLP) since the 1950s. Turing proposed the “Turing test” in [1] as a mechanism for knowledge judgement. In the late 1980s, people introduced machine learning algorithms, which promoted NLP to evolve into a newly developed stage. At that time, the operational capability of computer has been steadily increasing according to Moore’s Law, pushing NLP a significant step forward. In recent years, the technology of deep learning has been vigorously developed. It also has achieved good results in NLP. Because the traditional language processing method based on syntactic-semantic is complicated, with the establishment of large corpora defined as a lot of text which are arranged and collected by someone, and the rise of corpus linguistics defined as a scientific research in human language, the machine learning processing method has been one of the main focuses of NLP. On the other hand, researchers, scholars and domain experts have increasingly valued statistic data, particularly when the amount of data is big. The importance of vocabulary has been higher than before leading to the tendency of “lexicalism”, because machine learning can acquire rich linguistic hidden knowledge from big data. “Word” is the smallest independent language component. English has a natural boundary between two adjacent words. However, there is no obvious one between words and words in Chinese. Therefore, text segmentation has been the basis and

690

Z.-Q. Hong et al.

essential of Chinese NLP. Basically, Jieba uses a probabilistic language model to segment terms with the help of its specific dictionary which collected more than 20000 characters. Each character has a number of occurrence (number of times) and its own part of speech. All characters in this dictionary is structured as a Trie tree, which enables efficient word graph scanning as the input text contents. The corresponding sentence is generated based on all possible characters of the word in the dictionary. After that, a Directed Acyclic Graph (DAG) is produced and dynamic programming techniques are used to find the path with the maximum probability from the DAG. A maximum segmentation combination performed based on word frequencies can be then obtained. For characters absent from the dictionary, the Hidden Markov model (HMM) and Viterbi algorithm dynamic programming are employed to find new characters. Word2vec [2] can calculate the distance between word X and word Y by inputting “set of words” defined as a lot of text. It converts “words” into “vectors” so as to transfer the word contents into vectors. The distance between X and Y in vector space is the semantic similarity between the two words in the context. Recurrent Neural Network (RNN) [3], a neural network model, is suitable for processing sequential data, such as sentences/articles, speech and movies. There is a sequential relationship between this type of data. During the training process, the output from the previous point and the input data at current point are together input to current point to train the model. In other words, RNN is a model with memory. The features of Long Short-Term Memory (LSTM) are similar to those of RNN, except that it contains a Cell State to memorize previous messages. Each cell has three gates to adjust the strength of memorized messages. The input gate outputs a number between 0 and 1, in which 1 indicates a complete reservation, and 0 represents complete abandon; the forget gate determines which message should be forgotten; and the output gate controls which message can be output. By the forget gate, LSTM can regulate the convergence of the parameter variation gradient during training. According to the above, it is more memorable in long term than RNN.

3 The Architecture of the FNCaS In this section, we introduce the architecture of the FNCaS. 3.1

System Architecture

Figure 1 shows the architecture of the FNCaS, in which the system first collects news periodically from the Internet with a network crawler. The collected data are stored in a NoSQL database and then processed by NLP techniques. Next, a model integrated with LSTM model is trained. After that, it can analyze the political sentimental data obtained later. The results are also stored in the NoSQL database.

News Collection and Analysis on Public Political Opinions

691

Fig. 1. System architecture diagram.

3.2

News Content Analysis

The format of the news collected by the crawler is shown in Fig. 2.

Fig. 2. Format of news stored in the NoSql database.

Secondly, Jieba is invoked to segment terms from news contents and TF-IDF [4] algorithm is employed to extract keywords from news. Figure 3 shows the results of the term segmentation and the obtained keywords.

692

Z.-Q. Hong et al.

Fig. 3. News contents are segmented with Jieba and keywords are extracted by using TF-IDF.

Furthermore, we calculate the number of occurrences of keywords in all news collected in current month and produce the vocabulary statistics for single news web site. Figure 4 shows the frequencies of all identified terms, from which the popular things in the news in underlying month can be detected.

Fig. 4. Vocabulary statistics.

3.3

Word2vec Training

The kit of Word2vec used in this study is Gensim [5]. The training data includes the Wikipedia database, the PTT gossiping and some of the crawled news. After training, the results are stored in a file which will be input to the first layer of the LSTM, i.e., the embedding layer. 3.4

Deep Learning Training for Public Opinions

Before inputting the data into LSTM for training, we need to find the keywords that politically related to characters, i.e., candidates, in the news. We selected 1000 news for marking. As training data, the news contents positively related to our goal (i.e., the evaluation is positive) are all marked as 1, and the negative one is marked as 0.

News Collection and Analysis on Public Political Opinions

693

Secondly, we divide 1000 news into 800 training data and 200 validation data, and take the marks one-hot encoding, i.e., encode the mark 0 to [1, 0] and the mark 1 to [0, 1]. After that, we input the 800 training data info the previously trained Word2vec model, so that the corresponding vocabulary can be expressed in multiple dimensions and the vocabularies entered to the model is in their digit formats. If the model finds that an input word does not exist in the dictionary of the word2vec model, it skips the word. The purpose is to filter out those rarely-used vocabularies. The structure of the deep learning model established in this study is shown in Fig. 5. The first layer is Embedding layer, which applies the parameters obtained by using the Word2vec model to the neural network model. The LSTM layer is the second layer which is connected to two full-connection layers. In order to achieve a better accuracy, the activation function employed in first full-connection layer is Relu. Since this model is trained for classification, the activation function of the second fullconnection layer (i.e., output layer) is Softmax, aiming to expand the classification effect. We add Dropout between the two full-connection layers to avoid Overfitting during training. In addition, the parameter of LSTM layer is set to 64 and the Dropout rate is set to 0.4. Both are the preferred values after many of our attempts.

Fig. 5. The deep learning model architecture.

Now, Adam [6] which has been widely used in different domains is adopted as the optimizer. This optimizer uses squared value of gradient in the past training as the basis for adjusting current learning rate, and regulates current gradient adjusting speed with the past gradients, so as to perform the learning process smoothly. The loss function is Binary_Crossentropy (1) which is often used in 2D classifiers where yj is actual value of validation in training data and ybj is its predictive value. At last, the trained model is stored in database for later analysis. loss ¼ 

n X j¼1

ybj log yj þ ð1  ybj Þ logð1  ybj Þ

ð1Þ

694

Z.-Q. Hong et al.

4 Experiments and Discussion This section introduces our experiments and discusses their results. 4.1

Model Analysis

During the analysis stage, we firstly found out the news concerning “Guo-Yu Han” (Qi-Mai Chen) from all the collected news in the database. After that, 500 news are randomly chosen for each character/mayor candidate. Before feeding the news into LSTM for training, we mark the news as mentioned above. After many times of hyperparameters adjustment, the final training results are shown in Fig. 6. Both the training and validation accuracies are about 75% (see the right side of Fig. 6). Figure 7 shown the training results for reducing loss. The purpose is to minimize the overfitting and confirm that the loss rates of training and validation decrease synchronously.

Fig. 6. Model accuracy.

training

and

validation

Fig. 7. Model training and validation loss.

After further validation of the trained model and the subsequent analysis of big data, we additionally took 200 news to verify the accuracy of the trained model. The results are shown in Table 1 which is the corresponding Confusing matrix.

Table 1. Confusing matrix of the trained model.

Real

Positive

Negative

Positive

98

32

Negative

27

43

Predicted

News Collection and Analysis on Public Political Opinions

695

The data of Table 2 are derived from the data shown in Table 1. The validation accuracy is 70.5 (=(98 + 43)/200) %, the precision is 75.3 (=98/(98 + 32)) %, the recall is 78.4(=98/(98 + 27)) % and the F1 Score is 76.9(=2/((1/0.753) + (1/0.784)) %.

Table 2. Validation results of the model Evaluated item Percentage Accuracy 70.5% Precision 75.3% Recall 78.4% F1 score 76.9%

4.2

Data Analysis

The positive and negative terms we use are as follows: (1) Positive vocabularies: (supply), (economy), (development), (leading), (positive expectation), (win), (supporter), (news quantity), (fan), (success)。 (2) (2)Negative vocabularies: (defamation), (criticism), (false news), (call into question), (apologizing), (verbally attack), (attack), (cheating), (self-examination), (controversy)。 Table 3. Positive and negative vocabulary statistics per month right before and after the Mayor election in December 2018.

696

Z.-Q. Hong et al.

The statistic results of each month on positive and negative vocabularies are shown in Table 3, in which the second (third) row is the number of positive (negative) vocabularies in all collected news. Note that the first row is table header. We can see that the quantity of “Guo-Yu Han” is much higher than the quantity of “Qi-Mai Chen” as shown in Fig. 8. Therefore, the numbers of his positive and negative vocabularies are much higher than those of “Qi-Mai Chen”. However, if we calculated the number of positive (negative) vocabularies over the number of news, as shown in the fourth row (fifth row) in Table 3, i.e., the average number of positive (negative) vocabularies appear in each news as shown in Fig. 9, we can see that from September to November, the numbers of positive vocabularies of “Qi-Mai Chen” per news are individually higher than those of “Guo-Yu Han”. Next, we further individually sort the two candidates’ news by month, and then input the sorted results into LSTM to predict the possible results. It is clear that the positive level of “Qi-Mai Chen” is greater than that of the “GuoYu Han” in most of the concerned months. Also, the volume of “Guo-Yu Han” began to rise in January where election was held in Dec. 2018 both in Figs. 8 and 9. However, due to the end of election, the quantity of “Qi-Mai Chen” decreased sharply. Further, the monthly news statistics of each news website are put into the model for analysis, and the results are shown in Fig. 10 (Qi-Mai Chen) and Fig. 11 (Guo-Yu Han). Of course, we know that “Guo-Yu Han” won the election.

Fig. 8. News quantities of “Guo-Yu Han” and “Qi-Mai Chen”.

Fig. 9. Ratio of news of “Guo-Yu Han” and “Qi-Mai Chen” positive from Sept. 2018 to Feb. 2019.

Fig. 10. Analytical results of Qi-Mai Chen’s monthly news published by different news websites.

Fig. 11. Analytical results of Guo-Yu Han’s monthly news published by different news websites.

News Collection and Analysis on Public Political Opinions

697

The data presented in Figs. 10 and 11 are the same as those illustrated in Fig. 8, but the abnormal states, e.g., non-neutral news caused by insufficient data, are removed. Figure 10 also shows that before December, the positive rates of the news reported by the Central News Agency for both candidates were higher than those of others, and the positive rates of the news presented by Taiwan People News from September to February gradually increase. Figure 11 presents that the positive rates of all news announced by Apple Daily and Taiwan People News are lower than those of others. Now, we can conclude the general political tendency of these news media.

5 Conclusion and Future Studies The results presented in this study have a certain capability in analyzing a large amount of news, i.e., achieving macro results through the analysis on a large amount of data. In fact, it might not be the real trend of public opinions. But in some parts, from the analytical results, we can observe the political trends of public opinions. In the future, we can continue to decompose news of a specific domain and adjust the direction of the deep learning model. Through the news decomposition, we can distinguish who is the main character in a paragraph of a collected news, and then analyze the contents of news specifically. On the other hand, we also need to delete the stop words defined as the words that do not affect news semantics after removal. After we filter out the unnecessary words, the amount of data entering to the learning model were hugely reduced. We also consider that it is also possible for each news to be marked by the public to acquire a more publicly perceived result for training data. In addition, we can adjust the model or try other models, such as CNN, XGboost and Random Forest, to see whether they can achieve a higher accuracy or not. In terms of algorithms, the training model in this study is relatively simple, and no other NLP applications have been introduced. If we adopt a more efficient model or a better algorithm, in the future, we believe that the trained model will lead to more convincing results.

References 1. Turing, A.M.: Computing machinery and intelligence. Mind 59(236), 433–460 (1950) 2. Rong, X.: Word2vec parameter learning explained. eprint arXiv: 1411.2738, November 2014. https://arxiv.org/pdf/1411.2738.pdf. Accessed 27 June 2019 3. Sherstinsky, A.: Fundamentals of Recurrent Neural Network (RNN) and Long Short-Term Memory (LSTM) Network. eprint arXiv:1808.03314, August 2018. https://arxiv.org/pdf/ 1808.03314.pdf 4. Ramos, J.: Using TF-IDF to determine word relevance in document queries, January 2003. http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.121.1424&rep=rep1&type=pdf. Accessed 4 June 2019 5. Gensim (2009). Accessed 5 June 2019 6. Kingma, D.P., Ba, J.L.: Adam: a method for stochastic optimization. eprint arXiv:1412.6980, December 2014. https://arxiv.org/pdf/1412.6980.pdf. Accessed 10 June 2019

Mobile Physiological Sensor Cloud System for Long-Term Care Ping-Jui Chiang1(&), Heru Susanto2,3, Fang-Yie Leu1, and Hui-Ling Huang1 1

Department of Computer Science, Tunghai University, Taichung City 40704, Taiwan {g08350027,leufy}@thu.edu.tw, [email protected] 2 Research Center for Informatics, The Indonesian Institute of Sciences, Jakarta, Indonesia [email protected] 3 Information Management Department, Tunghai University, Taichung City 40704, Taiwan

Abstract. In this study, we propose a Mobile Physiological Sensor Cloud System for Long-term Care (MPCLC), the main functions of which are collecting carereceiver’s physiological data by using sensors and analyzing and reporting the carereceiver’s healthy condition. With the features of small in size, high convenience for use, and good immediacy of request response, the MPCLC can partially solve the problems of long-term care which are insufficiency of required labors and hard for tracking the results of traditional medical treatment. After the data measured by using sensors is collected, the data is sent to the cloud for storage. The health reports will be generated by the cloud, and delivered to caregivers and medical staffs for reference under their requests. This can save the time that people go and come between hospitals and the places where carereceivers stay for measuring patients’ psysilogical data. One of the other key functions is to prevent the carereceivers from disease in advance. Keywords: Long-term care
Immediate
Physiological data
Sensor
Cloud computing
Health reports

1 Introduction In the era of declining birthrate and aging, traditional long-term care models consume a huge volume of resources. The manpower that has been invested and can be invested is insufficient. In the near future, the demand will tremendously increase. However, many elderly people are not healthy. They just have not tracked their health immediately, but continuously endure their patience and discomfort in their everyday lives. Until the discomfort becomes a serious illness, it sometimes is too late for them to recover from the disease. In fact, we believe that if we can continuously and immediately detect an elderly’s uncomfortableness from his/her physiological data, and carefully take care of the elderly, we can then effectively compensate for the insufficiency of traditional longterm-care resources. © Springer Nature Switzerland AG 2020 L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 698–707, 2020. https://doi.org/10.1007/978-3-030-33506-9_64

Mobile Physiological Sensor Cloud System for Long-Term Care

699

Therefore, in this study, we propose a Mobile Physiological Sensor Cloud System for Long-term Care (MPCLC), the main function of which are obtaining the physiological data sensors. With the features of small in size, high convenience for use, and good immediacy of request/response, it can solve the problems of long-term care which are insufficiently of required labors and hard for tracking the results of traditional medical treatment. The rest of this paper is organized as followers Sect. 2 overviews the MPCLC. Section 3 introduces the MPCLC. Its system implementation is shown in Sect. 4. Section 5 concludes this paper and addresses our future students.

2 Background and Related Work 2.1

Environment for Long-Term Care

The indicators commonly used to identify dysfunctions and to assess long-term care needs and their services include: (1) Activity of Daily Living (ADLs), containing daily activities, such as carereceivers’ capabilities of eating, wearing clothes, etc.; (2) Informal Activity of Daily Living (IADLs), including carereceivers’ capabilities on cooking, washing, and taking medicine following physician’s instructions, etc.; (3) Degree of cognitive function, referring to carereceivers’ memory, orientation, abstraction, judgment, calculation, and language ability; If at least one of ADL and IADL is poor, long-term care is then required. However, an elderly often wishes to be cared at home. In viewing this, this study uses physiological sensors to monitor an elderly’s physiological data and responds properly when events occur. The following lists three key points of a smart-home care environment. (1) A barrier-free environment for elderly, called accessibility environment, which brings convenience and safety to life. (2) An IoT environment that employs physiological data, called IoT environment, which guarantees continuity physiological care. (3) The environment integrating the data collected for various services, called “data environment”, which is the basis of stable delivery of manpower and services. For the construction of smart-home care, the three key points are necessary. 2.2

Physiological Data Sensors

In this study, we employ blood pressure and blood sugar sensors, blood oxygen sensors and heart rhythm and electrocardiogram sensors to collect carereceiver’s physiological data which can reflect the health of the carereceiver. The blood pressure and blood sugar sensor as shown in Fig. 1 utilized consists of a host, an arm band and a blood collection needle (with blood test paper). The blood pressure measurement method is simple, like the operation of the common blood-pressure equipment on the market. The

700

P.-J. Chiang et al.

usage of the blood glucose measurement method is first extracting a small wound that can be bled from the epidermis of the carereceiver by using the blood collection needle, and then inserting the test paper into the host after the blood is absorbed. Then the blood sugar of the carereceiver will be shown.

Bluetooth: V2.1+EDR Power: AA battery or Electrical connection Company: Fora Care Inc. Fig. 1. Blood pressure and blood sugar sensors

Bluetooth: V2.1+EDR Power: AAA battery Company: NoninMedical, Inc. Fig. 2. The blood oxygen sensors

Figure 2 shows a blood oxygen sensor which consists of a host and finger sleeve. We first connect the finger sleeve to the host. The bulb inside the finger sleeve will light up, and then the blood oxygen concentration will be known based on the transmittance of the finger of the carereceiver. The feature of this sensor is that the power consumption can be adjusted according to the distance from the receiving end (smartphone) and the sensor to achieve the effect of power saving. Figure 3 shows the heart rhythm and electrocardiogram sensors, which consists of a host, sensing line and sensing patch. It is similar to the stethoscope used by doctors, and the patch needs to be pasted near carereceiver’s heart to measure the heart rate and electrocardiogram. The body temperature can also be measured simultaneously by this sensor.

Bluetooth: V2.1+EDR Power: Mini USB Company: Leadtek Research Inc. Fig. 3. The heart rhythm and electrocardiogram sensors

Mobile Physiological Sensor Cloud System for Long-Term Care

701

3 The MPCLC The MPCLC composed by three systems [3], A Wireless Body Sensor Network (WBSN), Data Collection and Classification System (DCaCS) and Healthcare Monitoring System (HMS). 3.1

The WBSN

The WBSN can be divided into two parts: Physiological data sensor and Smart phone of the carereceiver. The sensors transmit data to smartphone through Bluetooth protocol. 3.1.1 Physiological Data Sensor Basically, a sensor needs to connect to its mobile phone by using each other’s Media Access Control (MAC) and password to approve their mutual credentials. After that, the two will automatically connect to each other without the need of resetting again. When receiving data, a sensor first confirms whether physiological data has been successfully received or not. If not, the measurement is performed again until it successfully catches the data. Otherwise, the measurement results are sampled for digitization. After that, the packet conveying the physiological data is transmitted to the mobile phone. 3.1.2 Smart Phone After receiving the Bluetooth packet sent by the sensor, the smart phone first determines which sensor transmits the packet, and then retrieves the physiological data from the data packet. Finally, the concerned data is displayed on mobile phone and saved in the SQLite database installed in the mobile phone. 3.2

The DCaCS

The DCaCs consists of two sub systems: Data pre-processor and Data classification server. 3.2.1 Data Pre-processor Data pre-processor is responsible for collecting the data transmitted by the mobile phone, without filtering it. The purpose is to ensure all data to be received completely and correctly. If the data is processed immediately after DCaCS receives it, some packets may be unsuccessfully received due to the processing capability of DCaCS, particularly when a large number of sensor packets are transmitted to DCaCS at the same time, and the buffer at the receiving end overflows. This means some physiological data has been dropped. The receiving data is then transferred to the Pre-process Database for storage. 3.2.2 Data Classification Server Data Classification Server (DCS) retrieves data from Pre-process Database for data classification, induction and analysis. Because the Data preprocessor does not filter the

702

P.-J. Chiang et al.

data, the receiving data may be incomplete. For example, if the measuring date of physiological data is unknown, the date will be presumed and added with data transmission time, and then processed by the DCS. If some physiological data is lost, the server will fill it with null; if the data is hugely deviated from its normal value, the server will skip or delete the data. The processed data will be sent to the Physiological Database for storage. 3.3

The HMS

The HMS consists of two sub systems: Healthcare Cloud and Healthcare Monitoring Website. 3.3.1 Healthcare Cloud Healthcare Cloud is the Software as a Service application for cloud computing. It retrieves data from the Physiological Database and generates health reports and future health recommendations. We list some features that a health cloud has.

Fig. 4. The flow chart of data monitoring

(1) Data monitoring: Figure 4 shows the flow chart of our data monitoring. We observe physiological data of a carereceiver in the past 3–4 months, and compare the average value of the data with that provided by the Ministry of Health and Welfare, Taiwan. If some of the carereceivers’ physiological data are abnormal, the MPCLC will notify the corresponding caregivers and medical staffs. (2) Effectiveness of medical treatment: Figure 5 illustrates the flow chart of checking the effectiveness of specific medical treatment. If the carereceiver is taking a medicine treatment at his/her daily life, the MPCLC observes whether the medicine is effective, or the effectiveness is too strong or too weak. Of course, the side effects of the treatment are also detected and presented when necessary.

Mobile Physiological Sensor Cloud System for Long-Term Care

703

Fig. 5. The flow chart of effectiveness of medical treatment

3.3.2 Healthcare Monitoring Website The functions of our Healthcare Monitoring Website include short-term data checking, long-term tracking, health advisory, medical record enquiry, medical treatment records, and instant reminding. Short-term data checking and long-term tracking are all performed on the data retrieved from the Healthcare Cloud. The former uses the data within the past 30 days, while the latter is in the past 120 days. These physiological data can be those belonging to a single or multiple medical personnel for observing and tracking carereceivers’ health conditions. With these data, we can also compare the physiological data of the concerned carereceivers with those of others to uncover the possible infective diseases of the carereceivers. Health advisory is a report derived from the physiological data of a carereceiver stored in the healthcare cloud. Here, it refers to a situation when there is no urgency. It is convenient for medical personnel to enquire the health statues of the carereceiver, and detect his/her abnormalities much earlier, aiming to prevent the disease from worsening. Medical treatment records list the complete medical records of the carereceivers, including: gender, blood type, height, weight, physiological data, recent measurement time, date of recent medical visit, which medicine is being taken, time of return, and remarks. It is convenient for medical staff to enquire and track the care of the carereceivers. Instant reminding is the situation when the medical staff needs to contact carereceiver for further care. For example, the physiological data is serious anomaly, and the time of return is miss… etc.

4 System Implementation The implementation of the system is described below.

704

4.1

P.-J. Chiang et al.

Mobile Phone Function Planning and Measurement on iOS System

With excellent performance, long service life and simple user interface of iPhones, inn this study, we develop the mobile app in Swift language on the iOS system. Figure 6 shows the mobile phone interface. There are four buttons, including real-time measurement, short-term data, long-term data and health report.

Fig. 6. The mobile phone interfaces

(1) Real-time measurement Figure 7 shows the interface and results of real-time measurement. When the employed sensors start measuring physiological data for carereceivers, they transmit the data to the carereceiver’s smart phone through the Bluetooth. The mobile phone interprets the Bluetooth packet, grabs the physiological data represented by the corresponding bits, and displayed them on the mobile phone. The data is instantly displayed, and the data shown to caregivers are simultaneously uploaded to the database. In the iOS Bluetooth system, the Central side, which is the iOS App represents the central system of the measurement; the Peripheral sides send data gathered by the sensors to the Bluetooth device. The Bluetooth peripheral devices broadcast the peripheral device names, their main functions and other information with Advertising Packets. They also contain additional data about the types of data that the peripheral can provide. The central device’s responsibility is to scan these ad packages, find out which peripherals it cares about, and connect a device to receive data. The ad package is very small and cannot contain a large amount of data. The Central side needs to receive Bluetooth packets that carry the data we need from the peripheral device. The context of a Bluetooth packet contains services and characteristics: Services: which is a collection of data plus its associated behaviors used to describe a feature or the characteristics of a peripheral device. A peripheral device can provide more than one service. For example, the packets sent by a blood glucose sensor carry both the data of blood glucose and beating of heart. Characteristics: which carries specific information about the services provided by a peripheral device. e.g., a heart rate packet contains heart rates and other features, such as body sensor position, indicating where the sensor locates. Each service and characteristics are represented by a 16-bit or 128-bit UUID.

Mobile Physiological Sensor Cloud System for Long-Term Care

705

Fig. 7. The interface and results of real-time measurement.

(2) Short-term data Figure 8 shows the interface and results of short-term data. The main functions of this interface are to retrieve measured data in the last 30 days from the physiological databases, and display the contents on the mobile phone. If there is no measured data on a specific day, e.g., X, within the past 30 days, the data of the next day will be duplicated as the data of X.

(a)

(b)

(c)

Fig. 8. The interface and results of short-term data (a: Blood pressure and heart rate measurement; b: Blood oxygen measurement; c: Blood sugar measurement)

(3) Long-term data The main function of Long-term data is to capture measured data from the physiological database in the last 120 days and display it on the mobile phone with a table. Each day’s data is presented in a single column in the table. The top column of the table shows the average of the physiological data for the 120 data. (4) Health report The main function of Health report is to receive health reports sent by Healthy cloud. The purpose is to help the caregiver to better comprehend their carereceivers’ health statuses through the help of cloud computing, and adjust carereceivers’ living habits according to the report, aiming to find their abnormalities early and prevents the occurrence of diseases.

706

P.-J. Chiang et al.

4.2

Health Observation

4.2.1 Healthcare Monitoring Website Figure 9 shows the results of data monitoring. Health Cloud first grabs the health data from the website of Ministry of Health and Welfare, Taiwan, for example, the average value of a physiological data of Taiwanese. Then, the Health Cloud retrieves the data of a carereceiver for 120 days from the physiological database, stores it in an array form, calculates the average value of each physiological data, and observes its trend. If the data is abnormal, the carereceiver/caregiver are notified that the monitoring station cares about carereceiver’s health. At last, the website returns the calculation results to the Healthy Cloud for future reference. Data monitoring is also the basis for the next function.

Fig. 9. The results of data monitoring

Fig. 10. The results of effectiveness of medical treatment

4.2.2 Effectiveness of Medical Treatment Figure 10 shows the results of effectiveness of medical treatment. If a carereceiver has used drug to cope with the treatment, the system will gather the physiological data before the drug was taken, check the ideal range of the change, and the data after the drug is taken. It then comperes the data to judge the effectiveness of the drug. If the drug has a significant effect, it suggests Doctors to consider reducing the dose of the drug, replacing the drug with less potent drug, or even temporarily or permanently stop the drug. If the effectiveness of the drug is not as expected, it gives proper advice according to the real situation of the carereceiver. For example, if the decrease of drug does not make the carereceiver’s physiological data fall into the value as expected, even almost the same as before, we may increase the dose of the drug or to replace the drug with a more effective one. If the drug has an adverse effect on other physiological data, it suggests using another drug to relieve its side effects. If side effects have harmed the health of the carereceiver, discontinuing the drug and seeking other proper treatments will be suggested.

Mobile Physiological Sensor Cloud System for Long-Term Care

707

5 Conclusions and Future Studies We have completed the Bluetooth connection and packet interpretation for each physiological sensor and the mentioned mobile phone, and developed the interfaces and a small database on the mobile phone. In other words, The MPCLC is able to collect physiological data for individuals. Some results have been achieved. Next, our research will focus on the following items: (1) Cloud development: We have begun to write the code of the cloud function according to the developed flow chart, and improve the code to make the carereceiver’s health predictions more accurate and make the entire system work properly so that it can be utilized online; (2) Data Acquisition: We have entered the stage of implementation and testing. We will look for volunteers to test this system, attempting to make the MPCLC more complete; (3) Mobile phone interface landscaping: At present, the interface of our mobile phone only provides the basic functions and displays various data. We will further improve the functions of the interface so as to enhance its usage quality; (4) Enhancement of security: We will study the security of the transmitted data from physiological sensors to mobile phone and from the mobile phone to Healthcare Monitoring website to prevent the personal data of the carereceivers from being stolen during transmission and storage. The purpose is effectively protecting the data and privacy of the carereceivers.

References 1. Zhu, K.G.: iOS 11 program Design-Swift 4- Fst Developing Techniques, 200+ . Gototop Information Inc., Taipei City (2017). (in Chinese) 2. Zhong She, Y.-Q., Zhong, W.-J.: Cloud Computing. Tung Hua Book Co. Ltd, Taipei City (2013). (in Chinese) 3. Ho, C.L.: A smart phone-based wearable sensors for monitoring real-time physiological data. Comput. Electr. Eng. 65, 379–384 (2018) 4. Cheng, Y.H.: Implementation of Bluetooth LAN access profile with NAT & embedded web server functions. Department of Electronic Engineering, Chung Yuan Christian University (2001) 5. Chen, Y.L.: Cloud computing - a case study of clinic information system. Master thesis, Shu-Te University (2011)

The 8th International Workshop on Robot and Vehicle Interaction, Control, Communication and Cooperation (RVI3C-2019)

A Message Relaying Method with Enhanced Dynamic Timer Considering Decrease Rate of Neighboring Nodes for Vehicular-DTN Shogo Nakasaki1 , Makoto Ikeda2(B) , and Leonard Barolli2 1

Graduate School of Engineering, Fukuoka Institute of Technology, 3-30-1 Wajiro-higashi, Higashi-ku, Fukuoka 811-0295, Japan [email protected] 2 Department of Information and Communication Engineering, Fukuoka Institute of Technology, 3-30-1 Wajiro-higashi, Higashi-ku, Fukuoka 811-0295, Japan [email protected], [email protected]

Abstract. In recent years, the store-and-carry scheme has focused on recovery method which can be effectively applied to sparse and dense networks. In this paper, we propose a message relaying method with enhanced dynamic timer considering decrease rate of neighboring vehicles for Vehicular Delay/Disruption/Disconnection Tolerant Networking (DTN). From the simulation results, we found that the proposed method can reduce storage usage while maintaining high delivery rate. Keywords: Message relaying method Dynamic timer

1

· Vehicular DTN ·

Introduction

Recently, many researchers have been developing autonomous vehicles using 3D laser scanner and connected car technology. Lyft Inc. shares a large-scale dataset [11] using the raw sensor camera and 3D laser scanner by autonomous vehicles. Vehicles have been used for travel, logistics, car race and shopping, but now vehicles are going to be connected to the Internet and becoming terminals [5,9,10,14,15,28]. Vehicles will connect to the wireless network and can act not only as intermediate terminals but also as start-point and end-point. They will be used for various intelligent applications [7,8,16,22,24,27]. Delay/Disruption/Disconnection Tolerant Networking (DTN) is a message relaying method in inter-vehicle communication [12,20,21,23]. In Vehicular DTN, communication overhead and storage usage become a critical problem due to the DTN terminals duplicate bundles to others. c Springer Nature Switzerland AG 2020  L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 711–720, 2020. https://doi.org/10.1007/978-3-030-33506-9_65

712

S. Nakasaki et al.

In our previous work [13], we have proposed a Message Relaying Method with a Dynamic Timer considering Non-signal Duration (MRM-DTND) from neighboring vehicles in Vehicular DTN. We have shown that the dynamic timer can improve storage usage, but the performance for other parameters was decreased. In this paper, we propose a Message Relaying Method with an Enhanced Dynamic Timer considering the Decrease Rate (MRM-EDTDR) of neighboring vehicles for Vehicular DTN. The structure of the paper is as follows. In Sect. 2, we give the related work. In Sect. 3 is described the message relaying method considering the dynamic timer. In Sect. 4, we provide the description of the simulation system and the evaluation results. Finally, conclusions and future work are given in Sect. 5.

2

Related Work

DTN can provide a reliable internet-working for space tasks [2,3,6,18,26]. The space networks have possibly long delay, frequent link disconnection and frequent disruption. In DTN, the messages are stored and forwarded by nodes. When the nodes receive messages, they store the messages in their storage. After that, the nodes duplicate the messages to other nodes when it is transmitted. This technique is called message switching. The architecture is specified in RFC 4838 [4]. Epidemic routing is well-known routing protocol for DTN [17,25]. Epidemic routing uses two control messages to duplicate messages. Nodes periodically broadcast the Summary Vector (SV) message in the network. The SV contains a list of stored messages of each node. When the nodes receive the SV, they compare received SV to their SV. The nodes send the REQUEST message if received SV contains unknown messages. In Epidemic routing, consumption of network resources and storage usage become a critical problem. Because the nodes duplicate messages to neighbors in their communication range. Moreover, received messages remain in the storage, because the messages are continuously duplicated even if the destination receives the messages. Therefore, recovery schemes such as anti-packet or timer are needed to limit the duplicate messages. In the case of anti-packet, the destination broadcasts the anti-packet, which contains the list of messages that are received by the destination. Nodes delete the messages according to the anti-packet. Then, the nodes duplicate the anti-packet to other nodes. However, network resources are consumed by anti-packet. In the case of the timer, messages have a lifetime. The messages are punctually deleted when the lifetime of the messages is expired. However, the setting of a suitable lifetime is difficult. In this paper, we propose a message relaying method considering a dynamic timer to limit the duplicate messages in Vehicular DTN.

3

Message Relaying Methods Considering Dynamic Timer

In this section, we explain in detail two proposed message relaying methods considering dynamic timer.

A Message Relaying Method with Enhanced Dynamic Timer

3.1

713

Overview of Dynamic Timer

In this paper, we propose an enhanced dynamic timer to control the bundle drop deadline. In conventional Epidemic, a fixed value of lifetime is settled at the time of message generation. If the lifetime is expired, the vehicle deletes the bundle message in their storage. However, the conventional method does not consider network conditions around each vehicle such as vehicle density, distance to endpoint, and so on. In our method, each vehicle sets a dynamic timer for messages and the management of data is independent. We found that the MRM-DTND method can improve storage usage, but the performance for other parameters was decreased [13].

Fig. 1. Flowchart of MRM-EDTDR (The red frame shows new functions. The MRMDTND does not consider functions in the red frame).

3.2

Timer Setting in DTND

For both methods, each node periodically checks the number of received SV and measure a non-signal time (NT) from neighboring vehicles. If the current NT is

714

S. Nakasaki et al.

greater than maximum NT (NTmax ), NTmax will be updated. The formula of Dynamic Timer (DTimer) is: DTimer = NTmax + Interval,

(1)

where Interval indicates the check interval, which uses for checking the number of received SVs. When the number of received SVs is 0, MRM-DTND method resets the timer. In general, if the DTimer is set in the message, the lifetime is not reset. But, in MRM-DTND method, the DTimer reset is allowed even if the lifetime is reset. We consider the Interval to keep the message in the storage for checking the next SV. Even if the vehicle suddenly leaves the neighboring vehicles and there is no signal time, MRM-DTND method keeps the message. 3.3

Timer Setting in EDTDR

In the MRM-DTND method, there is a problem that the delivery ratio is decreased based on vehicle densities. Therefore, the MRM-EDTDR method improves network performance by changing the timer reset condition. We present the flowchart of MRM-EDTDR method in Fig. 1. The red frame shows the functions newly added to the MRM-DTND method. In the MRM-DTND method, the timer was reset only when the current number of neighboring nodes Nnow is 0. The MRM-EDTDR method considers the number of neighboring nodes measured last time (Nprev ). Also, the method calculates the decrease rate of neighboring nodes and compares the value with the Reset Threshold (RT). The timer is reset by the condition of Eq. (2): Nnow ≤ RT. Nprev

4

(2)

Road Network Model and Simulation Results

In this paper, we evaluate the MRM-EDTDR method for different vehicle densities. We implemented both MRM-DTND and MRM-EDTDR methods on the Scenargie [19] network simulator. 4.1

Scenario Setting

We consider a grid scenario with a maximum density of 250 vehicles/km2 . We show a road network model in Fig. 2. Table 1 shows the simulation parameters used on the Scenargie network simulator. Message start-point and end-point are static, and other vehicles move on the road based on the random way-point mobility model. The start-point sends bundle messages to end-point considering ITU-R P.1411 propagation model [1]. When the vehicle receives bundle messages, they store the bundle messages in their storage. After that, the vehicles duplicate the bundles to other vehicles. Therefore, we consider the interference of obstacles on 5.9 GHz band. We use three methods as follows:

A Message Relaying Method with Enhanced Dynamic Timer

715

Fig. 2. Road network model. Table 1. Simulation parameters. Parameter

Value

Simulation time (Tmax )

600 [s]

Area dimensions

1, 000 [m] × 1, 000 [m]

Density of vehicles

60, 80, 120, 250 [vehicles/km2 ]

Minimum speed (Vmin )

8.333 [m/s]

Maximum speed (Vmax )

16.666 [m/s]

Message start and end time 10 - 400 [sec] Message generation interval 10 [sec] Message size

1, 000 [bytes]

PHY model

IEEE 802.11p

Propagation model

ITU-R P.1411

Antenna model

Omni-directional

Rest threshold

0.5

1. Timer(Max): Epidemic activated timer method with the duration set to maximum delay, 2. MRM-EDTDR: Epidemic with proposed enhanced dynamic timer considering the decrease rate of neighboring nodes, 3. MRM-DTND: Epidemic with proposed dynamic timer considering the nonsignal duration. The non-signal duration means that there is no vehicle around. Before the simulation, we evaluated the delay performance using conventional Epidemic in this road model. We considered these results for setting the maximum delay. In the case of conventional Epidemic with Timer(Max) all bundles will be reach to the end-point. For simulations, we consider four evaluation parameters: delay, delivery ratio, storage usage and overhead. The delay indicates the transmission delay of the

716

S. Nakasaki et al.

Fig. 3. Delivery ratio

Fig. 4. Delay

message to reach the end-point. The delivery ratio indicates the ratio of bundles successfully received by end-point to the total number of bundles sent by the start-point. The storage usage indicates the average of the storage usage of each vehicle. The overhead indicates the number of times for sending duplicate messages. Each timer function will be activated after the simulation time is 60 seconds. In this simulation, RT sets 0.5. 4.2

Simulation Results

We show the simulation results of the delivery ratio for different methods in Fig. 3. For Timer(Max) and MRM-EDTDR method, the results of the delivery ratio for different vehicles reached 100%. The MRM-EDTDR method delivered all messages to end-point and the performance is better than the MRM-DTND method.

A Message Relaying Method with Enhanced Dynamic Timer

717

Table 2. Results of overhead for different methods. Density of vehicles 60

80

120

250

Timer(Max)

10,545 15,603 32,394 64,786

MRM-EDTDR

10,545 15,631 32,383 64,832

MRM-DTND

10,554 15,495 32,282 64,837

We present the simulation results of the delay for different methods in Fig. 4. The delay results were calculated from the messages that reached the end-point. In the case of dense networks, there are many intermediate vehicles. Thus, the results of the delay are decreased by increasing the density of vehicles. The MRM-EDTDR method decreased the delay for different vehicles compared with the MRM-DTND method. We observed that the delay performance was close to the Timer(Max).

Fig. 5. Storage usage for different methods.

The results of the overhead are shown in Table 2. We observed that the difference between the overhead performance is very small.

718

S. Nakasaki et al.

We present the simulation results of average storage usage in Fig. 5. The MRM-EDTDR method used much storage compared with the MRM-DTND method because the number of times to reset the timer is increased. But, we found that the difference of performance is small when vehicle densities is high. From 70 to 430 s in 120 vehicles/km2 , the results of storage usage of the MRMEDTDR method are less than Timer(Max). Thus, the results of the MRMEDTDR method are good because all messages reached the end-point. From this evaluation, we conclude that the MRM-EDTDR method has a high delivery ratio and can reduce storage usage.

5

Conclusions

In this paper, we proposed MRM-EDTDR method for Vehicular-DTN. We evaluated the proposed MRM-EDTDR method considering delivery ratio, delay, overhead and storage usage as evaluation metrics. From the simulation results, we found that the delivery ratio and delay of the MRM-EDTDR method are better than the MRM-DTND method. Also, the performance is close to Timer(Max) for all vehicle densities. In future work, we would like to solve the problem of increasing storage usage in sparse networks. We also would like to investigate the impact of RT parameter.

References 1. Rec. ITU-R P.1411-7: Propagation data and prediction methods for the planning of short-range outdoor radiocommunication systems and radio local area networks in the frequency range 300 MHz to 100 GHz. ITU (2013) 2. Araniti, G., Bezirgiannidis, N., Birrane, E., Bisio, I., Burleigh, S., Caini, C., Feldmann, M., Marchese, M., Segui, J., Suzuki, K.: Contact graph routing in DTN space networks: overview, enhancements and performance. IEEE Commun. Mag. 53(3), 38–46 (2015) 3. Caini, C., Cruickshank, H., Farrell, S., Marchese, M.: Delay- and disruptiontolerant networking (DTN): an alternative solution for future satellite networking applications. Proc. IEEE 99(11), 1980–1997 (2011) 4. Cerf, V., Burleigh, S., Hooke, A., Torgerson, L., Durst, R., Scott, K., Fall, K., Weiss, H.: Delay-tolerant networking architecture. IETF RFC 4838 (Informational), April 2007 5. Dias, J.A.F.F., Rodrigues, J.J.P.C., Xia, F., Mavromoustakis, C.X.: A cooperative watchdog system to detect misbehavior nodes in vehicular delay-tolerant networks. IEEE Trans. Ind. Electron. 62(12), 7929–7937 (2015) 6. Fall, K.: A delay-tolerant network architecture for challenged Internets. In: Proceedings of the International Conference on Applications, Technologies, Architectures, and Protocols for Computer Communications, SIGCOMM 2003, pp. 27–34 (2003) 7. Grassi, G., Pesavento, D., Pau, G., Vuyyuru, R., Wakikawa, R., Zhang, L.: VANET via named data networking. In: Proceedings of the IEEE Conference on Computer Communications Workshops (INFOCOM WKSHPS 2014), pp. 410–415, April 2014

A Message Relaying Method with Enhanced Dynamic Timer

719

8. Hou, X., Li, Y., Chen, M., Wu, D., Jin, D., Chen, S.: Vehicular fog computing: a viewpoint of vehicles as the infrastructures. IEEE Trans. Veh. Technol. 65(6), 3860–3873 (2016) 9. Kenney, J.B.: Dedicated short-range communications (DSRC) standards in the United States. Proc. IEEE 99, 1162–1182 (2011) 10. Lin, D., Kang, J., Squicciarini, A., Wu, Y., Gurung, S., Tonguz, O.: MoZo: a moving zone based routing protocol using pure V2V communication in VANETs. IEEE Trans. Mob. Comput. 16(5), 1357–1370 (2017) 11. Lyft: Dataset of lyft level 5, July 2019. https://level5.lyft.com/dataset/ 12. Mahmoud, A., Noureldin, A., Hassanein, H.S.: VANETs positioning in urban environments: a novel cooperative approach. In: Proceedings of the IEEE 82nd Vehicular Technology Conference (VTC-2015 Fall), pp. 1–7, September 2015 13. Nakasaki, S., Ikeda, M., Barolli, L.: A message relaying method with a dynamic timer considering non-signal duration from neighboring nodes for vehicular DTNs. Accepted, to appear in Proceedings of the 11th International Conference on Intelligent Networking and Collaborative Systems (INCoS-2019), September 2019 14. Ning, Z., Hu, X., Chen, Z., Zhou, M., Hu, B., Cheng, J., Obaidat, M.S.: A cooperative quality-aware service access system for social internet of vehicles. IEEE Internet Things J. 5(4), 2506–2517 (2018) 15. Ohn-Bar, E., Trivedi, M.M.: Learning to detect vehicles by clustering appearance patterns. IEEE Trans. Intell. Transp. Syst. 16(5), 2511–2521 (2015) 16. Radenkovic, M., Walker, A.: CognitiveCharge: disconnection tolerant adaptive collaborative and predictive vehicular charging. In: Proceedings of the 4th ACM MobiHoc Workshop on Experiences with the Design and Implementation of Smart Objects (SMARTOBJECTS-2018), June 2018 17. Ramanathan, R., Hansen, R., Basu, P., Hain, R.R., Krishnan, R.: Prioritized epidemic routing for opportunistic networks. In: Proceedings of the 1st International MobiSys Workshop on Mobile Opportunistic Networking (MobiOpp 2007), pp. 62–66 (2007) 18. R¨ usch, S., Sch¨ urmann, D., Kapitza, R., Wolf, L.: Forward secure delay-tolerant networking. In: Proceedings of the 12th Workshop on Challenged Networks (CHANTS-2017), pp. 7–12, October 2017 19. Scenargie: Space-time engineering, LLC. http://www.spacetime-eng.com/ 20. Stute, M., Maass, M., Schons, T., Hollick, M.: Reverse engineering human mobility in large-scale natural disasters. In: Proceedings of the 20th ACM International Conference on Modelling, Analysis and Simulation of Wireless and Mobile Systems (MSWiM-2017), pp. 219–226, November 2017 21. Theodoropoulos, T., Damousis, Y., Amditis, A.: A load balancing control algorithm for EV static and dynamic wireless charging. In: Proceedings of the IEEE 81st Vehicular Technology Conference (VTC-2015 Spring), pp. 1–5, May 2015 22. Tornell, S.M., Calafate, C.T., Cano, J.C., Manzoni, P.: DTN protocols for vehicular networks: an application oriented overview. IEEE Commun. Surv. Tutorials 17(2), 868–887 (2015) 23. Uchida, N., Ishida, T., Shibata, Y.: Delay tolerant networks-based vehicle-tovehicle wireless networks for road surveillance systems in local areas. Int. J. SpaceBased Situated Comput. 6(1), 12–20 (2016) 24. Urquiza-Aguiar, L., Igartua, M.A., Tripp-Barba, C., Calder´ on-Hinojosa, X.: 2hGAR: 2-hops geographical anycast routing protocol for vehicle-to-infrastructure communications. In: Proceedings of the 15th ACM International Symposium on Mobility Management and Wireless Access (MobiWac-2017), pp. 145–152, November 2017

720

S. Nakasaki et al.

25. Vahdat, A., Becker, D.: Epidemic routing for partially-connected ad hoc networks. Duke University, Technical report (2000) 26. Wyatt, J., Burleigh, S., Jones, R., Torgerson, L., Wissler, S.: Disruption tolerant networking flight validation experiment on NASA’s EPOXI mission. In: Proceedings of the 1st International Conference on Advances in Satellite and Space Communications (SPACOMM-2009), pp. 187–196, July 2009 27. Zhang, W., Jiang, S., Zhu, X., Wang, Y.: Cooperative downloading with privacy preservation and access control for value-added services in VANETs. Int. J. Grid Utility Comput. 7(1), 50–60 (2016) 28. Zhou, H., Wang, H., Li, X., Leung, V.C.M.: A survey on mobile data offloading technologies. IEEE Access 6, 5101–5111 (2018)

Prediction of RSSI by Scikit-Learn for Improving Position Detecting System of Omnidirectional Wheelchair Tennis Keita Matsuo(B) and Leonard Barolli Department of Information and Communication Engineering, Fukuoka Institute of Technology (FIT), 3-30-1 Wajiro-Higashi, Higashi-Ku, Fukuoka 811-0295, Japan {kt-matsuo,barolli}@fit.ac.jp

Abstract. The wheelchair with good performance for the aged and disabled people is attracting attention from the society. Also, the wheelchair can provide the user with many benefits, such as maintaining mobility, continuing or broadening community social activities, conserving energy and enhancing quality of life. The wheelchair body must be compact enough and should be able to make different movements in order to have many applications. In our previous work, we presented the design and implementation of an omnidirectional wheelchair. In this paper, we propose a position detecting system for improving the performance of omnidirectional wheelchair tennis. This is achieved by predicting RSSI value using Scikit-learn. The proposed system can find correctly the wheelchair position for avoiding the collision.

1

Introduction

Recently, the convenient facilities and equipments have been developed in order to satisfy the requirements of elderly people and disabled people. Among them, wheelchair is a common one which is used widely. A wheelchair can provide the user with many benefits, such as maintaining mobility, continuing or broadening community social activities, conserving energy and enhancing quality of life. Because of aged tendency of population and rapid growth in the number of the disabled people caused by diseases or injuries, the wheelchair with good performance for the aged and disabled people is attracting attention from the society. There are many research works on wheelchairs including wheelchair for recovery, climbing stairs, playing sport and multifunction [11]. Therefore, it is necessary to design a wheelchair with the feature of easy-walking, convenient-use, and small-radius-swerving because the wheelchair is often used in a relatively narrow and small room [13,14]. The wheelchair tennis has been recognized in the world and players using wheelchair tennis are increasing in many countries. But, when playing tennis a wheelchair user is required to quickly and accurately control the wheelchair. Also, there are needed sophisticated techniques to move the wheelchair and c Springer Nature Switzerland AG 2020  L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 721–732, 2020. https://doi.org/10.1007/978-3-030-33506-9_66

722

K. Matsuo and L. Barolli

strong physical strength for playing tennis. In particular, for the beginner who has not strong strength to move the wheelchair is very difficult to hit the ball. Thus, it is necessary to prepare an easy environment for tennis beginners. For this reason, we proposed an omnidirectional wheelchair for playing tennis. In this paper, we propose a position detecting system using RSSI (Received Signal Strength Indication). The RSSI is predicted by Scikit-learn. The proposed system can find correctly the wheelchair position in order to avoid the collision. The structure of this paper is as follows. In Sect. 2, we introduce the related work. In Sect. 3, we present our proposed omnidirectional wheelchair system for playing tennis. In Sect. 4, we discuss some implementation issues. In Sect. 5, we show the experimental environment and results. Finally, conclusions and future work are given in Sect. 6.

2

Related Work

Most of the work, for mobile robots has been done for improving the quality of life of disabled people. One of important research area is robotic wheelchairs. The persons having physical impairment often find it difficult to navigate the wheelchair themselves. The reduced physical function associated with the age or disability make independent living and playing sport more difficult. Many research works have been undertaken to reduce the problem of navigation faced by the physically and mentally challenged people and also older age persons. One of the suggestive measures are the development of a Brain Control Interface (BCI), that assist an impaired person to control the wheelchair using his own brain signal. The research proposes a high-frequency SSVEPbased asynchronous BCI in order to control the navigation of a mobile object on the screen through a scenario and to reach its final destination [3]. This could help impaired people to navigate a robotic wheelchair. The BCIs are systems that allow to translate in real time the electrical activity of the brain in commands to control devices, provide communication and control for people with devastating neuromuscular disorders, such as the Amyotrophic Lateral Sclerosis (ALS), brainstem stroke, cerebral palsy, and spinal cord injury [7]. One of the key issue in designing wheelchairs is to reduce the caregiver load. Some of the research works deal with developing prototypes of robotic wheelchairs that helps the caregiver by lifting function or which can move with a caregiver side by side [10,15]. The lifting function equipment facilitates easy and safe transfer from/to a bed and a toilet stool by virtue of the opposite allocation of wheels from that for a usual wheelchair. The use of lifting function and the folding of frames makes it more useful in indoor environments. Robotic wheelchair based on observations of people using integrated sensors can move with a caregiver side by side. This is achieved by a visual-laser tracking technique, where a laser range sensor and an omnidirectional camera are integrated to observe the caregiver. Another important issue for the design of wheelchair is the collision detection mechanism. The omnidirectional wheelchairs with collaborative controls ensures

Prediction of RSSI by Scikit-Learn for Improving Position Detecting System

723

better safety against collisions. Such wheelchairs possess high level of ability when moving over a step, through a gap or over a slope [2,8]. To achieve omnidirectional motion, vehicles are generally equipped with an omniwheel consisting of a large number of free rollers or a spherical ball wheel. The development of such omniwheels attempts to replace the conventional wheel-type mechanism. There are also other works which deal with vision design of robotic wheelchairs by equipping the wheelchair with camera for monitoring wheelchair movement and obstacle detection and pupil with gaze sensing [1,6]. Prototype for robotic wheelchairs have been suggested in various research works, which are exclusively controlled by eye and are used by different users, while proving robust against vibration, illumination change, and user movement [4,5]. In [16], in order to enable an older person to communicate with other people the assisting devices have been developed. The assisting device can improve the quality of life for the elderly and disabled people by using robotic wheelchairs. The head gesture recognition is performed by means of real time face detection and tracking techniques. The authors also developed a useful human-robot interface for RoboChair. Also, the application of detecting a position using RSSI has been an active research area in robotics. One of the applications is RSSI based Bluetooth positioning method [9]. In [12] is presented a method of indoor position detecting using WiFi and trilateration technique.

3

Proposed Omnidirectional Wheelchair for Playing Tennis

In this section, we describe the implementation of wheelchair for playing tennis. We show a conventional wheelchair in Fig. 1. The wheelchair should make 5 movements. This is only one example of using the wheelchair, but when the wheelchair is used for playing tennis is difficult to make movements. In order to deal with these problems, we propose an omnidirectional wheelchair as shown in Fig. 2. The implemented omnidirectional wheelchair and control unit is shown in Fig. 3. Our implemented omnidirectional wheelchair is very suitable for playing tennis. 3.1

Kinematics

For the control of the wheelchair are needed the omniwheel speed, wheelchair movement speed and direction. Let us consider the movement of the wheelchair in 2 dimensional space. In Fig. 4, we show the onmiwheel model. In this figure, there are 3 onmiwheels which are placed 120◦ with each other. The omniwheels are moving in clockwise direction as shown in the figure. We consider the speed for each omniwheel M1, M2 and M3, respectively. As shown in Fig. 4, the axis of the wheelchair are x and y and the speed is ˙ In this case, the moving speed of the v = (x, ˙ y) ˙ and the rotating speed is θ. wheelchair can be expressed by Eq. (1).

724

K. Matsuo and L. Barolli

Fig. 1. Moving of conventional wheelchair.

Fig. 2. Moving of our proposed wheelchair.

˙ V = (x, ˙ y, ˙ θ)

(1)

Based on the Eq. (1), the speed of each omniwheel can be decided. By considering the control value of the motor speed ratio of each omniwheel as linear and synthesising the vector speed of 3 omniwheels, we can get Eq. (2) by using Reverse Kinematics, where (d) is the distance between the center and the omniwheels. Then, from the rotating speed of each omniwheel based on Forward Kinematics, we get the wheelchair moving speed. If we calculate the inverse matrix of Eq. (2), we get Eq. (3). Thus, when the wheelchair moves in all directions (omnidirectional movement), the speed for each motor (theoretically) is calculated as shown in Table 1.

Prediction of RSSI by Scikit-Learn for Improving Position Detecting System

725

Fig. 3. Real implementation of omnidirectional wheelchair and control unit.

     M1   1 0 d x˙               1 √3     M2  =  − −  y˙  d 2    2             √  M3   − 1   θ˙  3 d 2 2   2   x˙   3 − 13 − 13  M1                 y˙  =  0 − √1 √1  M2      3 3           1  M   θ˙   1 1 3 3d 3d 3d 3.2

(2)

(3)

Control System of the Proposed Omnidirectional Wheelchair

For the control of the proposed omnidirectional wheelchair, we considered R8C38 CPU board from Renesas Electronics Corporation. This CPU board has a small size and high speed processing time. The core of the CPU has a maximum frequency of 20 MHz. It is equipped with a flash memory, which is easy to rewrite. The R8C38 board has the following features: • • • • • • •

8bit multi functions timer: 2, 16bit output competition timer: 5, Real time clock timer: 1, UART/clock synchronization type serial interface: 3 channels, 10bit A/D converter: 20 channels, 8bit D/A converter: 2 circuits, Voltage detected circuit,

726

K. Matsuo and L. Barolli

Fig. 4. Model of omniwheel. Table 1. Motor speed ratio. Direction (Degrees) Motor speed ratio Motor1 Motor2 Motor3 0

0.00

−0.87

0.87

30

0.50

−1.00

0.50

60

0.87

−0.87

0.00

90

1.00

−0.50

−0.50

120

0.87

0.00

−0.87

150

0.50

0.50

−1.00

180

0.00

0.87

−0.87

210

−0.50

−1.00

−0.50

240

−0.87

0.87

0.00

270

−1.00

0.50

0.50

300

−0.87

0.00

0.87

330

−0.50

−0.50

1.00

360

0.00

−0.87

0.87

• Number of output and input port: 75, • External interrupt input: 9. In Fig. 5 is shown the control system for the proposed omnidirectional wheelchair. The direction movement of the wheelchair is decided by the Joystick. The Analog-Digital Converter changes the analog value to a digital value needed for R8C38 board. The R8C38 board based on the Eq. (2) calculates the motors control value. Based on this value, the Pulse Width Modulation (PWM) generator generates an appropriate value for the control of each motor. The

Prediction of RSSI by Scikit-Learn for Improving Position Detecting System

727

Fig. 5. Control system for omnidirectional wheelchair.

number of rotation of each motor is detected by Pulse Counter and is sent to the R8C38 board in order to make a correct feedback control.

4

Implementation and Application Issues

In Fig. 6, we show a schematic illustration of position detecting system using RSSI for omnidirectional wheelchair tennis. The RSSI is used to calculate the omnidirectional wheelchair position. This time we predicted 3 spaces in the tennis court. In Fig. 7 is shown the implemented position detecting system using RSSI. The RSSI is predicted by Scikit-learn. The right side of the screen shows the image of tennis court. The Scikit-learn is an open source software for machine learning. The Scikitlearn has a number of algorithms that support vector machine, random forest, k-means clustering and neural network. We use neural network for predicting the RSSI. The court is divided in 3 areas (see. 1, 2, 3 of Fig. 7). The proposed position detecting system in Fig. 8 is composed by two stages: the training stage and the detection stage. In this paper, we used two RSSI receivers as shown in Fig. 9. The wheelchair has a smart phone or a tablet device for emitting the WiFi radio wave. Two receivers get the RSSI from the wheelchair. Then, the proposed system predicts the wheelchair position.

728

K. Matsuo and L. Barolli

Fig. 6. Position detecting system using RSSI for omnidirectional wheelchair tennis.

5

Experimental Environment and Results

We have shown the experimental environment in Fig. 9 and the experimental result of loss function in Fig. 10. The graph shows a good result, because the value of Loss is going close to 0 gradually. When the value of Time Step is over 100, the Loss value is almost 0 (the max number of learning times is 140). The accuracy of predicting rate is 99.66%.

Fig. 7. Implemented position detecting system using predicted RSSI.

Prediction of RSSI by Scikit-Learn for Improving Position Detecting System

Fig. 8. Proposed detecting architecture using Scikit-learn.

Fig. 9. Experimental environment using two RSSI receivers.

Fig. 10. Experiment results of loss function.

729

730

K. Matsuo and L. Barolli

Figure 11 shows the results of classification for 3 areas considering RSSI, which is predicted by Scikit-learn. The dots in the graph show the measured data using two RSSI receivers. Moreover, we show the results of predicting wheelchair positions by our proposed system in Fig. 12. The proposed system can predict the wheelchair positions almost correctly in 3 areas.

Fig. 11. Experiment results of classification for 3 spaces.

Fig. 12. The result of wheelchair position predicted by RSSI.

6

Conclusions and Future Work

In this paper, we introduced our implemented omnidirectional wheelchair for tennis. We showed some of the previous works and discussed the related problems and issues. Then, we presented in details the kinematics and the control system

Prediction of RSSI by Scikit-Learn for Improving Position Detecting System

731

for our implement omnidirectional wheelchair. We proposed a position detecting system for improving the performance of wheelchair tennis by predicting RSSI by Scikit-learn. The evaluation results show that the proposed system can find correctly the wheelchair position for avoiding collision. In the future work, we want to use 4 RSSI receivers in order to get the wheelchair position more correctly. The wheelchair should not only avoid the collision, but also the player should move to an accurately position automatically for the shot.

References 1. Arai, K., Mardiyanto, R.: A prototype of electric wheelchair controlled by eye-only for paralyzed user. J. Robot. Mechatron. 23(1), 66–74 (2011) 2. Carlson, T., Demiris, Y.: Collaborative control for a robotic wheelchair: evaluation of performance, attention, and workload. IEEE Trans. Syst. Man Cybern. Part B (Cybern.) 42(3), 876–888 (2012) 3. Diez, P.F., Mut, V.A., Perona, E.M.A., Leber, E.L.: Asynchronous BCI control using high-frequency SSVEP. J. Neuroeng. Rehabil. 8(1), 1–8 (2011). http://www. jneuroengrehab.com/content/8/1/39 4. Escobedo, A., Spalanzani, A., Laugier, C.: Multimodal control of a robotic wheelchair: using contextual information for usability improvement. In: Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems, pp. 4262–4267 (2013) 5. Gonzalez, J., Muaeoz, A., Galindo, C., Fernandez-Madrigal, J., Blanco, J.: A description of the SENA robotic wheelchair. In: Proceedings of the IEEE Mediterranean Electrotechnical Conference (MELECON-2006), pp. 437–440 (2006) 6. Gray, J., Jia, P., Hu, H.H., Lu, T., Yuan, K.: Head gesture recognition for handsfree control of an intelligent wheelchair. Ind. Robot: Int. J. 34(1), 60–68 (2007) 7. Grigorescu, S.M., L¨ uth, T., Fragkopoulos, C., Cyriacks, M., Gr¨ aser, A.: A BCIcontrolled robotic assistant for quadriplegic people in domestic and professional life. Robotica 30(03), 419–431 (2012) 8. Ishida, S., Miyamoto, H.: Collision-detecting device for omnidirectional electric wheelchair. ISRN Robot. 2013, 1–8 (2012) 9. Jianyong, Z., Haiyong, L., Zili, C., Zhaohui, L.: RSSI based Bluetooth low energy indoor positioning. In: 2014 International Conference on Indoor Positioning and Indoor Navigation (IPIN), pp. 526–533. IEEE (2014) 10. Kobayashi, Y., Kinpara, Y., Shibusawa, T., Kuno, Y.: Robotic wheelchair based on observations of people using integrated sensors. In: IROS, pp. 2013–2018 (2009) 11. Lu, T., Yuan, K., Zhu, H.: Research status and development trend of intelligent wheelchair. Appl. Technol. Robot 2, 1–5 (2008) 12. Mahiddin, N.A., Safie, N., Nadia, E., Safei, S., Fadzli, E.: Indoor position detection using WiFi and trilateration technique. In: The International Conference on Informatics and Applications (ICIA 2012), pp. 362–366 (2012) 13. Matsuo, K., Barolli, L.: Design and implementation of an omnidirectional wheelchair: control system and its applications. In: Proceedings of the 9th International Conference on Broadband and Wireless Computing, Communication and Applications (BWCCA-2014), pp. 532–535 (2014)

732

K. Matsuo and L. Barolli

14. Matsuo, K., Liu, Y., Elmazi, D., Barolli, L., Uchida, K.: Implementation and evaluation of a small size omnidirectional wheelchair. In: Proceedings of the IEEE 29th International Conference on Advanced Information Networking and Applications Workshops (WAINA-2015), pp. 49–53 (2015) 15. Mori, Y., Sakai, N., Katsumura, K.: Development of a wheelchair with a lifting function. Adv. Mech. Eng. 2012, 1–9 (2012) 16. Wang, H., Grindle, G.G., Candiotti, J., Chung, C., Shino, M., Houston, E., Cooper, R.A.: The personal mobility and manipulation appliance (PerMMA): a robotic wheelchair with advanced mobility and manipulation. In: 2012 Annual International Conference of the IEEE Engineering in Medicine and Biology Society, pp. 3324–3327 (2012)

Decentralized Mechanism for Hiring the Smart Autonomous Vehicles Using Blockchain Zain Abubaker1 , Muhammad Usman Gurmani1 , Tanzeela Sultana1 , Shahzad Rizwan2 , Muhammad Azeem1 , Muhammad Zohaib Iftikhar1 , and Nadeem Javaid1(B) 1

2

Department of Computer Science, COMSATS University Islamabad, Islamabad 44000, Pakistan [email protected] Department of Computer Science, COMSATS University Islamabad, Attock Campus, Attock, Pakistan

Abstract. Nowadays, technologies like Autonomous Vehicles (AVs) are influencing the ways of our traveling. This paper inspects closely the development of a decentralized blockchain-based mechanism for providing secure, reliable and real-time availability of AVs for the customers who want to do the ride. The AVs have many advanced control systems and sensors to detect a number of hurdles (unsafe design of vehicles, negligence of civilians, etc.) in the environment. Blockchain is a decentralized temper proof business protocol used to facilitate the users with transparent, reliable, secure and cost-effective solutions. The consensus mechanisms are used in blockchain for validation purposes. This paper uses the Proof of Work consensus algorithm for the validation of Demand Response (DR) events. It provides the mechanism for real-time monitoring and real-time supervision to the ride of the end-user. Furthermore, it briefly specifies that the AVs working with blockchain mechanisms provides real-time traffic information to the end-user. The blockchain-based mechanism provides secure services to the end-user. It also provides the mechanism of Peer to Peer (P2P) car-sharing that removes the need for any bank or any reliable authority. The proposed system is proved in the Ethereum environment by DR events in the network. The simulations portray that our system is much cost-effective, efficient and reliable to meet the demands of customers.

Keywords: Blockchain technology Services

1

· Autonomous Vehicles · Mobility ·

Introduction

The costs of transportation are rising gradually. Moreover, the environmental damage occur due to ordinary vehicles are very serious. These issues prompted c Springer Nature Switzerland AG 2020  L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 733–746, 2020. https://doi.org/10.1007/978-3-030-33506-9_67

734

Z. Abubaker et al.

the making of new methods for transportation such as AVs. AVs offer a large number of solutions to many issues in transportation. The AVs are driver-less and in this way, these AVs facilitate the people in a secure and flawless way. In the next few years, the AVs will become the standard for consumers. However, fully AVs are difficult for any individual to be an owner of it as these are very expensive. The people who want to hire a vehicle to reach at their desired place face many problems. First, they have to wait on the road for a taxi. Second, when they gets their taxi, the safety of their ride is compromised or not fully satisfied as there is human errors involved. The AVs vehicles working on the mechanism of blockchain satisfies all issues of customer. Autonomous Vehicles The AVs have many sensors to detect the road hurdles (unsafe design of vehicles, negligence of civilians etc.) very quickly. The AVs have capabilities to compute the distance between other vehicles and traffic signals on their own. These are also effective to communicate with other AVs and to store data. The sensors of AVs senses the traffic signals and respond well to them as compared to the drivers. There are a number of possible deficiencies that occurs with the taxi driver like they may drink and drive or may do distracted driving [7]. Using AVs these deficiencies definitely be removed from the transport system. In today’s traffic the blockage of routes is one of the main issues, the AVs performs well while operating around stable or moving obstacles. With all aforementioned capabilities of AVs, these vehicles have problem of storing data. These vehicle store data according to their capability and their capability to store data is not enough for large amount of vehicles. Therefore, we use the mechanism of blockchain to store data of vehicles to very large extent. 1.1

Motivation

The authors in [1] studied that blockchain based vehicular network is a reliable and robust model. They studied that the model operates well while some malicious activities are still damaging the network and reducing the performance of the network. The authors in [2] studied that sharing of resources of AVs is actually the combination of traditional car sharing and the services with the AVs. These shared vehicles provide cost-effective and convenient mobility ondemand services in real time [3]. The authors in [4] proposed the mechanism of self-driving autonomous cars which works well without the involvement of any human effort. Firstly, AVs were made in the US and Germany from 1980 to 2000 [5]. Satoshi Nakamoto proposed a decentralized model for P2P sharing of cash [6]. Our proposed vehicular network operates in a decentralized way to make the distributed transport management system more effective and reliable [1]. Our model uses the mechanism of smart share to tackle the issue of DR [2]. 1.2

Problem Statement

The searching for a driver of a taxi on the road frequently is not considered the convenient mobility on-demand service [1,3]. One more issue with the existing

Decentralized Mechanism for Hiring the Smart Autonomous Vehicles

735

system of [1] is that it does not provide transparency about the traveling route of the ride. This gives rise to concerns about security in the old system [1]. Moreover, the proposed model in [1] involves human deficiencies as the taxi drivers are used in this model. When there is an ordinary car with driver, there are security risks for the customer because he does not know the details of the driver available for his ride. For the ride, he has to trust in case of driver identification, because he has no other option. According to the situations of nowadays, this is really a security risk for customers. The autonomous vehicles are the proposed solution for this problem. The proposed model of [1] there is no mechanism to tackle with DR events. When there are a lot of people who want to do ride. When these people request over the network and number of vehicles to serve them are small, this raises a serious issue to deal the customers. To deal with a large number of requests with less number of vehicles, our proposed model uses the mechanism of smart share.

2

Related Work

Networks share their resources to provide services to the nodes. However, these networks have limited resources. To provide the services effectively, each and every node in the network should participate and remain active for assigned operations. With the increase in the data set and increasing size of the application, networks are not more efficient and scalable. In [1,8,14,15,19,23,24] the blockchain technology is used in the wireless sensors networks. Moreover, paper [18] uses Proof of Authority (PoA) consensus mechanism for validation purposes and consortium mechanism is used to make an alliance with other companies to achieve the common goals. Moreover, Ethereum Geth -11.811 is used for transactions in the proposed model. [15] uses the policy development process along with LoRaWAN, Ethereum Geth -1-1.811. [22] uses Proof of Collaboration (PoC) for authentication of nodes in the proposed network. In [18] all the simulations are done in Python 3.6, Smart Contract and Raspberry pi2. All the simulations in [24] are done using the mechanism of cloud-based data storage. The authors in [26] proposed a cloud based mechanism for secure services for IoT devices. In [2], the authors also perform simulations to prove secure distribution of services over the network. [8] provides a mechanism for a decentralized system, spectral efficiency Q-learning and exhaustive learning. [24] uses behavior chain and data chain which ensures high data rate and high reliability in the network. [22] proposed a mechanism for the sharing of data so that reuse the right of research information be overseen by utilizing the innovations. Moreover, the authors in [25] use blockchain mechanism over the smart grids for fair sharing of data in deregulated smart grids. The authors perform some simulations to prove fair data sharing in deregulated smart grids. [1,20–22] use the mechanism of blockchain in the vehicles to establish a secure, transparent and reliable network for the Internet of Vehicles (IoV). [1] and [22] use Proof of Work (PoW) consensus mechanism for validations purposes in the network. [1] and [22] both use Ethereum environments for the transaction of commodities between different entities and all the smart contracts are

736

Z. Abubaker et al.

designed using solidity in visual code. [20] provides us a mechanism for vehicles to overcome trust issues; moreover, the deadlock process is overcome in this method. [21] uses store transport data, sensory data, environmental data and insurance data of vehicles and performs trading in data. In [1] ordinary nodes, miner nodes, and controller nodes are used to achieve the robustness and adaptability. [22] uses mathematical puzzles and distributed consensus for a secure and reliable network. The blockchain is used on the Internet of Vehicles (IoVs) to establish the security, transparency and trust management between vehicles. In [8–12] the blockchain is used on the IoVs to establish the security, transparency and trust management between vehicles. The limitation of the model of [8] is that the usage of advanced management hub might reduce the performance of the overall system. In [9] the system model of trust policy is made to detect compromised Unmanned Aerial Vehicles (UAVs). The mechanism is made to detect wrong information when an official UAV is physically hijacked. The model explains that due to the dynamic topology of UAVs, these vehicles raise security challenges. Go Methods, ABS security UAV, NetLogo is used for its utility to support mobile ad-hoc network. Novel agent-based simulator ABS- security UAV is used to validate the model. [10] addresses that the accuracy and power of driving safety assessments are limited. Dynamic prediction of road safety in a city is not capable to provide safety on roads. This paper proposes the model of a Deep learning framework (DeepRSI) to conduct the prediction of real time road safety in order to improve the safety of vehicles. Mobile sensing data collection is used in VANETs to identify problems. Intel Core i7 machine with 32 GB Ram and NVIDIA TITAN X graphics card are used for simulations purposes. The limitation of this proposed model is that there is no mechanism to check the reliability of the prediction of road safety. The authors in [27] applied blockchain mechanism on under water, Water Sensor Network (WSN) and achieve efficient routing of energy. The authors in [28] use the mechanism of blockchain on the management of data over the ethereum network. The authors in [29] do monetization of data using data science over the IoT devices. The authors in [30] achieve the trustfulness in complex network using both blockchain and data science together. In [11] the blockchain based decentralized mechanism is proposed to handle the energy demand by controlling the number of transactions. It explains the issues that to update the ledger, a large amount of consumption of energy in the transaction of blockchain may cause some serious issues for vehicles. Distributed clustering is used to label every chain sequentially. The results are simulated in MATLAB TM. The model does not provide a fully optimal solution in handling the energy. In [12] P2P data sharing system is proposed to achieve accurate reputation management for high-quality data sharing for vehicles. It explains that the vehicular edge computing servers (Road Side Units) are not fully trusted and may cause serious security and privacy challenges. Consortium blockchain and smart contracts are used to achieve secure data storage. The authors in [31] balances the demand and response in smart grids using blockchain mechanism. The authors in [32] analyze and secure data using both broad fields data science and blockchain.

Decentralized Mechanism for Hiring the Smart Autonomous Vehicles

737

The authors in [33] introduce the mechanism of incentive for lightweight client based on blockchain. The authors in [34] introduce the node recovery scheme for wireless sensor network.

3

System Model

In [1], when the user selects any vehicle for the ride and starts the ride with this particular vehicle. During his travel, there is no mechanism to show the details of ride to the customer. This mechanism does not facilitate the user to keep an eye on the route during traveling. The blockchain provides the capability of transparency. In our proposed model, there is a mechanism to show each and every detail of the vehicle on the network and the location of each vehicle is also given. The user selects any vehicle according to his wish without any inference of the third party. This proposed model increases the user’s choice for transportation. However, in our proposed model, we have used the mechanism where all details of the routes are mentioned for the network. These details are also monitored by other miner nodes in the network. All miner nodes in the network are static nodes. These miner nodes are actually the Road Side Units (RSU). In this way, the model facilitates the customer with transparency about the details of travel. The proposed model provides the mechanism to establish communication between the smart vehicles and end users. Our proposed system model shown in Fig. 1 is similar to the system model proposed in [1]. In [1], the vehicles with drivers are used and this raises a big security concern for customers. Because, the user knows nothing about the driver and he has to travel with random driver. According to the situations of nowadays, there is a lot of probability that the driver may drink or drive, may do distract driving. The driver may harm the customer, this is really a big security concern. In this way, with such security risks, the customer had to travel with the driver. The issue with the system of [1] is that there is no mechanism to store and to show these details to the customer. Here, one important point to keep in mind is that all the way we use a mechanism to store the reputations’ details of the driver. These details are shown to the customer. When the user selects any vehicle for ride, there is also a chance that the driver do drink and drive or harm the customer. Therefore, our proposed model uses AVs as these vehicles are driver-less vehicles. So there is no security concern about the driver at all. This proposed model solves this issue of security as AVs are used and there is no driver involved. When the user interacts with the smart autonomous vehicle through the blockchain mechanism, then the user sees each and every detail of the vehicle about its previous rides and feels free to select any vehicle for his ride. After that the user selects any vehicle for his trip and he sends his location to the AVs, then this vehicle picks him from that particular location and drops him at his desired place. In this way, all the mechanism is under control of traveler as he selects any vehicle according to his wish. Figure 1 shows the system model in which a rider wants to do ride and he requests over the network for the ride. While receiving his request, all the

738

Z. Abubaker et al.

nodes in the network respond him back about their availability. Our proposed model provides two-way communication in the network. In this two-way communication, the customer picks the vehicle according to his priorities. Moreover, the vehicle also selects the customer according to its priority. This reduces the latency in the DR event. Moreover, to handle the issue of DR, our proposed model uses the mechanism of smart share. A vehicle using a smart share facilitates a lot of customers. When any customer is traveling from destination 1 to destination 2, on the way the AV can facilitate another person who also wants to go to destination 2. In this way, the smart share facilitates a lot of customers in less time. Therefore, the proposed system becomes better to deal with a large number of customers without any inconsistencies, because the proposed model provides two-way communication. In this way, this model does not only facilitates the customer/rider but also to the vehicles, that pick the customer according to their priorities. The system model shown in Fig. 1 works without the involvement of any third party and is owned by nobody. The reason for not having any involvement of the third party is that this proposed model is blockchain based and blockchain works in a decentralized way without involvement of any third party. This property of blockchain suits us to tackle the issue of extra cost. The rider has only to pay the vehicles for their services and has not to pay to the third party. In this way, the proposed model using blockchain provides us a cost-effective solution.

4

Methodology

We proposed an Intelligent Transport System (ITS) based on the mechanism of blockchain as shown in Fig. 1. This ITS system is made by taking the motivation from [1] and [22]. All the transactions and consensus are held by following the principles of the blockchain mechanism. Firstly, when a rider wants to go to his desired place. He requests over the network by giving all his details like location, from where he wants to take the ride and at which place he wants to reach. After submitting his request over the network, he has to wait for a while for the response. In the network, there are some AVs that respond to the request of the rider. If more then one vehicle respond to the customer then it is the choice of the customer to pick any vehicle according to his priorities. Moreover, each and every detail of the vehicle is given on the network. Then the user selects any vehicle according to his desire and comfort. When the user selects any vehicle then a consensus mechanism is established between the customer/rider and the vehicle in the form of the smart contract. The business rules are stored in the smart contract. Every entity contains its own smart contract. After the consensus is done, the AVs provide service to the customer and get the incentive as decided in the smart contract. In this way, the customer easily gets his services according to his desire. He picks any vehicle without any involvement of third party that interferes in selecting the vehicle. Moreover, during the ride of the customer, his ride is constantly monitored by miners. This capability of our proposed model provides us the transparency in the network. Finally, when the

Decentralized Mechanism for Hiring the Smart Autonomous Vehicles

739

Vehicular Cloud

Vehicular Cloud

Smart Vehicle

Smart Vehicle Miner Node

Controller Node

Mobile User

Ordinary Node

Mobile Users DR Data Exchange

Miner Node

Miner Node

Environmental Data Exchange

Fig. 1. The proposed model for communication and transaction between Autonomous Vehicles and End-Users

rider reaches his desired place and AV gets its reward/incentive, then miner nodes authenticate the exchange of services and rewards between them. The miner nodes also authenticate that the customer safely reached the desired place. This authentication is done using the PoW mechanism. When 51% of miner nodes authenticate the successful transaction between these entities then this transaction becomes part of the ledger and permanently stored in the ledger. Our blockchain based proposed model also balances the DR events. The demands of customers and the responses to their demands are balanced well in our proposed model. In the centralized approach, the customer has to rely on the third party, he has to wait until the third party provides him the vehicle. Sometimes, there are some drivers who are not interested to go to the specific route. First, the third party searches for interested drivers and then provides the vehicle to the customer. It is really a time consuming process and the customer has to wait until the third party finds any vehicle. However, in our proposed model, there is not any driver who influences the decision. Another issue of the centralized approach is that sometimes, there are a lot of vehicles in the network. It is only decided by the third party that which customer is given to which vehicles. This reduces the equality for any vehicle to pick the customer. In our proposed model, there is two-way communication. So, not only the customer picks the vehicle according to his priorities. However, the vehicle also picks the

740

Z. Abubaker et al.

customer according to its priorities. The request of the customer is deployed on the network where all the vehicles have equal chances to pick the customer and to own currency. When there are more then one vehicles are interested to give ride to the customer, In such situations, it is fully dependent on the customer to pick any vehicle. Usually the customer picks the vehicle that offers him better rates with better services. When all the vehicles and customers are involved in the network, then this reduces the delay in responding to the customer. This mechanism is helpful for both the customer and the vehicles. This helps to tackle the DR event in the proposed model. Moreover, the mechanism of smart share is also used to tackle the issue of DR. In the centralized approaches, the driver of vehicles is not allowed to pick any customer. First, the centralized authority takes the request from the customer and then forwards it to the driver. The driver of the vehicle is only allowed to say “yes” or “no” to the request. However, in our proposed model, the AVs have full authorization to select any customer. After showing his interest in the customer, the whole next process depends upon the consensus between customer and AVs. In our proposed model, when there are a number of requests on the network, and if the AVs select any customer, the network gives access to the AVs to communicate with this specific customer if this customer is still the part of the network. The access to communicate to the customer is not given to the AVs if the customer is not still part of the network, i.e, the customer leaves the network or selects another vehicle for a ride. The same mechanism is followed for the customer, a customer can select any vehicle only if this vehicle is present in the network. Moreover, the issue of DR is tackled by using the scheme of smart sharing [1]. 4.1

Smart Share

A vehicle using a smart share facilitates a lot of customers. As a customer wants to go to destination 2 and another customer also want to go to destination 2. They both can share their vehicle to go to the common destination (Destination 2). This scheme of smart share facilitates a lot of people in less time, resulting in response to the requests of the customers. The mechanism of smart share is helpful in two ways. First, when two or more customers are sharing the vehicle, the total cost is equally divided between them and in this way smart share provides cost effective solution. Second, it is helpful to tackle the DR issue. 4.2

Smart Contracts

The smart contracts are made to allow the trustworthy transaction between customer and service providers. The business rules are stored in this smart contract. First, the contract checks either the user has enough currency to travel at his desired place or not. When the user has sufficient currency then the access to travel is given to the user. On the other hand, the smart contract checks whether

Decentralized Mechanism for Hiring the Smart Autonomous Vehicles

741

the AVs have enough capabilities that they achieve the requirements of the customer, i.e, electricity power and path planning, etc. Once the smart contract checks the authentication of both, it gives access to both parties for the journey. 4.3

Mining

Once the successful travel is done, the details of the transactions broadcast on the blockchain network and the miner nodes in the network start mining to validate the transaction. All the miner nodes in the network are static nodes while the vehicles and the mobile user are considered as mobile nodes. The validation is done according to the consensus algorithm. PoW consensus algorithm is used in our proposed model. The miner nodes check either the person reached his desired place safely, means he got his services and the vehicle got its reward. Once the miner nodes have completed mining, the blockchain mechanism checks either the 51% nodes validate the transaction then this transaction permanently becomes the part of the blockchain ledger. The miner nodes are static nodes, these miner nodes are RSU in our proposed solution. The miner nodes should have particular capabilities that are needed for mining. 4.4

Hashing Algorithm

Each block in the blockchain is connected to its next block. The blocks are connected in this way that every node keeps the hash of its previous block. This hash is created by using different hash functions. In our proposed model, SHA256 hashing algorithm is used for the encryption of data of the block. When someone wants to alter any data in the block then the hash of this particular block changes. This leads to the change of hashes of all the next blocks. This capability of blockchain provides us a secure and reliable network. 4.5

Consensus Algorithm

Blockchain uses consensus algorithms for the validation of transactions done in the network. These consensus mechanisms help in checking the effectiveness of the transaction without the involvement of any third party [17]. Our proposed model uses the PoW consensus mechanism for validation purposes. It is used to authenticate the transactions and to produce new blocks in the blockchain. In the mechanism of PoW, the miner nodes compete with each other to complete the task given on the network for getting reward. These miner nodes performing PoW solve the mathematical puzzle. First, the nodes solve this puzzle and then get the incentives that has been decided earlier. The above proposed model makes every ride safe for the end user. The proposed model is blockchain based and the capabilities of blockchain solve many issues for end users.

742

5

Z. Abubaker et al.

Simulations and Results

The simulations are done using laptop which has following storage and computing capabilities: • 4.00 GB RAM and 500 GB ROM • 64-bit window 10 (Operating System). • Intel (R) Core(TM) M-5Y10c CPU @ 0.80 GHz 1.00 GHz In the centralized approach, there are many time slots when the number of vehicles and the number of requests are not equal. Sometimes, the number of vehicles are not enough to deal with the customer. So, the customer has to wait for a while. Figure 2 shows the results of centralized technique which is normally known as Careem Ride Service. Figure 3 shows the results of our proposed model. In our proposed model, the mechanism of smart share is used to deal with this issue. In this way, the customer without any wastage of time gets the service from vehicles. Figure 3 also shows the trade-off in DR events. When the number of requests of customers are very high and the number of vehicles in the network is small. Then, these vehicles are not able to deal with the requirements of the customer. In proposed blockchain based network, the customers have to wait for some while for the response. This shows the trade-off of time with demand response events. It may be also seen that the cost of the ride increases a little bit due to the high number of requests of customer. This shows the trade-off between time and DR as well as between cost and DR.

Centralized Demand Response Mechanism 120 100

Requests

80 60

No. of Vehicles

No. of Requests

40 20 0 1

3

5

7

9 11 13 15 17 19 21 23 Time (Hours)

Fig. 2. Centralized demand response mechanism

In centralized approach, the third party is involved and customers have to pay some charges to the third party. Moreover, the price is not stable for same amount of distance. Therefore, the customer has to pay more charges for the resources he used. In proposed model, there is no involvement of the third party. Moreover, before starting the ride, the amount that the customer has to pay to

Decentralized Mechanism for Hiring the Smart Autonomous Vehicles

743

Decentralized Demand Response Mechanism 120

Requests

100 80

60

No. of Vehicles

40

No. of Requests

20 0 1

3

5

7

9 11 13 15 17 19 21 23 Time (Hours)

Fig. 3. Decentralized demand response mechanism

the vehicle is predetermined. Therefore, our proposed model is considered to be better than the centralized approach. The simulation results of both centralized and decentralized approaches are shown in Fig. 4.

70

60

Cost

50

40

Decentralized Approach Centralized Approach

30

20 10 0 1

2

3 4 5 6 7 8 No. of TransacƟons

9 10

Fig. 4. Comparison of cost between centralized approach and decentralized approach

6

Conclusion and Future

In this paper, we propose the use of AVs together with decentralized blockchain based protocol. The blockchain based model provides us the mechanism in which the whole trip for the passenger is secure and transparent because the AVs are used in our proposed model. This removes the necessity of identification of the driver. Our blockchain based proposed model provides the whole information

744

Z. Abubaker et al.

about the route to the customer by real time traffic information. It makes the lowcost transaction easier by providing the mechanism of peer to peer car sharing which removes the need for any bank or any reliable authority. Moreover, the property of smart sharing between different nodes of the network helps to tackle the issue of DR. In proposed model, AVs are used and actually these vehicles are driver-less. Due to the use of these AVs, there are some issues that are given below: • Who will be responsible in case of an occurrence of an accident due to the systematic error of AVs? In our above described proposed model, it is very difficult to know the answer to this question that who is responsible for the accident and who should reimburse for the damage? • The AVs are actually designed to work using the information propagated in the environment. In proposed model, the vehicles use the sensitive information of the user like the address of home or location, etc. These are very private pieces of information, thus this creates major privacy issues. • The hackers can attack the autonomous vehicles and can easily get into the system of the vehicle. It controls the operations of the AVs, which creates a security concern. In the future, more work will be done to tackle these issues. Some studies are still needed in the future that the mechanism of rating is introduced in this network. The vehicles that provide a feasible time and comfortable service to the customer get a positive rating. Moreover, the vehicle that fails to provide feasible time and comfortable services and gets a negative rating. All the ratings of the vehicle should deploy on the network. This improves the services in the network.

References 1. Sharma, P.K., Moon, S.Y., Park, J.H.: Block-VN: a distributed blockchain based vehicular network architecture in smart City. JIPS 13(1), 184–195 (2017) 2. Fagnant, D.J., Kockelman, K.: Preparing a nation for autonomous vehicles: opportunities, barriers and policy recommendations. Transp. Res. Part A: Policy Pract. 77, 167–181 (2015) 3. Burns, L., Jordan, W., Scarborough, B.: Transforming Personal Mobility. The Earth Institute, Columbia University, New York (2013, accepted) 4. Howard, D., Dai, D.: Public perceptions of self-driving cars: the case of Berkeley, California. In: Transportation Research Board 93rd Annual Meeting, vol. 14, no. 4502, pp. 1–16 (2014) 5. Anderson, J.M., Nidhi, K., Stanley, K.D., Sorensen, P., Samaras, C., Oluwatola, O.A.: Autonomous vehicle technology: a guide for policymakers. Rand Corporation (2014, accepted) 6. Nakamoto, S.: Bitcoin: A peer-to-peer electronic cash system (2008, accepted) 7. Kpmg, C., et al.: Self-driving Cars: The Next Revolution. Kpmg, Seattle (2012, accepted)

Decentralized Mechanism for Hiring the Smart Autonomous Vehicles

745

8. Iqbal, R., Butt, T.A., Afzaal, M., Salah, K.: Trust management in social Internet of vehicles: factors, challenges, blockchain, and fog solutions. Int. J. Distrib. Sens. Netw. 15(1), 1550147719825820 (2019) 9. Garc´ıa-Magari˜ no, I., Lacuesta, R., Rajarajan, M., Lloret, J.: Security in networks of unmanned aerial vehicles for surveillance with an agent-based approach inspired by the principles of blockchain. Ad Hoc Netw. 86, 72–82 (2019) 10. Peng, Z., Gao, S., Li, Z., Xiao, B., Qian, Y.: Vehicle safety improvement through deep learning and mobile sensing. IEEE Netw. 32(4), 28–33 (2018) 11. Sharma, V.: An energy-efficient transaction model for the blockchain-enabled internet of vehicles (IoV). IEEE Commun. Lett. 23(2), 246–249 (2018) 12. Kang, J., Yu, R., Huang, X., Wu, M., Maharjan, S., Xie, S., Zhang, Y.: Blockchain for secure and efficient data sharing in vehicular edge computing and networks. IEEE Internet Things J. 5, 1389–1399 (2018) 13. Rahmadika, S., Ramdania, D.R., Harika, M.: Security analysis on the decentralized energy trading system using blockchain technology. Jurnal Online Informatika 3(1), 44–47 (2018) 14. Xu, Y., Wang, G., Yang, J., Ren, J., Zhang, Y., Zhang, C.: Towards secure network computing services for lightweight clients using blockchain. In: Wireless Communications and Mobile Computing (2018, accepted) 15. Lin, J., Shen, Z., Miao, C., Liu, S.: Using blockchain to build trusted LoRaWAN sharing server. Int. J. Crowd Sci. 1(3), 270–280 (2017) 16. Lin, D., Tang, Y.: Blockchain consensus based user access strategies in D2D networks for data-intensive applications. IEEE Access 6, 72683–72690 (2018) 17. Zhang, Y., Wen, J.: The IoT electric business model: using blockchain technology for the internet of things. Peer-to-Peer Netw. Appl. 10(4), 983–994 (2017) 18. Xu, C., Wang, K., Li, P., Guo, S., Luo, J., Ye, B., Guo, M.: Making big data open in edges: a resource-efficient blockchain-based approach. IEEE Trans. Parallel Distrib. Syst. 30(4), 870–882 (2018) 19. Novo, O.: Scalable access management in IoT using blockchain: a performance evaluation. IEEE Internet Things J. (2018, accepted) 20. Jiang, T., Fang, H., Wang, H.: Blockchain-based internet of vehicles: distributed network architecture and performance analysis. IEEE Internet Things J. 5, 4100– 4108 (2018) 21. Singh, M., Kim, S.: Branch based blockchain technology in intelligent vehicle. Comput. Netw. 145, 219–231 (2018) 22. Yang, Z., Yang, K., Lei, L., Zheng, K., Leung, V.C.M.: Blockchain-based decentralized trust management in vehicular networks. IEEE Internet Things J. 6(2), 1495–1505 (2018) 23. Dai, M., Zhang, S., Wang, H., Jin, S.: A low storage room requirement framework for distributed ledger in blockchain. IEEE Access 6, 22970–22975 (2018) 24. Zhang, G., Li, T., Li, Y., Hui, P., Jin, D.: Blockchain-based data sharing system for AI-powered network operations. J. Commun. Inf. Netw. 3(3), 1–8 (2018) 25. Samuel, O., Nadeem Javaid, M.A., Ahmed, Z., Imran, M., Guizani, M.: A blockchain model for fair data sharing in deregulated smart grids. In: IEEE Global Communications Conference (GLOBCOM 2019) (2019) 26. Rehman, M., Javaid, N., Awais, M., Imran, M., Naseer, N.: Cloud based secure service providing for IoTs using blockchain. In: IEEE Global Communications Conference (GLOBCOM 2019) (2019) 27. Mateen, A., Javaid, N., Iqbal, S.: Towards energy efficient routing in blockchain based underwater WSNs via recovering the void holes. MS thesis, COMSATS University Islamabad (CUI), Islamabad 44000, Pakistan, July 2019

746

Z. Abubaker et al.

28. Naz, M., Javaid, N., Iqbal, S.: Research based data rights management using blockchain over ethereum network. MS thesis, COMSATS University Islamabad (CUI), Islamabad 44000, Pakistan, July 2019 29. Javaid, A., Javaid, N., Imran, M.: Ensuring analyzing and monetization of data using data science and blockchain in loT devices. MS thesis, COMSATS University Islamabad (CUI), Islamabad 44000, Pakistan, July 2019 30. Syeda, H., Kazmi, Z., Javaid, N., Imran, M.: Towards energy efficiency and trustfulness in complex networks using data science techniques and blockchain. MS thesis, COMSATS University Islamabad (CUI), Islamabad 44000, Pakistan, July 2019 31. Zahid, M., Javaid, N., Rasheed, M.B.: Balancing electricity demand and supply in smart grids using blockchain. MS thesis, COMSATS University Islamabad (CUI), Islamabad 44000, Pakistan, July 2019 32. Noshad, Z., Javaid, N., Imran, M.: Analyzing and securing data using data science and blockchain in smart networks. MS thesis, COMSATS University Islamabad (CUI), Islamabad 44000, Pakistan, July 2019 33. Ali, I., Javaid, N., Iqbal, S.: An incentive mechanism for secure service provisioning for lightweight clients based on blockchain. MS thesis, COMSATS University Islamabad (CUI), Islamabad 44000, Pakistan, July 2019 34. ul Hussen Khan, R.J., Javaid, N., Iqbal, S.: Blockchain based node recovery scheme for wireless sensor networks. MS thesis, COMSATS University Islamabad (CUI), Islamabad 44000, Pakistan, July 2019

An Intelligent Approach for Resource Management in SDN-VANETs Using Fuzzy Logic Ermioni Qafzezi1(B) , Kevin Bylykbashi1 , Evjola Spaho2 , and Leonard Barolli3 1

Graduate School of Engineering, Fukuoka Institute of Technology (FIT), 3-30-1 Wajiro-Higashi, Higashi-Ku, Fukuoka 811–0295, Japan [email protected], [email protected] 2 Department of Electronics and Telecommunication, Faculty of Information Technology, Polytechnic University of Tirana, Mother Teresa Square, No. 4, Tirana, Albania [email protected] 3 Department of Information and Communication Engineering, Fukuoka Institute of Technology (FIT), 3-30-1 Wajiro-Higashi, Higashi-Ku, Fukuoka 811-0295, Japan [email protected]

Abstract. In this paper, we propose an intelligent approach for resource management in Vehicular Ad Hoc Networks (VANETs) using Software Defined Networking (SDN) and Fuzzy Logic (FL) approaches. We introduce a layered Cloud-Fog-Edge computing architecture in SDN-VANETs which is coordinated by the SDN Controller (SDNC). A fuzzy based system implemented in SDNC is used to make decisions on the processing layer of the VANETs application data. The decision is made by prioritizing the application requirements and by considering the available connections. We demonstrate in simulation the feasibility of our proposed system to improve the management of the network resources.

1

Introduction

Vehicular Ad Hoc Networks (VANETs) consists of a network of vehicles in which vehicles are capable of communicating among themselves in order to deliver valuable information such as safety warnings and traffic information. Nowadays, every car is likely to be equipped with various forms of smart sensors, wireless communication modules, storage and computational resources. The sensors will gather information about the road and environment conditions and share it with neighboring vehicles and adjacent roadside units (RSU) via vehicle-to-vehicle (V2V) or vehicle-to-infrastructure (V2I) communication. Meanwhile, cloud computing has been attracting organizations and individual users to transpose their data and services from local to remote cloud servers. In VANETs, cloud computing is considered as an appropriate solution for handling massive amounts of traffic data generated by smart cameras and sensors, especially for long term analytics. c Springer Nature Switzerland AG 2020  L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 747–756, 2020. https://doi.org/10.1007/978-3-030-33506-9_68

748

E. Qafzezi et al.

Recently, fog computing extends cloud more near to the user. This new architecture analyzes data close to devices for minimizing latency, decision making in real time and offloading massive traffic flow from the core networks. With edge computing, resources and services of computing, networking, storage and control capabilities are distributed anywhere along the continuum from the cloud to things [1]. By leveraging the fog/edge computing technology, a significant amount of computing power will be distributed near/to the vehicles. Therefore, most of data will be processed and stored at the fog/edge, which can minimize latency and ensure better quality of service for connected vehicles [2]. Although the integration of cloud fog and edge computing in VANETs is very promising to offer many services by offering scalable access to storage, networking and computing resources, this network architecture lacks mechanisms needed for resources and connectivity management as it controls the network in a decentralized manner. The prospective solution to solve these problems is by augmenting Software Defined Networking (SDN) with this architecture. In Fig. 1 we illustrate the topology of this novel VANET architecture which is composed of cloud computing data centers, fog servers with SDN Controllers (SDNCs), RSU Controllers (RSUCs), RSUs, Base Stations and vehicles. We also illustrate the infrastructure-to-infrastructure (I2I), V2I, and V2V communication links.

V2V links V2I links I2I links RSUC

SDNC BS RSUC

RSU

RSU coverage V Tx/Rx range

Fig. 1. Topology architecture of SDN-VANETs using Cloud, Fog and Edge Computing.

In this work, we focus on the resource management in this new architecture and propose an intelligent approach based on fuzzy logic. We present a CloudFog-Edge SDN-VANETs layered architecture which is coordinated by a fuzzy

An Intelligent Approach for Resource Management in SDN-VANETs

749

system implemented in the SDN modules. The proposed system called FSRM (Fuzzy-based System for Resource Management) decides which resources will be used by a particular vehicle based on its relative speed with the neighboring vehicles, the time-sensitivity of the application data and the size of the data to be processed. We evaluate the performance of FSRM by computer simulations. The remainder of the paper is as follows. In Sect. 2, we present an overview of Cloud-Fog-Edge Computing, SDN and VANETs. In Sect. 3, we describe the proposed fuzzy-based system. In Sect. 4, we discuss the simulation results. Finally, conclusions and future work are given in Sect. 5.

2

Background Overview

In this section we briefly introduce Cloud-Fog-Edge Computing and SDN as enabling technologies for full deployment and management of VANET applications and services. Moreover, we provide a short description of the VANET features, applications, characteristics and communication issues. 2.1

Cloud-Fog-Edge Computing

The notion of cloud computing started from the realization of the fact that instead of investing in infrastructure, businesses may find it useful to rent the infrastructure and sometimes the needed software to run their applications. The cloud computing is becoming a promising and prevalent service to replace traditional local systems due to its scalable access to computing resources. The benefits of cloud computing have motivated many researchers to focus on shifting the conventional VANETs to Vehicular Cloud Computing (VCC) by merging VANET with cloud computing [3–5]. VCC is a very appealing technology due to its features and capabilities in supporting a series of novel, relevant, or sensitive applications. Additionally, VCCs are designed to initiate objectives that directly match everyday transportation needs, such as enabling computational services at low cost to authorized users, reducing traffic congestion, and implementing services to improve road safety [6]. However, there still are requirements such as low latency, high throughput, location awareness and mobility support that VCC can barely fulfill. Fog computing is proposed as a solution to overcome these issues between the vehicles and the conventional cloud [7]. Similar to cloud, fog computing provides data, compute, storage and application services at the proximity of the vehicular nodes. The vehicular nodes are the edge of this layered architecture. With edge computing in VANETs, which means vehicles having resources and services of computing, networking, storage and control capabilities, a significant amount of data can be processed at/through the vehicles, consequently offloading massive traffic flow from the core networks.

750

2.2

E. Qafzezi et al.

Software Defined Networking

The core concept of SDN is the decoupling between the control plane and data plane, which provides dedicated mechanisms for resources and connectivity management. The first one is used for network traffic control and the latter for data forwarding. This separation will simplify network management that is extremely complicated when the number of nodes dramatically increases in such dynamic environment like VANET [8]. In addition, it will increase flexibility and programmability in the network by simplifying the development and deployment of new protocols and by bringing awareness into the system, so that it can adapt to changing conditions and requirements, i.e. emergency services [9]. This awareness allows SDN-VANET to make better decisions based on the combined information from multiple sources, not just individual perception from each node. Significant benefits of the incorporation of SDN are the reduction of interference, improvement of channels and wireless resources usage, as well as the routing of data in multi-hop and multi-path scenarios [9]. 2.3

VANETs

VANETs are a special case of Mobile Adhoc Networks (MANETs) in which mobile nodes are vehicles. In VANETs nodes (vehicles) have high mobility and tend to follow organized routes instead of moving at random. Moreover, vehicles offer attractive features such as higher computational capability and localization through GPS. VANETs have huge potential to enable applications ranging from road safety, traffic optimization, infotainment, commercial to rural and disaster scenario connectivity. Among these, the road safety and traffic optimization are considered the most important ones as they have the goal to reduce the dramatically high number of accidents, guarantee road safety, make traffic management and create new forms of inter-vehicle communications in Intelligent Transportation Systems (ITS). Despite the attractive features, VANETs are characterized by very large and dynamic topologies, variable capacity wireless links, bandwidth and hard delay constrains, and by short contact durations which are caused by the high mobility, high speed and low density of vehicles. In addition, limited transmission ranges, physical obstacles and interferences, make these networks characterized by disruptive and intermittent connectivity. To make VANETs applications possible, it is necessary to design proper networking mechanisms that can overcome relevant problems that arise from vehicular environments.

3

Proposed Fuzzy-Based System

In this section, we present the layered CFE SDN-VANETs architecture which is coordinated by the global intelligence provided by SDNC. In this architecture,

An Intelligent Approach for Resource Management in SDN-VANETs

751

SDNC manages not only the computing and storage resources of Fog, but also those of Edge and Cloud. It also controls the network behavior by selecting best routes and improving usage of channels and wireless resources. An illustration of this layered architecture is given in Fig. 2.

Fig. 2. Layered architecture of CFE SDN-VANETs.

Our proposed FSRM does not require the SDN-VANET to operate in a particular mode, i.e. central control mode, distributed control mode and hybrid control mode [9]. It supports either the central control and distributed control mode and consequently also the hybrid control mode. A vehicle that needs storage and computing resources for a particular application can use that of neighboring vehicles, fog servers or cloud data centers based on the application requirements. FSRM is implemented in the SDNC and in the vehicles which are equipped with SDN modules. If a vehicle does not have a SDN module, it sends the information to SDNC which sends back its decision. The FSRM uses the beacon messages received from the adjacent vehicles to extract information such as their current position, velocity, direction, and based on the application requirements, it decides the appropriate layer to run and process the application data. In the following we describe in detail the Fuzzy Logic Controller (FLC) of our proposed system. 3.1

Description of FLC

In this work, we use fuzzy logic to implement the proposed system. Fuzzy sets and fuzzy logic have been developed to manage vagueness and uncertainty in a reasoning process of an intelligent system such as a knowledge based system, an expert system or a logic control system [10–15].

752

E. Qafzezi et al.

The structure of the proposed Fuzzy System for Resource Management is shown in Fig. 3. For the implementation of our system, we consider three input parameters: Vehicle Relative Speed with Neighboring Vehicles (VRSNV), Time Sensitivity (TS) and Data Size (DS) to determine the Layer Selection Decision (LSD) value.

Fig. 3. FSRM structure.

Table 1. Parameters and their term sets for FSRM. Parameters

Term sets

Vehicle Relative Speed with Neighboring Vehicles (VRSNV)

Slower (Sl), Same (Sa), Faster (Fa)

Time Sensitivity (TS)

Low (L), Middle (Mi), High (H)

Data Size (DS)

Small (S), Medium (M), Big (B)

Layer Selection Decision (LSD)

Decision Level 1 (DL1), DL2, DL3, DL4, DL5

These three input parameters are not correlated with each other, for this reason we use fuzzy sets. The input parameters are fuzzified using the membership functions showed in Fig. 4(a), (b) and (c). In Fig. 4(d) are shown the membership functions used for the output parameter. We use triangular and trapezoidal membership functions because they are suitable for real-time operation. The term sets for each linguistic parameter are shown in Table 1. We decided the number of term sets by carrying out many simulations. In Table 2, we show the Fuzzy Rule Base (FRB) of FSRM, which consists of 27 rules. The control rules have the form: IF “conditions” THEN “control action”. For instance, for Rule 1: “IF VRSNV is Sl, TS is L and DS is S, THEN LSD is DL4” or for Rule 16: “IF VRSNV is Sa, TS is H and DS is S, THEN LSD is DL1”.

An Intelligent Approach for Resource Management in SDN-VANETs Sl

1.0

Sa

1.0 L

Fa

Mi

H

0.5

µ

µ

0.5

753

0.0 -0.5 -0.4 -0.3 -0.2 -0.1

0

0.1

0.2

0.3

0.4

0.0

0.5

0

0.1

(a) Vehicle Relative Speed with Neighboring Vehicles M

B

1.0 DL1

0.5

0.0

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

(b) Time Sensitivity

µ

µ

1.0 S

0.2

DL3

DL4

DL5

0.5

0.0

1

DL2

0

(c) Data Size

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

(d) Layer Selection Decision

Fig. 4. Membership functions. VRSNV = -0.4 unit 1 0.9

TS=0.05 TS=0.50 TS=0.95

0.9 0.8

0.7

0.7

0.6

0.6

LSD [unit]

0.8

0.5 0.4

0.5 0.4

0.3

0.3

0.2

0.2

0.1

0.1

0

0 0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0

0.1

0.2

0.3

0.4

DS [unit]

0.5

0.6

0.7

DS [unit]

(a) VRSNV = -0.4 unit

(b) VRSNV = 0 unit VRSNV = 0.4 unit

1

TS=0.05 TS=0.50 TS=0.95

0.9 0.8 0.7 LSD [unit]

LSD [unit]

VRSNV = 0 unit 1

TS=0.05 TS=0.50 TS=0.95

0.6 0.5 0.4 0.3 0.2 0.1 0 0

0.1

0.2

0.3

0.4

0.5 0.6 DS [unit]

0.7

0.8

(c) VRSNV = 0.4 unit

Fig. 5. Simulation results.

0.9

1

0.8

0.9

1

754

E. Qafzezi et al. Table 2. The fuzzy rule base of FSRM. Rule VRSNV TS DS LSD

4

1

Sl

L

S

DL4

2

Sl

L

M

DL5

3

Sl

L

B

DL5

4

Sl

Mi S

DL3

5

Sl

Mi M

DL4

6

Sl

Mi B

DL5

7

Sl

H

S

DL2

8

Sl

H

M

DL3

9

Sl

H

B

DL3

10

Sa

L

S

DL2

11

Sa

L

M

DL3

12

Sa

L

B

DL4

13

Sa

Mi S

DL1

14

Sa

Mi M

DL2

15

Sa

Mi B

DL3

16

Sa

H

S

DL1

17

Sa

H

M

DL1

18

Sa

H

B

DL1

19

Fa

L

S

DL4

20

Fa

L

M

DL5

21

Fa

L

B

DL5

22

Fa

Mi S

DL3

23

Fa

Mi M

DL4

24

Fa

Mi B

DL5

25

Fa

H

S

DL2

26

Fa

H

M

DL3

27

Fa

H

B

DL3

Simulation Results

In this section, we present the simulation results for our proposed system. The simulation results are presented in Fig. 5. We consider the VRSNV as a constant parameter. We change the TS value from 0.05 to 0.95 units. In Fig. 5(a) we consider the VRSNV value −0.4 in which we simulate the scenario where the vehicle is moving slower than other vehicles in its vicinity. Because it is hard to establish and maintain a connection between the vehicle and its neighbors, from applications which need real-time computing, only the smallest data can be processed in the edge. However, even if the data size is

An Intelligent Approach for Resource Management in SDN-VANETs

755

increased, these applications can be processed at the fog servers which offer low latency as well. As for delay tolerant applications, we can see that cloud layer is the layer where their data will be processed. The scenario where the vehicle moves with the same speed as its neighbors is considered in Fig. 5(b). We can see that all real-time applications are processed at the edge, even when the size of data is big. The vehicle will have the same adjacent vehicles for a while as they move with the same speed. Being in the range of each other for a time, creates the possibility of initiating virtual machines to these adjacent vehicles which can be used to process also a number of non realtime applications. Processing these delay tolerant vehicles at the edge and not at the cloud servers can offload an excessive traffic flow from the core networks. In Fig. 5(c), we consider the vehicle moving faster than adjacent vehicles. The same as the case when the vehicle moves slower, it is impossible for the vehicle to use the resources of its neighbors as it quickly moves out of their communication range. Therefore, it will use the fog servers for the real-time applications, and the cloud data centers for the delay tolerant ones.

5

Conclusions

In this paper, we presented a layered Cloud-Fog-Edge architecture for SDNVANETs and proposed a fuzzy-based system for layer selection in terms of data processing. We took into consideration three parameters: VRSNV, TS and DS. We evaluated the performance of proposed system by computer simulations. From the simulations results, we conclude as follows. • Highly time-sensitive data are processed always in the edge and fog layer no matter what their size is. • If the vehicle has the same relative speed with its neighbors, in the edge layer will be processed not only the highly time-sensitive data but also a considerable amount of delay tolerant data as to offload a massive traffic flow from the core networks. • If the vehicle moves relatively much slower/faster than neighboring vehicles and the data to be processed are delay tolerant, vehicles will process these data always in cloud layer. In the future, we would like to make extensive simulations to evaluate the proposed system and compare the performance with other systems.

References 1. Hu, Y.C., Patel, M., Sabella, D., Sprecher, N., Young, V.: Mobile edge computing-a key technology towards 5G. ETSI White Paper, vol. 11, no. 11, pp. 1–16 (2015) 2. Yuan, Q., Zhou, H., Li, J., Liu, Z., Yang, F., Shen, X.S.: Toward efficient content delivery for automated driving services: an edge computing solution. IEEE Netw. 32(1), 80–86 (2018)

756

E. Qafzezi et al.

3. Olariu, S., Khalil, I., Abuelela, M.: Taking VANET to the clouds. Int. J. Perv. Comput. Commun. 7(1), 7–21 (2011) 4. Olariu, S., Hristov, T., Yan, G.: The next paradigm shift: from vehicular networks to vehicular clouds. Mob. Ad Hoc Networking: Cutting Edge Dir. 56(6), 645–700 (2013) 5. Hussain, R., Son, J., Eun, H., Kim, S., Oh, H.: Rethinking vehicular communications: merging vanet with cloud computing. In: 4th IEEE International Conference on Cloud Computing Technology and Science Proceedings, pp. 606–609, December 2012 6. Boukerche, A., Robson, E.: Vehicular cloud computing: architectures, applications, and mobility. Comput. Netw. 135, 171–189 (2018) 7. Stojmenovic, I., Wen, S., Huang, X., Luan, H.: An overview of fog computing and its security issues. Concurrency Comput.: Pract. Exp. 28(10), 2991–3005 (2016) 8. Truong, N.B., Lee, G.M., Ghamri-Doudane, Y.: Software defined networking-based vehicular adhoc network with fog computing. In: IFIP/IEEE International Symposium on Integrated Network Management (IM), pp. 1202–1207 (2015) 9. Ku, I., Lu, Y., Gerla, M., Gomes, R.L., Ongaro, F., Cerqueira, E.: Towards software-defined VANET: architecture and services. In: 2014 13th Annual Mediterranean Ad Hoc Networking Workshop (MED-HOC-NET), pp. 103–110 (2014) 10. Kandel, A.: Fuzzy Expert Systems. CRC Press Inc., Boca Raton (1992) 11. Zimmermann, H.-J.: Fuzzy control. In: Fuzzy Set Theory and Its Applications, pp. 203–240. Springer, Heidelberg (1996) 12. McNeill, F.M., Thro, E.: Fuzzy Logic: A Practical Approach. Academic Press Professional Inc., San Diego (1994) 13. Zadeh, L.A., Kacprzyk, J.: Fuzzy Logic for the Management of Uncertainty. Wiley, Hoboken (1992) 14. Klir, G.J., Folger, T.A.: Fuzzy Sets, Uncertainty, and Information. Prentice Hall, Upper Saddle River (1988) 15. Munakata, T., Jani, Y.: Fuzzy systems: an overview. Commun. ACM 37(3), 69–77 (1994)

Tutorial Educating Developer of Reinforcement Learning Agent Using IDEAL Takahiro Uchiya(&), Kodai Shimano, and Ichi Takumi Nagoya Institute of Technology, Gokiso-cho, Showa-ku, Nagoya, Aichi 466-8555, Japan {t-uchiya,takumi}@nitech.ac.jp, [email protected]

Abstract. In recent years, network service users have requested various services along with the rapid development of network services. Agent-oriented computing methods provide a flexible system that responds to various service requests. Moreover, agents can study the best action from results of past actions through “Learning” skill. Nevertheless, training agent system developers to have the development skill of “Learning” agent takes a long time and great effort. As a solution to this difficulty, we propose a tutorial for training agent system developers to have the development skill of “Learning” agent. Then we verify the effectiveness of the tutorial through experimentation.

1 Introduction In recent years, along with the rapid development of network services, users have increasingly come to request various services. Agent-oriented computing, a flexible system that responds to those various service requests, is a technique for generating agents that operate autonomously according to behavior knowledge. Moreover, agents can infer the best action based on results of earlier actions. These characteristics, collectively designated as “Learning” skill, enable agents to operate efficiently and flexibly. Training agent system developers to develop “Learning” skill takes a long time and great effort because development of systems requires several techniques. This paper presents our proposal of a tutorial for training agent system developers. It cultivates the ability of development skill. We describes our verification of the tutorial effectiveness.

2 Related Research Hara et al. [1] designed and implemented a training system for an agent framework (TAF) for the introductory education of software agent system developers. Actually, TAF has a function, called an agent monitor, which visualizes individual agents’ operation. Users of TAF can obtain skills of agent system development methods from real experience because TAF comprehends agent operations using an agent monitor. © Springer Nature Switzerland AG 2020 L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 757–762, 2020. https://doi.org/10.1007/978-3-030-33506-9_69

758

T. Uchiya et al.

For introductory education of agent system developers, TAF is effective, but it does not deal with “Learning” skill. Therefore, TAF is inefficient for training agent system developers to develop agents with “Learning” skill.

3 Ideal 3.1

Overview

For this study, we use the Interactive Design Environment for Agent system with a Learning mechanism (IDEAL) [2] as a base environment. The Interactive Design Environment for Agent system (IDEA) [3] was first proposed to support agent designers who design and develop the agent system. In fact, IDEAL is an extended IDEA to support the development of a learning agent. Furthermore, IDEAL has an agent monitor, just as TAF does. IDEAL users can obtain skills of agent system development method from real experience because it comprehends agent operation by an agent monitor. Moreover, IDEAL has an automatic learning function that supports IDEAL users to include “Learning” skill for developing agent systems. In order to support the design and implementation of agent system, IDEAL has following five mechanisms (Fig. 1).

Fig. 1. IDEAL: interactive design environment of agent system with learning mechanism

Tutorial Educating Developer of Reinforcement Learning Agent

• • • • • 3.2

759

Mechanism of agent search support Mechanism of agent programming support Mechanism of agent simulation support Mechanism of agent registration support Learning mechanism Learning Mechanism

This mechanism has following three functions. Automatic Learning Function The rule priority is automatically updated by using the Q-Learning scheme or Profit Sharing scheme. The automatic learning mechanism is newly introduced as a mechanism to update priority, and it operates in cooperation with the inference mechanism built into existing agent framework. The automatic learning mechanism is composed by the action selection engine and the learning engine. • Action selection engine This engine selects one action. The e-greedy method and the soft max method are implemented as an action selection technique, and agent designer chooses one method when they start the learning agent. • Learning engine By using the update formula, this engine updates the priority of the executed rule that the action selection engine selected. Preservation and Reference Function of Learning Data After the learning process proceeds to some degree, the rule name and the priority of each rule are preserved with the file of Comma Separated Value as learning data. This file is called a learning data file. When the same agent works again, agent’s operation begins after reading the learning data file and setting the priority of each rule. Therefore, it is possible to interrupt or restart the learning act. Automatic Drawing in Graph and Preservation Function To visually confirm the appearance that the agent’s operation advances efficiently, an automatic drawing is performed as the graph of learning process. We assume that one trial is from the initial state to the target state, and the number of the execution of the rule of every one trial is displayed automatically by the graph form. Moreover, to confirm the past learning process, the function to preserve the graph data as the DAT file is provided.

4 Proposed Method This study examines a proposed tutorial for training agent system developers to develop agents with “Learning” skill. The tutorial comprises a basic knowledge component and an exercise component. Figure 2 presents an overview of the proposed method.

760

T. Uchiya et al.

Fig. 2. Overview of the proposed method

Basic Knowledge Component This component lets users obtain basic knowledge about agent systems (Sect. 1) and reinforcement learning (Sect. 2). The reinforcement learning section presents an overview of reinforcement learning and about Q-learning as a representative algorithm of reinforcement learning. In addition, this section explains the learning rate and discount rate parameters that are used in reinforcement learning. Moreover, this section presents an overview of the action selection algorithm used in reinforcement learning systems such as an e-greedy algorithm or a soft max algorithm. Exercise Component This component lets users do exercises in the tutorial with IDEAL (Sect. 3). This exercise comprises a 2 times 2 block maze problem and a 20 times 20 block maze problem (Fig. 3).

Fig. 3. Maze problems

(1) 2 times 2 block maze problem: Users do the 2 times 2 maze block maze problem first after learning with the basic knowledge component. Users comprehend the basic structure of the agent system that has “Learning” skill without unnecessary information by doing a simple maze

Tutorial Educating Developer of Reinforcement Learning Agent

761

problem such as the 2 times 2 block maze problem. Thereby, users gain a key to a development system that has a complicated structure. (2) 20 times 20 block maze problem: Users do more complicated maze problems such as a 20 times 20 block maze problem after completing a 2 times 2 block maze problem. Thereby, users comprehend the means for obtaining a strategy for reaching the goal with the shortest route by “Learning” skill. (3) Acquire parameter setting methods: This section shows users an example of results of a running program that solves the 20 times 20 maze problem with changing of the learning rate and the discount rate. Users change these parameters and run the program based on the example. Users can comprehend properties and roles of these parameters with real experiences because of the observation run results. Consequently, users acquire a parametersetting method to observe the learning progression change caused by changing the learning rate and discount rate.

5 Experiment and Evaluation We conducted an evaluation experiment to verify the effectiveness of the tutorial proposed in this research. The subjects of this experience are six students of this college. In this experiment, they learned development of an agent system that has “Learning” skill using the tutorial and IDEAL. After learning, we asked them to report their five-grade evaluation of the tutorial using a questionnaire. A. Evaluation Points We evaluated the basic knowledge component with Sects. 1 and 2. Additionally, we evaluated the exercise component with Sect. 3 for smoothness of learning progress, contents, and the number of the exercises and learning parameter setting method using the tutorial. Moreover, we evaluated the usability of the tutorial UI. B. Results The experiment result is presented in Table 1. In Sects. 1 and 2 assessing the obtaining basic knowledge component, we obtained a middle score. In Sect. 3 of evaluation of Table 1. Questionnaire results

762

T. Uchiya et al.

the exercise component, the smoothness of learning progress was evaluated highly. Evaluations of the contents and the number of the exercises are, respectively, 3.7 and 3.5. C. Considerations We ascertained that technical terms used in explanations in obtaining basic knowledge component are too difficult for beginners to understand. This difficulty is a cause of middle evaluation of obtaining basic knowledge component. This obstacle can be overcome by the addition of explanations of technical terms and examples that clarify explanations for beginners. In Sect. 3, maze problems are appropriate exercises to learn agent systems that have “Learning” skill. However users cannot understand the progress of obtaining a strategy by the 2 times 2 maze problem that has no dead end. Moreover, the exercise component has only maze problems. Consequently, we found that a variety of exercise is needed for training agent developer.

6 Conclusion This study assessed the proposed tutorial for training agent system developers to develop the skill for efficient development of learning agent. Future studies will undertake reconsideration of the contents of exercises such as a multi-armed bandit problem, explanations that are easy to understand for beginners and the tutorial UI.

References 1. Hideki, H., Konno, S., Sugawara, K., Kinoshita, T.: Design of TAF for training agent-based framework. IEICE J84-D-I(8), 1129–1139 (2001). (in Japanese) 2. Uchiya, T., Itazuro, S., Takumi, I., Kinoshita, T.: IDEAL: interactive design environment for agent system with learning mechanism. In: Proceedings of the 12th IEEE ICCI*CC, pp. 153– 160 (2013) 3. Uchiya, T., Maemura, T., Li, X., Kinoshita, T.: Design and implementation of interactive design environment of agent system. In: IEA/AIE 2007. LNAI, vol. 4570, pp. 1088–1097 (2007)

The 5th International Workshop on Advanced Techniques and Algorithms for Security and Privacy (ATASP-2019)

Trusted Remote Patient Monitoring Using Blockchain-Based Smart Contracts Hafiza Syeda Zainab Kazmi1 , Faiza Nazeer2 , Sahrish Mubarak3 , Seemab Hameed2 , Aliza Basharat2 , and Nadeem Javaid1(B) 1 Department of Computer Science, COMSATS University, Islamabad 44000, Pakistan [email protected] 2 Department of Computer Science, Government College Women University, Sialkot 51141, Pakistan 3 Department of Computer Science, University of Lahore, Lahore 54590, Pakistan http://www.njavaid.com

Abstract. With an increase in the development of the Internet of Things (IoT), people have started using medical sensors for health monitoring purpose. The huge amount of health data generated by these sensors must be recorded and conveyed in a secure manner in order to take appropriate measures in critical conditions of patients. Additionally, privacy of the personal information of users must be preserved and the health records must be stored in a secure manner. Possession details of IoT devices must be stored electronically for eradication of counterfeited actions. The emerging blockchain is a distributed and transparent technology that provides a trusted and unalterable log of transactions. We have made a healthcare system using blockchain-based smart contracts which support enrollments of patients and doctors in a health center thereby increasing user participation in remote patient monitoring. Our system monitors the patients at distant places and generates alerts in case of emergency. We have used smart contracts for authorization of its devices and provided a legalized and secure way of using medical sensors. Using the blockchain technology, forgery and privacy hack in healthcare settings is reduced, thereby increasing the trust of people in remote monitoring. We have provided a graphical comparison of costs that verifies the successful deployment of contracts.

Keywords: Remote patient monitoring Blockchain · Smart contracts · Privacy

1

· Healthcare · IoT ·

Introduction

In recent years, fast growing popularity and extensive development in Internet of Things (IoT) can be witnessed. IoT is being used in smart cities, smart c Springer Nature Switzerland AG 2020  L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 765–776, 2020. https://doi.org/10.1007/978-3-030-33506-9_70

766

H. S. Z. Kazmi et al.

cars, wearables, e-business and healthcare. Considerable increase in the number of medical patients has been observed in various countries. IoT and wearable devices have enhanced the patient monitoring quality and a large number of patients can be monitored remotely. Remote Patient Monitoring (RPM) allows the monitoring of patients outside the health centre thereby increasing the patient care and decreasing the appointments time and cost. The core functionality of RPM is the monitoring of patients through wearable devices and transmission of health readings for diagnosis and treatment. Healthcare devices are divided into the following types [1]: • Stationary: Devices having physical location e.g., remote chemotherapy • Embedded: Implanted devices in a body e.g., deep brain stimulation • Wearable: Body-worn devices e.g., insulin pump As RPM is growing world-wide, concerns about secure transmission of Electronic Health Record (EHR) is increased. The sensitive health data can be accessed by unauthorized parties, so there is a motivation to secure the medical data transmission [2]. SCs are used to maintain immutable log of the transactions being made in RPM. Automatic health notifications using blockchcin increases trust of patients in wearing medical sensors or devices.

2

Motivation and Problem Statement

The authors of [3] used blockchain for security and privacy preservation of EHR. The authors used private and consortium blockchains for tackling the privacy leakage issue of sensitive health data. Private blockchain is used to store the Protected Health Information (PHI) whereas, consortium blockchain maintains the indexes of the health record. The authors of [4] have proposed a model for sharing medical information exploiting the advantages of blockchain. They have used digital signatures for protection of medical information against forgery and unauthorized access. Medical information contains record number, date, time, doctor ID, patient name, patient address, clinical health status, certificate ID and the digital signature of the record. The authors concluded that blockchain technology is reliable and provides traceability for medical data sharing. Researchers are reluctant to share their data due to protection concerns. A mechanism for stating terms of reuse of digital content is presented in [5]. The authors used blockchain and SCs for research data rights management. They maintained the agreements regarding digital content between the authors and users in order to verify the reuse of data. Externally Owned Accounts (EOA) for protection of data are used. The authors of [6] used data masking for data privacy and implemented IPFS for a secure EHR. The patient data used for data masking consists of name, age, ID, address and disease. However, they have used data masking instead of encryption. As the data volume increases, data masking time will also increase.

Trusted Remote Patient Monitoring

767

The use of IoT devices based is increasing day by day thereby enhancing the comfort and lifestyles. The authors of [7] have suggested the use of blockchain technology for securing the IoT devices from tampering and unauthorized access. However, they have used the hyperledger for implementation instead of using ethereum platform. Hyperledger uses no cyptocurrency and the transactions are confidential, not transparent. Moreover, they have not considered authorizing the enterprise who made the device (manufacturer) and device user’s in order to avoid the counterfeited actions. Remotely monitoring the patients helps in decreasing the cost thereby increasing the patient care outside the health centres. The increased number of IoT devices poses various privacy and security issues in a healthcare setting where confidentiality of patients’ information must be maintained. The authors of [2] have used blockchain-based SCs for preserving the health data received from medical sensors. However, they have not maintained the profiles of patients and medical professionals that are enrolled in a health centre because of the privacy leakage issue and people will be unwilling to provide personal data. A forged or fake device can be risky for a patient and the log of device authorization must be maintained without involving a third party. The problems we have identified include: personal information privacy concerns and risky devices of patients. 2.1

Contributions

We have written the following blockchain-based SCs for healthcare system: • Patients and Doctors Enrollments: The personal information of patients and doctors is sent using EOAs due to privacy concerns. • Patients Health Monitoring: The health data of patient is analyzed and timely alerts are relayed to the patients, doctors and helath centres. For patients’ health tracking, we have implemented the following modular SCs: 1. Blood Pressure Monitor 2. Temperature Monitor 3. Blood Oxygen Monitor 4. Brain Inflammation Monitor • Enterprise: This SC will be initialized by the enterprise whenever a device is made and it will facilitate the tracking and maintenance of the device log. • IoT Device Authorization: The log of device’s original and new custodian records or licences are maintained along with IoT device details in a decentralized manner using SCs eliminating the participation of third party. The paper is organized as follows: Detailed literature review is given in Sect. 3. Section 4 describes the proposed methodology. The experimental results and evaluation of the proposed work is given in Sect. 5. Finally, Sect. 6 concludes the paper.

768

3

H. S. Z. Kazmi et al.

Related Work

Authors of [3] tackled the privacy and security issues of EHR sharing using the immutable blockchain technology. Private and consortium blockchains are used for PHI sharing thereby increasing the privacy. The data is encrypted with keyword search. The proposed scheme achieved better data security and control over data access. Medical research is increasing with an increase of medical accidents [4]. Healthcare is facing many threats like forgery, unauthorized access and record tracking. The authors used provided verification of the proposed solution and concluded that the medical information is reliable and traceable using blockchain. Their data recovery function helps save the medical information against alteration. Electronic Medical Records (EMRs) provide a way to store a huge amount of sensitive medical data yet it is difficult to share the personal data among health centres due to privacy concerns [6]. Blockchain provides a secure, trustworthy and tamper resistent maintenance of health records thereby enhancing data sharing. It is not feasible to store a huge amount of data on blockchain so, an IPFS storage is used to store the confidential data after masking. The solution provided data privacy due to data masking and the blockchain resources are saved using IPFS. Medical records are an essential element of our lives and a considerable increase can be witnessed in the medical big data [1]. RPM is based on the wearable sensors which is helpful in providing healthcare services to patients. There are many risks involved in the trafer of confidential data that can be life threatening for patients. The authors have tackled the privacy leakage issue using blockchain and the data generated by IoT devices is made anonymous. The authors of [7] have maintained immutable logs of the IoT devices configurations. The history of modifications is stored and made available for the administrators. The model helps enterprises in tracking the device configuration changes using the decentralized, secure and trusted blockchain technology. To avoid security vulnerabilities in RPM, a trackable and unchangeable transactions log must be maintained. The authors in [2] have used private blockchain to store the health record transactions. The health reading taken using sensors are evaluated based on threshold values. Health alerts are generated and sent to the patients and hospitals. The emerging blockchain technology helped greatly in protecting the EHR of patients. Protection of medical data is an important factor to be catered for smoothly executing the medical activities. The two main data protection strategies can be used; one is access control and other is encryption. Access control mechanisms can be applied on locally stored data however it can be tampered on local storage. The encryption of data using key has a disadvantage of losing the key in case of patient’s death. The authors in [8] have used Sibling Intractable Function Families (SIFF) that provides a shared key. Hyperledger fabric is used for implementation and better efficiency is achieved. People will be unwilling to participate in a RPM system due to the privacy hack issue. The authors in [9] have proposed a conceptual model to manage the

Trusted Remote Patient Monitoring

769

health data using the distributed blockchain technology. In traditional setting, patients were not allowed to view or manage their own data. The proposed model guaranteed the data integrity by allowing patients to gather PHI. Blockchain is a peer-to-peer network that eliminates the third party. The authors of [10] have worked on IoT-enabled WSNs and achieved efficient routing. The authors of [11– 20] have implemented blockchain in various domains like IoTs, healthcare, smart grids and crowd sensing networks. They have concluded that blockchain is an effective solution for data trading, remote patient monitoring, energy trading, malicious node detection, electric vehicles and IoT service provisioning.

4

Proposed Solution

In our scenario, medical sensors are embodied on patient’s body and the health readings are sent to the specific SC via an master device i.e., a smart phone. The patient profiles are managed by health centre using SCs. Patient’s health status is analyzed according to the data being received. Health data is stored on a decentralized IPFS storage. Patients and doctors are able to register or enrol themselves using the master device. The health centre is in charge to authorize a patient for a doctor. Additionally, IoT device possession details are also recorded in SCs. Whenever an enterprise manufactures a device, SCs are made by both the enterprise and the patient who takes possession of the device. The main SCs named patient monitoring, enrolments, enterprise and IoT device authorization are discussed below in detail. 4.1

Enrollment

Health centre initializes enrolments SC on the blockchain for initiating the doctors and patients’ registrations. The enrolments contract consists of enrolment, modification and authorization functions. As shown in Fig. 1, health centre entity generates a public and a private key. Then, it posts the SCs address on the smart phone for patients and doctors to get registered easily in a secure way. The patient and doctors register in a health centre using their own EOAs via SCs address using addpatient() and adddoc() functions. The information taken from patient and doctors includes id, name, address and age and is made secure using EOA due to privacy concerns. Personal information is made private so that patient and medical assistants do not suffer from confidential information theft. In this way, patients and doctors will not be reluctant to enroll themselves due to the fear of privacy leakage and participation in the health system will be increased. The enrolments contract also allows the modification of information of both patients and doctors using modifypatient() and modifydoc() functions. Also, only a specific doctor is allowed to check the health status of a patient. The health centre maintains a list of doctors and can authorize and deauthorize a doctor from monitoring a patient’s health using authorize() and deauthorize() functions. Patients can view their information and authorized doctors by means of EOA. The enrolment, modification and patient authorization details

770

H. S. Z. Kazmi et al.

can be seen or retrieved by patientdetails(), doctordetails(), authorizedpatientdetails() and deauthorizedpatientdetails() functions in enrolment SC. 4.2

Patient Monitoring

For patients’ monitoring, data received from the smart device is handled by the main SC named as HealthContractCaller. Then, the main patient monitoring or HealthContractCaller contract creates a specific contract for every individual device it is getting data from. The main contract is like a container that organizes and creates links among all devices and relevant subcontracts for patient monitoring as shown in Fig. 1. Authorized doctors are allowed to access patients’ information and will be able to change thresholds for monitoring purpose. For instance, if the smart device receives blood pressure data from a patient’s body sensor, the data will be sent to HealthContractCaller and subsequently, BloodPressureMonitor() function will be called for patient monitoring. Minimum and maximum blood pressure values will be sent by the device to this function and an object is created by this function. Then, the individual sub contract Blood Pressure Monitor will pass these values to its analyze() function in order to evaluate the received data. Response upon the incoming data is generated by subcontracts instead of regulating it to the main contract. If the analyze() function returns any other value other than zero (0) or “OK”, then an alert (e.g. high/low blood pressure) is sent to the patient, doctor and health centre for treatment. The subcontracts we have used to monitor patient status include: Heart Rate Monitor, Glucose Monitor, Blood Pressure Monitor, Temperature Monitor, Blood Oxygen Monitor and Brain Inflammation Monitor. The motivation of modular contracts i.e., Heart Rate Monitor and Blood Sugar Level is taken from [2]. Whereas, we have proposed the use of other four subcontracts. The stated subcontracts analyse the real time heart rate, sugar level, fever, oxygen level in blood and brain inflammation measured using the body sensor of the patient based on specific threshold values. These modular contracts provide uncomplicated, trouble-free and simple maintenance. These modules will allow a customized structure where any subcontract for a specific device can be changed without changing the functionality of others. 4.3

Enterprise and Device Authorization

There are two types of SCs for device authorization, one is of the enterprise and other is of the device custodian. Here, IoT device refers to the wearable body sensor of the patient. The patient having that IoT device is referred as custodian of the device. Device must be registered and the custody must be recognised. The patient who buys a device must get registered and the device credentials must be legalised. In traditional systems, the contracts were made by involving a third party e.g., a bank. However, third parties are run by people that can be deceitful. We have established device credential management by removing the third party through SCs. The original custodian or the enterprise who manufactured the device make a SC named newdevice() after the production of device as shown

Trusted Remote Patient Monitoring

771

Smart Contracts

Enterprise

IoT Device Device InformaƟon Enrollments Data PaƟent Monitoring

PaƟent and Doctor Enrollments

Device Manufactured

Health Alert Device Details and Transfer

Body Sensors

Enrollments Data

Health Readings

Get SC Address Enroll using EOA

Health Centres

Health Readings

Get SC Address Master Device

Enroll using EOA

Doctors

Device Details and Transfer Publish SC Address

Enterprise

Health Alert

PaƟent

Fig. 1. Blockchain-based healthcare system

in Fig. 1. Whenever a patient buys that medical device, it must make a contract to get registered as the custodian of device. The device custodian also initiates a SCs and stores device information like device name and device description. In this way, device management will be done by the patient. The device custodian can set access conditions and transfer the device possession to other parties in a decentralized manner. The transfer of possession function changes the possession using the current (registered) and new custodian (to be registered) address and change the credentials of the device. The updated IoT device and custody details will also be sent to the health centre.

5

Results

The specifications of the system used are: [email protected] GHz, 8 GB RAM, 64 bit operating system and X64-based processor. We have used ethereum platform and solidity language for writing our SCs. The contracts are made operational on the private blockchain using ethereum protocol. We have used open source web browser environment Remix to test, debug and deploy our SCs. Metamask browser extension is used for connectivity to distributed web. Whenever an ethereum transaction takes place on the blockchain, two types of costs are associated with it; one is the transaction cost and the other is execution cost. The blockchain network has the potential to increase trust by reducing the transaction costs because of its decentralized nature with no third party involved.

772

H. S. Z. Kazmi et al.

• Transaction cost: It includes the cost of data being sent, operations being performed and the storage of contract. Transaction cost is determined by gasUsed×gasPrice where gasPrice is specified by the user and gasUsed refers to the total gas used for operations. • Execution cost: This cost refers to the storage of local and global variables as well as the processing power for calculations. Figure 2 shows the transaction and execution costs of all SCs. SCs are shown on the x-axis and their gas consumption on y-axis. Enrollment of patients and doctors shows the costs about 2692790 gas and 1986938 gas in transaction and execution of the contract. Monitoring and IoT device SCs cost less gas as compared to other contracts because the number of inputs fed to the monitoring contract are less than the inputs fields given in enrolments. More gas consumption in enrolments depicts a huge internal storage because the more data sent to the contract, the more cost it takes. Enterprise contract deployment took 1308577 as transaction and 950029 as execution cost. Less costs are recorded in the deployment of IoT device and monitoring contracts that shows that these contracts are logically less complex.

Fig. 2. SCs deployment

Figure 3 shows the subcontracts being called by the main monitoring contract on x-aix and the gas consumption on y-axis. The reason behind the deployment of six subcontracts is to check the amount of gas consumption for patients having more than 2 body sensors. These modular contracts cost less than the main contract because breaking the contract up into subcontracts decreases the cost during interaction. There is a slight difference in all contracts costs because the modular concept makes the computation simple and the data types used in all modular contract are almost same. However, the subcontract consuming the

Trusted Remote Patient Monitoring

773

Fig. 3. Patient monitoring modular SCs deployment

Fig. 4. Enrollments functions costs

least transaction and execution gas is due to the reason that instances are using uint type instead of expensive types. This saves the blockchain from expensive storage of variables in terms of gas for a transaction. Figure 4 displays the costs of transaction and execution made by all functions of the enrolment SC. Adding the doctors and patient information cost about 236109 and 235845, respectively which is relatively high as compared to the costs of transactions in other functions. The execution costs of adding doctor

774

H. S. Z. Kazmi et al.

Fig. 5. IoT device functions costs

and patient are recorded as 209333 and 209069, respectively. The reason behind high costs is that the larger transactions require a huge amount of fee. Transaction costs of authorization, deauthorization, doctor modification and patient modification are 45832, 15788, 54365 and 54541, respectively. Execution costs of these four functions are 21744, 6700, 27589 and 27765. These functions consume less gas because smaller transactions are simpler to validate and consequently, consume less gas. Figure 5 displays the gas consumption by IoT device contract where the device contract is created and the possession is transferred from one custodian to the other. When the possession is transferred, new owner will be allowed to change the description of the device. The details are updated costing 30021 and 25357 transaction and execution fee. The possession is successfully transferred consuming 27398 transaction gas whereas the failed transaction ended up consuming 23164 transaction cost. When the transfer is successful, the execution cost is recorded as 5710 and if the same owner registers for the device again, the transfer is failed consuming 484 execution costs.

6

Conclusion and Future Work

Remote medical care rapidly increasing with an increase in the use of IoT devices. For improved health services, only the transfer of health status and patients personal information is not enough rather an immutable record should be maintained. We have used blockchain for a secure and permanent log of health and personal data of patients. The unchangeable nature of blockchain enables us to keep track of unauthorized alterations to healthcare system. We have written SCs using ethereum and provided patients and medical professionals with a secure way of enrolling themselves in a health centre. The health centre maintains the

Trusted Remote Patient Monitoring

775

list of enrolled patients and authorizes them to medical assistants for treatment. The medical device custody is verified through SCs and enabled the device custodian to transfer the possession of device to other patients. The results show the costs of all smart contracts and verify the successful deployment of the contracts. For the future work, we aim to implement prescription review system in which patients will be able to give reviews on doctor’s prescription. This system will help the hospitals to get an idea of the reputation of the doctors. We will also give a secure solution for medical data storage because blockchain is not suitable for a huge amount of storage.

References 1. Dwivedi, A.D., Srivastava, G., Dhar, S., Singh, R.: A decentralized privacypreserving healthcare blockchain for IoT. Sensors 19(2), 326 (2019) 2. Griggs, K.N., Ossipova, O., Kohlios, C.P., Baccarini, A.N., Howson, E.A., Hayajneh, T.: Healthcare blockchain system using smart contracts for secure automated remote patient monitoring. J. Med. Syst. 42(7), 130 (2018) 3. Zhang, A., Lin, X.: Towards secure and privacy-preserving data sharing in e-health systems via consortium blockchain. J. Med. Syst. 42(8), 140 (2018) 4. Han, S.H., Kim, J.H., Song, W.S., Gim, G.Y.: An empirical analysis on medical information sharing model based on blockchain. Int. J. Adv. Comput. Res. 9(40), 20–27 (2019) 5. P˜ anescu, A.T., Manta, V.: Smart contracts for research data rights management over the ethereum blockchain network. Sci. Technol. Libr. 37(3), 235–245 (2018) 6. Wu, S., Du, J.: Electronic medical record security sharing model based on blockchain. In: Proceedings of the 3rd International Conference on Cryptography, Security and Privacy, pp. 13–17. ACM, January 2019 7. Koˇsˇta ´l, K., Helebrandt, P., Belluˇs, M., Ries, M., Kotuliak, I.: Management and monitoring of IoT devices using blockchain. Sensors 19(4), 856 (2019) 8. Tian, H., He, J., Ding, Y.: Medical data management on blockchain with privacy. J. Med. Syst. 43(2), 26 (2019) 9. Rahmadika, S., Rhee, K.H.: Blockchain technology for providing an architecture model of decentralized personal health information. Int. J. Eng. Bus. Manag. 10, 1847979018790589 (2018) 10. Awais, M., Javaid, N., Imran, M.: Energy efficient routing with void hole alleviation in underwater wireless sensor networks. MS thesis, COMSATS University Islamabad (CUI), Islamabad 44000, Pakistan (2019) 11. Mateen, A., Javaid, N., Iqbal, S.: Towards energy efficient routing in blockchain based underwater WSNs via recovering the void holes. MS thesis, COMSATS University Islamabad (CUI), Islamabad 44000, Pakistan (2019) 12. Naz, M., Javaid, N., Iqbal, S.: Research based data rights management using blockchain over ethereum network. MS thesis, COMSATS University Islamabad (CUI), Islamabad 44000, Pakistan (2019) 13. Javaid, A., Javaid, N., Imran, M.: Ensuring analyzing and monetization of data using data science and blockchain in loT devices. MS thesis, COMSATS University Islamabad (CUI), Islamabad 44000, Pakistan (2019) 14. Kazmi, H.S.Z., Javaid, N., Imran, M.: Towards energy efficiency and trustfulness in complex networks using data science techniques and blockchain. MS thesis, COMSATS University Islamabad (CUI), Islamabad 44000, Pakistan (2019)

776

H. S. Z. Kazmi et al.

15. Zahid, M., Javaid, N., Rasheed, M.B.: Balancing electricity demand and supply in smart grids using blockchain. MS thesis, COMSATS University Islamabad (CUI), Islamabad 44000, Pakistan (2019) 16. Noshad, Z., Javaid, N., Imran, M.: Analyzing and securing data using data science and blockchain in smart networks. MS thesis, COMSATS University Islamabad (CUI), Islamabad 44000, Pakistan (2019) 17. Ali, I., Javaid, N., Iqbal, S.: An incentive mechanism for secure service provisioning for lightweight clients based on blockchain. MS thesis, COMSATS University Islamabad (CUI), Islamabad 44000, Pakistan (2019) 18. Khan, R.J.H., Javaid, N., Iqbal, S.: Blockchain based node recovery scheme for wireless sensor networks. MS thesis, COMSATS University Islamabad (CUI), Islamabad 44000, Pakistan (2019) 19. Samuel, O., Javaid, N., Awais, M., Ahmed, Z., Imran, M., Guizani, M.: A blockchain model for fair data sharing in deregulated smart grids. In: IEEE Global Communications Conference (GLOBCOM) (2019) 20. Rehman, M., Javaid, N., Awais, M., Imran, M., Naseer, N.: Cloud based Secure Service Providing for IoTs using Blockchain, in IEEE Global Communications Conference (GLOBCOM) (2019)

A Survey of Malicious HID Devices Songyin Zhao and Xu An Wang(&) Engineering University of PAP, Xi’an, Shaanxi, China [email protected], [email protected]

Abstract. As an interface between human and computers, human interface device is supported by most computer systems. On behalf of user, HID devices can complete many operations including many sensitive operations with high authority. Exploiting this feature, attackers have designed and produced many malicious HID devices, imitating the user’s control. Meanwhile, most systems neglect to consider this security issue, posing great challenges to information security. In this regard, this paper reviews the development of malicious HID devices with analysis of technologies used. According to the technical characteristics, these devices are classified into three categories: pure HID devices, composite devices with HID interface, malicious devices with wireless communication capabilities. Furthermore, this paper discusses the challenges and opportunities of related research.

1 Introduction As a kind of computer external bus that supports hot plugging, high speed, easy expansion, etc., USB (Universal Serial Bus) has become a necessary interface for computers and been widely used in peripheral devices. Meanwhile, with the advancement of hacker technology, USB has also turned into an important media of attacking computers, stealing privacy and confidential data. The public and enterprises are more familiar with security issues about U disk, which have received extensive attention and research. However, a new technology based on the USB HID (Human Interface Device) protocol has been developed to attack computer systems in recent years. By impersonating the user to send action reports of a keyboard, a mouse, etc., the HID device maliciously created or modified can do many malicious operations with the user’s privileges. But, the commonly used security software in computers, such as anti-virus software and intrusion detection systems, is still not aware of the attack. The BadUSB attack [1], which has caused widespread concern in the word, use this technology extensively. Exploiting the feature that USB device could support multiple classes, a HID interface is expanded to simulate human operations through firmware upgrade, reverse engineering, etc. without affecting normal functions of the device. Under the cover of the original equipment, the concealment of the attack is very good. Since then, conferences and journals in the field of information security have published some new attack methods and protection schemes. The research review focusing on malicious HID devices is still relatively rare. Nissim et al. [2] summarized 29 known USB attack technologies, described the features and costs of various attack © Springer Nature Switzerland AG 2020 L. Barolli et al. (Eds.): BWCCA 2019, LNNS 97, pp. 777–786, 2020. https://doi.org/10.1007/978-3-030-33506-9_71

778

S. Zhao and X. A. Wang

methods, including some malicious HID devices. But they didn’t point out the nature of their HID attacks, classified them into different categories along with other types of attacks. In order to facilitate researchers to understand malicious HID equipment and carry out related research work, this paper presents the research status of malicious HID devices by investigating related literature and related projects in GitHub, and points out the challenges and future research directions. The contributions mainly includes three aspects: (1), In-depth study of the development history and current status of malicious HID devices, including hardware, malware, concealment, and so on; (2), Analyzing the technical principle of malicious HID devices, and classifying these devices into three categories: pure HID devices, composite devices with HID interface, malicious devices with wireless communication capabilities, according to the technical characteristics; (3) Introducing existing protecting solutions against malicious HID devices briefly, and future research directions of the attack.

2 Basic Theory of Malicious HID Devices 2.1

HID Fundamentals

A HID [3] device or human interface device is a class of computer device that interacts directly with people, such as keyboards, mice, joysticks, etc. Initially, keyboards and mice are connected to the computer through the PS/2 interface. However, PS/2 does not support plug-and-play, which means inconvenient. With the great success and popularity of the USB bus, most of the keyboards and mice use USB instead now. In this process, the term “HID” was created. With the popularity of HID devices and their necessity in human-computer interaction, most of the modern operating systems basically integrate the HID device drivers. Standard USB HID devices, such as keyboards and mice will be recognized and configured properly without rebooting, like other USB devices. The configured HID device will interact with the computer system as a representative of the user. Firstly, it will detect the user’s actions on the device, such as pressing or releasing buttons, moving the mouse, etc. Secondly, it will encode these actions differently and send the corresponding codes to the computer. Finally, the computer will parse and perform the corresponding operations. Conversely, the computer can also control the state of some devices, such the LED indicator of CAPSLOCK. However, HID devices do not necessarily have a human interface. As long as the devices meet the HID specifications, they will be recognized and configured automatically. Subsequently, it’s difficult for a computer to distinguish between a real keyboard and a HID device that claim it’s a keyboard. 2.2

Basic Theory

Under the control of the user, HID devices can perform many operations on the computer system, including many highly privileged operations, such as changing the

A Survey of Malicious HID Devices

779

system configuration, accessing files and network. For computer systems, the HID device is the representative of the user. Each action of the device is trusted as the user’s action. Then, through good programming, a HID device that is not controlled by humans can perform every human operation theoretically. With this feature, hackers have designed and produced many malicious HID devices. While current computer systems have difficulty in distinguishing between malicious HID devices and normal HID devices, the malicious HID device can impersonate the user and thus send operating instructions as a keyboard or mouse to attack the connected computer system, As shown in Fig. 1. Facts have shown that it’s feasible to redirect network, download malicious tools, steal data, open back doors. These attack commands are usually pre-programmed into the firmware of malicious HID device by the attacker or dynamically loaded using remote communication.

Fig. 1. A malicious HID device controlled by the Attacker is recognized as a normal keyboard used by the User. Thus, the Attacker can manipulate the Computer as the User, which should not be admitted.

The attack can only be implemented with the malicious HID device connecting to the target computer, which is a limitation of this technology. To create connections, the attacker needs to camouflage malicious HID devices through social engineering, such as hiding it into USB desk lamps, keyboards, mice, even USB cables. Implanting malicious code into the firmware - the method of BadUSB is also a camouflage way. Of course, if the malicious HID device is used by a malicious insider, it will be much easier to implement the attack and more powerful.

3 Taxonomy of Malicious HID Devices As shown in Figs. 2, we’ve classified malicious HID devices into three categories based on their technical characteristics: (A) pure HID devices (green), (B) composite devices (blue), (C) wireless devices (green and blue).

780

S. Zhao and X. A. Wang

Malicious HID Devices

A Pure HID Devices

B with other USB device classes

Composite Devices

C with wireless communication

Wireless Devices

with wireless communication

Fig. 2. Categories of malicious HID devices and the relationships between them.

Pure HID devices are earliest malicious HID devices. This kind of malicious HID devices only support the HID protocol, without any other USB functions or wireless communication capabilities. Subject to the speed of HID protocol and the performance of hardware, pure HID devices can only perform some simple attacks. As a consequence, researchers added other functions like USB mass storage and networks to malicious HID devices, to make uploading of malicious programs or data theft faster, to hijack the network, to camouflage the attack, etc. We classify this type of malicious HID devices into (B)composite devices. Typical representatives have the unintended channel [4], BadUSB. The term “composite device” comes from the USB protocol [5], but is a general designation for multi-function devices with malicious HID interface in this paper. Both (A) and (B) malicious HID devices are inconvenient to control. (C) wireless devices solve this problem with wireless communication. Using (C) devices, attackers can dynamically load different attack payloads or select a better attack time.

4 Description of Malicious HID Devices In this section, malicious HID devices will be introduced with more details. 4.1

Pure HID Devices

Crenshaw (Irongeek) and Darren are probably the earliest personal who started designing malicious HID devices. Inspired by a Phantom Keystroker shown at Shmoocon 2010 in February, Crenshaw came up with the idea and talked to Daren, who had been working on such a project. Crenshaw posted it based on Teensy earlier in his website on 03/23/2010 [6], while Darren invented more powerful USB Rubber Ducky [7] with nicer packaging a few days later. Since then, research on attacks using the USB HID interface came out one after the other. 4.1.1 Teensy The Teensy [8] is a small USB development board based on programmable microcontroller. It is open source and compatible with Arduino Software & Libraries, which already support USB HID. Meanwhile, it is low-cost, only $7 to $24 ($18 to $27 in

A Survey of Malicious HID Devices

781

2010) depending on the version you choose. These features make it perfect for implementing a malicious HID device. Based on Teensy, Crenshaw created the PHUKD (Programmable HID USB Keyboard) at first. Later, he made a malicious mouse in which a USB hub and a microSD USB adapter were integrated with the Teensy. In this design, the time to begin attacking is controlled by a photodiode. Then, his malicious HID devices won’t work until the surrounding environment is dark, which means most people won’t be around and notice the process. In July, Jason Pisani etc. presented a similar work in the report “USB-HID Hacker Interface Design” at Black Hat 2010 [9], But he also discussed the possibility of using HID on wireless communication. Inspired by Crenshaw, Dave Kennedy created the Teensy USB HID Attack Vector and integrated it into Social-Engineer Toolkit (S.E.T) [10]. With Teensy and S.E.T, a malicious HID firmware can be automatically made in several steps of setting, much easier than before. Nikhil Mittal extended S.E.T in Kautilya [11] with customizable payloads which can be used to perform various pre-exploitation and post-exploitation activities, such as keylogger, information gathering, hashdump, and so on. Not limited to common computers, Tzokatziou et al. [12] successfully altered the behavior of SCADA (Supervisory Control and Data Acquisition) systems with malicious HID devices based on teensy. As SCADA systems are used in critical national infrastructures like railways, the damage can be much larger. All the attacks listed above are done by acting as a keyboard. Some of them may support acting as a mouse, but didn’t play a role in the attack. In contrast to these, USBdriveby [13] written by Sam Kamakar for Teensy 3.1 ($20) uses the movement and click of a fake mouse to evade the Little Snitch Firewall of OS X. Evilduino is a malicious HID device based on Arduino pro micro [14]. Its processor is ATMEGA32U4, the same with Teensy 2.0. Its functions are similar to PHUKD or Rubber Ducky. Its advantage is the price, at 5–10$, compared with the latter. 4.1.2 Rubber Ducky For good sales, the performance of Rubber Ducky is more powerful than Teensy 2.0 series. The processor is AT32UC3B1256, an Atmel 60 MHz 32-bit micro-controller, which can execute code more rapidly. The storage can be expanded with the micro SD card reader on board, which can store payloads. The plastic shell makes it looks like a general U disk and thus easier to be connected to the target unobtrusively. The firmware is open source and abundant, with 64 payloads in the project on GitHub. It’s also well supported with a forum, in which you can share your achievement and get help. These features make the hardware expensive, about $50. 4.1.3 Digispark Digispark [15] is a series of USB development board based on ATtiny85, a less powerful but cheaper processor. It’s also Arduino compatible, which means many useful resources available. The footprint is very small and thus easier to hide. In the Digistump Forums, CBcracker designed the digispark-mini [16], the cost of which is about $1. Moreover, there are some tools you can use to convert payloads of Rubber Ducky to Arduino Sketches for DigiSpark, such as Duck2spark [17], Dckuino.js [18].

782

4.2

S. Zhao and X. A. Wang

Composite Devices

To meet the needs of different devices flexibly, USB supports a variety of data transfer types and allows a device to support multiple functionalities. To load appropriate drivers, operating system or a USB host will read the class code information from the device connected. Some of the classes are widely used and well supported by many operating systems, like MSC (Mass Storage Class) class, CDC (Communications Device Class) class etc. At present, official operating systems will automatically load drivers and make configurations properly for these commonly used classes without any verification. However, this convenience brings risks. Researchers have extended the malicious HID devices with functions like USB mass storage and network, to make uploading of malicious programs or data theft faster, to hijack the network, to camouflage the attack, etc. 4.2.1 Unintended USB Channel The concept of “unintended USB channel” was created by John Clark, which means information transfers of USB that are not monitored by security software now. In addition to the Control Transfer used by keyboard, the Isochronous Transfer used by speakers is added in the hardware trojan to effectively exfiltrate data. The first step of the attack process is uploading an executable file by typing VBScripts with the keyboard function. As the keyboard function play such an important role in the attack, the hardware trojan is a malicious device obviously. Furthermore, the team built a model to estimate the threat magnitudes of the unintended channels from a throughput perspective. In another paper [19], they discussed the concealment of the hardware trojan in more detail, and analyzed the threats of insiders using it. The results indicated that the malicious device used by insiders can evade the current security system and need to be considered as part of internal attacks. 4.2.2 BadUSB The uniqueness of BadUSB [1] is that it’s implemented by reprograming the original benign device unlike other malicious HID devices. Karsten Nohl etc. noticed that the firmware of some USB devices can be modified or updated. Then, they found the leaked firmware and flash tool on the net, reverse-engineered and injected malicious functionality into the firmware. Exploiting the feature that USB device supports multiple classes, the normal functions of the original device can be maintained to hide the process of attack. Another contribution of BadUSB is 4 new attacks. They showed a new method of DNS redirection by assigning the DNS in DHCP over spoofed USB-Ethernet, internet traffic interception by overriding default gateway over USB-Ethernet, a method of virtual machine break-out by reprograming the USB device connected to a malicious one with a spoofed keyboard, a USB style boot-sector virus implemented by showing a concealed operating system image and a keyboard when the computer booting. In the attack scenarios of BadUSB, the function of keyboard is still essential to control the target. Similarly, Maskiewicz et al. [20] reprogramed a mouse in the same year.

A Survey of Malicious HID Devices

783

4.2.3 Bash Bunny The Bash Bunny [21] is a penetration test platform sold by Hak5. The hardware consists of a Quad-core ARM Cortex A7 SoC (System on Chip), 512 MB of DDR3 Memory and a desktop-class 8 GB SLC NAND Disk in a normal USB drive size. On the powerful hardware, a full featured Linux is running. With BASH and many Linuxbased tools, advanced logic and conditions can be easily programmed in Bunny Script. By emulating combinations of USB devices, like Ethernet, serial, storage and keyboards, the Bash Bunny can finish many complex attacks, in which keystroke injection plays an important role. Meanwhile, it’s very easy to use as a commercial product at $99. Unlike the other malicious HID devices, the Bash Bunny payload text files do not need to be specially encoded and can be deployed by copying to a folder. The payloads can also be swapped easily by a selector switch on the board. Like the Rubber Ducky, there is a forum for the big community. 4.3

Wireless Devices

4.3.1 URFUKED Inspired by Crenshaw, Monta Elkins [22] made a wireless version of PHUKD called URFUKED (Universal RF USB Keyboard Emulation Device). By integrating a Sparkfun RF (radio frequency) Transmitter with Teensy, URFUKED supports choosing the timing and payloads of the attack. 4.3.2 TURNIPSCHOOL/COTTONMOUTH-1 TURNIPSCHOOL [23] is presented as an open version of COTTONMOUTH-I which is an ANT (Advanced Network Technology) product for NSA (the United States National Security Agency). Both TURNIPSCHOOL and COTTONMOUTH-I have the same structure of a USB hub, a microcontroller and a RF module in an USB cable connector. The differences are the performance and price. TURNIPSCHOOL uses a C1111 which contains an 8051 microcontroller and a RF transceiver, while COTTONMOUTH-I [24] uses a HOWLERMONKEY RF Transceiver and a TRINITY which contains an ARM9 microcontroller, FPGA, Flash and SDRAM memories in a miniaturized Multi-ChipModule. To attack the target, TURNIPSCHOOL can emulate as a keyboard under radio control. Once successful, it can exfiltrate data through the RF transceiver. In addition to TURNIPSCHOOL, COTTONMOUTH-I could eavesdrop communications between the target and peripheral devices, which means better concealment and more complex attacks could be achieved. Another particularity about COTTONMOUTH-I is that it has a mature over-the-air protocol to communicate with other COTTONMOUTH devices. Given the complexity of software and hardware, Cottonmouth-1 cost $1,015 K per 50 units, while a DIY TURNIPSCHOOL could be made at $20. The O.MG Cable [25] developed by Mike Grover is another work similar to TURNIPSCHOOL. It is smaller, packaged better, but more expensive, $200 for now. In future, a production release would be available at about $100 in the Hak5 shop. He also sells an original version without wireless communication capabilities as DemonSeed DIY build kits at $20. A few days after the original version shown on twitter, Olaf Tan, Kevin Mitnick et al. [26] demonstrated another malicious USB cable called USBHarpoon. The

784

S. Zhao and X. A. Wang

function of USBHarpoon is just the same as DemonSeed, as Kevin Mitnick was the man asked MG to build such a cable. 4.3.3 WHID Injector The hardware of WHID Injector [27] is a USB stick that utilizes an ESP-12S WiFi module with a serial connection to an Atmega32u4 microcontroller. The software framework used are ESPloitV2 and USaBuse. The WiFi module allows the device to host its own access point or connect to an existing network. This allows users to upload and pick between payloads or even type out “live payloads” without uploading a file. The device also supports upgrading the firmware over WiFi, deleting payloads, reformatting the file system, WiFi and basic configuration, and more. To exfiltrate data, a HID RAW device or USB-Serial device will be emulated to work with the application infiltrated into the target. For an elite version, a micro SD slot and microphone are added. The SD slot makes it possible to fully act as USB dead drop. The microphone can monitor the surrounding sounds to enhance the concealment of the attack activity. 4.3.4 P4wnP1 P4wnP1[28] is an attack platform like the Bash Bunny, which was declared a few days later than the initial P4wnP1 according to MaMe82. The hardware is a low-cost Raspberry Pi Zero (about $5) or Raspberry Pi Zero W (about $11), which is less powerful but much cheaper than the Bash Bunny. On Raspberry Pi Zero, a Sigle-core ARM11 SoC, 512 MB of memory and micro SD card slot are welded in 65  30 5 mm. Raspberry Pi Zero W adds WiFi and Bluetooth 4.1 module. The software is based on Raspbian Stretch Lite, a very basic Linux for the hardware. Compared to the Bash Bunny, P4wnP1 supports covert HID channel, Mouse emulation and interactive DuckyScript execution. The payloads could be triggered branched based on callbacks of some events, such as the signal from target to trigger the LED indicator of CAPSLOCK, the signal to report the network status of target. For Rapsberry Pi Zero W, P4wnP1 has a successor called P4wnP1 A.L.O.A. (A Little Offensive Appliance). P4wnP1 A.L.O.A. [29] could reconfigure the USB stack without reboot. The limited DuckyScript is also replaced with HIDScript to support more complicated attacks. Now, P4wnP1 A.L.O.A. have been integrated in KALI Linux. 4.3.5 MouseJack Unlike the malicious devices above, MouseJack [30] attacks the target by sending special radio signals to interfere with the USB dongle used by wireless keyboards or mice. The wireless keyboards or mice affected commonly use vulnerable protocols operating in the 2.4 GHz ISM (Industrial Scientific Medical) band. By modifying the firmware of a USB dongle works in the 2.4 GHz ISM band, it’s possible to sniff or inject the radio packets. Thus, the mouse movements and keystrokes to attack the target can be wirelessly sent to the USB dongle of the benign keyboards or mice.With a USB dongle called the Crazyradio PA, the Bastille Threat Research Team found some products of AmazonBasics, Dell, Gigabyte, HP, Lenovo, Logitech, Microsoft were affected, while other devices may be affected too.

A Survey of Malicious HID Devices

785

5 Discussion and Conclusion Through analyzing the malicious HID devices above, we can see the important role HID based attacks playing in the security risks associated with USB, which is the first step in many complex attacks. In the future, researchers will continue to improve malicious HID devices in three ways: (1) placing more powerful hardware in smaller package to achieve better concealment; (2) designing new attacks; (3) finding methods for bypassing inspection and protective measures. The continuous development makes it more and more difficult for a normal human or organization to identify and protect from malicious devices. At present, the research of the protection measures is mainly focused on the reinforcement of USB protocol, checking the USB enumeration attribute, keyboard(mouse) dynamics analysis, etc. But the versatility, reliability, cost is still not ideal for widespread application. More effort is needed, as it will be an ongoing battle between the protection measures and malicious HID devices. Acknowledgements. This work is supported by the National Key Research and Development Program of China Under Grants No. 2017YFB0802000, National Cryptography Development Fund of China Under Grants No. MMJJ20170112, the Natural Science Basic Research Plan in Shaanxi Province of china (Grant Nos. 2018JM6028), National Nature Science Foundation of China (Grant Nos. 61772550, 61572521, U1636114, 61402531), Engineering University of PAP’s Funding for Scientific Research Innovation Team (grant no. KYTD201805).

References 1. Nohl, K., Lell, J.: BadUSB-on accessories that turn evil. Black Hat USA 1, 9 (2014) 2. Nissim, N., Yahalom, R., Elovici, Y.: USB-based attacks. Comput. Secur. 70, 675–688 (2017) 3. Human_interface_device-wikipedia. https://en.wikipedia.org/wiki/Human_interface_device 4. Clark, J., Leblanc, S., Knight, S.: Compromise through USB-based hardware trojan horse device. Future Gener. Comput. Syst. 27(5), 555–563 (2011) 5. Universal Serial Bus Specification Reversion 2.0 (2000). http://www.usb.org 6. Programmable HID USB keystroke dongle: using the teensy as a pen testing device. http:// www.irongeek.com/i.php?page=security/programmable-hid-usb-keystroke-dongle 7. USB Rubber Ducky - Hak5. https://shop.hak5.org/products/usb-rubber-ducky-deluxe 8. Teensy USB Development Board – PJRC. https://www.pjrc.com/teensy/ 9. Pisani, J., Carugati, P., Rushing, R.: USB-HID hacker interface design. BlackHat Briefings, July 2010 10. Social-Engineer Toolkit v0.6.1 Teensy USB HID Attack Vector – TrustedSec. https://www. trustedsec.com/2010/08/social-engineer-toolkit-v0-6-1-teensy-usb-hid-attack-vector/ 11. Kautilya: Teensy beyond shells. https://media.blackhat.com/bh-ad-11/Mittal/bh-ad-11Mittal-Kautilya_Teensy_Beyond_Shell-WP.pdf 12. Tzokatziou, G., Maglaras, L.A., Janicke, H., He, Y.: Exploiting SCADA vulnerabilities using a human interface device. Int. J. Adv. Comput. Sci. Appl. 234–241 (2015) 13. Kamkar, S.: USBdriveby: exploiting USB in style. http://samy.pl/usbdriveby/ 14. EvilDuino – SlideShare. https://www.slideshare.net/Rashidferoz1/evilduino 15. Digispark USB Development Board – Digistump. http://digistump.com/products/1

786

S. Zhao and X. A. Wang

16. A smaller, cost-reduced digispark (