Internet Economics: Models, Mechanisms And Management 168108547X, 9781681085470

The internet represents a rapidly evolving set of technologies which is central to the development of a modern economy.

353 51 3MB

English Pages 222 Year 2017

Report DMCA / Copyright

DOWNLOAD FILE

Polecaj historie

Internet Economics: Models, Mechanisms And Management
 168108547X,  9781681085470

Table of contents :
CONTENTS......Page 6
FOREWORD......Page 10
PREFACE......Page 12
Organization......Page 13
ACKNOWLEDGEMENTS......Page 16
1.1. INTRODUCTION......Page 18
1.2. INDUSTRY STRUCTURE......Page 21
1.4. INTERNET PRICING......Page 22
1.6. PRICING CONGESTION......Page 24
1.7. COMPETITIVE BIDDING IN LARGE COMMUNICATION MARKETS......Page 26
1.8. INTERNET AND TELECOMMUNICATIONS REGULATION......Page 27
1.9. INTERNET COMMUNICATION TECHNOLOGIES: ATM AND B-ISDN......Page 28
1.10. BURSTINESS......Page 29
1.12. SIMPLE ECONOMICS OF INTERNET RESOURCE ALLOCATION AND PRICING......Page 30
1.13. BANDWIDTH-BUFFER TRADEOFF......Page 31
1.14. PRICING THE RESOURCES......Page 32
The Price Function......Page 33
1.15. PRICE ELASTICITIES......Page 34
1.16. RESOURCE ALLOCATION ALGORITHM......Page 35
REFERENCES......Page 36
2.1. INTRODUCTION......Page 38
2.2. THE INTERNET AS A REFLECTION OF THE ECONOMY......Page 41
Multiple Agent-Network Supplier Interaction......Page 42
2.3. INTERNET RESOURCES......Page 44
ATM and B-ISDN......Page 45
Traffic in B-ISDN......Page 46
Service Discipline......Page 47
Limiting Complexity......Page 48
2.5. MODELLING APPROACHES......Page 49
Optimal Allocation and QoS......Page 50
Scheduling and Pricing Mechanisms......Page 51
Network and Server Economies......Page 52
Allocation and Pricing Models......Page 54
Specific Problems of Economic Resource Allocation......Page 55
2.6. NETWORK ECONOMY......Page 56
Utility Parameters......Page 57
Packet Loss......Page 58
2.7. EQUILIBRIUM PRICE AND CONVERGENCE......Page 59
2.8. EXAMPLE OF TWO AGENTS AND ONE SUPPLIER......Page 62
CONCLUSION......Page 64
A. The Network Economy......Page 65
B. The Server Economy......Page 67
REFERENCES......Page 70
3.1. INTRODUCTION......Page 73
Network Routing......Page 75
Transaction Processing......Page 76
The Network Economy......Page 78
Price Equilibrium......Page 80
Proposition 3.2......Page 81
Agent Routing and Admission......Page 82
Admission Control......Page 83
Multiple Agent Network Supplier Interaction......Page 84
Proposition 3.4......Page 87
Transaction Routing......Page 88
APPENDIX: PROOF OF PARETO OPTIMAL ALLOCATIONS......Page 89
REFERENCES......Page 91
Internet Economics of Distributed Systems......Page 92
4.1. INTRODUCTION......Page 93
Pricing and Performance......Page 94
4.3. MECHANISM DESIGN APPROACHES......Page 95
Server Economy: Architecture for Interaction......Page 96
Performance Requirements......Page 97
4.4. ALLOCATION AND PRICING MODELS......Page 98
4.5. THE DATA MANAGEMENT ECONOMY......Page 99
Universal Access......Page 100
Quality-of-Service Characteristics......Page 101
4.7. DISCUSSION......Page 102
APPENDIX: SERVICE ARCHITECTURES FOR THE INTERNET ECONOMY......Page 104
Specialized Features in Centralized and Decentralized Models......Page 107
REFERENCES......Page 108
5.1. INTRODUCTION......Page 110
A Simple Mechanism Design......Page 112
Packet Loss......Page 113
Loss Probability Requirement: Utility Function......Page 114
Max and Average Delay Requirements......Page 115
Proposition 5.1......Page 116
Proposition 5.2......Page 117
Tail Probability Requirements: Utility Functions......Page 118
5.3. SERVICE ECONOMY: ARCHITECTURE FOR INTERACTION......Page 121
Performance Requirements......Page 122
CONCLUSION......Page 123
REFERENCES......Page 124
Network Platforms......Page 126
6.1. INTRODUCTION......Page 127
Assumptions and Implications......Page 130
Platform Utility......Page 131
Consumer Utility......Page 132
6.3. REVIEW OF PLATFORM ECONOMICS......Page 133
1. Amazon Web Services Platform......Page 136
3. Social Network Platform......Page 137
CONCLUSION......Page 138
REFERENCES......Page 139
7.1. INTRODUCTION......Page 142
Industrial Driving Forces in the IoT Context......Page 144
7.3. SPECIFIC TECHNOLOGIES AND USAGE......Page 146
7.4. BREADTH OF APPLICATION AREAS......Page 149
Healthcare......Page 150
7.5. SECURITY OF THINGS......Page 151
7.6. ECONOMIC BENEFITS......Page 152
7.7. FUTURE DIRECTIONS......Page 154
Smart Home Network......Page 156
REFERENCES......Page 158
8.1. INTRODUCTION......Page 161
8.2. BIG DATA DIMENSIONS......Page 166
Heterogeneity......Page 168
Quality......Page 169
Security and Privacy......Page 170
8.4. BIG DATA ANALYTICS AND SECURITY CHALLENGE......Page 171
8.5. BIG DATA ANALYTICS FOR INDUSTRY 4.0......Page 174
8.6. STATISTICAL AND COMPUTATIONAL NEEDS FOR BIG DATA......Page 177
CONCLUSION......Page 181
REFERENCES......Page 182
Internet, Innovation and Macroeconomics......Page 184
9.1. INTRODUCTION......Page 185
9.2. BASICS OF NETWORK ECONOMY......Page 186
9.3. ECONOMIC TRANSFORMATION......Page 187
9.4. ASSESSING THE TRANSFORMATION......Page 190
9.5. THE PRODUCTIVITY PARADOX......Page 193
9.6. GROWTH PROCESSES......Page 198
9.7. THE GLOBAL NETWORK ECONOMY......Page 200
Intangible Assets......Page 202
CONCLUSION......Page 203
REFERENCES......Page 204
ABBREVIATIONS......Page 207
GLOSSARY......Page 210
SUBJECT INDEX......Page 220

Citation preview

Internet Economics: Models, Mechanisms and Management Authored by Hans W. Gottinger STRATEC Munich, Germany

Internet Economics: Models, Mechanisms and Management Author: Hans W. Gottinger ISBN (Online): 978-1-68108-546-3 ISBN (Print): 978-1-68108-547-0 © 2017, Bentham eBooks imprint. Published by Bentham Science Publishers – Sharjah, UAE. All Rights Reserved. First published in 2017.

BENTHAM SCIENCE PUBLISHERS LTD.

End User License Agreement (for non-institutional, personal use) This is an agreement between you and Bentham Science Publishers Ltd. Please read this License Agreement carefully before using the ebook/echapter/ejournal (“Work”). Your use of the Work constitutes your agreement to the terms and conditions set forth in this License Agreement. If you do not agree to these terms and conditions then you should not use the Work. Bentham Science Publishers agrees to grant you a non-exclusive, non-transferable limited license to use the Work subject to and in accordance with the following terms and conditions. This License Agreement is for non-library, personal use only. For a library / institutional / multi user license in respect of the Work, please contact: [email protected]

Usage Rules: 1. All rights reserved: The Work is the subject of copyright and Bentham Science Publishers either owns the Work (and the copyright in it) or is licensed to distribute the Work. You shall not copy, reproduce, modify, remove, delete, augment, add to, publish, transmit, sell, resell, create derivative works from, or in any way exploit the Work or make the Work available for others to do any of the same, in any form or by any means, in whole or in part, in each case without the prior written permission of Bentham Science Publishers, unless stated otherwise in this License Agreement. 2. You may download a copy of the Work on one occasion to one personal computer (including tablet, laptop, desktop, or other such devices). You may make one back-up copy of the Work to avoid losing it. The following DRM (Digital Rights Management) policy may also be applicable to the Work at Bentham Science Publishers’ election, acting in its sole discretion: ●



25 ‘copy’ commands can be executed every 7 days in respect of the Work. The text selected for copying cannot extend to more than a single page. Each time a text ‘copy’ command is executed, irrespective of whether the text selection is made from within one page or from separate pages, it will be considered as a separate / individual ‘copy’ command. 25 pages only from the Work can be printed every 7 days.

3. The unauthorised use or distribution of copyrighted or other proprietary content is illegal and could subject you to liability for substantial money damages. You will be liable for any damage resulting from your misuse of the Work or any violation of this License Agreement, including any infringement by you of copyrights or proprietary rights.

Disclaimer: Bentham Science Publishers does not guarantee that the information in the Work is error-free, or warrant that it will meet your requirements or that access to the Work will be uninterrupted or error-free. The Work is provided "as is" without warranty of any kind, either express or implied or statutory, including, without limitation, implied warranties of merchantability and fitness for a particular purpose. The entire risk as to the results and performance of the Work is assumed by you. No responsibility is assumed by Bentham Science Publishers, its staff, editors and/or authors for any injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or from any use or operation of any methods, products instruction, advertisements or ideas contained in the Work.

Limitation of Liability: In no event will Bentham Science Publishers, its staff, editors and/or authors, be liable for any damages, including, without limitation, special, incidental and/or consequential damages and/or damages for lost data and/or profits arising out of (whether directly or indirectly) the use or inability to use the Work. The entire liability of Bentham Science Publishers shall be limited to the amount actually paid by you for the Work.

General: 1. Any dispute or claim arising out of or in connection with this License Agreement or the Work (including non-contractual disputes or claims) will be governed by and construed in accordance with the laws of the U.A.E. as applied in the Emirate of Dubai. Each party agrees that the courts of the Emirate of Dubai shall have exclusive jurisdiction to settle any dispute or claim arising out of or in connection with this License Agreement or the Work (including non-contractual disputes or claims). 2. Your rights under this License Agreement will automatically terminate without notice and without the need for a court order if at any point you breach any terms of this License Agreement. In no event will any delay or failure by Bentham Science Publishers in enforcing your compliance with this License Agreement constitute a waiver of any of its rights. 3. You acknowledge that you have read this License Agreement, and agree to be bound by its terms and conditions. To the extent that any other terms and conditions presented on any website of Bentham Science Publishers conflict with, or are inconsistent with, the terms and conditions set out in this License Agreement, you acknowledge that the terms and conditions set out in this License Agreement shall prevail. Bentham Science Publishers Ltd. Executive Suite Y - 2 PO Box 7917, Saif Zone Sharjah, U.A.E. Email: [email protected]

CONTENTS FOREWORD ........................................................................................................................................... i PREFACE ................................................................................................................................................ Organization ............................................................................................................................ CONFLICT OF INTEREST ......................................................................................................... ACKNOWLEDGEMENTS ...........................................................................................................

iii iv vii vii

CHAPTER 1 THE EVOLVING INTERNET: TECHNOLOGY, REGULATION AND PRICING 1.1. INTRODUCTION ................................................................................................................... 1.2. INDUSTRY STRUCTURE ..................................................................................................... 1.3. TELECOMMUNICATIONS AND THE INTERNET ......................................................... 1.4. INTERNET PRICING ............................................................................................................ 1.5. QUALITY OF SERVICE (QOS) ........................................................................................... 1.6. PRICING CONGESTION ...................................................................................................... 1.7. COMPETITIVE BIDDING IN LARGE COMMUNICATION MARKETS .................... 1.8. INTERNET AND TELECOMMUNICATIONS REGULATION ...................................... 1.9. INTERNET COMMUNICATION TECHNOLOGIES: ATM AND B-ISDN ................... 1.10. BURSTINESS ........................................................................................................................ 1.11. VIRTUAL CHANNEL .......................................................................................................... 1.12. SIMPLE ECONOMICS OF INTERNET RESOURCE ALLOCATION AND PRICING ......................................................................................................................................... 1.13. BANDWIDTH-BUFFER TRADEOFF ............................................................................... 1.14. PRICING THE RESOURCES ............................................................................................. The Price Function .................................................................................................................. Resource Balancing ................................................................................................................ 1.15. PRICE ELASTICITIES ........................................................................................................ 1.16. RESOURCE ALLOCATION ALGORITHM .................................................................... REFERENCES ...............................................................................................................................

1 1 4 5 5 7 7 9 10 11 12 13

CHAPTER 2 NETWORK ECONOMIES FOR THE INTERNET –CONCEPTUAL MODELS 2.1. INTRODUCTION ................................................................................................................... 2.2. THE INTERNET AS A REFLECTION OF THE ECONOMY ......................................... Agents and Network Suppliers ............................................................................................... Multiple Agent-Network Supplier Interaction ........................................................................ 2.3. INTERNET RESOURCES ..................................................................................................... ATM and B-ISDN ................................................................................................................... Traffic in B-ISDN ................................................................................................................... Congestion Control ................................................................................................................. Service Discipline ................................................................................................................... 2.4. THE RATIONALE OF ECONOMIC MODELS IN NETWORKING .............................. Decentralization ...................................................................................................................... Limiting Complexity ............................................................................................................... Pricing and Performance ......................................................................................................... Usage Accounting, Billing and Dimensioning ....................................................................... Administrative Domains ......................................................................................................... Scalability ............................................................................................................................... 2.5. MODELLING APPROACHES ............................................................................................. Optimal Allocation and QoS ................................................................................................... Scheduling and Pricing Mechanisms ...................................................................................... Network and Server Economies .............................................................................................. Allocation and Pricing Models ...............................................................................................

21 21 24 25 25 27 28 29 30 30 31 31 31 32 32 32 32 32 33 34 35 37

13 14 15 16 17 17 18 19

Specific Problems of Economic Resource Allocation ............................................................ 2.6. NETWORK ECONOMY ........................................................................................................ Utility Parameters ................................................................................................................... Packet Loss ............................................................................................................................. 2.7. EQUILIBRIUM PRICE AND CONVERGENCE ............................................................... Competitive Pricing Algorithm (CPA) ................................................................................... 2.8. EXAMPLE OF TWO AGENTS AND ONE SUPPLIER .................................................... CONCLUSION ............................................................................................................................... APPENDIX ...................................................................................................................................... A. The Network Economy ...................................................................................................... B. The Server Economy .......................................................................................................... REFERENCES ...............................................................................................................................

38 39 40 41 42 45 45 47 48 48 50 53

CHAPTER 3 NETWORK ECONOMIES FOR THE INTERNET: FURTHER DEVELOPMENTS .................................................................................................................................. 3.1. INTRODUCTION ................................................................................................................... 3.2. TWO EXAMPLES OF NETWORK OPERATIONS .......................................................... Network Routing ..................................................................................................................... Transaction Processing ........................................................................................................... 3.3. A MODEL OF NETWORK AND SERVER ECONOMY .................................................. The Network Economy ........................................................................................................... Price Equilibrium .................................................................................................................... Proposition 3.1 ........................................................................................................................ Proposition 3.2 ........................................................................................................................ Agent Routing and Admission ................................................................................................ Admission Control .................................................................................................................. The Server Economy ............................................................................................................... Agents and Network Suppliers ...................................................................................... Multiple Agent Network Supplier Interaction ............................................................... Proposition 3.3 .............................................................................................................. Proposition 3.4 .............................................................................................................. Transaction Routing ...................................................................................................... CONCLUSION ............................................................................................................................... APPENDIX: PROOF OF PARETO OPTIMAL ALLOCATIONS .......................................... REFERENCES ...............................................................................................................................

56 56 58 58 59 61 61 63 64 64 65 66 67 67 67 70 70 71 72 72 74

CHAPTER 4 INTERNET ECONOMICS OF DISTRIBUTED SYSTEMS .................................... 4.1. INTRODUCTION ................................................................................................................... 4.2. THE RATIONALE OF ECONOMIC MODELS IN NETWORKING .............................. Decentralization ...................................................................................................................... Pricing and Performance ......................................................................................................... Organizational Domains ......................................................................................................... Scalability ............................................................................................................................... 4.3. MECHANISM DESIGN APPROACHES ............................................................................. Network and Server Economies .............................................................................................. Server Economy: Architecture for Interaction ........................................................................ Access and Dissemination ...................................................................................................... Performance Requirements ..................................................................................................... Performance ............................................................................................................................ 4.4. ALLOCATION AND PRICING MODELS .......................................................................... Allocation Principles ............................................................................................................... 4.5. THE DATA MANAGEMENT ECONOMY .........................................................................

75 76 77 77 77 78 78 78 79 79 80 80 81 81 82 82

4.6. STRATEGIC INTERNET MANAGEMENT ISSUES ........................................................ Universal Access ..................................................................................................................... Congestion Problems .............................................................................................................. Quality-of-Service Characteristics .......................................................................................... Internet and Telecommunications Regulation ........................................................................ 4.7. DISCUSSION ........................................................................................................................... CONCLUSION ............................................................................................................................... APPENDIX: SERVICE ARCHITECTURES FOR THE INTERNET ECONOMY ............... 1. Centralized Read-Write (RW) Architecture ....................................................................... 2. Centralized Transfer-Access (TA) Architecture ................................................................. 3. Decentralized Index Based (IB) Architecture ..................................................................... Specialized Features in Centralized and Decentralized Models ............................................. Performance Model for RW, TA and IB Architectures .......................................................... Comparison of Response Time ............................................................................................... REFERENCES ...............................................................................................................................

83 83 84 84 85 85 87 87 90 90 90 90 91 91 91

CHAPTER 5 GENERALIZED QUALITY OF SERVICE ON QUEUEING NETWORKS FOR THE INTERNET ..................................................................................................................................... 5.1. INTRODUCTION ................................................................................................................... A Simple Mechanism Design ................................................................................................. 5.2. UTILITY AND QUEUEING PARAMETERS ..................................................................... Packet Loss ............................................................................................................................. Loss Probability Requirement: Utility Function ..................................................................... Loss Probability Constraints ................................................................................................... Max and Average Delay Requirements .................................................................................. Proposition 5.1 ........................................................................................................................ Proposition 5.2 ........................................................................................................................ Tail Probability Requirements: Utility Functions ................................................................... 5.3. SERVICE ECONOMY: ARCHITECTURE FOR INTERACTION ................................. Access and Dissemination ...................................................................................................... Performance Requirements ..................................................................................................... Performance ............................................................................................................................ CONCLUSION ............................................................................................................................... REFERENCES ...............................................................................................................................

93 93 95 96 96 97 98 98 99 100 101 104 105 105 106 106 107

CHAPTER 6 NETWORK PLATFORMS .......................................................................................... 6.1. INTRODUCTION ................................................................................................................... 6.2. TWO-SIDED PLATFORMS .................................................................................................. Assumptions and Implications ................................................................................................ Platform Utility ....................................................................................................................... Consumer Utility ..................................................................................................................... 6.3. REVIEW OF PLATFORM ECONOMICS .......................................................................... 6.4. PLATFORM OPERATIONS ................................................................................................. 1. Amazon Web Services Platform ......................................................................................... 2. IP Multimedia Systems (IMS) Platform ............................................................................. 3. Social Network Platform ..................................................................................................... CONCLUSION ............................................................................................................................... REFERENCES ...............................................................................................................................

109 110 113 113 114 115 116 119 119 120 120 121 122

CHAPTER 7 THE INTERNET OF THINGS AND THE INDUSTRIAL INTERNET ................. 125 7.1. INTRODUCTION ................................................................................................................... 125 7.2. BACKGROUND RESEARCH ON INTERNET OF THINGS ........................................... 127

Industrial Driving Forces in the IoT Context .......................................................................... 7.3. SPECIFIC TECHNOLOGIES AND USAGE ....................................................................... 7.4. BREADTH OF APPLICATION AREAS ............................................................................. Smart Infrastructure ................................................................................................................ Healthcare ............................................................................................................................... Supply Chains/Logistics ......................................................................................................... 7.5. SECURITY OF THINGS ........................................................................................................ Privacy .................................................................................................................................... 7.6. ECONOMIC BENEFITS ........................................................................................................ 7.7. FUTURE DIRECTIONS ......................................................................................................... SUMMARY AND CONCLUSION ............................................................................................... APPENDIX: SMART HOME SKELETON DESIGN – AN ILLUSTRATIVE EXAMPLE Smart Home Network ............................................................................................................. REFERENCES ...............................................................................................................................

127 129 132 133 133 134 134 135 135 137 139 139 139 141

CHAPTER 8 THE INTERNET, DATA ANALYTICS AND BIG DATA ....................................... 8.1. INTRODUCTION ................................................................................................................... 8.2. BIG DATA DIMENSIONS ..................................................................................................... 8.3. THE 3 VS: ISSUES AND CHALLENGES ........................................................................... Heterogeneity .......................................................................................................................... Scale ........................................................................................................................................ Timeliness ............................................................................................................................... Complexity .............................................................................................................................. Quality ..................................................................................................................................... Security and Privacy ............................................................................................................... 8.4. BIG DATA ANALYTICS AND SECURITY CHALLENGE ............................................. 8.5. BIG DATA ANALYTICS FOR INDUSTRY 4.0 .................................................................. 8.6. STATISTICAL AND COMPUTATIONAL NEEDS FOR BIG DATA ............................. CONCLUSION ............................................................................................................................... REFERENCES ...............................................................................................................................

144 144 149 151 151 152 152 152 152 153 154 157 160 164 165

CHAPTER 9 INTERNET, INNOVATION AND MACROECONOMICS ..................................... 9.1. INTRODUCTION ................................................................................................................... 9.2. BASICS OF NETWORK ECONOMY .................................................................................. 9.3. ECONOMIC TRANSFORMATION ..................................................................................... 9.4. ASSESSING THE TRANSFORMATION ............................................................................ 9.5. THE PRODUCTIVITY PARADOX ...................................................................................... 9.6. GROWTH PROCESSES ........................................................................................................ 9.7. THE GLOBAL NETWORK ECONOMY ............................................................................ Intangible Assets ..................................................................................................................... Information Markets ............................................................................................................... CONCLUSION ............................................................................................................................... REFERENCES ...............................................................................................................................

167 168 169 170 173 176 181 183 185 186 186 187

ABBREVIATIONS .................................................................................................................................. 190 GLOSSARY ............................................................................................................................................. 193 SUBJECT INDEX .................................................................................................................................... 203

i

FOREWORD A front-page leader in The Economist (28 January, 2017) showcases how in an era of protectionism, global companies are now in retreat. On the other hand, the world has become even more globalized and inter-connected, breeding a new generation of global enterprises and nimble, innovative small-and-medium-sized businesses, thanks to the pervasive internet and its integrated technologies and platforms. China alone now boasts of some 700 million netizens, more than double the entire population of the United States. The number of internet users is similarly rising across the globe, including Latin America and Africa, thanks to the ubiquitous smart phone and supporting networks. Nothing illustrates the power of internet-economics better than President Trump’s apparent agreement that Jack Ma’s Alibaba internet-driven global business empire could help create a million American jobs by selling US goods and services to China and the rest of Asia, Trump’s China-bashing and protectionist rhetoric notwithstanding. According to Klaus Schwab, Founder and Executive Chairman of the World Economic Forum, instead of simple digitization (the Third Industrial Revolution), innovative combination of technologies (the Fourth Industrial Revolution) is upending business models, labor markets, socio-political matrix, and is reshaping our economic, social, cultural, and human environments. A key trend is the development of technology-enabled platforms that combine both demand and supply to disrupt existing industry structures, such as the “sharing” or “on demand” internet economy. The dynamics are multiplied by technology breakthroughs in artificial intelligence, robotics, the Internet of Things, autonomous vehicles, 3-D printing, nanotechnology, biotechnology, materials science, energy storage, and quantum computing. In this topsy-turvy world of the 21st century, understanding internet economics, its complexity and operations, as well as its models, mechanisms and management, would be invaluable to practicing network designers and engineers as well as to industry managers and academic researchers. For them and broader policy formulators, Dr Hans W. Gottinger’s book is a treasure-trove of timely, meticulous research. Readers will find scholarly treatments of a variety of interrelated topics, including internet supply and demand, supply chain analytics, distributed market agencies, Quality of Service (QoS), network modelling and security, aggregated outcomes, output analysis covering financial services, healthcare and legal services, total factor productivity, Big Data, cloud computing, and much more.

ii

I have no hesitation in commending Dr Gottinger’s book for all who are interested in exploring the complex, fast-moving, and integrated world of internet economics.

Andrew K P Leung Andrew Leung International Consultants, The e-Centre, European Centre for e-Commerce and Internet Law, Berkshire Publishing Group, Massachusetts, USA

iii

PREFACE From its very early beginning as a project at ARPA (now US. DARPA), the Internet started out as a small scale communication paradigm for a well-established and regulated organization, the military, that over decades, expanded to virtually all dimensions of human life and activity, and seemingly transmitted borderless throughout the world. In substance and impact, it was more than evolutionary telecommunications tools and devices. As economics described the various interactions of human endeavor targeted at social and commercial activities through information and communication, the Internet as a communication system would facilitate and expand economic interaction throughout an ever evolving network economy. This text makes an attempt to provide an integrated view of internet economics which ranges from modeling internet structures akin to economy-wide modelling, from mechanism design to pricing, cost and service models in view of the changing technology structure of the Internet. From this, a major outcome is to merge networks and economies on a conceptual and practical level and highlight specific application areas such as service provisioning, cloud computing, commerce and business security, network externalities, social media and more recent enlargements to the Internet of Things (IoT), the Industrial Internet, Data Analytics and Big Data. Contentwise we attempt to pursue a middle-of-the-road path in that we don’t shy away from indicating technical issues but embed them in a larger context of relevance to economics and the economy system. We look at the design of the Internet as a reflection of economic mechanisms in universal communication. Resource allocation would involve pricing and cost structures and face network access and regulation. Most approaches of resource allocation mechanism have used a performance model of the resource where the very concept of the resource is defined in terms of measurable qualities of the service such as utilization, throughput, response time (delay) and the like. Optimization of resource allocation is defined in terms of these measurable qualities. The operations research (OR) approach is to design a system which takes into account the diverse quality-of-service (QoS) requirements of users and therefore, uses multi-objective or multi-criteria (utilities)optimization techniques to characterize and compute optimum allocations. Economic (mechanism design) modelling of computer and communication resource sharing uses a uniform paradigm described by two level modelling: QoS requirements as inputs into a performance model that is subject to economic optimization. In the process one transforms QoS requirements of users into a performance (example: queueing service model). This model establishes quantifiable parameterization of resource allocation. For example, average delay QoS requirement, when based on queueing models, is a function of resources, bandwidth and buffer, and user traffic demands. These parameters are then used to establish an economic optimization model. We consider foremost decentralized models of network and server economies , where we show efficient QoS provisioning and Pareto allocation of resources (network and server resources) among agents and suppliers, which are either network routes or servers (content providers). It is shown how prices for resources are set at the suppliers based on the QoS demands from the agents. A closed view of Internet -based distributed computer systems reveals the complexity of the organization and management of the resources and services they provide. The complexity arises from the system size (e.g. number of systems, number of users) and heterogeneity in applications (e.g. online transaction processing, e-commerce, multimedia, decision support, intelligent information search) and resources (CPU, memory, I/O bandwidth, network

iv

bandwidth and buffers, etc.) In a large distributed system, the set of systems, users and applications is continuously changing. We address some of the management issues of providing Quality of Service (QoS), pricing, and efficient allocation of resources (computational resources) in networks and systems facilitated through economic mechanism design. The complexity increases on the expansion of three dimensions: (i) on a web scale through multi-sided platform interactions, (ii) through the almost limitless expansion of the Internet of Things (IoT) and the Industrial Internet, and (iii) the evolvement of Big Data through volume, velocity and variety and their private, business and public use. The book serves as an introductory, evolutionary account of Internet Economics, from micro to macro. It could be of interest to practicing network designers and engineers as well as to industry and academic researchers. Industrial economists in telecommunications, media and IT services are interested in Internet scale, scope and economics who wish to explore the subject matter further might benefit. The text is also suitable for graduation level courses in operations research, management science, and economics and public affairs programs. Several chapters could serve as supplements to industrial economics, intensive courses on specific topics, as well as complementary texts for mainstream economics or management courses.

Organization In what follows, we briefly highlight the contents of the individual chapters. Chapter 1 traces the evolution of the Internet in its historic setting, the emergence of Internet institutions, technologies, architectures and services. From this, an Internet industry was developed with major dominance in the telecommunications world. Some key economic issues involve Internet pricing, Quality of Service (QoS) provisioning, congestion management, Internet regulation and the backbone network technologies such as Asynchronous Transfer Mode (ATM) in view of traffic management and congestion control. The relation between service discipline and the varieties of the bandwidth-buffer tradeoff are discussed. In Chapter 2, we consider the Internet as a market design for information with major actors supplying and demanding information, and supply and demand tending toward equilibrium. Concepts of information flows and delivery provisions would reflect game-theoretic models of trade and exchange economies where the mechanism (design) of information flows would adopt the behavior and outcome of mathematical queueing systems. Of particular interest are large scale systems of the Internet type that act and perform as decentralized distributed systems of multi-level control. A prototype economic optimization model has been put forward for a network economy with variation on traffic classes, utility parameters, queueing types and equilibrium approaches. Chapter 3 considers specific examples of network economies with network routing and transaction processing. Any of these examples capture a structural model of the network economy with Pareto optimality and price equilibrium for agents competing for resources from emerging suppliers. A routing algorithm establishes the dynamic nature of of session arrival and departure. Chapter 4 focusses on economic management of web services in distributed multi-media systems. We dig deeper into mechanism design approaches tracing them back to classical

v

economic mechanisms of market designs and information economics which I refer to as Hayek-Hurwicz mechanism design – due to Austrian-British economist F.A. Hayek and American economist L. Hurwicz. The basic idea of market agents as computationally efficient human agents induced by incentive compatibility and selfishness has been rediscovered and reapplied in rigorous methodological form by computer scientists merging algorithmic game theory, computability and network complexity. Distributed algorithmic mechanism design (DAMD) for internet resource allocation in distributed systems is akin to an equilibrium converging market based economy where selfish agents maximize utility and firms seek to maximize profits and the state keeps an economic order providing basic public goods and public safety. A distributed algorithmic mechanism design thus consists of three components: a feasible strategy space at the network nodes for each agent (or autonomous system), an aggregated outcome function computed by the mechanism and a set of multi-agent prescribed strategies induced by the mechanism. A distributed algorithmic mechanism design being computationally efficient in a large decentralized internet economy is a powerful paradigm to substantiate claims by Hayek (1945) that an industrialized economy based on market principles has an overall better output and growth performance (static and dynamic) than socialist type economies of a similar nature and scale. That puts historically the socialist planning debate in a new light which ironically, by some proposals, has been conducted on the basis of computational feasibility and superiority. Conclusion: Best economic coordination through markets producing maximal social welfare is supported by computational efficiency in computer science. Applications relate to a data management economy. In Chapter 5, we broaden the criteria for QoS performance and guaranteed service through reputation systems hinging on trust and belief, and as the Internet matures in promptness, reliability, accessibility and foremost security for a certain targeted QoS level which we termed as ‘Generalized QoS level’. GQoS should include emphasis on security, reputation and trust. Chapter 6 opens up a new property of Internet enabled communication and computation that is intrinsincally linked to the interactive social and commercial use of websites through the World Wide Web (WWW), i.e. two-sided or multi-sided platforms. We provide a model of platform operations that involves a sequential decision process very much alike a dynamic programming algorithm (DPA) for selecting functionalities of the platform. This serves as a simple approximation procedure for building the optimal design of a platform. The platform business in a vertically integrated supply chain as well as toward product development is a good example of facilitating ‘increasing returns mechanisms’ (IRM), as one can follow the evolution of the Amazon platform in a commercial context but also on the growth of social media like Facebook and LinkedIn. Such a business growth would not have been facilitated without a dedicated, universal, easily accessible and low cost network economy provided by the Internet. Chapter 7 shows how the Internet could lead to a radical expansion to the Internet of Things (IoT) with everything being connected up to a scale of aggregate capacity and complexity. It could evolve into a network where not only each node would be a computational device but by itself, it would also be an intelligent computing device that could replace human supervision and control through Artificial Intelligence (AI). The Internet of Things (IoT) as being embedded in the ‘Industrial Internet’ is a new paradigm shift that comprehensively affects computers and networking technology. This is also recently referred to the buzzword ‘Industry 4.0’. This technology is going to increase the utilization followed by Bandwidth of

vi

the Internet. More and more (intelligent) devices in this network are connected to the Internet through various combinations of sensor networks. Chapter 8 focusses on huge online data generation through an expanded Internet (IoT and the Industrial Internet) and using the flux of information for monitoring, predicting, controlling and decision making. During the time of applying methods of statistical inference and statistical decisions some 70 years ago, information derived from data collection was considered costly. Models were built where information was linked to payoff relevance of a decision making criterion (utility or payoff function), therefore statistical information was handled to satisfy these criteria. Now as masses of online data through IoT are produced at relatively low costs, all these data could be quickly aggregated for business or government decisions. Statisticians have coined a term, ‘value of perfect information’, which was set up to integrate data points, collection and analysis through statistical inferential models i.e.,exploratory data analysis (EDA) or through statistical decision models . For example, achieving this goal is quite challenging to gather all the data for perfect information. What really is subsumed under ‘Big Data’ (BD) qualifies under a few main characteristics: (i) BD is primarily network generated on a large scale by volume, variety and velocity and comprises large amounts of information at the enterprise and public level, in the categories of terabytes (1012 bytes), petabytes (1015) and beyond of online data. (ii) BD consists of a variety and diversity of data types and formats, many of them dynamic, unstructured or semistructured and are hard to handle by conventional statistical methods. (iii) BD is generated by disparate sources as in interactive application through IoT from wireless devices, sensors, streaming communication generated by machine-to-machine interactions. The traditional way of formatting information from transactional systems to make them available for ‘statistical processing’ does not work in a situation where data are in huge volumes from diverse sources, and where even the formats could be changed. While Internet economic effects in micro structures, that is, on enterprise and industry levels, have been our prevailing concern in previous chapters, Chapter 9 addresses network effects on a macro scale involving productivity, growth, and the business cycle. The network effect is strongly facilitated by computerization and information technologies (ITs). Ubiquitous computerization pervades many sectors of the economy, and communication by network technologies such as the Internet (the network is the computer) is a strong catalyst. Eventually, through this synergy, most sectors of the economy will be impacted by network effects. Thus, networking and computerization have far-reaching impacts on the pace and path of the economy but they could also make the economy more vulnerable to economic shocks and security breaches. We address three important issues: Networks and productivity, endogeneous growth and increasing returns. Examples of some productivity- enhancing activities on the enterprise level which aggregates to impacting total factor productivity on a macro scale are (i) computerization of ordering along the supply chain, (ii) Internet-based procurement systems , and (iii) computer-based supply chain integration. There are many other examples of pervasive internet applications that trickle down to aggregate increases in productivity. The relationship between technology and productivity used for the United States, on the economy or sector level, found little evidence of a relationship in the 1980s. Capital investment between 1977 and 1989 rose several hundred per cent but was barely reflected in a rise in output per worker. There was this famous saying by Nobel prize-winning economist Robert Solow (1987): “You can see the computer age everywhere except in productivity statistics”.

vii

In the 1990s, such a positive relationship on the firm level was established empirically. On a short-term basis: one-year difference in IT investments vs. a one-year difference in firm productivity should be benchmarked by benefits equal to costs. However, benefits are supposed to rise by a factor of 2–8 in forecasting future benefits (in productivity growth). Economic techniques focus on the relatively observable aspects of investment, such as price and quantity of computer hardware in the economy, but neglect intangible investments in developing complementary new products, services, markets, and business processes. The recent World Bank World Development Report (2016,) clearly states for the macro economy as three key factors of growth: (i) Inclusion – through international trade, (ii) efficiency – through capital utilization and (iii) innovation through competition. Current statistics typically treat the accumulation of intangible capital assets, new production systems, and new skills as expenses rather than as investments. This leads to lower levels of measured outputs in the periods of net capital accumulation. Output statistics miss many of the gains of IT brought to consumers such as variety, speed, and convenience. For instance, US productivity figures do not take account quality changes, in particular, in services industries: (a) financial services sectors, (b) health care, and (c) legal services (online information). As a time-dimensional cross-sectional review and synthesis of Internet economic and technological issues, this text naturally covers some overlapping areas , in particular, in Chapters 1 to 5. It also serves the purpose to make each chapter self-contained and selfserving on the development of the topics.

CONFLICT OF INTEREST The author (editor) declares no conflict of interest, financial or otherwise.

ACKNOWLEDGEMENTS The material in the book is for the most part the product of multi-year efforts on working and consulting network economies and innovation processes. In preparing the previous text ‘Networks, Competition, Innovation and Industrial Growth’, New York: NovaScience (2016), I was made to believe that network economics in terms of Internet Economics reveals some unique features of combining networks with technological change. From economic point of view, it is amazing how many positive network externalities are connected with the Internet but also with its growth and ubiquitness how many dark sides (negative network externalities) are proliferating so that one ultimately talks about high level of regulations in response to a potential cyber war against people, corporations and nations. Contrary to a widespread belief of economics as a dismal science, we tend to emphasize the positive aspects of Internet Economics though we do not deny some dark aspects with the implicit assumption that the net economic effect in most cases is positive. With portions of the book, e.g. Chaps. 2-5, 7, we rely on the previous work of mine covered in the following articles: 1.“Network Economics for the Internet – Application Models”, iBusiness 3, 2011,313-322. 2.“Network Economies for the Internet”, Modern Economy 3, 2012, 408-423. 3.“Quality of Service on Queueing Networks for the Internet”, iBusiness 5, 2013, 1-12.

viii

4.“Internet Economics of Distributed Systems”, Transactions on Networks and Communication 2(6), 2014, 56-72. 5.“Supply-Chain Coopetition”, International Journal of Business and Economics Research 4(2), 2015, 67-71. 6.“Vertical Competition and Outsourcing in a Supply Chain”, International Journal of Business and Economics Research 4(6), 2015, 315-322. 7.“Krohn-Rhodes Complexity on Decision Rules”, Advances in Social Sciences Research Journal 3(4), 2016, 30-43. All of this background material can be found on www.researchgate.net_gottinger or www.hansgottinger.com I want to specially mention and thank Prof. Stanley Reiter (Northwestern Univ., Evanston) on conversations regarding his work on Computation and Complexity, and the late Professor Leonid Hurwicz (Univ. of Minnesota) on economic mechanism design that relates to Chaps. 2-5. Also the insights of Prof. B. Vöcking, RWTH Aachen helped me to advance on applications to algorithmic game theory. Chap. 8 has been joint work with Dr. Soraya Sedkaoui at Montpellier University. The figures associated with Chaps. 2 and 3 have been provided by Mrs. Sabine Spiesz of Spiesz Design, Neu-Ulm. My special thanks to all of them.

Hans W. Gottinger STRATEC Munich, Germany E-mail: [email protected]

Internet Economics: Models, Mechanisms and Management, 2017, 1-20

1

CHAPTER 1

The Evolving Internet: Technology, Regulation and Pricing “The Age of Networked Intelligence is an age of promise. ... It is not just an age of linking computers but of internetworking human ingenuity” Don Tapscott, The Digital Economy (2015) Abstract: This chapter covers the evolution of the Internet in its historic setting, the emergence of Internet institutions, technologies, architectures and services. From this an Internet industry developed as taking major dominance in the telecommunications world expanding to a global context. Key economic issues involve Internet pricing, Quality of Service (QoS) provisioning, congestion management, Internet regulation and the backbone network technologies such as Asynchronous Transfer Mode (ATM) in view of traffic management and congestion control. The relation between service discipline and the varieties of the bandwidth-buffer tradeoff is also discussed. It shows improvements being made to effectively operate and manage networks that may support performance and guarantees in serving heterogeneous users with a wide range of requirements.

Keywords: Asynchronous Transfer Mode (ATM), Bandwidth-Buffer, Burstiness, Competitive Bidding, Congestion Control, Internet Protocol (IP), Internet Pricing, Internet Service Provider (ISP), Price Elasticities, Resource Allocation Algorithm, Quality of Service (QoS), Standardization, Service Discipline, Service Level Arrangement (SLA), Telephony Model, Transmission Control Protocol (TCP), Virtual Channel, Voice over IP (VOIP). 1.1. INTRODUCTION In its early formation, the Internet was generally viewed as an experimental special purpose communication network linking research and defense related establishments in a US national and increasingly global context. In the 1990s, the Internet was increasingly seen as a principal pathway to a future information and communication structure. The new focus was on a general purpose information infrastructure, not limited to a special community of users, but raising the problems of universal service for the population at large and also involving business activities (Hatfield et al., 2005). By then, the problem was not only Hans W. Gottinger All rights reserved-© 2017 Bentham Science Publishers

2 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

posed as an issue of connectivity, as it was with the landline telephone system, but as one linked to the access of information. The Internet was sometimes termed as a ‘bottom-up infrastructure’ which relates to establishments and investments in local area networks (LANs) driven by distributed computing in military, university-type and enterprise networks. In this regard, it contrasts with online service systems such as ’Minitel’ offered by France Telecom, which essentially was a top down activity by a natural monopoly. The US Department of Defense Advanced Research Projects Agency (DARPA) initially created the Internet to be a military network (Kleinrock, 1996; Ryan, 2010). It placed a high priority on considerations that have little relevance to commercial networks (such as survivability in the face of attack) and a low priority on considerations that are critical to a network’s commercial success (such as efficiency and cost accountability). Due the emergence of the Internet as a public Internet the technological and economic environments surrounding the Internet, have changed since the mid-1990s. Four major changes have impacted the network since then: i. Increase in the number and diversity of end users. A small population of technologically knowledgeable researchers in the early Internet has been replaced by a user base that is much larger, more diverse, and less technologically sophisticated. ii. Increase in the diversity and intensity of applications. Early Internet applications such as email and file transfers required relatively little bandwidth and were not particularly sensitive to variations in network performance. Modern applications, such as videoconferencing (through voice of internet protocol (VOIP)) and online gaming demand greater bandwidth, security and performance. iii. Increase in the variety of technologies. When the Internet first emerged, almost everyone was connected to it via desktop computers attached to a landline connection. Dial-up modems have now given way to a multitude of end use networking technologies including cable modem systems, digital subscriber lines (DSL), fiber-to-the home (FTTH), and wireless broadband (WB). New technologies for transmission and traffic flows vary widely in terms of their available bandwidth, reliability, mobility and susceptibility to local congestion. The number of devices used to connect to the Internet has diversified as well and now includes laptops, smartphones, specialized devices (such as e-readers, tablets) and radio frequency identification tags (RFID). And further expansion is on the way through increased interaction of network devices and software agents involving the Internet of Things (IoT). iv. The emergence of more complex business relationships. The topology of the Internet began as as a three-level hierarchy consisting of backbones, regional

The Evolving Internet

Internet Economics: Models, Mechanisms and Management 3

internet service providers (ISPs) and last mile providers. Over time, the networks comprising the Internet began to enter into a much more diverse set of business relationships., like private peering, multihoming, content delivery networks (CDNs), and server farms. These changes are placing increasing demand on the Internet to develop new architectural principles better suited to meet diverse demands that end users are placing on the network. In the US, Internet Service Providers (ISPs) have been classified as ‘enhanced service providers’ by the Federal Communication Commission (FCC) and with the advent of the Obama Administration regulations, the attempts of regulations have increased significantly. In a more global context, in particular Europe and major parts of Asia, there are increasing attempts to regulate internet services, through various institutional means up to completely control the Internet. But with changing technology and industry structure, this will also be more difficult to accomplish. A primary concern in regulating universal access to the Internet had been the issue of pricing its services, the maintainance of competition among providers and strengthening incentives for investments into the network infrastructure. In the early pre-1995, the Internet possible policy emerged in identifying the issues toward a workable model (Majumdar,Vogelsang and Cave, 2005): 1. Charging by access to telecommunications capacity, e.g. flat rate pricing and keeping distance independent pricing(‘the death of distance paradigm’), 2. Consider network externalities in the economics and growth of networks, e.g. positive like scale, and negative like congestion effects, 3. Introduce usage based linear pricing, 4. Introduce usage-based nonlinear prices. The evolution of Internet pricing poses interesting concerns. Flat-rate pricing has been one of the factors that promoted the Internet to expand at a dramatic rate. It has enabled low-cost dissemination, beta testing and refinement of new tools and applications.The strength of many of these tools is their ability to operate in and rationalize a distributed and heterogeneous structure making it easier to identify and retrieve information sources. Flat rate pricing and usage based pricing made the Internet attractive to mass communication with many new users. However, as the Internet expanded and its dynamics in terms of usage, business and technologies proliferated new requirements of pricing through quality of service (QoS) provisioning emerged.

4 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

1.2. INDUSTRY STRUCTURE From the emergence of the public Internet in the mid 1990s, it was fair to say that no one in particular owns the Internet but that virtually hundreds of companies own a small part of it. And an increasing number of companies have been requesting a share notably for e-commerce and social media. Beyond the established telephone companies, cable TV companies have also realized that the Internet had passed their interactive TV rivals. In a deregulatory environment, they were racing to offer Internet and other related services on their network. In this new world, commercial online service companies were positioning themselves to gain market share in a fast growing dynamic market but faced an uncertain future because offering services without content was not enough. In the same vein, telephone companies after staying on the sideline were fighting to regain positions. They were more than ever convinced that the Internet is a real threat to the telephone model to which one could only respond taking it as an opportunity. Then, it was already possible to make telephone calls through VOIP on the Internet from specially equipped computers; the spread of multimedia PCs and faster Internet connections could make this common place. At the same time, companies were turning an increasing amount of their telephone traffic into digital data and sending it through private data networks saving up to half their telecoms costs, regulation permitting, this traffic could eventually move to the Internet. The telephone companies’ initial choice was whether they participate in the cannibalization of their revenues or watch it happen. At this moment, neither the Internet nor any digital network, could handle all the world’s telephone calls. But credible predictions suggested that by the turn of the millenium, the Internet would carry more data than the voice networks. In a growing Internet, a few things were bound to happen. First, the telephone companies might be able to hook up enough customers to be able to impose the telephone model on the Internet under which they had thrived for so long. That means settlements, usagebased prices and the like. They might be supported by technical capacity limits to the Internet, to the extent that future demand outstripped capacity, to the extent that we would see bottlenecks and then slowing down and even halting of traffic through the network (Bohn et al., 1994). Second, it could also mean the creation of a more secure, faster and more efficient but higher priced Internet that is used as an orderly business network instead of the cheaper, more chaotic consumer network with minimal interconnection between them. But this scenario would fail to the extent that the customers would prefer the independent providers, and the telecoms would compete for a market with increasingly integrated services (voice, data and video).

The Evolving Internet

Internet Economics: Models, Mechanisms and Management 5

1.3. TELECOMMUNICATIONS AND THE INTERNET The nature of the Internet flies in the face of traditional telecommunication policies within regulated environments. Early on the Internet has grown in unregulated environments and in parts has been subsidized. Given its history until the late 1980s the networks of the Internet were all noncommerical. Unlike telephone networks and cable systems which are both propriety and facility based, the Internet is a network of networks assembled from leased lines, routers, switches and hub-to-hub backbone interconnection. Low costs of leased lines have fostered rapid growth in the US relative to other world regions. The Internet offers a radical shrinking of communication costs underpricing conventional communication forms (such as telephone and fax). The cost-based environment of the Internet offers significant economies of scale though decreasing costs at the margin would tend to strengthen a ‘natural monopoly’ and therefore induce regulatory intervention. This would be counteracted by lowering the barriers to entry into the industry and strengthen competition among Internet Service Providers (ISP). Also, in the development of the Internet ‘economies of scale’ led to the formation of cooperative-style frameworks. The technology of the Internet permits efficient aggregation of traffic and sharing of network resources. 1.4. INTERNET PRICING The Internet has overturned the conventional wisdom on building telecoms networks, as it has challenged the telephone pricing system. The first is the, death of distance paradigm’. One reason why distant messages cost no more than local ones is that the Internet, even if it runs on phone lines, uses them through packet switching much more effeciently than voice calls do. A traditional voice call is an analog signal, called circuit switching, which needs a lot of electronic space to avoid interference, so it takes up an entrire line for the duration of the call. By contrast, the Internet is digital so its data bits can be suitably compressed. Second, the Internet data are split into packets which do not need a line for themselves. Packets from many sources are mixed up by computer and pipelined in a major packet stream.The router at the other end of the line receives each one, reads its address and sends it into the right direction.When you send a message on the Internet, you are sharing a plentiful and cheap resource, the entire bandwidth on the line. Third, telecoms pricing, though propagated as usage based pricing, has no clear economic justification. A large part of the price of a telephone call (sometimes more than 40 percent) goes to the recipient’s telephone company for taking it to

6 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

the last few miles. Through a complicated accounting scheme known as ‘settlements’, telecoms companies exchange billions of dollars each year to pay for the local component of international calls. So long as the average usage of sending messages through the Internet will not exceed capacity it appears that it would cost as much sending one email as hundreds of emails that is nothing. Of course, each email does cost somebody something because it consumes a tiny bit of bandwidth (as a resource) the total of which is constituting capacity of data pipelines. But since the Internet providers have no way of billing for such incrementally small consumption they have settled for a rough approximation instead. Bearing better economic tools as presented further ahead in this text, as a rule of thumb, they multiplied the number of their subscribers by the average network usage to calculate the capacity they need to lease which would bet he basis of their calculation how much to charge. But as the Internet evolved, deepened and continuously expanded usage changes. In its early stages the Internet had been mostly a world of text which is an efficient way to communicate. But people hsve increasingly flooded the Internet with byte hungry multimedia. The World Wide Web (WWW) since its activation after 1989 accounted for an increasing share of the traffic on the Internet and swallows more bandwidth than any other service. The major shifts in Internet usage suggests that its early architectural principles from the mid-1990s may no longer be appropriate today. Restructuring the Internet to best fit today’s usage requires responses to end use changes (Yoo, 2013). ●







Standardization: With the diversity of consumers, agents and their composition changing the optimal level of standardization must be adjusted accordingly.As consumer demand and technologies comprising the network become more heterogeneous, network protocols, topology and and business relationships will have to become more varied. Governance: The Internet originally relied on cooperative behavior of its users but with the increase in the number and heterogeneity of end users, the proliferation of social media, government censureship and cybercrime needs increased reliance on more formal modes of governance. Security and Congestion: The need of coordinating and controlling information about the actions of multiple users and negative network externalities in terms of security and congestion would put the focus on enhanced network service management. Internet Pricing: The complexity of Internet growth would be incompatible with a simple set of pricing relationships. The increasing heterogeneity of end users’ bandwidth consumption, the need to manage congestion, the growing importance of advertising as a revenue source are leading the industry toward a

The Evolving Internet



Internet Economics: Models, Mechanisms and Management 7

more diverse array of pricing arrangements. Dominance of Intermediation: While the Internet’s early architecture demonstrated the ability to bypass gatekeepers and allow agents to connect directly with others the transformation from person-to-person communication to a platform for mass communication has necesssitated intermediation. It takes advantage of new architectural approaches to delivering content.

1.5. QUALITY OF SERVICE (QOS) With the Internet we observe a single quality of service (QoS): ‘best effort packet service’ though subsequent protocols, IPng (IP next generation such as IPv4, IPv6), have broadened performance criteria. Packets are transported first, served with no guarantee of success. Some packets may experience severe delays, while others may be dropped and never arrive. Different kind of data place different demands on network services . Email and file transfer requires 100 percent accuracy but can easily tolerate delay. Real-time voice broadcasts require much higher bandwidth than file trasfers, and can tolerate minor delays but cannot tolerate significant distortion. Real-time video broadcasts have very low tolerance for delay and distortion. Because of these different requirements, network allocation algorithms should be designed to treat different types of traffic differently but the user must truthfully indicate which type of traffic he/she is preferring, and this would only likely happen though incentive compatible pricing schemes. An example of such a scheme for asynchronous transfer modes (ATM) will follow next in Internet Communication Technologies (ICTs). Network pricing has been looked at early as a mechanism design problem . The user can indicate the ‘type’ of transmission and the workstation in turn reports this type to the network. To ensure truthful revelation of preferences, the reporting and billing mechanism must be incentive compatible. 1.6. PRICING CONGESTION The social cost of congestion is a result of the existence of network externalities. Charging for incremental capacity requires usage information. We need a measure of the user’s demand during the expected peak period of usage over some period, to determine the share of the incremental capacity requirement. In principle it might seem that a reasonable approach would be to charge a premium price for usage during the predetermined peak periods (a positive price if the basic usage price is zero), as is routinely done for electicity pricing (Wilson,1993, Chaps.1,10,12; Laffont and Tirole, 2000, on telecommunications pricing). However, in terms of Internet usage, peak demand periods are much less predictable than for other utility services. Since the use of computers would allow to schedule some activities during off-peak hours, in addition to different time

8 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

zones around the globe, we face the problem of shifting peaks. A broad review of cost and pricing issues in interconnected networks is provided by Gupta et al. (2005). By identifying social costs for network externalities the early suggestion by MacKie-Mason and Varian (1995) was directed toward a scheme for internalizing this cost as to impose a congestion price that is determined by a realtime Vickrey auction. The scheme requires that packets should be prioritized based on the value that the user puts on getting the packet through quickly. To do this, each user assigns his/her packets a bid measuring his/her willingness-to-pay (indicating effective demand) for immediate servicing. At congested routers packets are priorized based on bids. In line with the design of a Vickrey auction, in order to make the scheme incentive compatible, users are not charged the price they bid, but rather are charged the bid of the lowest priority packet that is admitted to the network. It is well-known that this mechanism provides the right incentives for truthful revelation. Such a scheme has a number of desirable characteristics. In particular, not only do those users with the highest cost of delay get served first, but the prices also send the right signals for capacity expansion in a competitive market for network services. If all of the congestion revenues are reinvested in new capacity, then capacity will be expanded to the point where its marginal value is equal to its marginal cost. This also fits into the framework of capacity pricing (Wilson, 1993, Chap. 11). A conceptual implementation of such a scheme based on competitive bidding for Internet directed ATM networks is described in Sec. 1.11. More universal price discrimination schemes within tiered services have been suggested as ‘Paris Metro Pricing’(PMP), Odlyzko, 1999). The PMP scheme separates the network into independent subnetworks that behave similarly but charge their customers at different rates. In view of resource allocation schemes we observe: The world wide web (WWW), network computing (NC) and electronic commerce (EC) have been crucial factors in creating early problems of interconnective congestion on the Internet. This has given rise to pessimistic concerns on the ‘Tragedy of the Commons’ of resource use with a congestion type gridlock as a consequence . In a more optimistic technologicsl perspective congestion would not be considered a longer term issue on the belief that new fiber-optic cables on the Internet backbone would boost bandwidth in the foreseeable future together with wireless technologies. In their otherwise detailed research on resource pricing of the Internet, Chinese authors Ku et al. (2014) were argueing that the ‘Tragedy of the Commons’ paradigm is applicable to the Internet where selfish behavior of the agents destroys the resource base of the Commons so that the aggregate costs (damages) to all interconnections would exceed the benefits. This (static) viewpoint puts the Internet into a public domain, servicing and pricing as a public utility where

The Evolving Internet

Internet Economics: Models, Mechanisms and Management 9

resource ‘rationing’ with commensurate pricing would be a proper tool to avoid the collaps on the Internet commons. However, in analogy to the Coasean view of resource allocation (Hurwicz, 1995), the Internet as an open domain and utilized by private agents through property rights would make market mechanism and efficient competitive pricing the proper conceptual mechanism design.In such a framework usage based pricing would be a more likely contender than flate rate pricing. 1.7. COMPETITIVE MARKETS

BIDDING

IN

LARGE

COMMUNICATION

Prices in real-world communication markets cannot be updated continuously. The efficient price is determined by comparing a list of user bids to the available capacity and determing the cutoff price. In fact, packets arrive not all at once but over time, and thus it would be necessary to clear rhe market periodically based on some accumulation of bids. The efficiency of such a scheme then depends on how costly it is to frequently clear the market and on how persistent the periods on congestion are. For example, one objection against such a scheme has been that the users may not be able to know in advance how much network utilization will cost them. It is sufficient if they know how much they can afford since the user’s bid always controls the maximum network usage costs they would be willing to pay. That is, since most users are willing to tolerate some delay for email, file transfer and the like, most traffic should be able to go through with acceptable delays at zero congestion prices, but time critical traffic will typically pay a positive price. When the network is congested enough to have a positive congestion price, these users will pay the cost in units of delay unless they raise their bidding. One may also expect that in a competitive market for network services, fluctuating congestion prices would result in creating ‘wholesale’ traders which would repackage the services and offer them at a guaranteed price to end-users, essentially establishing a futures market for network services.Some auction implementation problems may arise, such as where and at which entry points auctions will be accessed. In practice, networks have multiple gateways, each subject to differing states of congestion. Here coordination must be achieved between separate auctions. Empirical work must determine the optimal rate of market-clearing and inter-auction information sharing, given the costs and delays of real-time communication. A potentially serious problem for almost any usage pricing scheme is how to determine the correct billing of either a sender or receiver. With a telephony model (i.e., the PublicSwitched Telephone Network, PSTN) it is clear that in most cases the originator of a call should pay.

10 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

However, in a packet switching network, both sides originate their own packets, and in a connectionless network there is no mechanism for identifying party’s B packets, that were solicited as responses to a session initiated by party A. For example, one major use of the Internet is for file retrieval from public archives. If the originator of each packet were charged for that packet’s congestion cost, then the providers of free public goods (the file archives) would pay nearly all the congestion charges induced by a user’s file request. Either the public archive provider would need a billing mechanism to charge requesters for the congestion charges, or the network would need to be engineered so that it could bill the correct party. In principle, this problem can be solved by billing schemes like reverse charges but in a packetized network it may cause high transaction costs. 1.8. INTERNET AND TELECOMMUNICATIONS REGULATION The growth of data networks like the Internet provides an additional source for regulatory reform and deregulation of telecommunications. In the conventional regulatory structure local telephone companies, as natural monopolies were regulated. However, these companies face ever-increasing competition from data network services. For example, the fastest growing component of telephone demand some time ago had been for fax transmission. But fax technology is better suited to packet switching networks, and faxes as voice calls (Voice over Internet Protocol or VOIP) are increasingly transmitted over the Internet. As integrated services networks (ISN) emerge, they will provide an alternative for voice calls and videoconferencing as well. Internet transport itself is currently unregulated but services provided over telecommunication carriers are not. This principle has never been consistently applied to telephone companies since their services over fixed telephone lines also used to be regulated. There have been increasing demands, sometimes supported by established telecommunication carriers that similar regulatory requirements should apply to the Internet. One particular claim is ‘universal access’ to Internet services (Kahin and Keller, 1995), that is, the provision of basic Internet access to all citizens at a very low price or even for free. What is a basic service, and should its provision be subsidized? For example, should there be an approriate access subsidy for primary and secondary schools? A related question is whether the government should provide some data network services as public goods. A particular interesting question concerns the interaction between pricing schemes and market structure for telecommunication services. If competing Internet

The Evolving Internet

Internet Economics: Models, Mechanisms and Management 11

service providers offer only connection pricing, inducing increasing congestion, would other service providers be able to attract high value multimedia users by charging usage prices but offering effective congestion control? On the other hand, would a flat rate connection price provider be able to undercut usage-price providers by captsizing a large share of baseload customers who would prefer to pay for congestion with a delay rather than with a fee. Could this develop into a fragmented market with different Internets? These developments may have profound impacts to shape a future telecommunications industry which may be taken over by different layers of the Internet. 1.9. INTERNET COMMUNICATION TECHNOLOGIES: ATM AND BISDN The Internet and Asynchronous Transfer Mode (ATM) have strongly positioned themselves for defining the future information infrastructure (Kleinrock, 1996). The Internet is successfully operating one of the popular information systems, the World Wide Web, which suggests that the information highway is forming on the Internet. However, such a highway is limited in the provision of advanced multimedia services such as those with guaranteed quality of service (QoS). Guaranteed services are easier to support in ATM technology. Its capability far exceeds that of the current Internet, and it is expected to be used as the backbone technology for the future information infrastructure. ATM proposes a new communications paradigm . It allows integration of different types of services such as digital voice, video and data in a single network consisting of high speed links and switches. It supports a Broadband Integrated Services Digital Network (B-ISDN), so that ATM and B-ISDN are sometimes used interchangeably, where ATM is referred to as the technology and B-ISDN as the underlying technical standard. ATM allows efficient utilization of network resources, and simplifies the network switching facilities compared to other proposed techniques in that it will only require one type of switching fabric (packet switch). This simplifies the network management process (Pandya and Sen, 1999; Stallings, 2002). The basic operation of ATM, and generally of packet-switched networks, is based on statistical multiplexing. In order to provide QoS, the packets need to be served by certain scheduling (service) disciplines. Resource allocation algorithms depend heavily on the scheduling mechanism deployed. The scheduling is to be done at the entrance of the network as well as the switching points. The term ’cell’ designates the fixed-size packet in ATM networks. ATM allows variable bit rate sources to be statistically multiplexed. Statistical multiplexing produces more efficient usage of the channel at the cost of possible congestion at the buffers of an ATM switch. When the congestion persists, buffer overflow occurs, and cells are discarded (or packets are dropped). Therefore, resources (i.e. bandwidth and buffer space) need to be carefully allocated to meet the cell loss and the delay

12 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

requirements of the user. The delay and the cell loss probability that the user wishes the network to guarantee are referred to as the QoS parameters. Although the Internet is capable of transporting all types of digital information, it is difficult to modify the existing Internet to support some features that are vital for real time communications. One important feature to be supported is the provision of performance guarantees. The Internet uses the Internet Protocol (IP), from IPv4 to IPv6, in which each packet is forwarded independently of the others. The Internet is a connectionless network where any source can send packets any time at speeds that are neither monitored nor negotiated. Congestion is bound to happen in this type of network. If congestion is to be avoided and real-time services are to be supported, then a negotiation (through pricing or rationing) between the user and the network is necessary. ATM is a connection-oriented network that supports this feature. A virtual channel is established, and resources are reserved to provide QoS prior to data transfer. This is referred to as channel establishment. In order to support connectionless services, e.g. IP over ATM, connectionless servers (CLS) can be placed at ATM switches where ATM cells are forwarded over particular permanent virtual circuits to the CLS, and the ISDN address is used to do regular routing. 1.10. BURSTINESS Burstiness is an important property in travel characteristics to minimize in ATM networks. The burstiness curve b(s) defines the necessary buffer space (for a particular input traffic m(t) in the intermediate nodes for a given service rates to meet the zero cell loss requirement. The concept of burstiness curve (Low, 1992) could be defined as such: let m(t) be the source traffic rate (in bits/sec.) on t ε [0,T], where T is the total duration of the connection, then: b(s) = sup ∫ [ m(r) – s] dr 0≤s≤t≤T

Where b(s) is the burstiness of input m(t) at fixed service rates. Under normal circumstances b(s) is assumed to be nonnegative, convex and strictly decreasing for s < M, where M is the peak rate of traffic. Fig. (1.1) shows an example of a burstiness curve with the burstiness budget line (BBL). The burstiness curve represents the buffer size necessary to avoid cell losses at each service rates. When a bandwidth-buffer space pair (s,b(s)) on the burstiness curve is used for resource allocation, there will be no cell loss.

The Evolving Internet

Internet Economics: Models, Mechanisms and Management 13

b (s) B

A C BBL

0



s

M

Fig. (1.1). Burstiness Curve b(s).

1.11. VIRTUAL CHANNEL In a connection-oriented packet switched network, a virtual channel must be established, and resources need to be reserved prior to the actual data transfer. A virtual channel identifier in the header of each packet is used at the switching points to avoid routing on a packet-by-packet basis. Resources need to be reserved to provide guaranteed QoS. The needed resources are calculated on the basis of traffic characteristics and the QoS requirements by the admission control, and enables the admission control algorithm to estimate the minimum necessary resources for the connection. Congestion control games will be referred to Chap. 2. 1.12. SIMPLE ECONOMICS OF INTERNET RESOURCE ALLOCATION AND PRICING A resource allocation algorithm should allow an efficient utilization of network resources. Resource allocation has several aspects in ATM networks. First, resources such as bandwidth and buffer space can be allocated during each channel establishment. This is the allocation of resources from the fixed supply. These fixed supplies can be expanded by adding more link capacity and more buffer space. This is called network capacity planning. Then there is an implicit way of allocating resources to different groups of users by charging differently by application type. This is called the billing approach. In what follows, unlike network capacity planning, nor with approaches to resource allocation based on charging policies we are dealing with resource allocation to a channel alone at the time when the channel is established and consider the network’s supply of resources fixed.

14 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

On the background that the number of subscribers to digital services has been increasing exponentially there is a lack of mechanisms, however, to provide guaranteed QoS and the scarce resources for packet switched network (e.g. Internet, ATM) are starting to surface as serious problems with the emerging potential of a ‘data tsunami’. The Internet community is trying to accommodate the need for guaranteed services by providing signalling and admission control, modifying the Internet Protocol format and upgrading router functionality to support prioritized scheduling. The issues in providing guaranteed services in ATM are similar to those in the Internet, in that admission control needs to be supported, and switches need to provide prioritized scheduling. Various algorithms for resource allocation have been proposed, however, simple solutions that seem to be practical in engineering suffer from inefficiencies in resource allocation, and complex solutions seem to be impractical though in principle they may achieve better resource utilization. This suggests an economic approach of congestion pricing based on economic principles of competitive bidding, auctioneering and shadow pricing familiar in the economics of regulation. In search for a solution one has to take advantage of theoretical results, yet, one must work in a practical framework that results in the efficient utilization of network resources. 1.13. BANDWIDTH-BUFFER TRADEOFF The bandwidth-buffer tradeoff can be seen from Fig. (1.1). Bandwidth can be traded for buffer space and vice versa to provide the same QoS, as long as the resource pair falls on the burstiness budget line (BBL). If a bandwidth is scarce, then a resource pair that uses less bandwidth and more buffer space should be used. Resource pricing is targeted to exploit this tradeoff to achieve efficient utilization of the available resources. The pricing concept for a scarce resource is well-known in economics, but in the context to exploit the bandwidth-buffer tradeoff it is due to Low and Varaiya (1993). They use non-linear optimization theory to determine centralized optimal (shadow) prices in large networks. With respect to large scale application, however, the complex optimization process limits the frequency of pricing updates, which causes inaccurate information about available resources. In order to make pricing in the context of a bufferbandwidth tradeoff more adjustable and flexible it should be based on decentralized pricing procedures according to competitive bidding in large markets where prices will be optimal prices if the markets are efficient. This would also allow flexible pricing which results in accurate representation of available resources in that prices are updated as the instant connection request arrives. The subsequent procedure suggested is based on distributed pricing as a more feasible alternative to optimal pricing.

The Evolving Internet

Internet Economics: Models, Mechanisms and Management 15

1.14. PRICING THE RESOURCES During the admission process of a new connection, a bandwidth-buffer pair is to be chosen in each node. If the algorithm takes advantage of the resource tradeoff and picks more of what is more available, and less of what is less, than the resources can be used most efficiently. If the pair were picked randomly, the bandwidth in a node might run out before the buffer space does or vice versa. If, however, the algorithm is such that the utilization of bandwidth and buffer space are balanced then this results in inefficient utilization of both resources. The resource availability can be communicated to the user by pricing the resources where prices are normalized per unit bandwidth (i.e. cells per second) and per unit buffer (cells). For example, if bandwidth becomes scarce, then the price of a unit bandwidth is increased, thus encouraging the network to use more buffers. During the channel admission process, the pricing information is used to improve the total network usage. As can be seen from the figure, when one of the suppliers decreases, the price of the corresponding resources increases, thus discouraging the use of that resource. Separate price curves for bandwidth and for buffer space are assigned at every output port. The curves should be shaped so that, when the amount of the resource available is lower, the price per unit resource is higher. This means that the pricing (or inverse demand) function P(q) should be strictly decreasing where q is the quantity of the resource that is unallocated, and P(q) is the price per unit quantity at that quantity. The P(q) curves for each of the resources (i.e. bandwidth and buffer space) are downward sloping as in Fig. (1.2) where prices could be determined at any instant. In order to achieve this we need to specifically (i) define the price function and (ii) use the concept of resource balancing. P (q) e e e

1

1

2

q

(resources left)

Fig. (1.2). Comparison of Price Curves with Different Elasticities.

1

e

0

e

2

16 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

The Price Function A constant elasticity function is chosen for both bandwidth and buffer space pricing. This means that the price elasticity (that is, how fast the price changes as the available quantity changes) is a constant. The elasticity is defined to be: ε (q) = - (dP(q)/P) /(dq/q)

A constant elasticity function has the property that at any point on the price curve the elasticity is constant (i.e. ε (q) = ε for all q). A function that satisfies this condition is P(q) = q -ε. A price curve with higher elasticity responds more slowly to a change in the quantity left than a price curve with a lower elasticity. This is depicted in Fig. (1.2). The price elasticity ε should reflect the demands of a particular port for a particular resource. Each port at each node (switch) has its own demand levels. A demand level is defined as the total resource request normalized by the total resource. Since network management is responsible for monitoring traffic information we assume that traffic (demand) statistics are available. Denote the bandwidth and buffer space demand levels by DA and DB respectively, and D‾A, D‾B represent the average demand levels that capture their trend. Since the demand levels D‾A and D‾B reflect the utilization of resources, these parameters can be used to determine the elasticity. If the demand level is high, then a lower elasticity should be assigned to the curve and vice versa. Since the demand levels are normalized (i.e. D‾A, D‾B [0, 1]), the elasticities can be computed as: εA= 1 - D‾A, εB = 1 - D‾B

By allowing elasticities to differ each output port enables the port with higher demand to have more control in selecting the resource pair. This is so because prices of the port with low elasticity will be higher even when the resources available are exactly the same. The control ensures that the most congested (or most likely to be congested) port is able to select a resource pair to optimize resource utilization. The design of a resource allocation algorithm will include a mechanism to show how the ε’s are selected. The resource pricing should depend on various factors, and these can be incorporated in the functions by carefully choosing the parameters. The generalized price functions to be adopted for each of the resources are: for bandwidth h (qA) = α(qA/γ)-εA

The Evolving Internet

Internet Economics: Models, Mechanisms and Management 17

for buffer space K (qB) = β(qA/δ)-εB

where qA and qB are the bandwidth and buffer space resources left, respectively. How to pick the parameters α,β,γ,δ and the ε’s is discussed next. Resource Balancing The reason for pursuing balancing the resource usage is that this promotes depletion of both resources rather than just one. The admission control algorithm tries to ensure the balanced usage of the resources by choosing the appropriate resource pair. When the resources are balanced, the pair of bandwidth matches that of buffer space. This means that: h (qA) = k (qB) α (QA) –εA = β (QB)-εB

where QA and QB equal qA/γ and qB/δ, respectively. When the load is balanced and a single normalized unit of each resource is left (QA = QB = 1), then this reduces to α =β. 1.15. PRICE ELASTICITIES Previously it was shown that changing the bandwidth unit ratio affected the resource utilization levels and the admittance levels. In turn, we show how changing the price elasticities can affect the same two quantities. The resulting change in resource utilization responds to a change in the bandwidth- buffer unit ratio. In order to achieve high resource utilizations the unit ratio needs to be approximately near its optimal value. When changing the price elasticities, a small change in elasticities can result in big changes in resulting resource utilizations. By simulation one can show that varying the elasticities can result in better resource utilization, however, the elasticities have to be picked carefully to achieve the optimum. Among the more general results one can state (other things being equal): 1. The admittance level drops quickly as the price elasticity for buffer space increases. 2. Maximum resource utilizations are achieved at a critical buffer price elasticity, however, the difference in connection admittance with the buffer price elasticity is very low. 3. Changing the elasticities can result in unsatisfactory traffic because it can put one price curve far above the other price curve, resulting in low price

18 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

utilizations and low admittance levels. In order to avoid this one can set the buffer and bandwidth price elasticities equal at every node: εA = εB. 1.16. RESOURCE ALLOCATION ALGORITHM The mechanism for resource allocation proposed here is resource balancing. The usage of the resources is balanced so that, for example, if less bandwidth is available, then less bandwidth is assigned. This is achieved by pricing the resources. The resource pair is chosen from the burstiness traffic model in such a way as to minimize the price difference between bandwidth and buffer space (i.e. h – k). Formally, it means:

With the constraint that (s,b(s)) falls on the burstiness (bound) curve for the given traffic source. The algorithm avoids the minimization computation. It first calculates the prices for both resources based on the resources left via the pricing function. If the bandwidth price is greater than the buffer price (h > k) it simply selects the minimum bandwidth and maximum buffer space point on the burstiness budget line (B,B’). If the bandwidth price is less than the buffer price (h < k), then the maximum bandwidth and minimum buffer space point on the burstiness (bound) curve (i.e. (M,0) is selected. This is summarized as: if (h > k) select the resource pair (B,B’), else select the resource pair (M,0)

In principle, if we know the exact traffic characteristics of the source and the resource capacities, then we can calculate the resource pair on the burstiness (bound) curvet hat results in the most efficient utilization of the available resources. The algorithm we propose here achieves the maximum admittance through a linear combination of the resource pair (B,B’) and (M,0). For example, if the bandwidth and buffer capacities are 2.4Gb/s and 32K cells, respectively, and the source traffic is characterized by the burstiness bound model (10Mb/s., 200cells, 15Mb/s), then a simple calculation shows that allocating the resource pair (1l.5Mb/s, l50cells) results in a maximum of 213 active connections. That is, we can solve for the number of connections with resource pairs (B,B’) and (M,0) by equating their linear combinations to the total resource capacities.

The Evolving Internet

Internet Economics: Models, Mechanisms and Management 19

The amounts of resources are expressed in Mb/s for bandwidth and cells for buffer space. In the case of one traffic type, solving the linear combination reduces to solving two equations with two unknowns. The above equation yields a = 160 and b = 53. This means there are 160 maximum buffer space allocations and 53 maximum bandwidth allocations which result in a total of 213 connections. The algorithm achieves the maximum through resource pricing. In general, if we let traffic type i be represented by the burstiness bound model (Bi, Bi’,Mi), i  I, where I is the total number of different traffic types, the total number of maximum buffer space (B, B’) and maximum bandwidth (M, 0) allocations for each traffic type i be ai and bi, and the total available bandwidth and buffer resource be BW and BU, respectively, then the proposed algorithm results in a set of a b that maximizes the resource allocations a = (a1, a2,...,aI) and b = (b1, b2 ...bI). Formally, this means that:

In a brute force approach, without resource pricing, in order to achieve the maximum resource utilization, the network needs to know the exact traffic patterns and depends on the available resource capacities at each node. Also, as the total number of traffic types increases, the calculation of the proper resource choice becomes difficult, and there will be a need to separate resource assignments by carefully examining the incoming traffic types. In fact that this is dependent on the resource capacities will make the allocation process difficult because resource capacities may differ from node to node. Without prior knowledge of incoming traffic, the proposed resource allocation algorithm automatically results, for all practical purposes, in the optimal utilization of the resources via price balancing. REFERENCES Bohn, R., Braun, H.-W., Claffy, K.C., Wolff, S. (1994). Mitigating the coming Internet Crunch: Multiple Service Levels via Precedence, Technical Report Super Computer Center, San Diego, CA. Gupta, A., Stahl, D.O., Whinston, A.B. (2005). Pricing Traffic on Interconnected Networks: Issues, Approaches, and Solutions Hatfield, D.N., Mitchell, B.M., Srinagesh, P. (2005). Emerging Network Technologies Hurwicz, L. (1995). What is the Coase Theorem? Japan World Econ, 7, 49-74. [http://dx.doi.org/10.1016/0922-1425(94)00038-U] Kahin, B., Keller, J. (1995). Public Access to the Internet. Cambridge, Ma.: MIT Press. Kleinrock, L. (1996). Technology Issues in the Design of the NREN In: Kahin, B., (Ed.), Building Information Infrastructure, Technology and Public Policy. New York: McGraw Hill. Laffont, J.J, Tirole, J. (2009). Competition in Telecommunications. Cambridge, MA: MIT Press.

20 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

Low, S. (1992). Traffic Control in ATM Networks Proc.Globecom. Low, S., Varaiya, P. (1993). A New Approach to Service Provisioning in ATM Networks IEEE/ACM Trans. Net, 1(Nov) MacKie-Mason, J.K., Varian, H.R. (1995). Pricing the Internet In: Kahin, B., Keller, J., (Eds.), Public Access to the Internet. Cambridge, MA.: MIT Press. Majumdar, S.K., Vogelsang, I., Cave, M.E. (2005). Technology Evolution and the Internet (Vol. 2). Amsterdam: North Holland.Handbook of Telecommunications Economics Odlyzko, A. (1999). Paris Metro Pricing for the Internet Proc. First ACM Conference on Electronic Commerce, 140-147. [http://dx.doi.org/10.1145/336992.337030] Pandya, A.S., Sen, E. (1999). ATM Technology for Broadband Telecommunications Networks. Boca Raton, Fl.: CRC Press. Ryan, J. (2010). A History of the Internet and the Digital Future. London: Reaktion Books. Stallings, W. (2002). High-Speed Networks and Internets:Performance and Quality of Service (2nd ed.). Upper Saddle River, N.J.: Prentice Hall. Wilson, R. (1993). Nonlinear Pricing. Oxford, New York: Oxford Univ. Press. Xu, K., Zhong, Y., He, H. (2014). Internet Resource Pricing Models. New York: Springer. [http://dx.doi.org/10.1007/978-1-4614-8409-7] Yoo, C.S. (2013). The Dynamic Internet: How Technology, Users, and Businesses are transforming the Network. Washington, D.C.: AEI Press.

Internet Economics: Models, Mechanisms and Management, 2017, 21-55

21

CHAPTER 2

Network Economies for the Internet –Conceptual Models “In the ‘real world ’the invisible hand of free markets seems to yield surprisingly good results for complex optimization problems. This occurs despite the many underlying difficulties: decentralized control, uncertainties, information gaps, limited computational power, etc. One is tempted to apply similar market-based ideas in computational scenarios with similar complications, in the hope of achieving similarly good results.” N. Nisan, Algorithms for Selfish Agents (STACS 99) Abstract: The Internet is framed as a market design for information with major actors supplying and demanding information, and supply and demand tending toward equilibrium, as a paradigm of a distributed computational system. Concepts of information flows and delivery provisions would reflect game-theoretic models of trade and exchange economies where the mechanism (design) of information flows would adopt the behavior and outcome of mathematical queueing systems. Of particular interest are large scale systems of the Internet type that act and perform as decentralized distributed systems of multi-level control. A prototype economic optimization model is put forward for a network economy with variation in traffic classes, utility parameters, queueing types and equilibrium approaches.

Keywords: Algorithmic Mechanism Design (AMD), Auctions, B-ISDN, Broadband, Buffer Space, Complexity, FIFO, Game-theoretic System, Interdomain Routing, Mechanism Design, Multi-Agent System (MAS), Network Economy, Networking, Optimum Allocations, Pareto Efficient Allocations, Quality of Service, Queueing Model, Traffic Classes, TELNET, Utility Parameters. 2.1. INTRODUCTION Most studies of resource allocation mechanisms have used a performance model of the resource, where the very concept of the resource is defined in terms of measurable qualities of the service such as utilization, throughput, latency, response time (delay) and so on. Optimization of resource allocation is defined in terms of these measurable qualities. One novelty introduced by the economic approach is to design a system which takes into account the diverse Quality of Hans W. Gottinger All rights reserved-© 2017 Bentham Science Publishers

22 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

Service (QoS) requirements of users, and therefore use multi-objective (utilities) optimization techniques to characterize and compute optimum allocations. Economic and algorithmic mechanism design, modelling of computer and communication resource sharing uses a uniform paradigm described by two level modelling: QoS requirements as inputs into a performance model that is subject to economic optimization. As a first step, we transform QoS requirements of users to a performance model a queueing service model. Queueing and networks and queueing in networks are very natural phenomena and determine the economic value of networks. Networks are vulnerable by queueing in the essential hubs and nodes as excessive queueing in flows can result in torn or fragmented networks and therefore negatively impact the scale of large networks (Albach, 1993). A performance model establishes quantifiable parameterization of resource allocation. For example, average delay QoS requirement, when based on a first-in first-out (FIFO) queueing model, is a function of resources, bandwidth and buffer, and user traffic demands. These parameters are then used to establish an economic optimization model. The question of whether the resource is a piece of hardware, a network link, a software resource such as a database or a server, or a virtual network entity such as a TCP connection is not of primary importance. The first modelling transformation eliminates the details and captures the relevant behaviors and the optimization parameters. We consider a decentralized model of network and server economies, where we show efficient QoS provisioning and Pareto allocation of resources (network and server resources) among agents and suppliers, which are either network routers or servers (content providers).We show how prices for resources are set at the suppliers based on the QoS demands from the agents. With advances in computer and networking technology, numerous heterogeneous computers can be interconnected to provide a large collection of computing and communication resources (Robertazzi, 1994). These systems are used by a growing and increasingly heterogeneous set of users which are identified with the present Internet. A macroscopic view of distributed computer systems reveals the complexity of the organization and management of the resources and services they provide. The complexity arises from the system size (e.g. no. of systems, no. of users) and heterogeneity in applications (e.g. online transaction processing, ecommerce, multimedia, intelligent information search) and resources (CPU, memory, I/O bandwidth, network bandwidth and buffers, etc.). The complexity of resource allocation is further increased by several factors. First, in many distributed systems, the resources are in fact owned by multiple

Network Economies for the Internet

Internet Economics: Models, Mechanisms and Management 23

organizations. Second, the satisfaction of users and the performance of applications are determined by the simultaneous application of multiple resources. For example, a multimedia server application requires I/O bandwidth to retrieve content, CPU time to execute server logic and protocols, and networking bandwidth to deliver the content to clients. The performance of applications may also be altered by trading resources. For example, a multimedia server application may perform better by releasing memory and acquiring higher CPU priority, resulting in smaller buffers for I/O and networking but improving the performance of the communication protocol execution . Finally, in a large distributed system, the set of systems, users and applications is continuously changing. In line with identifying the strategic factors of Internet enterprises, we address some of the issues of Quality of Service (QoS) and pricing, and efficient allocation of resources (computational resources) in networks and systems. We review and consider problems of Quality of Service (QoS) provisioning, resource allocation and pricing in computer networks and information systems. We consider such complex systems as economies where multiple classes of users (consumers) compete for resources and services from suppliers: network providers and information providers (servers). The resources under contention are network bandwidth and buffers, server processing rate and memory. We model and solve problems using economic principles of market mechanism, where resources are priced by suppliers based on demand, and users buy resources to satisfy their Quality of Service (QoS) needs. We focus on the issues of QoS provisioning, and the reasons in choosing economic paradigms to solve some of the problems. Economic resource allocation in networks relate to computational models of networks, as developed in the works of Radner (1993), Mount and Reiter (2002, Chap.4), van Zandt (1998), see also Gottinger (2009, Chap. 9). Here they emanate from certain types of queueing systems, as Kleinrock (1976) reported on generalized networks. Network resource allocation and pricing could be considered at as a mechanism design problem (Hurwicz and Reiter, 2006). More specific mechanism design approaches for distributed networks and grid-type systems are covered by Narahari et al. (2009) and Neumann et al. (2010) see also Meinel and Tison (1999). Massive complexity makes traditional approaches to resource allocation impractical in modern distributed systems such as the Internet. Traditional approaches attempt to optimize some system-wide measure of performance (e.g. overall average response time, throughput etc.). Optimization is performed either

24 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

by a centralized algorithm with complete information, or by a decentralized consensus based algorithm. The current and future complexity of resource allocation problems described makes it impossible to define an acceptable system-wide performance metric. What single, system-wide performance metric adequately reflects the performance objectives of multimedia server and an online transaction processing system? Centralized or consensus based algorithms are impractical in dynamic systems owned by multiple organizations. Resource allocation complexity due to decentralizations and heterogeneity is also present in economic systems. In general, modern economies allocate resources in systems whose complexity overwhelms any algorithm or technique developed for computer systems. As in economic mechanism design, we focus on the similarities between complex distributed systems and economies. Competitive economic models could provide algorithms and tools for allocating resources in distributed computer systems (Deng and Graham, 2007), in particular Ackermann et al. (2007), Chen et al. (2007), Iong et al. (2007).The tools used are algorithmic mechanism design, dynamic game theory, complexity and computational equilibrium. There is another motivation that has come about due to the commercialization of the Internet. The debate has begun in the area of multiple QoS levels, tiered services and pricing (Rouskas, 2009). How should pricing be introduced to provide many service levels in the Internet? Should pricing be based on access cost or should it be based on usage and QoS received by the end user? Will usage based pricing help the Internet economy grow, and help in accounting and improve the efficiency of the Internet, and make users benefit much more? Similar issues are being investigated in the ATM networking community. We address some of the issues of QoS and pricing, and efficient allocation of resources (computational resources) in networks and systems. 2.2. THE INTERNET AS A REFLECTION OF THE ECONOMY The economic system would serve an underlying network model for the Internet. As the economy grows in scale, scope, and diversification, this correspondingly reflects the digital economy and its parts: the Internet. In a network model, growth, dynamics and vitality of the Internet could be understood as a result of the combined effect of the economic benefits of the network’s positive (network) externalities, the Internet’s technical efficiency through statistical sharing, and the policy benefits of the Internet’s interoperability. The economic model schematically consists of the following players: Agents (human, software) and Network Suppliers. Consumers or user classes: Consumers request quality adjusted services on demand or quality of services (QoS). Each

Network Economies for the Internet

Internet Economics: Models, Mechanisms and Management 25

user class has several sessions (or user sessions). Users within a class have common preferences. User classes have QoS preferences such as preferences over packet-loss probability, max/average delay and throughput. Users within a class share resources. Agents and Network Suppliers Each user class is represented by an agent. Each agent negotiates and buys services (resource units) from one or more suppliers. Agents demand resources in order to meet the QoS needs of the user classes. Network providers have technology to partition and allocate resources (bandwidth and buffer) to the competing agents. In this competitive setting, network providers (suppliers) compete for profit maximization or satisficing. Multiple Agent-Network Supplier Interaction Agents present demands to the network suppliers. The demands are based on their wealth and QoS preferences of their class. The demand by each agent is computed via utility functions which represent QoS needs of the user classes. Agents negotiate with suppliers to determine the prices. The negotiation process is iterative, where prices are adjusted to clear the market; supply equals the demand. Price negotiation could be done periodically or depending on changes in demand. Each agent in the network is allocated a certain amount of buffer space and link capacity. The buffer is used by the agent for queueing packets sent by the users of the class. A simple first-in-first-out (FIFO) queueing model is used for each class. The users within a class share buffer and link resources. Agent and supplier optimality: Agents compete for resources by presenting demand to the supplier. The agents, given the current market price, compute the affordable allocations of resources (assume agents have limited wealth or budget). The demand from each agent is presented to the supplier. The supplier adjusts the market prices to ensure that the demand equals the supply. The main features of the economic model are: ●





Characterization of class QoS preferences and traffic parameters via utility functions, and computation of demand sets, given the agent wealth and the utility function. Existence and computation of Pareto optimal allocations for QoS provisioning, given the agent utility functions. Computation of equilibrium price by the supplier based on agent demands, and conditions under which price equilibrium exists. Price negotiation mechanisms

26 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

between the agents and suppliers. The variety and scale of market designs for network markets as deeply rooted in micro-economic models of the economy (originally Arrow and Debreu, 1954), their interoperability and computation (Scarf,1973) have been the basis of internet economic workshops (Deng and Graham, 2007). With the economic system paradigm of human and artificial agents, the functionability of software agents for future communication systems (Hayzelden and Bigham, 1999; Shoam and Leyton-Brown, 2009) could be naturally embedded through multi-agent systems (MAS). From this perspective, the aspects of computability and intelligent agents will be focussed on as agent technology needs to be implemented in software. Software agents will foster distributed artificial intelligence techniques and distributed computational systems that conversely relate to counterparts in decentralized economic mechanism design (Hurwicz and Reiter, 2006), corresponding incentive systems (Radner, 1987) and economic computational systems (Scarf, 1973). We also could conceive economic systems as complex adaptive systems (Holland, 2013) that “… learn or adapt in response to interactions with other agents”. From the perspective of computer science, in networks, the concept of ‘market based control’ emerged (Clearwater, 1996). Multi-agent systems (MAS) as embedded in the market-type economic structures provide special features that include decentralization or distributed systems, interacting agents and involving some sort of resource allocation mechanism. Complex resource allocation and load balancing in computerized systems would be subsumed under market based control (Wellman, Ferguson et al., 1996). Market-based control is viewed as a distributed multi-agent system to attribute to the agents particular knowledge, preferences, abilities and rational behavior that would be isomorphic to a computable distributed economic system of a similar nature, scale and scope. We can use simulations along the lines of market-based controls in distributed resource allocation mechanisms. The paradigm of an interactive economic system with interactive agents causing positive and negative network externalities brings game theory into full play. Such as congestion games with priorities to load balancing and resource sharing (Ackermann et al., 2007a) extended to two-sided network markets, Chap.6, interdomain routing through algorithmic mechanism design and distributed computing (Hall et al., 2007), or (security) mechanism design on trust networks (Ghosh et al., 2007).

Network Economies for the Internet

Internet Economics: Models, Mechanisms and Management 27

2.3. INTERNET RESOURCES The evolution of Internet pricing poses interesting resource allocation problems. Flat-rate pricing has been a key condition that allowed the Internet to expand very fast. In the earlier stages of Internet development flat rate pricing has been introduced earlier and was more prevalent in the US than in Europe, which, next to the Internet history originating in the US partially, explains higher user diffusion rates in the US than in (most of) Europe among private users. The expansion has been significantly facilitated through ‘bandwagon effects’ in network economies (Rohlfs, 2005; Faulhaber, 2005). The net has grown in scope and complexity, in the earlier phase sometimes but not every time and everywhere with growth rates beyond expectation (Odlyzko, 2004). With engineering advances in network (management) technologies, it become more obvious that other pricing schemes able to deal with severe congestion and deadlock would come forward. New pricing schemes should not only be able to cope with a growing Internet traffic but also be able to foster application development and deployment vital to service providers. Usage-based pricing of this new kind should make the Internet attractive to many new users. Casual users would find it more affordable, while business users would find a more stable environment. Several routes with different capacities exist between a source and a destination (Fig. 2.1). Classes of traffic compete for resources/services between the source and the destination. The packet switches contain buffer and link resources which are shared. Traffic classes l

r

k

Agents

100 Mbps, buff Packet switches

125 Mbps, buff Source

Destination

(Multimedia server Database server Web server)

150 Mbps

Fig. (2.1). Traffic classes, agents to generate multiple routes between sources and destination.

Without an incentive to economize on usage, congestion and higher exposure to security breaches could become quite serious. The problem is more serious for

28 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

data networks like the Internet than for other congestible transportation network resources because of the tremendously wide range of usage rates. A single user at a modern workstation can send a few bytes of email or put a load of hundreds Mbps (megabits per second) on the network, for example, by downloading videos demanding as much as 1 Mbps. In the 1990s, the maximum throughput on given backbones used to be only 100 Mbps, so it was clear that even a few users with relatively inexpensive equipment could block the network (Economides, 2005; Stallings, 2002, Chap.6). We witnessed these problems in the US after America Online (AOL) introduced a flat fee for its Internet services and experienced a massive and persistent breakdown with disgruntled customers. A natural response by shifting resources to expand technology will be expensive and not a satisfactory solution in the long run. Aside from breakthrough technological advances such as broadband that could mitigate but not eliminate the problem many proposals relied on voluntary efforts to control congestion. Many participants in congestion discussions suggest that peer pressure and user ethics will be sufficient to control congestion costs. But as MacKie-Mason and Varian (1995) suggest, we essentially have to deal with a problem like the ‘the tragedy of the commons’, that is overgrazing the commons, e.g. by overusing a generally accessible communication network. A few proposals would require users to indicate the priority they want each of the sessions to receive, and for routers to be programmed to maintain multiple queues for each priority class. The success of such schemes would depend on the users’ discipline to stick to the assigning of appropriate priorities to some of their traffic. On the other hand, priority scheduling flies in the face of ‘net neutrality’ now requested by law in some OECD countries, that is, that all internet traffic should be delivered at the same speed and reliability. There are no effective sanction and incentive schemes that would control such traffic, and therefore such a scheme is liable to be ineffective. This is why alternative pricing schemes have gained foremost attention and various approaches and models have been discussed in the network community (Shenker, 1995,1996; Wang et al., 1997). ATM and B-ISDN As mentioned in Section 1.9, ATM allows efficient utilization of network resources, and simplifies the network switching facilities compared to other proposed techniques in that it will only require one type of switching fabric (packet switch). This simplifies the network management process (Ackermann et al., 2007b). The basic operation of ATM, and generally of packet-switched networks, is based on statistical multiplexing. In order to provide QoS, the packets need to be served by certain scheduling (service) disciplines. Resource allocation algorithms depend heavily on the scheduling mechanism deployed. The scheduling is to be done at the entrance of the network as well as the switching

Network Economies for the Internet

Internet Economics: Models, Mechanisms and Management 29

points. The term ‘cell’ designates the fixed-size packet in ATM networks. ATM allows variable bit rate sources to be statistically multiplexed. Statistical multiplexing produces more efficient usage of the channel at the cost of possible congestion at the buffers of an ATM switch (Fig. 2.2).

Switching fabric

Output buffer

Link controller (scheduler) Output links

Fig. (2.2). Packet switch (node) with output links and output buffers.

Traffic in B-ISDN In a B-ISDN environment, high-bandwidth applications such as video, voice and data are likely to take advantage of compression. Different applications have different performance requirements, and the mechanisms to control congestion should be different for each class of traffic. Classification of these traffic types is essential in providing efficient services. There are two fundamental classes of traffic in B-ISDN: real-time and non real-time (best effort) traffic. The Internet can support data traffic well but not real-time traffic due to the limitations in the functionality of the protocols. B-ISDN needs to support both non-real-time and real-time traffic with QoS guarantees. Most data traffic requires low cell loss, but is insensitive to delays and other QoS parameters. Applications such as TELNET require a real-time response and should therefore, be considered real-time applications, the same applies to Voice of Internet Protocol (VOIP). Video is delay-sensitive and, unlike Telnet, requires high bandwidth. High throughput and low delay are required due the ATM switches for the network to support video services to the clients. This puts a constraint on the ATM switch design in that switching should be done in hardware and the buffer sizes should be kept reasonably small to prevent long delays. On the other hand, best effort traffic tends to be bursty, and its traffic characteristics are hard to predict. This puts another, opposite constraint on an ATM switch, which requires large buffers at the switching point, further complicating its design (Fig. 2.3).

30 Internet Economics: Models, Mechanisms and Management

Switch I

l i

Hans W. Gottinger

Resource partitions

l i k l i

Flow of information Content providers

k

k Class K Agent K

l i

k Switch N

Fig. (2.3). Resource partitions for K agents on N link suppliers.

Congestion Control Statistical multiplexing can offer the best use of resources, however, this is done at the price of possible congestion. Congestion in an ATM network can be handled basically in two ways: reactive control and preventive control . Reactive control mechanisms are commonly used in the Internet, where control is triggered to alleviate congestion after congestion has been detected. Typical examples of reactive control are (i) explicit congestion notification (ECN), (ii) node to node flow control and (iii) selective cell discarding. In the more advanced preventive control approach, congestion is avoided by allocating the proper amount of resources and controlling the rate of data transfers by properly scheduling cell departures. Some examples of preventive control mechanisms are (i) admission and priority control, (ii) usage parameter control and (iii) traffic shaping. Reactive and preventive control can be used concurrently, but most reactive controls are unsuitable for high-bandwidth real-time applications in an ATM network, since reactive control simply is not fast enough to handle congestion in time. Therefore, preventive control is more appropriate for high-speed networks. Service Discipline Traffic control occurs at various places in the network. First, the traffic entering the network is controlled at the input, second, the traffic is controlled at the switching nodes. In either case, traffic is controlled by scheduling the cell

Network Economies for the Internet

Internet Economics: Models, Mechanisms and Management 31

departures. There are various ways how to schedule departure times, and these mechanisms are part of service disciplines. The service discipline must transfer traffic at a given bandwidth by scheduling the cells and make sure that it does not exceed the buffer space reserved (or the delay bound assigned) for each channel. These functions are usually build into the hardware of the ATM switch and into the switch controller. When implementing a service discipline in an ATM network, it is important to choose it simple enough that it can be easily integrated into an ATM switch. However, the discipline must support the provision of quality of service guarantees. This also means that the service discipline is responsible for protecting ‘well-behaved’ traffic from the ‘ill-behaved’ traffic and must be able to provide certain levels of QoS guarantees. The service discipline also needs to be flexible enough to satisfy the diverse requirements of a variety of traffic types, and to be efficient, that is, to permit a high utilization of the network. Various service disciplines have been proposed, and many of them have been investigated thoroughly and compared. An important class is that of disciplines used in rate-allocating servers. 2.4. THE RATIONALE OF ECONOMIC MODELS IN NETWORKING There are intrinsic interfaces between human information processing and networking that show the usefulness of economic modelling. Decentralization In an economy, decentralization is provided by the fact that economic models consist of agents which selfishly attempt to achieve their goals. There are two types of economic agents: suppliers and consumers. A consumer attempts to optimize its individual performance criteria by obtaining the resources it requires, and is not concerned with system-wide performance. A supplier allocates its individual resources to consumers. A supplier's sole goal is to optimize its individual resources to consumers. A supplier's sole goal is to optimize its individual satisfaction (profit) derived from its choice of resource allocation to consumers. Models of economic agents are summarized in the Appendix. Limiting Complexity Economic models provide several interesting contributions to resource sharing algorithms. The first is a set of tools for limiting the complexity by decentralizing the control of resources. The second is a set of mathematical models that can yield several new insights into resource sharing problems.

32 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

Pricing and Performance Most economic models introduce money and pricing as the technique for coordinating the selfish behavior of agents. Each consumer is endowed with money that it uses to purchase required resources. Each supplier owns a set of resources, and charges consumers for the use of its resources. The supplier prices its resources based on the demand by the agents, and the available supply. Consumers buy resources or services such that the benefit they receive is maximized. Consumer-agents buy resources based on maximizing performance criteria. As a whole the system performance is determined by some combination of the individual performance criteria. Usage Accounting, Billing and Dimensioning By using economic models for service provisioning in distributed systems, accounting for QoS becomes an important task for suppliers, as they have to keep track of the resource usage in order to price the resources, and thereby charge or bill the users for QoS. In addition, pricing can be used to understand the user demands and thereby dimension systems appropriately. Administrative Domains Often large distributed systems and computer networks spread over several domains, the control of resources is shared by multiple organizations that own distinct parts of the network. In such an environment, each organization will have a set of services that it supports. Economic principles of pricing and competition provide several valuable insights into decentralized control mechanisms between the multiple organizations and efficient service provisioning. Scalability A key issue in designing architectures for services in large computer networks and distributed systems is scalability. With the ever growing demand for new services, flexible service architectures that can scale to accommodate new services is needed. Economic models of competition provide, in a natural fashion, mechanisms for scaling services appropriately based on service demand and resource availability. 2.5. MODELLING APPROACHES Most studies of resource allocation mechanisms have used a performance model of the resource, where the very concept of the resource is defined in terms of measurable qualities of the service such as utilization, throughput, response time (delay) and so on. Optimization of resource allocation is defined in terms of these

Network Economies for the Internet

Internet Economics: Models, Mechanisms and Management 33

measurable qualities. One novelty introduced by the economic approach is to design a system which takes into account the diverse QoS requirements of users, and therefore use multiobjective (utilities) optimization techniques to characterize and compute optimum allocations. Economic modelling of computer and communication resource sharing uses a uniform paradigm described by two level modelling: QoS requirements as inputs into a performance model that is subject to economic optimization. In the first step, one transforms QoS requirements of users to a performance (example: queueing service model). This model establishes quantifiable parameterization of resource allocation. For example, average delay QoS requirement, when based on a FIFO queueing model, is a function of resources, bandwidth and buffer, and user traffic demands. These parameters are then used to establish an economic optimization model. The question of whether the resource is a piece of hardware, a network link, a software resource such as a database or a server, or a virtual network entity such as a TCP connection is not of primary importance. The first modelling transformation eliminates the details and captures the relevant behavior and the optimization parameters. Our approach evolves in the following sequence. Many users present QoS demands, which are translated into demands on resources based on a performance model. The suppliers compute the optimal allocations based on principles of economic optimization and market mechanisms. Once the optimization is done, the results provide inputs to mechanisms for QoS provisioning, such as scheduling of resources and admission of users in networks and load balancing in distributed systems. We present briefly an overview of the problems and contributions. Optimal Allocation and QoS We motivate and solve a problem of allocating resources and providing services (QoS) to several classes of users at a single link. The resources at the link are buffer space and bandwidth. The link (network provider) prices per unit buffer and bandwidth resources. The consumers (user traffic classes), via economic agents, buy resources such that their QoS needs are satisfied. The network provider prices resources based on demand from the consumers. The ingredients are as follows: ●



Economic models: we use competitive economic models to determine the resource partitions between user traffic classes, which compete to obtain buffer and bandwidth resources from the switch suppliers. Optimal allocations using economic principles: we look for Pareto optimal allocations that satisfy QoS needs of agents. Agents represent QoS via utility

34 Internet Economics: Models, Mechanisms and Management







Hans W. Gottinger

functions which capture the multiple performance objectives. Pricing based on QoS: we compute equilibrium prices based on the QoS demands of consumers. Prices are set such that the market demand and supply are met. Prices help in determining the cost of providing a service. Priorities: using the economic framework, we show a simple way to support priority service among the user-classes (or agents). Decentralization: we show a natural separation between the interactions of the user-classes (represented by agents) and the network switch suppliers. The interaction is purely competitive and market based. This decentralization promotes scalable network system design.

Scheduling and Pricing Mechanisms We consider a dynamic system where sessions arrive and leave a traffic class, and demand fluctuates over time. In such a setting, we investigate practical mechanisms, such as packet level scheduling to provide bandwidth and buffer guarantees, admission control mechanisms to provide class QoS guarantees, practical pricing to capture the changing demand, and charging mechanisms for user sessions within a class. ●





Scheduling algorithms for class based QoS provisioning: we provide novel scheduling mechanisms, which allocates bandwidth and buffer for meeting the demand from traffic classes. The scheduling mechanism allocates bandwidth, which is computed from the economic optimization. Admission Region and Control: we compute the admission control region of the agents on the economic model. Due to the natural separation between who controls the admission of sessions into the traffic class, the admission region can be determined. Proposing simple pricing models which capture the changing demand, and are easy to implement. We also propose novel QoS based charging mechanisms for sessions in a class with applications to charging in ATM Networks and Integrated Services Internet.

We first consider a network economy, of many parallel routes or links, where several agents (representing user classes) compete for resources from several suppliers, where each supplier represents a route (or a path) between a source and destination. Agents buy resources from suppliers based on the QoS requirements of the class they represent. Suppliers price resources, independently, based on demand from the agents. The suppliers connect consumers to information providers, who are at the destination; the flow of information is from information providers to the consumers. We formulate and solve problems of resource allocation and pricing in such an environment.

Network Economies for the Internet

Internet Economics: Models, Mechanisms and Management 35

We then consider a server economy in a distributed system. Again, we use a similar model of interaction between agents and suppliers (servers). The servers sell computational resources such as processing rate and memory to the agents for a price. The prices of resources are set independently by each server based on QoS demand from the agents. Agents represent user classes such as transactions in database servers or sessions for Web servers that have QoS requirements such as response time. Using such economic models, our contributions are as follows: ●





We propose a decentralized model of network and server economies, where we show efficient QoS provisioning and Pareto allocation of resources (network and server resources) among agents and suppliers, which are either network routes or servers (content providers). We show how prices for resources are set at the suppliers based on the QoS demands from the agents. We propose alternative dynamic routing algorithms and admission control mechanisms based on QoS preferences by the user classes for the network economy, and we propose a simple way to perform transaction routing. We also show static optimal routing policies by the agents for the network and server economies.

Network and Server Economies Consider a large scale distributed information system with many consumers and suppliers. Suppliers are content providers such as web servers, digital library servers, multimedia database and transaction servers. Consumers request for and access information objects from the various suppliers and pay a certain fee or no fee at all for the services rendered. Consider that third party suppliers provide information about suppliers to consumers in order to let consumers find and choose the right set of suppliers. Access and dissemination: consumers query third-party providers for information about the suppliers, such as services offered and the cost (price). Likewise, suppliers advertise their services and the costs via the third party providers in order to attract consumers. Consumers prefer an easy and simple way to query for supplier information, and suppliers prefer to advertise information securely and quickly across many regions or domains. For example, consider a user who wishes to view a multimedia object (such as a video movie). The user would like to know about the suppliers of this object, and the cost of retrieval of this object from each supplier. Performance requirements: users wish to have good response time for their search results once the queries are submitted. However, there is a tradeoff. For more

36 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

information about services offered, advanced searching mechanisms are needed, but at the cost of increased response time. In other words, users could have preferences over quality of search information and response time. For example, users might want to know the service costs in order to view a specific information object. In large networks, there could be many suppliers of this object, and users may not want to wait forever to know about all the suppliers and their prices. Instead, they would prefer to get as much information as possible within a certain period of time (response time). From the above example, in order to let many consumers find suppliers, a scalable decentralized architecture is needed for information storage, access and updates. Naming of services and service attributes of suppliers becomes a challenging issue when hundreds of suppliers spread across the globe. A simple naming scheme to connect consumers, across the internet, with information about suppliers is essential. The naming scheme must be extensible for new suppliers who come into existence. A name registration mechanism for new suppliers and a de-registration mechanism (automatic) to remove non-existent suppliers is required. In addition, naming must be hierarchical, domain based (physical or spatial domains) for scalability and uniqueness. Inter-operability with respect to naming across domains is an additional challenging issue not covered here. The format of information storage must be simple enough to handle many consumer requests quickly within and across physical domains. For better functionality and more information, a complex format of information storage is necessary, but at the cost of reduced performance. For example, a consumer, in addition to current service cost, might want to know more information such as the cost of the same service during peak and off-peak hours, the history of a supplier, its services, and its reputation, in order to make a decision. This information has to be gathered when requested. In addition, the storage formats must be interoperable across domains. Performance: a good response time is important to make sure consumers get the information they demand about suppliers within a reasonable time period, so that decision-making by consumers is done in a timely fashion. In addition, the design of the right architectures for information storage and dissemination is necessary for a large scale market economy to function efficiently. Using the previous example, consumers and suppliers would prefer an efficient architecture to query for and post information. Consumers would prefer good response time in obtaining the information, and suppliers prefer a secure and fast update mechanism to provide up-to-date information about their services.

Network Economies for the Internet

Internet Economics: Models, Mechanisms and Management 37

Security in transferring information and updating information at the bulletin boards (name servers) is crucial for efficient market operation and smooth interaction between consumers and suppliers. For this the third party suppliers (naming services) have to provide authentication and authorization services to make sure honest suppliers are the ones updating information about their services. Allocation and Pricing Models In economic models, there are two main ways to allocate resources among the competing agents. One of them is the exchange based economy and the other is the price based economy. In the exchange based economy, each agent is initially endowed with some amounts of the resources. They exchange resources until the marginal rate of substitution of the resources is the same for all the agents. The agents trade resources in the direction of increasing utility (for maximal preference). That is, two agents will agree on an exchange of resources (e.g. CPU for memory) which results in an improved utility for both agents. The Pareto optimal allocation is achieved when no further, mutually beneficial, resource exchanges can occur. Formally, an allocation of resources is Pareto optimal when the utility derived by the competing economic-agents is maximum. Any deviation from this allocation could cause one or more economic agents to have a lower utility (which means the agents will be dissatisfied). In a price based system, the resources are priced based on the demand, supply and the wealth in the economic system. The allocations are done based on the following mechanisms. Each agent is endowed with some wealth. Each agent computes the demand from the utility function and the budget constraint. The aggregate demand from all the agents is sent to the suppliers who then compute the new resource prices. If the demand for a resource is greater than its supply, the supplier raises the price of the resource. If there is surplus supply, the price is decreased. The agents again compute their demands given the current prices and present the demand to the suppliers. This process continues iteratively until the equilibrium price is achieved where demand equals the supply (Fig. 2.4). Bidding and auctioning resources is another form of resource allocation based on prices. There are several auctioning mechanisms such as the Sealed Bid Auction, Dutch Auction, and English Auction. The basic philosophy behind auctions and bidding is that the highest bidder always gets the resources, and the current price for a resource is determined by the bid prices.

38 Internet Economics: Models, Mechanisms and Management (a)

Hans W. Gottinger (b)

Bulletin board

Tatonnement process Agents

Agent 1 Advertise prices

Agent 2 Demand

Agent I

Demand Resources are priced Switch supplier

New prices Advertise

Supplier

Bulletin board

Agent K

Fig. (2.4). Economic players: K agents competing for resources from a supplier.

Specific Problems of Economic Resource Allocation Some of the interesting problems encountered when designing an economic based computer network are discussed and stated: ●











How do agents demand resources? This is a fundamental question regarding the agents preferences on the resources they consume. Are there smooth utility functions that can capture the agents preferences of the resources? Are there utility functions that can capture the diversity of the agents preferences? How are the prices adjusted to clear the economy or to clear the markets? In an economic model, efficient allocation of resources occurs when the demand equals the supply at a certain equilibrium price vector. What rational pricing mechanisms do the suppliers adopt? This question raises issues on pricing mechanisms that will attract agents (consumers). How do suppliers provide price guarantees to agents? This is a fundamental question in advertising and providing price guarantees to agents. Delays in information about prices, and surges in demand can cause prices to vary. Therefore agents can make bad decisions. What are the protocols by which the consumers and suppliers communicate to reserve resources? What are the general allocation principles? Can economic models give insight into the allocation mechanisms that can cause the computer system to reach equilibrium? Can these principles be used practically to evolve the computer system in a way that price equilibrium can be achieved?

Network Economies for the Internet

Internet Economics: Models, Mechanisms and Management 39

2.6. NETWORK ECONOMY On the basis of the general specifications laid out in Sec. 2.2 we proceed to a problem formulation for a network model. The network is composed of nodes (packet switches) and links. Each node has several output links. Each output link is associated with an output buffer. The link controller, at the output link, schedules packets from the buffers and transmits them to other nodes in the network. The switch has a buffer controller that can partition the buffer among the traffic classes at each output link. We assume that a processor on the switch aids in control of resources. We have confined ourselves to problems for a single link (output link) at a node, but they can be applied to the network as well. Let B denote the output buffer of a link and C be the corresponding link capacity. Let {ck,bk}be the link capacity and buffer allocation to class k on a link, where k ε [1,K]. Let p = {pc,pb} be the price per unit link capacity and unit buffer at a link, and wk be the wealth (budget) of traffic class k. The utility function for TCk is Uk = f (ck,bk,Trk). The traffic of a class is represented by a vector of traffic parameters (Trk) and a vector of QoS requirements (such as packet loss probabilities, average packet delay and so on.). Agent (TC: traffic class) buys resources from the network at the given prices using its wealth. The wealth constraint of agent TCk is: pb * bk + pc *ck ≤ wk. A budget set is the set of allocations that are feasible under the wealth constraint (budget constraint). The budget set is defined as follows: B(p) = (x : x ε X, p x ≤ wk}

(1)

Computation of demands sets: The demand set for each agent is given by the following: Φ (p) = ( x : x ε B(p) ,U(x’), ∀x’ ε B(p)}

(2)

The goal of TCk is to compute the allocations that provide maximal preference under wk and p. Each TCk performs the following to obtain the demand set (defined above): Find: {c b } k,

such that: max Uk= f(ck,bk,Trk)

k

(3)

40 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

Constraints: pk bk + pc ck ≤ wk ck ε [0,C], bk ε [0,B] Utility Parameters For every traffic class we show a general utility function which is a function of the switch resources; buffer (b) and bandwidth (c). The utility function could be a function of the following: ● ● ● ● ●

Packet loss probability Ut = g (c,b,Tr) Average packet delay Ud = h (c,b,Tr) Packet tail utility Ut = v (c,b,Tr) Max packet delay Ub = f (b,bT) Throughput Uc = g (c,cT)

The variables b and c in the utility functions refer to buffer space allocation and link bandwidth allocation. In the utility functions Ub and Uc; the parameters bT and cT are constants. For example, the utility function Ub = f (b,bT) for max packet delay is simply a constant as b increases, but drops to 0 when b = bT and remains zero for any further increase in b (Fig. 2.5a and b). (a)

(b)

Max delay constraint

QoS constraint Buffer

Loss probability contraint

QoS set Maximal QoS

Buffer

C

QoS allocation A

Wealth line Link capacity

B Link capacity

Fig. (2.5). Agent QoS set, given constraint (a) Loss probability and max delayconstraints (b).

We look at utility functions which capture packet loss probability of QoS requirements by traffic classes, and we consider loss, max-delay and throughput requirements. After this we proceed to utility functions that capture average delay requirements, followed by utility functions that capture packet tail utility

Network Economies for the Internet

Internet Economics: Models, Mechanisms and Management 41

requirements. We also could give examples of utility functions for agents with many objectives; agents have preferences over several QoS parameters: U =  ( U1, U2, …) 

(4)

Packet Loss The phenomenon of packet loss is due to two reasons: the first, packets arrive at a switch and find that the buffer is full (no space left), therefore, are dropped. The second is that packets arrive at a switch and are buffered, but they do not get transmitted (or scheduled) in time, then they are dropped. A formal way of saying this: for real-time applications, packets, if delayed considerably in the network, do not have value once they reach the destination. In this context queueing modelling comes to the fore as the invention of the Swedish engineer A.K. Erlang when he considered such a system as one model for the behavior of telephone systems in the early twenty century. We consider K agents, representing traffic classes of Markovian queueing type M/M/1/B in which we have exponential interarrival times, exponential service times and a single server with a finite buffer B competing for resources from the network provider. The utility function is packet loss utility (U1) for the user classes. We choose the M/M/1/B model of traffic and queueing for the following reasons: ●



The model is tractable, where steady state packet loss utility is in closed-form, and differentiable. This helps in demonstrating the economic models and the concepts. There is continuous interest in M/M/1/B or M/D/1/B models for multiplexed traffic (such as video), where simple histogram based traffic models capture the performance of queueing in networks (Kleinrock, 1976; Kleinrock and Gail, 1996).

For more complex traffic and queueing models (example of video traffic) we can use tail utility functions to represent QoS of the user class instead of loss utility. In the competitive economic model, each agent prefers less packet loss, as the more packet loss, the worse the quality of the video at the receiving end. Let each agent TCk have wealth wk, which it uses to purchase resources from network provider. Let each TC transmit packets at a rate λ (Poisson arrivals), and let the processing time of the packets be exponentially distributed with unit mean. Let c,b be allocations to a TC. The utility function U for each TC is given as follows:

42 Internet Economics: Models, Mechanisms and Management

⎧ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ (1 − λ )( λ )b ⎪ c c U = f (c, b, λ ) = ⎨ ⎪1 − ( λ ) + 1 + b c ⎪ ⎪ 1 ⎪ b + 1 ⎪ ⎪ ( −1+ )( λ )( λ ) b c c ⎪ ⎪ − 1 + ( λ )1 + b ⎪⎩ c

Hans W. Gottinger

for λ< c ,λ=c, λ>c resp.

(5)

The above function is continuous and differentiable for all c ε [0,C], and for all b ε [0,B]. We assume that b ε R for continuity purposes of the utility function. 2.7. EQUILIBRIUM PRICE AND CONVERGENCE Pareto efficient allocations are such that no traffic class can improve upon these allocations without reducing the utility of one or more traffic classes. The more formal definition and implication of Pareto efficiency in this context can be taken from Hurwicz (1986). The set of Pareto allocations that satisfy the equilibrium conditions forms the Pareto surface. Each agent computes the demand set, which is a set of allocations, that maximizes the preference under the wealth constraint. The demand set is obtained by minimizing the utility function under the budget constraint. The Lagrangian is given below with L as the Lagrange multiplier. min[f(c,b,λ) – L ∗(pc ∗c + pb - w)]

(6)

c ≥ 0, b ≥ 0 The function f(c,b,λ) is smooth, strictly convex and compact, thus the demand set is just one element [1,2]. Using the Kuhn-Tucker optimality conditions, the optimal resource allocation vector is obtained, where L is the Lagrange multiplier

Network Economies for the Internet

Internet Economics: Models, Mechanisms and Management 43

∂U ∂U = L ∗ pc , = L ∗ pb ∂c ∂b

(7)

From this the equilibrium condition is obtained. This condition states that the marginal rate of substitution is equal among the competing traffic classes. The economic problems are to establish competitive equilibrium, compute the equilibrium prices p*c, p*b and Pareto optimal allocations. The equilibrium condition is shown as follows: ∂U / ∂c p *c = ∂U / ∂b p b*

(8)

Using the utility function given by equation (5), the price ratio is given below. From the equation, it is evident that the price ratio is a function of the resource allocations and the traffic parameter λ.

pc = g (λ, c, b) = pb

λb − λb + cb c λ (λ − c)c log( ) c

−λ+λ

(9)

This equation can be rewritten in the following way, where function N has a nice interpretation. It is the ratio of the effective queue utilization (ρ(1-U)) to the effective queue emptiness (1 – ρ *(1-U)), where ρ = λ/c.

pc N −b ρ ∗ (1 − U ) = where N = pb c log ρ 1 − ρ ∗ (1 − U )

(10)

This can also be interpreted as the effective number in an equivalent M/M/1 queueing system, where the system utilization is ρ = ρ (1-U). For an M/M/1 ρ . system, the average number in the system is 1− ρ The following gives the equilibrium condition for K agents competing for resources from a single network provider. From this condition, and the resource constraints. the Pareto allocations and the corresponding equilibrium price ratios can be computed.

44 Internet Economics: Models, Mechanisms and Management

-

Hans W. Gottinger

pc b1 − N 1 b2 − N 2 bk − N K = = = ... = pb c1 log( ρ1) c 2 log( ρ 2) cK log( ρK )

(11)

Using the buffer constraint b1 + b2 + b3+ ...+ bk = B, the equilibrium price ratio and optimal buffer allocation for each agent i can be represented by the following equations: p ∗c = p b∗

b i = Ni -

∑ N −B ∑ (c log( ρ )) i

i

i

i

(∑ i N i − B)(c i log( ρ i ))



i

(12)

i

(c i log( ρ i ))

)

(13)

The issue of determining the equilibrium prices, so that supply equal demand for different types of utility functions can use convex optimization tools (Chen, Ye and Zhang, 2007). Fig. (2.6) next shows the tradeoff allocations in an Edgeworth Box Diagram. Class (2) Pareto surface Iso-quants for class (2) B Buffer A

Iso-quants for class (1) Wealth line

Class (1)

Link capacity

Fig. (2.6). Edgworth Box diagram. Link capacity along the x-axis and buffer space alongthe y-axis. TC1 and TC2 compete for resources at both ends of the resource box. Any point (c, b) in the box represents an allocation {c, b} to class (1) and anallocation {C – c, B – b} to class (2). The wealth line is the wealth constraintof both the TCs. Isoquants are contours of a function f (c, b, λ) = β(constant).

Network Economies for the Internet

Internet Economics: Models, Mechanisms and Management 45

Competitive Pricing Algorithm (CPA) 1. Set initial prices: pc = poc, pb = pob 2. Compute demand set, ie. find minimum of [Ui = fi(ci,bi, i)] i [1,K] given pcci+pbbi ≤ wi (wealth constraint) 3. Demand: Dc =

∑ ii==1K

ci,Db =

∑ ii==1K b . i

4. If (Dc - C) ) 0, then, pc = pc – (+)∆c If (Db - B) )0, then, pb = pb – (+)∆b Go back to step (2) 5. Else if D c = C at p c, and D b = B at p b, then the equilibrium is attained and prices are at equilibrium The algorithm computes iteratively the equilibrium prices in a competitive economy using the utility functions. Given the wealth in the economy, the prices converge to a point on the Pareto surface, which can be computed using the firstorder conditions. There is a minimum price pε, that each traffic class has to pay, if the equilibrium prices are lower than p i. Once the prices are computed, the network service provider releases the resources to the agents. 2.8. EXAMPLE OF TWO AGENTS AND ONE SUPPLIER We consider two agents, representing traffic classes of the M/M/1/B model. The utility function is shown in equation (5). The agents have wealth w1 and w2 respectively. The agents compete for resources, which then are used to provide services to users. Two Classes: We consider two competing traffic classes. Using the equilibrium conditions, the equilibrium price ratio is given by: P c∗ N 1−b1 N 2−b 2 = = ∗ P b c1 log ρ 1 c 2 log ρ 2

(14)

The above equation states that at equilibrium, the log of the ratio of utilizations of the two traffic classes is equal to the ratio of the time to evacuate the residual buffer space of the traffic classes. Rewriting the above equation:

46 Internet Economics: Models, Mechanisms and Management

log ρ 1 N 1−b1 c2 . = N 2−b 2 log ρ 2 c1

Hans W. Gottinger

(15)

By using the resource constraints c1 + c2 = C and b1 + b2 = B, the equilibrium conditions become a function of just two variables. The Pareto surface is the set of allocations that satisfy equation (14). The function Ni and Ui (for all i ε {1,2}) have several interesting properties for different values of ρi. We study the properties of these functions for various regions of ρ1 and ρ2, where ρ1 and ρ2 are utilizations of TC1 and TC2 respectively. ●





ρ1 1, ρ2 >1: The allocated capacity is less than the mean rates of TC1 and TC2. We consider the case where the buffer tends to infinity. limb1→∞ N1 = ∞, limb1→∞ U1 = 0. The quantity b1 – N1 < 0 for b1 ε[0, ∞). ρ1 → 1, ρ2 → 1: The quantity is N1 = b1, N2 = b2. The equilibrium condition for offered loads equal to 1 is

b1 ∗(b1 +1) b 2 ∗(b 2 +1) = . 2 *λ 1 2 *λ 2

Several other cases such as ρ1 > 1, ρ2 = 1 are omitted, but are essential in determining the Pareto surface. For the two competing traffic classes, the following relation between the utility functions of the traffic classes with respect to the Pareto optimal allocations is obtained:

log U 1 (b1 − N 1 ) c2 = . log U 2 c1 (b 2 − N 2 )

(16)

log U 1 (b1 − N 1 ) c 2 = . log U 2 (b 2 − N 2 ) c1

(17)

This relation has an interesting physical meaning: The loss of the ratio of the utilities of the traffic classes is equal to the ratio of the time to evacuate the

Network Economies for the Internet

Internet Economics: Models, Mechanisms and Management 47

residual buffer in the queues. The residual buffer is simply: bi – Ni, where Ni is given by equation (10). CONCLUSION We demonstrate the application of economic tools to resource management in distributed systems and computer networks. Concepts of analytical economics were used to develop effective market based control mechanisms, and to show the allocation of resources are Pareto optimal. As applications of market based control in Clearwater (1996) convincingly show how market based mechanisms facilitate resource with very little ‘centralized’ information. Methodologies of decentralized control of resources, and pricing of resources are based on QoS demand of users. We bring together economic models and performance models of computer systems into one framework to solve problems of resource allocation and efficient QoS provisioning. Such a scheme can be applied to pricing services in ATM networks and Integrated Services Internet of the future. There are drawbacks to this form of modeling where several agents have to use market mechanisms to decide where to obtain service (which supplier?). If the demand for a resource varies substantially over short periods of time, then the actual prices of the resources will also vary causing several side effects such as indefinite migration of consumers between suppliers. This might potentially result in degradation of system performance where the resources are being underutilized due to the bad decisions (caused by poor market mechanisms) made by the users in choosing the suppliers. Unlike economies, the resources in a computer system are not easily substitutable. The future work is to design robust market mechanisms and rationalized pricing schemes which can handle surges in demand and variability, and can give price guarantees to consumers over longer periods of time some of which have been discussed by Spulber an Yoo (2009, Chap.12). Another drawback is that resources in a computer system are indivisible resulting in non-smooth utility functions, which may yield sub-optimal allocations, and potential computational overhead. In summary, economic models are useful for designing and understanding computer network systems. The Internet currently connects hundreds of millions of users and thousands of sites. Several services exist on many of these sites, notably the World Wide Web (WWW) which provides access to various information sources distributed across the Internet. Many more services

48 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

(multimedia applications, commercial transactions) are to be supported in the Internet. To access this large number of services, agents have to share limited network bandwidth and server capacities (processing speeds). Such large-scale networks require decentralized mechanisms to control access to services. Economic concepts such as pricing and competition can provide some solutions to reduce the complexity of service provisioning and decentralize the access mechanisms to the resources. APPENDIX A. The Network Economy The network consists of V nodes (packet switches) and N links. Each node has several output links with an output buffer. The resources at output link are transmission capacity (or link capacity) and buffer space. The link controller at the output link schedules packets from the buffer. This is based on how the buffer is partitioned among the traffic classes and the scheduling rule between the traffic classes. Sessions are grouped into traffic classes based on similar traffic characteristics and common QoS requirements. Sessions that belong to a class share buffer and link resources, and traffic classes compete for resources at a packet switch. Each session arrives to the network with a vector of traffic parameters Tr, vector of QoS requirements and wealth. A session is grouped or mapped to a corresponding traffic class. A traffic class has common QoS requirements, and we consider QoS requirements per traffic class rather than per session. Once a session is admitted along a path (a route), it will continue along that path until it completes. Each agent k performs the following to obtain the demand set on each link. The allocations are buffer (b) and bandwidth (c) on each link for each agent. The wealth is distributed across the links by each agent to buy resources. That is, the problem is to find pairs {c*k, b*k} such that max Uk = f(ck,bk,Trk), constraints pbbk + pcck ≤ wk. In the above formulation, each agent k buys resources from each link. The allocation for agent k is c*k = { ck*1, ck*2, ..., ck*N } and b*k = { bk*1, bk*2, ..., bk*N }. An agent can invest wealth in either some or all the links. We assume that at each link there is competition among at least some of the agents for buying resources. As previously, Gottinger (2009), we show a general utility function which is a function of the switch resources: buffer (b) and bandwidth (c) . A utility function of the agent could be a function of: • Packet loss probability Ut = g (c,b,Tr)

Network Economies for the Internet

Internet Economics: Models, Mechanisms and Management 49

• Average packet delay Ud = h (c,b,Tr) • Packet tail probability Ul = v (c,b,Tr) • Max packet delay Ub = f (b) • Throughput Uc = g (c) We consider that an agent will place demands for resources based on a general utility function, which is a combination of the various QoS requirements: U = ϕ(c,b,Tr) = x1Ul + xdUd + xbUb + xcUc + xtUt

where Ul is the packet loss probability utility function, Ud is the average delay utility function, Ut is the packet tail probability, Ub is the utility function for maxdelay requirements, and Uc is for bandwidth (throughput) requirements. x1, xd, xb, xc, xt are constants. Agents could use such a utility function . As long as the convexity property with respect to buffer b and bandwidth c holds. Pareto optimal allocations and price equilibria exist. However, if they are not convex, then depending on the properties of the functions, local optimality and price equilibrium could exist. To show the main ideas for routing and admission control, we use packet loss probability as the main utility function (Ul), which means we assume that x1 from the above equation are the only constant and the rest are zeros. For doing this, we need first some further specifications of the loss probability. We later show results for Pareto optimality and price equilibrium, and then we propose routing and admission control algorithms. In general, one can assume that agents, on behalf of user classes, demand for resources from the link suppliers based on the utility function shown above. The agent uses the utility function to present the demand for resources over the whole network of parallel links. Loss Probability Specifications. At each output link j the resources are buffer space Bj and link capacity Cj . Let {cjk, bjk} be the link capacity and buffer allocation to class k on link j where k ε [1,K] . Let pjc and pjb be the price per unit link capacity and unit buffer respectively at link j, and wk be the wealth (budget) of a traffic class k. For a link j from the source to the destination, the packet loss probability (utility) for traffic class k is given by the following: Ulk = Ploss = 1 − Πj=1N (1 − Pkj)

50 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

where Pkj is the packet loss probability at link j of agent k. The goal of the agent is to minimize the packet loss probability under its wealth or budget constraints . If the traffic classes have smooth convex preferences 1 with respect to link capacity and buffer allocation variables at each link, then the utility function Ulk is convex with respect to the variables. B. The Server Economy We now discuss the server economy where servers offer processing resources and memory to agents representing user classes. The agents compete for these resources and buying as much as possible from suppliers. The agents perform load balancing based on the QoS preferences of the class it represents. The economic model consists of the following players: Agents and Server Suppliers, Consumers or user classes and Business. User sessions within a class have common preferences. User classes have QoS preferences over average delay and throughput, and in some cases completion times of sessions (deadlines). Users within a class share resources at the servers. Agents and Network Suppliers: Agents represent user classes. An agent represents a single user class. Agents negotiate with the supplier and buy resources from service providers. Agents on behalf of user classes demand for resources to meet the QoS needs. Suppliers compete to maximize revenue. Suppliers partition and allocate resources (processing rate and memory) to the competing agents. Multiple Agent Network Supplier Interaction: Agents present demands to the suppliers . The demands by agents are based upon their wealth and user class preferences. The demand by each agent is computed via utility functions which represent QoS needs of the class. Agents negotiate with suppliers to determine the prices . The negotiation process is iterative where prices are adjusted to clear the market. Price negotiation could be done periodically or depending on changes in demand. The agent and network supplier become service providers in the market. The role of the supplier is to provide technologies to sell resources (buffer and bandwidth units) and to partitioning them flexibly based on the demand by the agents. The agents transform the goods (buffer and bandwidth) and provide QoS levels to the user-classes. The agents strive to maximize profits (minimize buying costs) by using the right utility functions and the right performance models in order to provide QoS to the user-class. More users within a user-class implies more revenue for the agent. The agent is decoupled from the traffic class and the supplier .

Network Economies for the Internet

Internet Economics: Models, Mechanisms and Management 51

In this economy, user classes are transaction classes that send transactions to database servers for processing . The transaction processing time at each of the server is based on the type of transaction . Consider K classes of transactions and each class is represented by an agent (economic agent). In the economy, the agents negotiate with the servers for server capacity. We assume that transactions of any class can run on any of the database servers. Therefore, agents negotiate with all the servers for server thruput (or processing speed). A model where K agents compete for services in a transaction processing system, each class could do the following based on its preferences on average delay and thruput: (i) each agent i can minimize its average response time under throughput constraints, (ii) each agent i can maximize throughput of its transactions under an average delay constraint, (iii) each agent i can look at a combination of QoS requirements and have preferences over them. Therefore, each class can choose either one of these preferences and let the agent control the flow of transactions through the system. The problem now becomes a multi-objective optimization problem as every agent is trying to maximize its benefit in the system based on the class of QoS preferences. Consider that the classes wish to choose various objectives, the the utility function assumes U = xdUd + xlUl where Ud is the utility function for average delay and Ul is the utility function for throughput, and xd and xl are constants. Consider that there are requirements for transaction completion time. Instead of scheduling transactions to meet deadlines, we try to minimize the number of transactions that have missed the deadlines (in a stochastic sense). Consider that each transaction class is assigned a service queue at each server, then we try to minimize the probability of the number of transactions of a class exceeding a certain threshold in the buffer. This is the tail probability P(X >b) where X is the number of transactions of a class in a queue at a server, and b is threshold is threshold for the number in the queue, beyond which transactions miss deadlines. If we include this QoS requirement, then the above utility function will be U = xdUd + xlUl + xtUt where Ut is the tail probability utility function and xt is a constant. Pareto Optimality: We now have a simple formulation for classes competing for server capacity (processing rate) in order to minimize average delay (or average response time). The utility function is simply U = xd Ud as the rests of the constants are zero. Let pj be the price per unit processing rate at server j. The maximum processing rate at server j is Cj . The problem therefore for each agent is the following: find {cij*} such that min Ud = { ∑j 1N Wij } with constraints ∑j 1N λij = γi ∀i , ∑1N cij ∗ pj ≤ wi ∀i =

=

52 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

In the above problem definition, each agent will try and minimize the utility function under the wealth conbstraint and under the throughput constraint. This constraint is necessary to make sure that positive values of throughput are obtained as a result of the optimization. The transaction agents compete for processing rate at each server, and transaction servers compete for profit. The objectives of the transaction classes are conflicting as they all want to minimize their average response time. In the above formulation Wij = λij / (cij − λij) this is the average number of class i transactions in queue at system j. The average delay in the system for each class i is simply the average number in the system divided by the overall throughput ∑j=1N λij. The main goal of the agent representing the transaction class is to minimize a utility function which is simply the average number in the overall system . This will also minimize the average delay or average response time of the transaction class. Proposition 1. The utility function Ud is convex with respect to the resource allocation variable cij where λij ∈ [0,cij) , and cij ∈ (0, Cj] The proof follows from Gottinger (2009). The utility function Ud is discontinuous when λij = cij . Demand Set: The demand set for an agent i, given the prices (pj of server j) of the processing rates (or capacities) at the servers is { ci1, ci2, ..., ciN } over all the servers. We use the standard techniques of optimization to find the demand set, which is given as follows for all j ∈ [1,N] cij = λij + ((wi − ∑j=1N λij pj) / ∑j=1N√ λij pj) √ λij / pj

Price Equilibrium: Once the demand set is obtained, then using the wealth constraints, we can solve for the equilibrium price . This is not easily tractable. However, numerical results can be computed using the tatonnement process whereby agents compute the demand set, given the processing rate prices by each server. An iteration process between the agents and the servers takes place. This will converge to an equilibrium price, when demand equals the supply which is ∑i =1K cij = Cj. We now state formally the result for K agents competing for processing resources from N servers.

Network Economies for the Internet

Internet Economics: Models, Mechanisms and Management 53

Proposition 2 Consider K agents competing for processing resources from N servers. If the utility function of these agents is Ud and the performance model at the servers is an M/M/1 model, then price equilibrium and Pareto optimality exist. The proof of this proposition is the same as described in Gottinger (2009). The utility function Ud is continuous and decreasing convex with respect to the allocation variables cij. The function is discontinuous when λij = cij. Due to this, Pareto allocations or price equilibrium may not exist. However, we solve this problem by stating that the agents, when they present their demands, have to make sure that the transaction throughput rate λij at a server has to be lower than the capacity allocation cij. If this is not met, then the price iteration process or the tatonnement process will not converge. We assume that the servers know the transaction thruput or arrival rate from each agent during the iteration process. REFERENCES Ackermann, H., Goldberg, P.W., Mirrokni, V.S., Röglin, H., Vöcking, B. (2007a). A Unified Approach to Congestion Games and Two-Sided Markets In: Deng, X., Graham, F.C., (Eds.), Third International Workshop, WINE 2007, Berlin, New York: Springer.30-41. [http://dx.doi.org/10.1007/978-3-540-77105-0_7] Ackermann, H., Briest, P., Fanghänel, A., Vöcking, B. (2007b). Who Should Pay for Forwarding Packets In: Deng, X., Graham, F.C., (Eds.), Third International Workshop, WINE 2007, Berlin, New York: Springer.208-219. Albach, H. (1993). Zerrissene Netze (Torn Networks). Berlin: Wissenschaftszentrum Berlin. Arrow, K.J., Debreu, G. (1954). Existence of an Equilibrium for a Competitive Economy. Econometrica, 22(3), 265-290. [http://dx.doi.org/10.2307/1907353] Chen, L., Ye, Y., Zhang, J. (2007). A Note on Equilibrium Pricing as Convex Optimization In: Deng, X., Graham, F.C., (Eds.), Third International Workshop, WINE 2007, San Diego, Berlin, New York: Springer.716. [http://dx.doi.org/10.1007/978-3-540-77105-0_5] Clearwater, S.H. (1996). Market-Based Control, Singapore: World Scientific. Deng, X., Graham, F.C. (2007). Internet and Network Economics Third International Workshop, WINE 2007San Diego, Berlin, New York: Springer. Economides, N. (2005). The Economics of the Internet Backbone In: Majumdar, S.K., Vogelsang, I., Cave, M.E., (Eds.), Handbook of Telecommunications Economics Chapter 9. Vol. 2 Technology Evolution and the Internet.(pp. 375-412). Amsterdam: North Holland: Faulhaber, G.R. (2005). Bottlenecks and Bandwagons: Access Policy in the New Telecommunications In: Majumdar, S.K., Vogelsang, I., Cave, M.E., (Eds.), Handbook of Telecommunications Economics Chapter 12. Vol. 2 Technology Evolution and the Internet.(pp. 488-516). Amsterdam: North Holland: Ferguson, D.F., Nikolaou, C., Sairamesh, J., Yemini, Y. (1996). Economic Models for Allocating Resources in Computer Systems In: Clearwater, S.H., (Ed.), Market-Based Control Chapter 7. Vol. 2 World Scientific.(pp. 156-183). Singapore: Ghosh, A., Mahdian, M., Reeves, D.M., Pennock, D.M., Fugger, R. (2007). Mechanism Design on Trust

54 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

Networks In: Deng, X., Graham, F.C., (Eds.), Third International Workshop, WINE 2007, Berlin, New York: Springer. 257-268. Gottinger, H.W. (2009). Strategic Economics in Network Industries. New York: NovaScience. Hall, A., Nikolova, E., Papadimitriou, C. (2007). Incentive-Compatible Interdomain Routing with Linear Utilities In: Graham, F.C., (Ed.), Third International Workshop, WINE 2007, Berlin, New York: Springer.232-244. Hayzelden, A.L., Bigham, J. (1999). Software Agents for Future Communications Systems. Berlin: Springer. [http://dx.doi.org/10.1007/978-3-642-58418-3] Hurwicz, L. (1986). On Informational Decentralization and Efficiency in Resource Allocation Mechanism In: Reiter, S., (Ed.), MAA Studies in Mathematics 25 The Mathematical Association of America. Hurwicz, L., Reiter, S. (2006). Designing Economic Mechanisms. Cambridge: Cambridge Univ. Press. [http://dx.doi.org/10.1017/CBO9780511754258] Iong, S., Man-Cho-So, A., Sundararajan, M. (2007). Stochastic Mechanism Design In: Deng, X., Graham, F.C., (Eds.), Third International Workshop, WINE 2007, San Diego, Berlin, New York: Springer.269-280. Kleinrock, L. (1976). Queueing Systems (Vol. 2). Computer Applications, New York: Wiley-Interscience. Kleinrock, L., Gail, R. (1996). Queueing Systems, Problems and Solutions. New York: Wiley-Interscience. Low, S., Varaiya, P. (1993). A New Approach to Service Provisioning in ATM Networks IEEE Trans. on Networking. MacKie-Mason, J.K., Varian, H.R. (1995). Pricing the Internet Public Access to the Internet (pp. 269-314). Cambridge, Ma.: MIT Press. Meinel, C., Tison, S. (1999). STACS 99, 16th Annual Symposion Theoretical Aspects of Computer Science. Trier, Germany, Berlin: Springer. Mount, K.R., Reiter, S. (2002). Computation and Complexity in Economic Behavior and Organization. Cambridge: Cambridge Univ. Press. [http://dx.doi.org/10.1017/CBO9780511754241] Narahari, Y., Garg, D., Narayanam, R., Prakash, H. (2009). Game Theoretic Problems in Network Economics and Mechanism Design Solutions. London: Springer. Neumann, D., Baker, M., Altmann, J., Rana, O.F. (2010). Economic Models and Algorithms for Distributed Systems. Basel: Birkhaeuser. [http://dx.doi.org/10.1007/978-3-7643-8899-7] Nisan, N., Ronen, A. (2001). Computationally Feasible VCG Mechanisms. Games Econ. Behav, 35, 166196. [http://dx.doi.org/10.1006/game.1999.0790] Odlyzko, A.M. (2004). Intermet traffic growth: Sources and implications Univ. of Minnesota, 1-15. Available at: www.dtc.umn.edu/odlyzko Paxon, V. (1994). Growth Trends in TCP/IP IEEE Communications Magazine. Radner, R. (1987). Decentralization and Incentives Information, Incentives and Economic Mechanisms (pp. 3-47). Minneapolis: University of Minnesota Press. Radner, R. (1993). The Organization of Decentralized Information Processing. Econometrica, 62, 11091146. [http://dx.doi.org/10.2307/2951495] Robertazzi, Th.G. (1994). Computer Networks and Systems, Queueing Theory and Performance Evaluation (2nd ed.). New York: Springer. Rohlfs, J.H. (2005). Bandwagon Effects in Telecommunications In: Majumdar, S.K., Vogelsang, I., Cave, M.E., (Eds.), Handbook of Telecommunications Economics Chapter 3. Vol. 2 Technology Evolution and the

Network Economies for the Internet

Internet Economics: Models, Mechanisms and Management 55

Internet.(pp. 81-115). Amsterdam: North Holland: Rouskas, G.N. (2009). Internet Tiered Services. New York: Springer. [http://dx.doi.org/10.1007/978-0-387-09738-1] Scarf, H. (1973). Computation of Economic Equilibria. New Haven: Yale Univ. Press. Shenker, S. (1995). Service Models and Pricing Policies for an Integrated Services Internet In: Kahin, B., Keller, J., (Eds.), Public Access to the Internet (pp. 315-337). Cambridge, MA: MIT Press. Shenker, S., Clark, D., Estrin, D., Mendelson, H.D., Herzog, S. (1996). Pricing in Computer Networks: Reshaping the Research Agenda. Journal of Telecommunications Policy, 20(3), 183-201. [http://dx.doi.org/10.1016/0308-5961(96)00002-X] Shoham, Y., Leyton-Brown, K. (2009). Multi-Agent Systems. Cambridge: Cambridge Univ. Press. Spulber, D.F., Yoo, C.S. (2009). Networks in Telecommunications, Economics and Law. Cambridge: Cambridge Univ. Press. [http://dx.doi.org/10.1017/CBO9780511811883] Stallings, W. (2002). The Scheduling and Organization of Periodic Associative Computation: Efficient Networks, Review of Economic Design 3 High-Speed Networks and Internets: Performance and Quality of Service (2nd edition., pp. 93-127). Upper Saddle River, N.J.: Prentice Hall Van Zandt,T. Wang, Q., Peha, J.M., Sirbu, M.A. (1997). Optimal Pricing for Integrated Services Network In: McKnight, L.M., Bailey, J.P., (Eds.), Internet Economics. Cambridge, MA: MIT Press. Wellman, M.P. (1996). Market Oriented Programming: Some Early Lessons In: Clearwater, S.H., (Ed.), Market-Based Control. Singapore: World Scientific. Wilson, R. (1993). Nonlinear Pricing. Oxford: Oxford Univ. Press.

56

Internet Economics: Models, Mechanisms and Management, 2017, 56-74

CHAPTER 3

Network Economies for the Internet: Further Developments “… the very same architecture – a decentralist model – is why the Internet is such a scalable phenomenon today. It cannot be controlled the way politicians would like to think, with laws or bombs. Messages get through, one way or another.” Nicholas Negroponte, Being Digital (1995) Abstract: We consider specific examples of network economies with network routing and transaction processing. Any of these examples capture a structural model of the network economy with Pareto optimality and price equilibrium for agents competing for resources from emerging suppliers. A routing algorithm establishes the dynamic nature of of session arrival and departure.

Keywords: Admission Control, Agent Routing, A Server Economy, Burstiness, e-Commerce, Loss probability, Network Economy, Network Routing, Optimal Allocation, Packet probability utility function, Pareto Optimality, Price Equilibrium, QoS demands, Server Economy, Traffic Classes, Transaction Processing. 3.1. INTRODUCTION Following a conceptual framework of network economics, we solved two problems of allocating resources and providing services (QoS) to several classes of users in view of network links and the collection of servers. In the first case, we address the informational links in its supply/demand infrastructure, while in the second case, we focus on the transaction based aspect of the internet, recently identified with e-commerce on the business- to -consumer as well as business-t-business dimension. For both, we start with some stylized examples that reflect the present internet structure. We first consider a network economy of many parallel routes or links, where several agents (representing user classes) compete for resources from several suppliers, where each supplier represents a route (or a path) between source and destination. Agents buy resources from suppliers based on the QoS of the class Hans W. Gottinger All rights reserved-© 2017 Bentham Science Publishers

Network Economies for the Internet

Internet Economics: Models, Mechanisms and Management 57

they represent. A schematic presentation of sessions request from servers is pictured in Fig. (3.1). Suppliers price resources independently based on demand from the agents. The suppliers connect consumers to information providers who are at the destination, the flow of information is from information providers to consumers. We formulate and solve the problems of resource allocation and pricing in such an environment. We then consider a server economy in a distributed system. Again we use a similar model of interaction between agents and suppliers (servers). The servers sell computational resources such as processing rate and memory to the agents for a price. The prices of resources are set independently by each server based on QoS demands from the agents. Agents represent user classes such as transactions in database servers or sessions for Web servers that have QoS requirements such as response time. Servers

Session classes

Agents

1

Examples: Database servers Multimedia servers Web servers

1 1 2

r 10 10

5

Fig. (3.1). Several sessions (session classes) request for transaction or multimedia services from servers. The servers offer response-time services for information retrieval. Agents represent user classes where a single agent represents a single user class.

This chapter is organized as follows. In Section 3.2, we present two examples, one of simple network routing, the other on network transactions that provide similar platforms of network allocation decisions. Based on a simple decentralized model for the network economy, we apply the principles of economic optimization between agents and suppliers in Section 3.3. First, we outline a structural model

58 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

of the network economy with Pareto Optimality and price equilibrium for agents competing for resources from emerging suppliers. We present a routing algorithm which considers the dynamic nature of session arrival and departure. Some results for optimal allocation and for the routing mechanism are presented. Next, we present the server economy, and show the Pareto optimal allocations and price equilibrium when agents are competing for resources from servers (suppliers). Correspondingly, we apply transaction routing policies to handle the dynamics of user behavior. Conclusions are presented in Section 3.4. 3.2. TWO EXAMPLES OF NETWORK OPERATIONS Network Routing The first example shows a network representing many user classes or types of users wishing to access specific content providers. The users have a choice of routes to connect to the servers. Several routes exist between the source and the destination. At the destination, there are various kinds of content providers, such as databases, digital libraries and web servers. Each route is independent (parallel) and they have different amounts of resources. The resources are buffer and bandwidth. For simplicity, we assume that each route is a single link between a source and a destination, and we assume that each route has one packet switch that buffers packets and transmits them. User classes have several QoS requirements such as packet loss utility, maximum end-to-end delay and average packet delay. The QoS requirements are due to applications such as digital video libraries, access to multimedia databases and web servers. Sessions are set up between the source and the destination along one of the routes to access the content providers. The applications, for smooth operation, demand a certain QoS from the network routes and the end-nodes (content providers), for an end-to-end QoS. For example, video applications generate bursty traffic, and this can lead to packet loss in the network depending on the allocation of network resources for the video sessions. Video applications can tolerate a certain amount of packet loss, but beyond a threshold, the QoS of the video at the user workstation will deteriorate. In addition, maximum delay requirement is necessary to design buffer play-out strategies for smooth operation of the video application at the user workstation. Let b(s) be the burstiness curve of input m(t), the source traffic rate, at fixed service rates. Under normal circumstances, b(s) is assumed to be nonnegative, convex and strictly decreasing for s smaller than the peak rate of traffic. The burstiness curve then represents the buffer size necessary to avoid cell losses at each service rates. When a bandwidth-buffer space pair (s,b(s)) on the burstiness curve is used for resource allocation, there will be no cell loss.

Network Economies for the Internet

Internet Economics: Models, Mechanisms and Management 59

From the demand side, the demand at the network changes due to random arrivals and departures of user sessions of the traffic classes. The new session arrival may require the user class to acquire more resources to ensure a certain QoS level. In addition, when a new session arrives, a decision has to be made as to which route to choose from. In the example, resources are finite, and therefore have to be used efficiently in order to provide QoS. The traffic classes are allocated resources only for a certain period of time. The main reason being that session arrival and departure rates could change, causing fluctuations in demand, and therefore, resources have to be re-allocated to meet the change in demand. The traffic classes can re-negotiate for resources once their ownership of resources expires. From the supply side, consider that each route is a supplier, and let each traffic class be represented by an agent. The agent on behalf of the class negotiates for resources from the suppliers based on QoS requirements of the class. Each supplier has to guarantee buffer and bandwidth resources, depending on the demand from the agents. The supplier has to ensure efficient utilization of the network resources, so that the resource limits are fully exploited given the QoS requirements of classes. The task of the agents is to represent the QoS needs of the traffic class, given a certain performance framework from the supplier. Every time a session of a certain traffic class arrives, a decision must be made on which route to take between the source and destination. This depends on the agent, who can choose a route based on preferences of the traffic class, and the available resources in the routes. Therefore, dynamic mechanisms are necessary to ensure the right decision making in routing a newly arrived session. In a dynamic network the available resources at each route could be different, and in addition there is competition from other agents who have similar tasks to perform. With many routes between source and destination, the routing or placing of sessions along a route or a link must be done in a decentralized fashion. This is necessary to handle many routes and many traffic classes, each of which could have diverse QoS requirements. A framework to decentralize the various functions or tasks in admitting and routing sessions, and scheduling to switch bandwidth and buffer among the traffic classes is a challenging problem. In addition, the framework must ensure flexible QoS provisioning and promote efficient utilization of resources. Transaction Processing In this example, users request services from the content providers, and users are grouped into classes. The user classes are transactions classes for databases or just sessions for computation or information retrieval, which request for access services from one or more of the servers (content providers).

60 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

Consider a transaction processing system, where transactions that arrive are routed to one of many systems in order to satisfy performance objectives such as average response time or per-transaction deadlines. In commercial online transaction processing systems, it is very common for transactions to be processed on heterogeneous servers which have different operating systems, database management systems, hardware and software platforms, and a host of various communication protocols. Transactions are grouped into transaction classes, transactions in the same class have common workload characteristics and performance objectives. Transactions arrive at random times to their respective classes and therefore need to be routed dynamically to one of the servers. Each transaction class could have different preferences over the performance objectives and they have different processing requirements from the servers. In a transaction processing system, it is quite difficult to match the quantities of resources for an efficient usage with the diverse QoS requirements of user classes. For example, a queue could be assigned to each class at each server in order to provide various service levels, or a queue at each server could be shared among the classes. For a queue that is shared by many classes, the complexity of service provisioning increases as transactions from each class have to be distinguished in order to provide service levels. The allocation mechanism determines the throughput of each queue and the buffer allocation at the server. In addition, efficiency could mean a server wide performance measure of session level throughput, given the QoS requirements of the transaction classes. In order to handle many transaction classes and provide access to various services, the control of resources must be decentralized for the reasons of efficiency and transparency. Each server (supplier) has to offer resources such as processing, memory and input/output, and services such as average response time and throughput. This cannot be done in a centralized fashion, if we consider all the servers, instead decentralized mechanisms are needed to distribute user sessions (transaction sessions) among the servers and provide the QoS needs of classes. In addition, each server has to implement practical mechanisms, such as processor scheduling, to partition resources among the various transaction classes or provide priority services among the classes. In this example, when a user session arrives, the problem of choosing a server in order to provide a service is needed. Consider each class is represented by an agent. If a new session arrives the agent has to know if there are enough available resources to provide the required QoS. Consider that agents represent transaction classes, and they compete for resources from the various databases, and remove this burden from the servers. The problem for the agents are to choose the right server and to make sure QoS is guaranteed to the class it represents, and use the

Network Economies for the Internet

Internet Economics: Models, Mechanisms and Management 61

allocated resources judiciously. This implies mechanisms for optimal routing need to be designed. With random arrival and departure of user sessions, the agent must handle routing and admission of sessions in a dynamic way. The problem of efficient resource management by the agents and optimal allocation of resources by the servers, due to changing demand is challenging. The allocation of resources cannot be static and time-periods of renegotiation of resources and services will affect the way routing and admission of sessions is done. In addition, servers will have to adapt to changing demand in order to reflect the new allocations. For example, consider that depending on the time of day, demand at the servers fluctuate, and demands could be independent from server to server. The challenge is in determining the time-intervals based on demand. 3.3. A MODEL OF NETWORK AND SERVER ECONOMY The Network Economy The network consists of V nodes (packet switches) and N links. Each node has several output links with an output buffer. The resources at output link are transmission capacity (or link capacity) and buffer space. The link controller at the output link schedules packets from the buffer. This is based on how the buffer is partitioned among the traffic classes and the scheduling rule between the traffic classes. Sessions are grouped into traffic classes based on similar traffic characteristics and common QoS requirements. Sessions that belong to a class share buffer and link resources, and traffic classes compete for resources at a packet switch. Each session arrives to the network with a vector of traffic parameters Tr, vector of QoS requirements and wealth. A session is grouped or mapped to a corresponding traffic class. A traffic class has common QoS requirements, and we consider QoS requirements per traffic class rather than per session. Once a session is admitted along a path (a route), it will continue along that path until it compeletes. Each agent k performs the following to obtain the demand set on each link. The allocations are buffer (b) and bandwidth (c) on each link for each agent. The wealth is distributed across the links by each agent to buy resources. That is, the problem is to find pairs {c*k, b*k} such that max Uk = f(ck,bk,Trk), constraints pbbk + pcck = wk. In the above formulation, each agent k buys resources from each link. The allocation for agent k is c*k = { ck*1, ck*2, ..., ck*N } and b*k = { bk*1, bk*2, ..., bk*N }. An agent can invest wealth in either some or all the links. We assume that at each

62 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

link there is competition among at least some of the agents for buying resources. Following Gottinger (2009, Chap. 9), we show a general utility function which is a function of the switch resources: buffer (b) and bandwidth (c). A utility function of the agent could be a function of: ● ● ● ● ●

Packet loss probability Ut = g (c,b,Tr) Average packet delay Ud = h (c,b,Tr) Packet tail probability Ul = v (c,b,Tr) Max packet delay Ub = f (b) Throughput Uc = g (c)

We consider that an agent will place demands for resources based on a general utility function, which is a combination of the various QoS requirements: U = f(c,b,Tr) = x1Ul + xdUd + xbUb + xcUc + xtUt

Where Ul is the packet loss probability utility function, Ud is the average delay utility function, Ut is the packet tail probability, Ub is the utility function for maxdelay requirements, and Uc is for bandwidth (throughput) requirements. x1, xd, xb, xc, xt are constants. Agents could use such a utility function. As long as the convexity property with respect to buffer b and bandwidth c holds. Pareto optimal allocations and price equilibria exist. However, if they are not convex, then depending on the properties of the functions, local optimality and price equilibrium could exist. To show the main ideas for routing and admission control, we use packet loss probability as the main utility function (Ul), which means we assume that x1 from the above equation are the only constant and the rest are zeros. For doing this, we need first some further specifications of the loss probability. We later show results for Pareto optimality and price equilibrium, and then we propose routing and admission control algorithms. In general, one can assume that agents, on behalf of user classes, demand for resources from the link suppliers based on the utility function shown above. The agent uses the utility function to present the demand for resources over the whole network of parallel links. Loss Probability Specifications. At each output link j the resources are buffer space Bj and link capacity Cj. Let {cjk, bjk} be the link capacity and buffer allocation to class k on link j where k ε [1, K] . Let pjc and pjb be the price per unit link capacity and unit buffer respectively at link j, and wk be the wealth (budget) of a traffic class k. For a link j from the source to the destination, the packet loss probability (utility) for traffic class k is given by the following:

Network Economies for the Internet

Internet Economics: Models, Mechanisms and Management 63

Ulk = Ploss = 1 − Πj 1N (1 − Pkj )  =

(1)

Where Pkj is the packet loss probability at link j of agent k. The goal of the agent is to minimize the packet loss probability under its wealth or budget constraints. If the traffic classes have smooth convex preferences with respect to link capacity and buffer allocation variables at each link, then the utility function Ulk is convex with respect to the variables. This is under the assumption that general queueing systems have convex packet loss probability functions (see Harel and Zipkin (1987) which means convex preferences in link capacity and buffer space. Once proved, this can be a very useful property in designing resource allocation mechanisms for general network topologies. Price Equilibrium Let each TC (represented by an agent) transmit packets at a rate λ (Poisson arrivals), and let the processing time of the packets be exponentially distributed with unit mean. Let c, b be allocations to a TC. The utility function (packet loss probability for M/M/1/B) applies to queues in specific communication networks. The ground-laying work without recourse to an explicit utilitarian formulation of probabilistic loss has been done by Kleinrock (1976, Chap.5). He mentions some of the principal problems of Internet related communication networks: “The analysis of stochastic flow in store-and-forward networks suffers under many of the combinatorial burdens from network flow theory as well as many of the probabilistic burdens from queueing theory. As a result, the efficient design of a computer-communication network is an extremely complex task, and the number of design parameters and operating modes is considerable.”(Kleinrock, 1976, 299).          (1   )(  )b  c c U  f (c, b,  )    1  ( )  1  b c   1 b 1   (1)(  )(  ) b c c    1  (  )1  b  c

for < c ,c, c resp.

(2)

64 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

Let U for each TC at each link be given by (5), as in Chap. 2. The allocation variables at each node for each traffic class are c (link capacity) and b (buffer space). The utility function is continuous and differentiable for all c ε [0, C], and for all b ε [0, B] . We assume that b ε  for continuity purposes of the utility function. With agents competing for resources in a network of parallel links, the overall utility function U can be obtained by using the utility function above. We have the following theorem. Proposition 3.1 The packet loss probability function for agent k shown in (1), assuming an M/M/1/B model for each link, is decreasing convex in ckj for ckj ε [0,Cj], and decreasing convex in bkj, ∀ bkj ε [0,Bj]. Proof. See Appendix The goal of each agent is to maximize the preference (which is minimizing packet loss probability) under the budget constraint. Each traffic class computes a demand set using the wealth constraints and the current prices. The demand set can be computed using Langrange multiplier techniques. Using the utility function given by (2) and the first order equilibrium conditions, the price ratio at each link j is given by the following: ∂Uk/∂ckj pcj Nkj − bkj ρkj ∗ (1 − Pkj) ⎯⎯⎯⎯ = ⎯ = ⎯⎯⎯⎯⎯ , Nkj = ⎯⎯⎯⎯⎯⎯⎯⎯ ∂Uk/∂bkj pbj ckjlogρkj 1 − ρkj ∗ (1 − Pkj)

(3)

where function Nkj is the ratio of the effective queue utilization (ρkj * (1 − Pkj)) to the effective queue emptiness 1 − ρkj * (1 − Pkj) and ρkj = λkj / ckj . Consider K traffic classes of M/M/1/B type competing for resources (link and buffer) in a network of parallel links. Then the following theorem is stated: Proposition 3.2 Let each traffic class k have smooth convex preferences represented by the utility function shown in (2) Given that Σ1K ci = C and Σ1K bi = B for all i,k ε[1,K], then the Pareto surface exists. Given the wealth constraint wk of the traffic classes, the Pareto optimal allocation and the price equilibrium exist.

Network Economies for the Internet

Internet Economics: Models, Mechanisms and Management 65

The proof is based on the fact that the utility functions are decreasing convex and smooth in the resource space (preferences are convex and smooth). The proof is essentially the same as in Gottinger (2009, Chap.9) for Pareto optimality, except that the preferences are shown here to be convex in link capacity and buffer space at each link (given the traffic parameters of each traffic class at each link) in the network of parallel links using the M/M/1/B model. Agent Routing and Admission For a session that arrives to a traffic class in the network, the agent has several routes to choose from between the source and the destination. The agent can choose a route that benefits the traffic class it joins. This means that an agent is searching for the right set of service providers (or links) to reach the destination. Several interesting questions arise in a market economy with many users and suppliers: will the network economy be efficient in service provisioning? What are the negotiation protocols between the users and the suppliers so that services are guaranteed? What is the session blocking probability per class, given session arrival and average session holding time per class? The static description of the problem is simply to find the best allocation of sessions among the suppliers. The allocation can satisfy efficiency criteria such as throughput of the number of sessions per class admitted to the overall network. For example, consider the static case that Ai is the session arrival rate (Poisson with distribution) for class i, and Ωi is the average session holding time of sessions of class i. Let Agent be allocated cij link capacity on link j. Let the maximum number of sessions that can be admitted per link j for class i, such that certain QoS level is satisfied. Let the space be {ni1, ni2..., niN}. Then the problem for the agent is simply to determine the flow of sessions among the network of links, given the above parameters. Formally, the agent has to find the following: find {ρi1, ρi2, ..., ρiN }, minimize 1 - Π1N (1 - Pblock) given that {cij, bij} and constraints ∑1N ρij = ρi where ρij = Ai / Ωi for all j ∈ [1,N] is the session level utilization, and ∑1N Aij = Ai and ρi = Ai / Ωi . For the main goal of agent i is to maximize the throughput of the sessions through the network. This is one of the many efficiency requirements of agent i. We now discuss dynamic routing algorithms for each agent i that routes a session over the network along one of the routes (or links). The dynamic routing algorithms depend on the state of the network portion owned by agent i. For example, the routing decisions will be made based on the number of sessions currently active on each of the links for class i. Consider a parallel link network as explained in Sect. 3. 2 where each link is a supplier. The routing algorithm is as follows: The agent representing the traffic

66 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

class will choose the supplier which can give a better QoS for the overall class. This means that the suppliers in the network are ordered in the decreasing order of preference by the agent based on the utility derived by joining them. This routing algorithm is described as follows for a network consisting of several parallel links between a source and a destination. If a session of type TCk arrives, then it is routed to a link j which gives the maximum preference (maximum QoS) to the class from among the set of suppliers. The routing mechanism yields the guideline: route to j such that max Uk (Φ(p)) for all j. Where (Φ(p) is the demand at supplier j. Uk (Φ(p)) is the overall utility derived by traffic class k if the session joins supplier j. This mechanism essentially states that the agent will choose the service provider which gives the maximum utility (or in this case minimal packet loss probability) to the traffic class. The routing algorithm (by the agent) first computes Pkj (Φ(p)) for all the suppliers, where Pkj (Φ(p)) is the packet loss probability of traffic class k at link j in the parallel link network. The agent then ranks the class utility derived by joining a supplier, and then ranks them in decreasing order of preference. Admission Control The agent will admit the session on one of the many routes (links) provided the QoS of the traffic class it joins is honored. If the agent has the same preference over a subset of the suppliers (a tie for suppliers), then one of them will be chosen at random. If all the sessions of a traffic class are identical (same traffic load and parameters), then the agent can compute the admission space, and the number of sessions that can be admitted without violating the QoS constraints of the class over all the links. Formally, find {n1*, n2*, ..., nN* } given that {cij*, bij* } with constraints qic = {qi1c, qi2c, ... }. The agent has to find the maximum (nj*) of admissible sessions at each link, given the Pareto allocation for agent i on each link j, and given the QoS constraints of the class i (qic) which, for example, could be packet loss probability, max-delay and average delay requirement per class. Several interesting questions that arise under the class welfare based routing: what is the session level blocking probability per class, given the session arrival rate and average holding time per class? How does it depend on the session arrival rate and holding time? Does this routing algorithm balance the loads in such a fashion that a traffic class benefits in the long run by the routing algorithm? We study some of the questions numerically. We use simulations where 2 traffic classes (2 agents) compete for resources in a two node (link) parallel network, i.e. just two suppliers. The sessions of traffic class k arrive to the network at a Poisson rate of γk. The session holding time is exponentially distributed with mean νk (k ∈

Network Economies for the Internet

Internet Economics: Models, Mechanisms and Management 67

[1, 2]). Each session of class k arriving has average packet arrival rate λk (traffic parameters are Poisson arrivals) and the mean service time is exponentially distributed with mean one. The state space is basically a Markov chain with four parameters {n11, n21, n12, n22} representing the number of sessions of traffic class k at each of the links. However, for each agent the state space is 2 dimensional. Numerical studies indicate that the routing algorithms are stable. The results can be obtained using simulations and Markov chain models of 2 user classes and 2 suppliers (links). The session blocking probability is in the order of 1/107 for an offered load (at the session or call level) γk/νk = 2.0. It is evident that the dynamic algorithm is better than the static one as we increase ρ = A/ Ω. The agent that routes a session can choose one of the links dynamically based on the state of the link. Let class 1 be allocated bandwidth and buffers in such a way that 10 sessions can be admitted to supplier 1 and 15 to supplier 2 for agent 1. This admission region assumes that a packet loss probability of 1/108 is the QoS requirement by class 1. The Server Economy We now discuss the server economy where servers offer processing resources and memory to agents representing user classes. The agents compete for these resources and buying as much as possible from suppliers. The agents perform load balancing based on the QoS preferences of the class it represents. The economic model consists of the following players: Agents and Server Suppliers, Consumers or user classes and Business. User sessions within a class have common preferences. User classes have QoS preferences over average delay and throughput, and in some cases completion times of sessions (deadlines). Users within a class share resources at the servers. Agents and Network Suppliers Agents represent user classes. An agent represents a single user class. Agents negotiate with the supplier and buy resources from service providers. Agents on behalf of user classes demand for resources to meet the QoS needs. Suppliers compete to maximize revenue. Suppliers partition and allocate resources (processing rate and memory) to the competing agents. Multiple Agent Network Supplier Interaction Agents present demands to the suppliers. The demands by agents are based upon their wealth and user class preferences. The demand by each agent is computed via utility functions which represent QoS needs of the class. Agents negotiate with suppliers to determine the prices. The negotiation process is iterative where prices

68 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

are adjusted to clear the market. Price negotiation could be done periodically or depending on changes in demand. The agent and network supplier become service providers in the market. The role of the supplier is to provide technologies to sell resources (buffer and bandwidth units) and to partitioning they flexibly based on the demand by the agents. The agents transform the goods (buffer and bandwidth) and provide QoS levels to the user-classes. The agents strive to maximize profits (minimize buying costs) by using the right utility functions and the right performance models in order to provide QoS to the user-class. More users within a user-class implies more revenue for the agent. The agent is decoupled from the traffic class and the supplier. Control by agent

t

Agent I

Non-sharing model

λII

Class I

t

c(II )

i λKI

i

k

Control by server Rate scheduling C(I )

c(KI )

k Class K

Agent K

t λIK λKN

c(IN ) C(N )

i k

c(KN )

Fig. (3.2). Non-sharing model: K agents compete for services in a transaction processing system with N servers.

In this economy, user classes are transaction classes that send transactions to database servers for processing. The transaction processing time at each of the server is based on the type of transaction. Consider K classes of transactions and each class is represented by an agent (economic agent). In the economy, the

Network Economies for the Internet

Internet Economics: Models, Mechanisms and Management 69

agents negotiate with the servers for server capacity. We assume that transactions of any class can run on any of the database servers. Therefore, agents negotiate with all the servers for server throughput (or processing speed). A model where K agents compete for services in a transaction processing system, and where transaction classes do not share the service queues is shown in Fig. (3.2). Each class could do the following based on its preferences on average delay and throughput: (i) each agent i can minimize its average response time under throughput constraints, (ii) each agent i can maximize throughput of its transactions under an average delay constraint, (iii) each agent i can look at a combination of QoS requirements and have preferences over them. Therefore, each class can choose either one of these preferences and let the agent control the flow of transactions through the system. The problem now becomes a multi-objective optimization problem as every agent is trying to maximize its benefit in the system based on the class of QoS preferences. Consider that the classes wish to choose various objectives, the utility function assumes U = xdUd + xlUl where Ud is the utility function for average delay and Ul is the utility function for throughput, and xd and xl are constants. Consider that there are requirements for transaction completion time. Instead of scheduling transactions to meet deadlines, we try to minimize the number of transactions that have missed the deadlines (in a stochastic sense). Consider that each transaction class is assigned a service queue at each server, then we try to minimize the probability of the number of transactions of a class exceeding a certain threshold in the buffer. This is the tail probability P(X >b) where X is the number of transactions of a class in a queue at a server, and b is threshold is threshold for the number in the queue, beyond which transactions miss deadlines. If we include this QoS requirement, then the above utility function will be U = xdUd + xlUl + xtUt where Ut is the tail probability utility function and xt is a constant. Pareto Optimality: We now have a simple formulation for classes competing for server capacity (processing rate) in order to minimize average delay (or average response time). The utility function is simply U = xd Ud as the rests of the constants are zero. Let pj be the price per unit processing rate at server j. The maximum processing rate at server j is Cj. The problem therefore for each agent is the following: find {cij*} such that: min Ud = { ∑j 1N Wij} with constraints ∑j 1N λij = γi ∀i , ∑1N cij ∗ pj ≤ wi ∀i =

=

In the above problem definition, each agent will try and minimize the utility function under the wealth conbstraint and under the throughput constraint. This constraint is necessary to make sure that positive values of throughput are

70 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

obtained as a result of the optimization. The transaction agents compete for processing rate at each server, and transaction servers compete for profit. The objectives of the transaction classes are conflicting as they all want to minimize their average response time. In the above formulation Wij = λij / (cij − λij) this is the average number of class i transactions in queue at system j. The average delay in the system for each class i is simply the average number in the system divided by the overall throughput ∑j=1N λij. The main goal of the agent representing the transaction class is to minimize a utility function which is simply the average number in the overall system. This will also minimize the average delay or average response time of the transaction class. Proposition 3.3 The utility function Ud is convex with respect to the resource allocation variable cij where λij ∈ [0, cij), and cij ∈ (0, Cj] The utility function Ud is discontinuous when λij = cij . Demand Set: The demand set for an agent i, given the prices (pj of server j) of the processing rates (or capacities) at the servers is {ci1, ci2, ..., ciN } over all the servers. We use the standard techniques of optimization to find the demand set, which is given as follows for all j ∈ [1,N]: cij = λij + ((wi − ∑j=1N λij pj) / ∑j=1N√ λij pj) √ λij / pj. Price Equilibrium: Once the demand set is obtained, then using the wealth constraints, we can solve for the equilibrium price. This is not easily tractable. However, numerical results can be computed using the tatonnement process whereby agents compute the demand set, given the processing rate prices by each server. An iteration process between the agents and the servers takes place. This will converge to an equilibrium price, when demand equals the supply which is ∑i =1K cij = Cj. We now state formally the result for K agents competing for processing resources from N servers. Proposition 3.4 Consider K agents competing for processing resources from N servers. If the utility function of these agents is Ud and the performance model at the servers is an M/M/1 model, then price equilibrium and Pareto optimality exist.

Network Economies for the Internet

Internet Economics: Models, Mechanisms and Management 71

The proof of this proposition is the same as described in Gottinger (2009, Chap.9). The utility function Ud is continuous and decreasing convex with respect to the allocation variables cij. The function is discontinuous when λij = cij. Due to this, Pareto allocations or price equilibrium may not exist. However, we solve this problem by stating that the agents, when they present their demands, have to make sure that the transaction throughput rate λij at a server has to be lower than the capacity allocation cij . If this is not met, then the price iteration process or the tatonnement process will not converge. We assume that the servers know the transaction throughput or arrival rate from each agent during the iteration process. Transaction Routing The static routing problem for the agent i, once the allocation of processing rates at the N servers is done for agent i, can be formulated as: Find {λij} such that min {∑j=1N Wij} with constraints ∑j=1N λij = γi ∀i . Here, Wij is the average response time for agent i traffic when sent to server (supplier) j. We use a simple M/M/1 model of the queueing system, where Wij = λij/ (cij − λij) (average number of agent i transactions in server j). This or the average delay can be minimized (the same result will be obtained for either one of them). The optimal arrival rate vector to the servers or the optimal flow of transactions, assuming a Poisson distribution for arrivals with rate λij is given by:  cij ij  cij    ( j1N cij  ij) j1N cij This result gives the optimal flow of transactions of class i to the servers, given the capacity allocation to the agent i. Using this, a simple random routing policy which can split transaction traffic optimally can be designed. This policy does not assume the current state of the servers, the number of transactions of agent i queued for service at server j. A simple, but well known routing algorithm is illustrated here. Dynamic Routing Algorithm: The Join Shortest Queue (JSQ) algorithm routes transactions of class i to a system j found by obtaining the minimum of the following:

72 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

Qi1 + 1 Qi2 + 1 QiN + 1 Min {⎯⎯⎯ , ⎯⎯⎯ , .... , ⎯⎯⎯ } ci1 ci2 ciN Where Qij is the queue length of server j or the number of transactions of class i queued up for service at server j. If there are ties, then one of the queues is picked at random (or with equal probability). CONCLUSION We have developed a decentralized framework for QoS provisioning based on economic models. A new definition of QoS provisoning based on Pareto efficient allocation s is given. These allocations are not only efficient (from a Pareto sense) but also satisfy the QoS constraints of competing traffic classes (or users). We have shown that Pareto optimal allocations exist in a network economy (parallel link network), and a methodology is provided for the network service provider to price services based on the demands placed by the users. Prices are computed based on the load of the traffic classes and the corresponding demand. Furthermore, a dynamic session routing algorithm is coupled with admission control mechanisms to provide QoS to the traffic classes, for a network as well as a server economy. Future research should address several issues related to the dynamics of the overall system. For example, if we assume time is divided into intervals, and during each time interval prices of resources are stable. Then price negotiation is done between the agents and the suppliers at the beginning of the time interval. However, each supplier (server) could have time intervals which are different from the rest. This can cause the agents to negotiate with each supplier independently. Economic models can provide several new insights into resource sharing and QoS provisioning in future networks and distributed systems which will connect millions of users and provide a large number of serves. Pricing and competition can provide solutions to reduce the complexity of service provisioning and efficiently utilize the resources. APPENDIX: PROOF OF PARETO OPTIMAL ALLOCATIONS We start by giving the first derivative of P with respect to the buffer variable b:

Network Economies for the Internet

Internet Economics: Models, Mechanisms and Management 73

ρb(1−ρ)log[ρ] P' = ⎯⎯⎯⎯⎯⎯⎯ (1− ρb)2

limc

→λ

1 P' = − −⎯⎯ (1+ b)2

(A.1)

Where ρ= λ/c. This function is negative for all b ∈ [0, B] and for all ρ > 0. The second derivative with respect to be yields

P'' =

(1− ρb) ρb (log[ρ] )2 (1− ρ) ⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯ (1− ρb)3

limc P'' = →λ

2 −⎯⎯ (1+ b)2

(A.2)

This function is positive for all b ∈ [0, B] and all ρ > 0. Similarly, the function P can be shown to be continuous (smooth) and decreasing convex in c for all c ∈ [0, C], by rewriting function P to the following: 1 P = ⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯ 1+ c/λ + (c/λ)2 + .... + c/λ)b

(A.3)

From this the first derivative can be shown to be negative and the decond derviative to be positive for all c ∈ [0, C], hence the proof. In the system of parallel links the overall packet loss probability (QoS parameter) for a traffic class k is given as folllows: Uk = Ploss, k = 1 − ∏j 1N (1 − Pkj )  =

(A.4)

74 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

Where Pkj is the packet loss probability of TCk on link j (or supplier). This utility function has the same properties as captured in Proposition 3.2. REFERENCES Gottinger, H.W. (2009). Strategic Economics in Network Industries. New York: NovaScience Publ., New York. Harel, A., Zipkin, P.H. (1987). Strong Convexity Results for Queueing Systems’. Oper. Res., 35, 405-418. [http://dx.doi.org/10.1287/opre.35.3.405] Kleinrock, L. (1975). Queueing Systems (Vol. 2). Computer Applications, New York: Wiley Interscience.

Internet Economics: Models, Mechanisms and Management, 2017, 75-92

75

CHAPTER 4

Internet Economics of Distributed Systems “The field of mechanism design … has shown how by carefully constructing economic mechanisms to provide the proper incentives, one can use selfish behavior to guide the system toward a socially desirable outcome …being also computationally efficient with least costs” J. Feigenbaum, M. Schapira, and S. Shenker, Distributed Algorithmic Mechanism Design (2007) Abstract: We focus on economic management of web services in distributed multimedia systems. We dig deeper into mechanism design approaches tracing them back to classical economic mechanisms of market designs and information economics which I refer to as Hayek-Hurwicz mechanism design – due to Austrian-British economist F.A. Hayek and American economist L. Hurwicz. The basic idea of market agents as computationally efficient human agents induced by incentive compatability and selfishness have been rediscovered and reapplied in rigorous methodological form by computer scientists merging algorithmic game theory, computability and network complexity. Distributed algorithmic mechanism design (DAMD) for internet resource allocation indistributed systems is akin to an equilibrium converging market based economy where selfish agents maximize utility and firms seek to maximize profits and the state keeps an economic order providing basic public goods and public safety.A distributed algorithmic mechanism design thus consists of three components: a feasible strategy space at the network nodes for each agent or autonomous system, an aggregated outcome function computed by the mechanism and a set of multi-agent prescribed strategies induced by the mechanism. A distributed algorithmic mechanism design being computationally efficient in a large decentralized Internet economy is a powerful paradigm to substantiate claims by Hayek (1945) that an industrialized economy based on market principles has an overall better output and growth performance (static and ynamic) than socialist type economies of a similar nature and scale. Best economic coordination through markets producing maximal social welfare is supported by computational efficiency in computer science. Applications relate to a data management economy.

Keywords: Algorithmic Game Theory (AGT), Autonomous System (AS), Congestion Problems, Distributed Algorithmic Mechanism Design (DAMD), Hayek-Hurwicz Mechanism Design, Incentive Compatibility, Information Economics, Mechanism Design, Multi-media Systems, Network Complexity, Scalability, Server Economy, Vickrey Auction. Hans W. Gottinger All rights reserved-© 2017 Bentham Science Publishers

76 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

4.1. INTRODUCTION A macroscopic view of decentralized (distributed) computer systems reveals the complexity of the organization and management of the resources and services they provide. The complexity arises from the system size (e.g. number of systems, number of users), heterogeneity in applications (e.g. online transaction processing, e-commerce, multimedia, intelligent information search, and auctions) and resources (CPU, memory, I/O bandwidth, network bandwidth and buffers, etc.). The complexity of resource allocation is further increased by several factors. First, in many distributed systems, like the present day web, the resources are in fact owned by multiple organizations, at least in leading OECD economies. Second, the satisfaction of users and the performance of applications are determined by the simultaneous application of multiple resources. For example, a multimedia server application requires I/O bandwidth to retrieve content, CPU time to execute server logic and protocols, and networking bandwidth to deliver the content to clients. The performance of applications may also be altered by trading resources. For example, a multimedia server application may perform better by releasing memory and acquiring higher CPU priority, resulting in smaller buffers for I/O and networking but improving the performance of the communication protocol execution (Gupta et al., 1997). Finally, in a large distributed system, the set of systems, users and applications are continuously changing. In this chapter, we address some of the issues in managing Quality of Service (QoS) and pricing, and efficient allocation of resources (computational resources) in networks and systems. QoS demand very well fits into the classical econometric demand analysis as advanced by H. Wold (1953). The structure of this chapter is as follows: Sec. 4.2. exhibits some broader design criteria on large scale networks which underlie the heterogeneity of Internet based resource allocation and use. Also it shows the major components of an interface architecture with which an ‘economically enhanced resource manager’ is confronted (Macias et al., 2010). Sec. 4.3. indicates more broadly the scope of mechanism design approaches that link economic modelling to computational resources, at the interface of economics, computer and management science. Sec. 4.4. deals with a specific class of problems arising in mechanism design that show how in resource allocation processes, pricing schemes have to be made ‘incentive compatible’.

Internet Economics

Internet Economics: Models, Mechanisms and Management 77

Sec. 4.5. relates to the basic structure of a data management economy which more recently features in major application areas as in grid computing, cloud computing, sponsored search auctions, broadcast protocols, and other areas like procurement auctions, spectrum auctions, dedicated communication networks, supply chain formation and social networks. Strategic management issues emerging through resource provisioning and pricing are covered in Sec. 4.6. Discussions follow in Sec. 4.7. Some examples for service architectures relating to large scale distributed systems are sketched in the Appendix. 4.2. THE RATIONALE OF ECONOMIC MODELS IN NETWORKING There are intrinsic interfaces between human information processing and networking that show the usefulness of economic modelling (as advanced early by Ferguson et al. (1996)). For designing resource allocation and control mechanisms in complex distributed systems and networks several goals need to be considered and could be traced in the literature in more detail, as described by Clearwater (1996), Shenker et al. (2007), Deng and Graham (2007), and Neumann et al. (2010). With reference to Sec. 2.4, we recall the relevant features of organizational economics for completeness. Decentralization In an economy, decentralization is provided by the fact that economic models consist of agents which selfishly attempt to achieve their goals. Suppose there are two types of economic agents: suppliers and consumers. A consumer attempts to optimize its individual performance criteria by obtaining the resources it requires, and is not concerned with system-wide performance. A supplier allocates its individual resources to consumers. A supplier's sole goal is to optimize its individual resources to consumers and to optimize its individual satisfaction (profit) derived from its choice of resource allocation to consumers. Pricing and Performance Most economic models introduce money and pricing as the technique for coordinating the selfish behavior of agents. Each consumer is endowed with money that it uses to purchase required resources. Each supplier owns a set of resources, and charges consumers for the use of its resources. The supplier prices its resources based on the demand by the agents, and the available supply.

78 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

Consumers buy resources or services such that the benefit they receive is maximized. Consumer-agents buy resources based on maximizing performance criteria. As a whole, the system performance is determined by some combination of the individual performance criteria. Organizational Domains Often large distributed systems and computer networks spread over several domains, the control of resources is shared by multiple organizations that own distinct parts of the network. In such an environment, each organization will have a set of services that it supports. Economic principles of pricing and competition provide several valuable insights into decentralized control mechanisms between the multiple organizations and efficient service provisioning. Scalability A key issue in designing architectures for services in large computer networks and distributed systems is scalability. With the ever growing demand for new services, flexible service architectures that can scale to accommodate new services is needed. Economic models of competition provide, in a natural fashion, mechanisms for scaling services appropriately based on service demand and resource availability. 4.3. MECHANISM DESIGN APPROACHES Network allocation and pricing could be considered as part of the mechanism design theory (Hurwicz and Reiter, 2006) and in differential form coping with dynamic mechanism design by Williamson (2008). In a relevant economic historical context, the justification for linking market mechanism to computational resource allocation may be attributed to the Austrian/British economist F.A.Hayek (1945), therefore, what we suggest is an Internet based distributed system as a sort of Hayekian mechanism design. More specific mechanism design approaches for distributed networks and grid-type systems are covered by Narahari et al. (2009) and Neumann et al. (2010), see also Meinel and Tison (1999). In the context of computational resources, specifically, an algorithmic mechanism design (AMD) uses a computational platform with an output specification and agents’ preferences represented by utilities (Nisan, 1999). In a distributed algorithmic mechanism design (DAMD) as with algorithmic mechanism design (AMD), there are dual concerns of incentive compatibility and computational complexity but as Shenker et al. (2007,365) explains:

Internet Economics

Internet Economics: Models, Mechanisms and Management 79

“DAMD’s measure of complexity is quite different from AMD’s, because the computation is distributed. Any measure of the complexity of a distributed algorithm executed over an interconnection network T must consider at least five quantities: the total number of messages sent over T, the maximum number of messages sent over any one link in T, the maximum size of a message, the local computational burden at each node, and the storage required at each node. …. We will use the term network complexity to refer to these.” Along this path of DAMD an expansion in terms of cooperative problem-solving, individual and market rationality would become encoded through distributed artificial intelligence (DAI) which would enhance economic design and planning (Huhns, 1997; Hayzelden and Bigham, 1999). Network and Server Economies We first consider a network economy, of many parallel routes or links, where several agents (representing user classes) compete for resources from several suppliers, where each supplier represents a route (or a path) between a source and destination. Agents buy resources from suppliers based on the QoS requirements of the class they represent. Suppliers price resources, independently, based on demand from the agents. The suppliers connect consumers to information providers, who are at the destination; the flow of information is from information providers to the consumers. This formulates and solves the problems of resource allocation and pricing in such an environment. Following this, we consider a server economy in a distributed system. Again, we use a similar model of interaction between agents and suppliers (servers). The servers sell computational resources such as processing rate and memory to the agents for a price. The prices of resources are set independently by each server based on QoS demand from the agents. Agents represent user classes such as transactions in database servers or sessions for Web servers that have QoS requirements such as response time. Examples are given in Gottinger (2013). Server Economy: Architecture for Interaction We also consider a large scale distributed information system with many consumers and suppliers. Suppliers are content providers such as web servers, digital library servers, and multimedia database and transaction servers. Consumers request for and access information objects from the various suppliers and pay a certain fee or no fee at all for the services rendered. Consider that third party suppliers provide information about the suppliers to consumers in order to let consumers find and choose the right set of suppliers.

80 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

Access and Dissemination Consumers query third-party providers for information about the suppliers, such as services offered and the cost (price). Likewise, suppliers advertise their services and the costs via the third party providers in order to attract consumers. Consumers prefer an easy and simple way to query for supplier information, and suppliers prefer to advertise information securely and quickly across many regions or domains. For example, consider a user who wishes to view a multimedia object (such as a video movie). The user would like to know about the suppliers of this object, and the cost of retrieval of this object from each supplier. Performance Requirements Users wish to have good response time for their search results once the queries are submitted. However, there is a tradeoff. For more information about the services offered, advanced searching mechanisms are needed, but at the cost of increased response time. In other words, users could have preferences over quality of search information and response time. For example, users might want to know the service costs in order to view a specific information object. In large networks, there could be many suppliers of this object, and users may not want to wait forever to know about all the suppliers and their prices. Instead, they would prefer to get as much information as possible within a certain period of time (response time). From the above example, in order to let many consumers find suppliers, a scalable decentralized architecture is needed for information storage, access and updates. Naming of services and service attributes of suppliers becomes a challenging issue when hundreds of suppliers spread across the globe. A simple naming scheme to connect consumers, across the Internet, with information about suppliers is essential. The naming scheme must be extensible for new suppliers who come into existence. A name registration mechanism for new suppliers and a de-registration mechanism (automatic) to remove non-existent suppliers is required. In addition, naming must be hierarchical, domain based (physical or spatial domains) for scalability and uniqueness. Inter-operability with respect to naming across domains is an additional challenging issue not covered in this paper. The format of information storage must be simple enough to handle many consumer requests quickly within and across physical domains. For better functionality and more information, a complex format of information storage is necessary, but at the cost of reduced performance. For example, a consumer, in addition to current service cost, might want to know more information such as the

Internet Economics

Internet Economics: Models, Mechanisms and Management 81

cost of the same service during peak and off-peak hours, the history of a supplier, its services, and its reputation, in order to make a decision. This information has to be gathered when requested. In addition, the storage formats must be interoperable across domains. Performance A good response time is important to make sure consumers get the information they demand about suppliers within a reasonable time period, so that decisionmaking by consumers is done in a timely fashion. In addition, the design of the right architectures for information storage and dissemination is necessary for a large scale market economy to function efficiently. Using the previous example, consumers and suppliers would prefer an efficient architecture to query for and post information. Consumers would prefer good response time in obtaining the information, and suppliers prefer a secure and fast update mechanism to provide up-to-date information about their services. Security in transferring information and updating information at the bulletin boards (name servers) is crucial for efficient market operation and smooth interaction between consumers and suppliers. For this the third party suppliers (naming services) have to provide authentication and authorization services to make sure honest suppliers are the ones updating information about their services. 4.4. ALLOCATION AND PRICING MODELS In economic models, there are two main ways to allocate resources among the competing agents. One of them is the exchange based economy and the other is the price based economy. In the exchange based economy, each agent is initially endowed with some amounts of the resources. They exchange resources until the marginal rate of substitution of the resources is the same for all the agents. The agents trade resources in the direction of increasing utility (for maximal preference). That is, two agents will agree on an exchange of resources (e.g. CPU for memory) which results in an improved utility for both agents. The Pareto optimal allocation is achieved when no further, mutually beneficial, resource exchanges can occur. Formally, an allocation of resources is Pareto optimal when the utility derived by the competing economic-agents is at the maximum. Any deviation from this allocation could cause one or more economic agents to have a lower utility (which means the agents will be dissatisfied). In a price based system, the resources are priced based on the demand, supply and the wealth in the economic system. The allocations are done based on the following mechanism. Each agent is endowed with some wealth. Each agent computes the demand from the utility function and the budget constraint. The aggregate demand from all the agents is sent to the suppliers who then compute

82 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

the new resource prices. If the demand for a resource is greater than its supply, the supplier raises the price of the resource. If there is surplus supply, the price is decreased. The agents again compute their demands given the current prices and present the demand to the suppliers. This process continues iteratively until the equilibrium price is achieved where demand equals the supply. Bidding and auctioning resources is another form of resource allocation based on prices. There are several auctioning mechanisms such as the Sealed Bid Auction, Dutch auction, and English Auction. The basic philosophy behind auctions and bidding is that the highest bidder (or in the Vickrey auction the second highest bidder) always gets the resources, and the current price for a resource is determined by the bid prices. Allocation Principles What are the general allocation principles? Can economic models give insight into the allocation mechanisms that can cause the computer system to reach equilibrium? Can these principles be used practically to evolve the computer system in a way that price equilibrium can be achieved? Even devoting the entire WINE 2007 proceedings to those issues, with active participation and guidance of K. Arrow, H. Scarf and C. Papadimitriou (in Deng and Graham, 2007), still many practical issues of implementation haven’t been yet finally resolved . 4.5. THE DATA MANAGEMENT ECONOMY Unlike the flow control and load balancing economies where users maximize an utility function to compute the required allocation, this economy considers data migration, replication and pricing strategies for a data management economy as evidenced by large scale e-commerce facilitated through new platforms in grid computing, cloud computing and related application areas (Kushida et al., 2011). The problem of data migration, storage and replication is formulated in an economic setting. Transactions that enter the system for service are charged by the processors for read and write access to data objects. Processors also lease resources to other processors to make profit using the revenue they earn. The distributed system consists of N processing nodes connected via links. Each processor Pi (i [1,N]) has rate ri at which it can process operations on local data. A link eij connects processor Pi to Pj. There are M data object denoted by D1, D2, ...., DM . S(Di) defines the size of Di in bytes. The economy treats these as abstract data objects. In a real system, they could correspond to relations, tuples, files, records or any other data structure. The data management problem is to minimize the mean transaction response time with the following as control variables.

Internet Economics ● ● ●

Internet Economics: Models, Mechanisms and Management 83

Number of copies of data object Assignment of copies to processing nodes Pricing strategies of suppliers

In the data management economy there are four types of agents. The consumers are transactions, and the suppliers are data object managers, local data agents and processors as through cloud computing. The economy functions in the following way. Each transaction T that arrives has an allocation of money MT. Transactions pay to access data at a processor Pi. Data access is provided by the processor by leasing copies of data objects from data object managers. The local data agents act as an intermediary between a processor Pi and the object managers (remote). Two economic factors cause the data management economy to adapt the number of read copies of each object Dj to the read/write ratio. These are: ●

● ●

The total revenue that all processors earn by selling Read(Dj) decreases as the initially set price of the agents given its wealth pw increases The read lease price for Dj increases linearly with the number of copies c(j) The data management economy uses decentralized decision making to compute the number of read copies of each object. The business strategies of the processors are decoupled, and Pi uses only local information to estimate its revenue. The economy adapts itself to any read/write ratio without any external intervention. The economy is not completely self tuning, however, there is a subtle interaction between the following factors: (i) lease price function, (ii) transaction arrival rates.

4.6. STRATEGIC INTERNET MANAGEMENT ISSUES Universal Access A primary concern in regulating universal access to the Internet, next to security, had been the issue of pricing its services, the maintaining of competition among providers and strengthening incentives for private investment into the network infrastructure. Possible options emerged in identifying the issues toward a workable model: i. charging by access to telecommunications capacity, e.g., flat rate pricing and keeping distance independent pricing ii. consider network externalities in the economics and growth of networks iii. introduce usage-based linear prices iv. introduce usage-based nonlinear prices The evolution of Internet pricing poses interesting problems. Flat-rate pricing has been one of the factors that promoted the Internet to expand at a dramatic rate. It

84 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

has enabled low-cost dissemination, beta-testing and refinement of new tools and applications. The strength of many of these tools is their ability to operate in and rationalize a distributed and heterogeneous structure making it easier to identify and retrieve information sources. The increased demand that has arisen due to the power and new resources these tools have brought to the Internet (and in view of lagging a corresponding capacity expansion due to advanced fiber-optic technology) is likely to create more gridlock and a need for a new pricing model. This despite new regulatory proposals on “net neutrality” emerging, usage based pricing and service charges or more specific content pricing should make the Internet attractive to many new users and also incentivize innovation driven product development on the net. One paradox of usage based pricing is that its implementation may actually cost more on a transaction basis than the underlying cost of transport. Therefore, it very much depends on network accounting capabilities as a critical implementation tool. Congestion Problems A natural response by shifting resources to expand technology will be expensive and not necessarily a satisfactory solution in the long run. Some proposals rely on voluntary efforts to control congestion. Others have suggested that we essentially have to deal with the problem of overgrazing the commons, e.g. by overusing a generally accessible communication network. A few proposals would require users to indicate the priority they want each of the sessions to receive, and for routers to be programmed to maintain multiple queues for each priority class. If priority class is linked to the value the users attach to it, one could devise schemes of priority pricing. This is where application of mechanism design could help. At congested routers, packets are prioritized based on bids. In line with the design of a Vickrey (seal-bid) auction, in order to make the scheme incentive compatible, users are not charged the price they bid, but rather are charged the bid of the lowest priority packet that is admitted to the network. It is well-known that this mechanism provides the right incentives for truthful revelation. Such a scheme has a number of desirable characteristics. In particular, not only do those users with the highest cost of delay get served first, but the prices also send the right signals for capacity expansion in a competitive market for network services. If all of the congestion revenues are reinvested in new capacity, then capacity will be expanded to the point where the marginal value is equal to its marginal cost. More recently, game-theoretic approaches adopt a unified view even for two-sided markets (Ackermann et al. (2007), Deng and Graham (2007)). Quality-of-Service Characteristics Because of different customer requirements, network allocation algorithms should

Internet Economics

Internet Economics: Models, Mechanisms and Management 85

be designed to treat different types of traffic differently but the user must truthfully indicate which type of traffic(s) he is preferring, and this would only happen through incentive compatible pricing schemes. QoS can be affected by various factors, both quantitative (network latency, CPU performance,…) and qualitative, among the latter could proliferate reputation systems that hinge on trust and belief in a certain QoS level being achieved, resulting in a service level arrangement (SLA) comprising service reliability and user satisfaction (Anandasivam and Neumann, 2010). Internet and Telecommunications Regulation In contrast to traditional telecommunications services Internet transport itself is currently minimally regulated but services provided over telecommunication carriers are not. This principle has never been consistently applied to telephone companies since their services over fixed telephone lines also used to be regulated. There have been increasing demands, sometimes supported by established telecommunication carriers that similar regulatory requirements should apply to the Internet. One particular claim is “universal access” to Internet services, that is, the provision of basic Internet access to all citizens at a very low price or even for free. What is a basic service, and should its provision be subsidized? For example, should there be an appropriate access subsidy for primary and secondary schools? A related question is whether the government should provide some data network services as public goods. A particular interesting question concerns the interaction between pricing schemes and market structure for telecommunications services. If competing Internet service providers offer only connection pricing, inducing increasing congestion, would other service providers be able to attract high value ‘multimedia’ users by charging usage prices but offering effective congestion control? On the other hand, would a flat rate connection price provider be able to undercut usage-price providers by capturing a large share of baseload customers who would prefer to pay for congestion with delay rather than with a fee. Could this develop into a fragmented market with different Internets? These developments may have profound impacts to shape a future telecommunications industry which may be taken over by different structured layers of the Internet. 4.7. DISCUSSION In this chapter we focus on applications of mechanism design to resource management problems in distributed systems and computer networks. These

86 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

concepts are used to develop effective market based control mechanisms, and to show that the allocation of resources are Pareto optimal. The emphasis here is on management implications given the economics of the Internet. We follow novel methodologies of decentralized control of resources, and pricing of resources based on varying, increasingly complex QoS demands of users. We bring together economic models and performance models of computer systems into one framework to solve problems of resource allocation and efficient QoS provisioning matching large-scale e-commerce applications. The methods can be applied to pricing services in ATM networks and (wireless) Integrated Services Internet of the future. We address some of the drawbacks to this form of modelling where several agents have to use market mechanisms to decide where to obtain service (which supplier?). If the demand for a resource varies substantially over short periods of time, then the actual prices of the resources will also vary causing several side effects such as indefinite migration of consumers between suppliers. This might potentially result in degradation of system performance where the resources are being underutilized due to the bad decisions (caused by poor market mechanisms) made by the users in choosing the suppliers. As in real economies, the resources in a computer system may not easily be substitutable. The future work is to design robust market mechanisms and rationalized pricing schemes which can handle surges in demand and variability, and can give price guarantees to consumers over longer periods of time some of which have been discussed by Spulber and Yoo (2009, Chap.12). Another drawback is that resources in a computer system are indivisible resulting in nonsmooth utility functions which may yield sub-optimal allocations, and potential computational overhead. In addition to models for QoS and pricing in computer networks, we are also working towards designing and building distributed systems using market based mechanisms to provide QoS and charge users either in a commercial environment or in a private controlled environment by allocating quotas via fictitious money (charging and accounting) by central administrators. In summary, economic based management is useful for implementing and operating internet-type systems. The Internet currently connects hundreds of millions of users and thousands of sites. Several services exist on many of these sites, notably the World Wide Web (WWW) which provides access to various information sources distributed across the Internet. Many more services (multimedia applications, commercial transactions) are to be supported in the Internet. To access this large number of services, agents have to share limited network bandwidth and server capacities (processing speeds). Such large-scale networks require decentralized mechanisms to control access to services.

Internet Economics

Internet Economics: Models, Mechanisms and Management 87

Economic/managerial concepts such as pricing and competition can provide some solutions to reduce the complexity of service provisioning and decentralize the access mechanisms to the resources. CONCLUSION We explore name service architectures for disseminating information about suppliers and their services to consumers, and look at the main properties of these architectures. We use analytical models to compute the expected response time for consumers to access for information in each architecture. We compare the three architectures in terms of performance, security and flexibility. The economic models of networks and systems, and the corresponding mechanisms described previously can use the framework mentioned to allocate resources and provision services in a real environment. APPENDIX: SERVICE ARCHITECTURES FOR THE INTERNET ECONOMY In designing market based frameworks for distributed systems one would like to look at corresponding architectures which let consumers find information about suppliers and their services, and let suppliers advertise QoS information about the services they offer and the corresponding costs. Consider a large scale distributed information system with many consumers and suppliers. Suppliers are content providers such as web servers, digital library servers, and multimedia database and transaction servers. Consumers request for and access information objects from the various suppliers and pay a certain fee or no fee at all for the services rendered. Consider that third party suppliers provide information about suppliers to consumers in order to let consumers find and choose the right set of suppliers. Access and dissemination: consumers query third-party providers for information about the suppliers, such as services offered and the cost (price). Likewise, suppliers advertise their services and the costs via the third party providers in order to attract consumers. Consumers prefer an easy and simple way to query for supplier information, and suppliers prefer to advertise information securely and quickly across many regions or domains. For example, consider a user who wishes to view a multimedia object (such as a video movie). The user would like to know about the suppliers of this object, and the cost of retrieval of this object from each supplier.

88 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

Performance requirements: users wish to have good response time for their search results once the queries are submitted. However, there is a tradeoff. For more information about services offered, advanced searching mechanisms are needed, but at the cost of increased response time. In other words, users could have preferences over quality of search information and response time. For example, users might want to know the service costs in order to view a specific information object. In large networks, there could be many suppliers of this object, and users may not want to wait forever to know about all the suppliers and their prices. Instead, they would prefer to get as much information as possible within a certain period of time (response time). From the above example, in order to let many consumers find suppliers, a scalable decentralized architecture is needed for information storage, access and updates. Naming of services and service attributes of suppliers becomes a challenging issue when hundreds of suppliers spread across the globe. A simple naming scheme to connect consumers, across the internet, with information about suppliers is essential. The naming scheme must be extensible for new suppliers who come into existence. A name registration mechanism for new suppliers and a de-registration mechanism (automatic) to remove non-existent suppliers is required. In addition, naming must be hierarchical, domain based (physical or spatial domains) for scalability and uniqueness. Inter-operability with respect to naming across domains is an additional challenging issue not covered in this paper. The format of information storage must be simple enough to handle many consumer requests quickly within and across physical domains. For better functionality and more information, a complex format of information storage is necessary, but at the cost of reduced performance. For example, a consumer, in addition to current service cost, might want to know more information such as the cost of the same service during peak and off-peak hours, the history of a supplier, its services, and its reputation, in order to make a decision. This information has to be gathered when requested. In addition, the storage formats must be interoperable across domains. Performance: a good response time is important to make sure consumers get the information they demand about suppliers within a reasonable time period, so that decision-making by consumers is done in a timely fashion. In addition, the design of the right architectures for information storage and dissemination is necessary for a large scale market economy to function efficiently. Using the previous example, consumers and suppliers would prefer an efficient architecture to query for and post information. Consumers would prefer good response time in

Internet Economics

Internet Economics: Models, Mechanisms and Management 89

obtaining the information, and suppliers prefer a secure and fast update mechanism to provide up-to-date information about their services. Security in transferring information and updating information at the bulletin boards (name servers) is crucial for efficient market operation and smooth interaction between consumers and suppliers. For this the third party suppliers (naming services) have to provide authentication and authorization services to make sure honest suppliers are the ones updating information about their services. Architecture Models. For our architecture and design, we choose the existing, operational Internet Domain Name Service (DNS) for reasons of scalability, simplicity, efficiency and performance, and for its distributed architecture. DNS has a simple hierarchical structure for uniquely naming internet hosts across the globe. DNS uses this naming in finding information about hosts located anywhere in the Internet. The naming space is divided among administrative domains, and within each domain, the naming is done independently. DNS is a simple distributed architecture for storing information about hosts in the Internet. The name service has a database that keeps several resource records (RRs) for each host, indexed by the host domain name. One such RR is the IP address of a host indexed by the hostname. The RR is used commonly for mapping domain names (hostnames) to IP addresses for networking between hosts (example: email). In addition to this widely used RR, there are several other types of RRs which store more information about a host, and its characteristics. The Internet is divided into domains. Each domain is controlled by a primary name server (NS) and some secondary name servers which replicate the primary NS database for better response time. Within the DNS naming tree we can add any number of service nodes, which have RRs for storing IP addresses and RRs for service parameter information which is stored in the TXT record of the node. For each server, the TXT RR describes in a simple way (string), the service attribute value pairs. Within the new DNS functionality and naming schemes, the customer can submit complex queries which can be based on attributes and other information. A customer could also ask information about services in other domains or zones. This means that the DNS engine has to query other name servers for information regarding the services. This querying can be done in a recursive fashion between the primary name servers to obtain information from other domains, similar to the way it is done for IP addresses of hosts in other domains.

90 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

We explore three architectures to store and retrieve information about various suppliers. The architectures are designed using the functionality offered by the Internet Domain Naming Service. 1. Centralized Read-Write (RW) Architecture Each supplier (host) is registered at the primary NS, which maintains the whole database (DB) of supplier information in the RRs. The TXT RR stores information about services offered by suppliers and its service attributes. Each supplier updates DB securely at NS using Public Key Methods. NS contains information about each supplier. Consumers, via the Web, query NS for service information about each supplier. 2. Centralized Transfer-Access (TA) Architecture Each supplier is a primary of its local domain. Each supplier keeps its information local (in the DB). This way the information is updated locally by the supplier and is secure. Suppliers belong to a global primary DNS (NS). 3. Decentralized Index Based (IB) Architecture Each supplier maintains its own DB. The DB contains the services offered and prices, and the time periods where prices are fixed and the expiry dates. Each supplier is registered at the primary NS for the domain. The registration of the supplier is done in a secure fashion. A Registration Server exists and authenticates, using private and public key techniques, the digital signature of each host. The IP address of each supplier is stored in the primary name server. Also, the primary NS maintains a list of IP addresses for each service that is being offered in that domain. Specialized Features in Centralized and Decentralized Models The resource records of the node services show that www, video, gopher, ftp … are the services offered in this domain. One can use these keywords and find more information about the specific services, and suppliers offering these services and their corresponding service attributes. WWW based access to supplier information: consumers have an access to the supplier information via the worldwide-web interface. All the consumers see is a list of categories of services offered or a simple keyword based search, where the keywords should match with the services being offered in a domain. For example, a user can click on Netscape and obtain all the information about services offered in a domain. Once this is done, a user can pick a specific service and ask for the list of suppliers that offer this service. The requests are submitted via the cgi-bin interface of the www. The

Internet Economics

Internet Economics: Models, Mechanisms and Management 91

responses come back in a form that can be viewed by the Web browser. Performance Model for RW, TA and IB Architectures 1. Centralized RW: We assume a simple model to study the performance the model is based on M/M/1 with (two classes of traffic) queueing system. Read requests from consumers in a domain arrive at the primary at a certain Poisson rate λr, and update requests or updates arrive at the primary at a rate λw which is also a Poisson distribution. The average service rate of the read request is μr which is exponentially distributed, and the average service rate of the update requests is μw. Let C be the processing rate of the primary name server. Then the average delay in queueing and service for each request (whether read or write) is Delay and μNS. 2. Centralized TA In this model the primary NS services customer queries (all the load). In the simple model, the name server spends some time ion answering queries, and periodically polls the suppliers for information or any updates. We model such a system as an M/M/1 queueing system user queries for reads and writes and at a certain rate the secondaries transmit to the primary, and we assume that the rate has a Poisson distribution model. 3. Distributed Index Based Access. The primary name server acts as a simple router of requests to the suppliers, who respond with the final answers. Customers query the primary NS, and get a list of suppliers offering a service. They then query each supplier in that list and get more information about their services. User read requests are first processed at the primary and then routed to the suppliers for more information. The overall request rate remains the same as in previous models. This model is distributed, as the processing of a query is done by suppliers. Therefore, the response time will be lower on an average to the customers compared to the other architectures. Comparison of Response Time Model 1 has a lower response time compared to model 2. This is because in model 2, the primary NS spends some time polling for update information from the suppliers. For model 3, we consider that the read requests are split evenly among the suppliers, likewise we consider that the update frequency is the same for each supplier, for the sake of simplicity. As expected model 3 gives a better response time. REFERENCES Ackermann, H., Goldberg, P.W., Mirrokni, V.S., Röglin, H., Vöcking, B. (2007). A Unified Approach to Congestion Games and Two-Sided Markets.

92 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

[http://dx.doi.org/10.1007/978-3-540-77105-0_7] Deng, X., Graham, F.C. (2007). Internet and Network Economics. Third International Workshop. WINE 2007San Diego, Berlin, New York: Springer. Ferguson, D.F., Nikolaou, C., Sairamesh, J., Yemini, Y. (1996). Economic Models for Allocating Resources in Computer Systems In: Clearwater, S., (Ed.), Market-Based Control: A Paradigm for Distributed Resource Allocation, Singapore: World Scientific. [http://dx.doi.org/10.1142/9789814261371_0007] Gottinger, H.W (2013). Quality of Services for Queueing Networks of the Internet iBusiness 5, 1-12. [http://dx.doi.org/10.4236/ib.2013.53012] Gupta, A., Stahl, D.O., Whinston, A.B. (1997). Priority Pricing of Integrated Services Networks In: McKnight, L.W., Bailey, J.P., (Eds.), Internet Economics. Cambridge, MA: MIT Press. Hayek, F.A. (1945). The Use of Knowledge in Society. Am. Econ. Rev., 35, 519-530. Hayzelden, A.L., Bigham, J. (1999). Software Agents for Future Communications Systems. Berlin: Springer. [http://dx.doi.org/10.1007/978-3-642-58418-3] Huhns, M.N. (1997). Distributed Artificial Intelligence. Los Altos, Ca.: Morgan Kaufmann. Hurwicz, L., Reiter, S. (2006). Designing Economic Mechanisms. Cambridge: Cambridge Univ. Press. [http://dx.doi.org/10.1017/CBO9780511754258] Macias, M., Smith, G., Rana, O., Guitart, J., Torres, J. (2010). Enforcing Service Level Agreements using Economically Enhanced Resource Manager In: Neumann, D., Baker, M., Rana, O.F., Altmann, J., (Eds.), Economic Models and Algorithms for Distributed Systems (pp. 109-127). Basel: Birkhaeuser. Meinel, C., Tison, S. (1999). STACS 99, 16th Annual Symposion Theoretical Aspects of Computer Science, Trier, Germany. March, Berlin: Springer. Narahari, Y., Garg, D., Narayanam, R., Prakash, H. (2009). Game Theoretic Problems in Network Economics and Mechanism Design Solutions. London: Springer. Nisan, N. (1999). Algorithms for Selfish Agents, Mechanism Design for Distributed Computation. Nisan, N., Roughgarden, T., Tardos, E., Vazirani, V.V. (2007). Algorithmic Game Theory. Cambridge, New York: Cambridge Univ. Press. [http://dx.doi.org/10.1017/CBO9780511800481] Low, S, Varaiya, P.A. (1993). New Approach to Service Provisioning in ATM Networks IEEE Trans. Networking , 7-14. [http://dx.doi.org/10.1109/90.251913] Kushida, K.E., Murray, J., Zysman, J. (2011). Diffusing the Fog: Cloud Computing and Implications for Public Policy, Berkeley Roundtable on the International Economy (BRIE). BRIE Working Paper 197 Shenker, S., Feigenbaum, J., Schapiro, M. (2007). Distributed Algorithmic Mechanism Design Algorithmic Game Theory (pp. 363-384). Cambridge: Cambridge Univ. Press. Spulber, D.F., Yoo, C.S. (2009). Networks in Telecommunications, Economics and Law. Cambridge: Cambridge Univ. Press. [http://dx.doi.org/10.1017/CBO9780511811883] Williamson, S.R. (2008). Communication in Mechanism Design, A Differential Approach. Cambridge: Cambridge Univ. Press. Wold, H. (1953). Demand Analysis: a Study in Econometrics. New York: Wiley.

Internet Economics: Models, Mechanisms and Management, 2017, 93-108

93

CHAPTER 5

Generalized Quality of Service on Queueing Networks for the Internet “QoS …has to comprise all necessary aspects of business resulting in a ‘trust relationship’ between customer and provider.This can be confidentiality,data integrità,non-repudiation,accountability,…” D. Neumann et al. Eds., Economic Models and Algorithms for Distributed Systems (2010) Abstract: Broadening the criteria for QoS performance and guaranteed service through reputation systems hinge on trust and belief. As the Internet matures in promptness, reliability, accessibility and foremost security for a certain targeted QoS level, this is termed as ‘Generalized QoS level’. GQoS includes emphasis on security, reputation and trust. Invoking economic design principles in the present day Internet have provided measurable ingredients for QoS as they identified performance criteria such as average response time, maximum response time, throughput, application failure probability and packet loss. QoS can be affected by various factors, both quantitative (i.e., latency, CPU performance, storage capacity etc.) and qualitative that proliferate through reputation systems hinging on trust and belief .As the Internet matures in promptness, reliability, availability and foremost security for a certain quality service level targeted, it embraces a GQoS. level.

Keywords: Bandwidth Allocation, Border Gateway Protocol (BGP), Buffer, Bulletin Board, Burstiness Curve, Cloud Computing, CPU Performance, Generalized Quality of Service (GQoS), Load Balancing, Loss Probability Constraints, Mechanism Design, Network Allocation Algorithm, Packetloss Probability, Performance Model, Reputation System, Security Level, Service Level Arrangement (SLA), Social Networks, Traffic Class, Truthful Revelation of Preferences, Utility Functions. 5.1. INTRODUCTION In the context of congested networks for Internet traffic, the phenomenon of packet loss is due to two reasons: the first, packets arrive at a switch and find that the buffer is full (no space left), and therefore are dropped. Secondly, is that packets arrive at a switch and are buffered, but they do not get transmitted (or scheduled) in time, then they are dropped. An economic way of saying this is that: Hans W. Gottinger All rights reserved-© 2017 Bentham Science Publishers

94 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

for real-time applications, packets, if delayed considerably in the network, do not have value once they reach the destination, a job that misses the deadline may have no value at all. The sort and variety of delay can severely impact the operability and efficiency of a network and therefore is of eminent interest for economic analysis. A way to look at the network economy is to invoke mechanism design principles supporting market mechanisms (Deng et al., 2007; Neumann et al., 2010). The present chapter builds on previous work and expands on Quality of Service principles, Service Level Arrangements (SLAs) and management procedures toward creating demand for service in the Internet economy. Network allocation algorithms should be designed to treat different types of traffic differently but the user must truthfully indicate which type of traffic he/she is preferring, and this would only happen with a mechanism design allowing incentive compatible pricing schemes, which is at the very core of economic mechanism design. Invoking economic design principles into the present day Internet have provided measurable ingredients for QoS as they identified performance criteria such as average response time, maximum response time, throughput, application failure probability and packet loss. QoS can be affected by various factors, both quantitative (i.e., latency, CPU performance, storage capacity etc.) and qualitative that proliferate through reputation systems hinging on trust and belief and as the Internet matures in promptness, reliability, availability and foremost security for a certain quality service level targeted, thus embracing a ‘Generalized Quality of Service’ level . If you relate performance to utility as a multiplicative factor, performance varies over the range of utility in [0,1]. This would result in service level arrangements (SLAs) comprising service reliability and user satisfaction (Macias et al., 2010). To ensure truthful revelation of preferences, the reporting and billing mechanism must be incentive compatible. Generalized QoS services should include emphasis on security, reputation and trust among agents and network providers. QoS comprises more than technical properties. In the business world, it is essentially a sub-notion of service level arrangement (SLA) and would include all necessary aspects of business resulting in a ‘trust relationship’ between customer and provider such as security protection. Designing mechanisms to rate the reputation of agents in multiagent systems is a topic that has received much attention more recently in QoS modelling. For example, one can make use of a social network where agents ask others in the community about the reputability of agents with which they had past experience (Zhang and Cohen, 2007). When social networks are used, agents need

Generalized Quality of Service

Internet Economics: Models, Mechanisms and Management 95

to assess the reputation of those agents that provide information as well. Often the setting on trust modelling is relevant for ecommerce, where buying agents select selling agents. In this context, an agent would accumulate a reputation rating somewhere in between the lowest and the highest value. This will contribute to the design of distributed systems in cooperative environments. A Simple Mechanism Design There are K agents in an Internet economy that collectively generate demand competing for resources from a supplier. The supplier itself announces prices entering a bulletin board accessible to all agents (as a sort of transparent market institution). In a simple form of a trading process, we could exhibit a “tatonnement process” on a graph where the agents set up a demand to the supplier who advertise prices on a Bulletin Board which are converted into new prices in interaction with the agents. The tatonnement process in economics is a special form of an algorithmic mechanism design (AMD), introduced by Nisan and Ronen (2001), that in modern computer science emerged as an offspring to algorithmic game theory (Nisan et al., 2007). The approach to mechanism design would enable the users of applications to present their QoS demands via utility functions, defining the system performance requirements. The resource allocation process involves economic actors to perform economic optimization given scheduling policies, load balancing and service provisioning. Along this line, such approaches have been the basis of business models for grid and cloud service computing (Mohammed et al., 2007). Distributed algorithmic mechanism design (DAMD) for internet resource allocation in distributed systems is akin to an equilibrium converging market based economy where selfish agents maximize utility and firms seek to maximize profits and the state keeps an economic order providing basic public goods and public safety (Feigenbaum et al., 2007). In fact, to make the link to economic mechanism design, this is a sort of Hayek type mechanism limited to a special case of a diversified Internet economy (Myerson, 2006). As an example of the operation of DAMD, we only mention a common routine protocol, the Border Gateway Protocol (BGP), for the Internet interdomain routing. This is a path vector protocol because the BGP allows adjacent nodes to exchange information through messages involving a route announcement that displays the entire path to the destination. In view of the DAMD, it satisfies the following desirable properties. If implemented honestly and rationally, the protocol should maximize social welfare. The protocol should be incentive

96 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

compatible so that agents are not motivated to deviate from compliance, from the actions they are asked to perform. The exact form of incentive compatability would depend on the nature of a game-theoretic solution concept (economic equilibrium). 5.2. UTILITY AND QUEUEING PARAMETERS In Chaps. 3 and 4 we showed a general utility function which is a function of the switch resources; buffer (b) and bandwidth (c). The utility function could be a function of the following: ● ● ● ● ● ●

Packet loss expected utility Ut = g (c,b,Tr) Average packet delay Ud = h (c,b,Tr) Packet tail utility Ut = v (c,b,Tr) Max packet delay Ub = f (b,bT) Throughput Uc = g (c,ct) Security level Us[Pr(|m-m'|)] where m' is a compromised message accessible by other agents, and |m-m'| = 0 is true message preserving. We could conceive this as a security parameter λ in a crytographic setting for Multiparty Computation (MPC) Protocols (Dodis and Rabin, 2007).

The variables b and c in the utility functions refer to buffer space allocation and link bandwidth allocation. In the utility functions Ub and Uc; the parameters bT and cT are constants. For example, the utility function Ub = f(b,bT) for max packet delay is simply a constant as b increases, but drops to 0 when b = bT and remains zero for any further increase in b. The last entry, security level, indicates a security or trust parameter that can be built into a DAMD implementable QoS service. We consider utility functions which capture packet loss probability of QoS requirements by traffic classes, and we consider loss, max-delay and throuput requirements. After this we proceed to utility functions that capture average delay requirements, followed by utilità functions that capture packet tail utility requirements. We also could give examples of utility functions for agents with multiple objectives; agents have preferences over several QoS parameters. Packet Loss The phenomenon of packet loss is due to two reasons: the first, packets arrive at a switch and find that the buffer is full (no space left), therefore, are dropped. The second is that packets arrive at a switch and are buffered, but they do not get transmitted (or scheduled) in time, then they are dropped. A formal way of saying

Generalized Quality of Service

Internet Economics: Models, Mechanisms and Management 97

this: for real-time applications, packets, if delayed considerably in the network, do not have value once they reach the destination. A proper way to deal with it is through queueing systems. We consider K agents, representing traffic classes (TC) of M/M/1/B type, competing for resources from the network provider. The utility function is packet loss utility (U1) for the user classes. M/M/1/B model is a standard queueing model with B for a given service time distribution for the following reasons. We choose the M/M/1/B model of traffic and queueing for the following reasons. The model is tractable, where steady state packet loss probability is in closed form, and differentiable. This helps in demonstrating the economic models and concepts. Markovian type models such as M/M/1/B or M/D/1/B for multiplexed traffic (such as video) are appropriate where simple histogram based traffic models capture the performance of queueing in networks. In this case we look at the simple queueing system, for example with Poisson inputs, and exponential interarrival and general service time distribution B at a single server (Kleinrock and Gail, 1996, 7, 21). For more complex traffic and queueing models (example of video traffic) we can use tail utility functions to represent QoS of the user class instead of loss utility. In the competitive economic model, each agent prefers less packet loss, as the more packet loss, the worse the quality of the video at the receiving end. Let each agent TCk have wealth wk, which it uses to purchase resources from network provider. Let each TC transmit packets at a rate λ (Poisson arrivals), and let the processing time of the packets be exponentially distributed with unit mean. Let c, b be allocations to a TC. The utility function U for each TC is then a function of the allocations and the Poisson arrival rate. Loss Probability Requirement: Utility Function In view of queueing discipline, we consider K agents, representing traffic classes of M/M/1/B type, competing for resources from the network provider. The utility function is packet loss probability (Ut) for the user classes. Analogously, for more complex traffic and queueing models (say, example of video traffic) we can use tail probability functions to represent QoS of the user class instead of loss probability. Let each TC transmit packets at a rate λ (Poisson arrivals), and let the processing time of the packets be exponentially distributed with unit mean. Let c, b be allocations to a TC. The utility function U for each TC is given as follows:

98 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

= { (1 - λ/c)(λ/c)b/ (1 - (λ/c))1+b, if λ < c U = f(c,b,λ) = { 1/ (b + 1), if λ = c ={ (- 1 + λ/c)/(λ/c)(λ/c)b/- 1+ (λ/c)1+b, if λ > c

The above function is continuous and differentiable for all c ∈ [0, C], and for all b ∈ [0,B]. We assume b ∈ ℜ for continuity purposes of the utility function. Loss Probability Constraints The loss probability constraint is defined as follows: it is the set of (bandwidth, buffer) allocations { x: x ∈ X, U(x) ≤ Lc } where U(x) is the utility function (loss probability function where lower loss is better) and Lc is the loss constraint.The preferences for loss probability are convex with respect to buffer and link capacity. Computation of the QoS Surface by Supplier: assume that the supplier knows the utility functions of the agents, which represent the QoS needs of the traffic classes, then the supplier can compute the Pareto surface, and find out the set of Pareto allocations that satisfy the QoS constraints of the two agents. This set could be a null set, depending on the constraints and available resources. The QoS surface can be computed by computing the points A and B with a bandwidth-buffer space pair on the burstiness curve used for resource allocation. The burstiness curve represents the buffer size necessary to avoid cell losses at each service rate level. Point A is computed by keeping the utility of (say) class 1 constant at its loss constraint and computing the Pareto-optimal allocation by maximizing the preference of (say) class 2. Point B can be computed in the same way. The QoS surface is the set of allocations that lies in [A, B]. The same technique can be used to compute the QoS surface when multiple classes of traffic compete for resources. There are situations where the loss constraints of both the traffic classes cannot be met. In such cases, either the demand of the traffic classes must go down or the QoS constraints must be relaxed. This issue is treated as an admission control problem, where new sessions are not admitted if the loss constraints of either class is violated. Max and Average Delay Requirements A max delay constraint simply imposes a constraint on the buffer allocation, depending on the packet sizes. If the service time at each switch for each packet is

Generalized Quality of Service

Internet Economics: Models, Mechanisms and Management 99

fixed, the max delay is simply the buffer size or a linear function of buffer size. Once the QoS surface for loss probability constraints are computed, then the set of allocations that meet the buffer constraint will be computed. This new set will provide loss and max delay guarantees. A traffic class will select the appropriate set of allocations that meet the QoS requirements under the wealth constraint. A class of interesting applications would require average delay constraints on an end-to-end basis. Some of these applications include file transfers, image transfers, and lately Web based retrieval of multimedia objects. Consider a traffic model such as M/M/1/B for each traffic class, and consider that several traffic classes (represented by agents) compete for link bandwidth and buffer resources at a link with QoS demands being average delay demands. Let us now transform the average delay function into a normalized average delay function for the following reasons: average delay in a finite buffer is always less than the buffer size. If a user class has packet loss probability and average delay requirements, then buffer becomes an important resource, as the two QoS parameters are conflicting with respect to buffer. In addition, the switch buffer needs to be partitioned among the traffic classes. Another way to look at this: a user class can minimize the normalized average delay to a value that will be less than the average delay constraint. This normalized average delay function for an M/M/1/B performance model, for an agent, could be shown as: { [ λ/c(1-λ/c) − b (λ/c)1+b/1-(λ/c)b ]/ λb, λb) = (λ/c)b+1

(1)

The system assumes that λ < c. From the above equation, the tail probability is decreasing convex with respect to c as long as λ < c., and is decreasing convex with respect to b as long as λ < b. Consider agents using such a utility function for obtaining buffer and bandwidth resources, then using the convexity property and the regions of convexity being (λ < c.) Using the equilibrium condition, as derived in Gottinger (2009, Chap.9.7), we obtain for Pareto optimal allocation and price equilibrium:

102 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

pc/pb = (b1+1)/c1log(λ1/ c1) = (b2+1)/c2log(λ2/ c2) = … = (bn+1)/cnlog(λn/ cn)

(2)

We assume K agents competing for buffer and bandwidth resources, with tail probability requirements as shown in (1) For the case of two agents in competition, the equilibrium condition is as follows: log ρ1/log ρ2 = [(b1+1)/c1 ] [ c2 / (b2+1)] with ρ = λ/c

(3)

For equilibrium in network economies we can interpret (3) as the ratio of the logs of the utilizations of the classes is proportional to the ratio of the time spent in clearing the buffer contents. (b) Tail Probability with On-Off Models In standard performance models the utility functions are derived using simple traffic models such as Poisson, with the mean arrival rate as the main parameter. Here we use on-off (bursty) traffic models in relation to the competitive economic model. The traffic parameters are mean and variance in arrival rate. We show how the traffic variability has an impact on the resource allocation, and in general the Pareto surface at a link. We assume an ATM type network where packet sizes are of fixed size (53 bytes). On-off models are commonly used as traffic models in ATM networks (Kleinrock, 1996). These traffic sources transmit ATM cells at a constant rate when active and nothing when inactive. The traffic parameters are average burst length, average rate, peak rate, and variances in burst length. The traffic models for such sources are on-off Markov sources (Ross, 1970). A source in a time slot (assuming a discrete time model) is either “off” or “on”. In the on state it transmits one cell and in the off state it does not traansmit any cell. When several such (homogeneous or heterogeneous) sources feed into an infinite buffer queue, the tail distribution of the queue is given by the following formula: Pr(X > b) = h(c,b,ρ,Cv2)g(c,b,ρ,Cv2))-b

Where h(c,b,ρ,Cv2) and g(c,b,ρ,Cv2) are functions of traffic parameters and link capacity c. Such functions are strictly convex functions in c and b. These functions are currently good approximations to packet loss probability in finite buffer systems, where packet sizes are of fixed size. These approximations become very close to the actual cell (packet) loss for very large buffers. The utility function is as follows: A TC consists of S identical (homogeneous) on-off sources which are

Generalized Quality of Service

Internet Economics: Models, Mechanisms and Management 103

multiplexed to a buffer. Each source has the following traffic parameters: {T,rp,ρ, Cv2} where T is the average on period, rp is the peak rate of the source, Cv2 is the squared coefficient of variation of the on period, and ρ is the mean rate. The conditions for a queue to form are: S rp > c (peak rate of the TC is greater than the link capacity) and S rpρ < c (mean rate less than link capacity). The packet tail distribution of the queue when sources are multiplexed into an infinite buffer queue then has the form U = S rpρ/c [(1+ 2(c - S rpρ)/ S rpρ(1 - ρ)2 (Cv2 + 1)T]-b

(4)

Using a numerical example, we use two traffic classes (with the same values). There are S1 = S2 = 10 sessions in each traffic class, T = 5, rp = 1, ρ = 0.5. Using the constraints c1 + c2 = 60 and b1 + b2 = 100, the Pareto surface is obtained. As Cv2 increases from 1 to 20, the Pareto surface tends to show that buffer space and link capacity are becoming more and more valuable. The equilibrium price ratios p(c)/p(b) vs. Cv2 increase as Cv2 increases. A higher Cv2 implies a higher cell loss probability and therefore more resources are required, therefore a higher price ratio (link capacity is more valuable compared to buffer). Specific Cases Now we consider some specific cases of agents with different QoS requirements. (a) Loss and Average Delay Consider the following case, where two agents have different QoS requirements, one of them (agent 1) has a packet loss probability requirement and the other (agent 2) has an average delay requirement. We assume that the network supplier has finite resources, C for link bandwidth and B for buffer. Using the properties of loss probability and average delay with respect to bandwidth and buffer, the Pareto optimal solution is simply: all buffer to agent 1, as agent 2 does not compete for link buffer. The competition is for link bandwidth between agent 2 and agent 1. Let w1 be the wealth of agent 1, annd w2 for agent 2, then the equilibrium prices of buffer and bandwidth are the following: pb = pbf and pc = (w1 + w2)/ C. Since there is no competition for buffer space, the cost of the buffer is simply the fixed cost pbf. The Pareto allocations are {B, C w1/ (w1 + w2)} for agent 1 and {0,C w2/ (w1 + w2)} for agent 2.

104 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

(b) Loss and Normalized Average Delay Consider the following case, where agent 1 and agent 2 have preferences on loss probability and normalized average delay requirements (transforming average delay requirements into normalized average delay requirements). In this case, the two agents have different utility functions, however, their preferences are such that more buffer and more bandwidth is required and this causes the agents to compete for both resources. The utility function for agent 1 is as follows: U1 = γ1Uloss + (1 - γ1) Udelay, where γ1 ∈ [0,1]

The utility function for agent 2 is as follows: U2 = γ2Uloss + (1 - γ2) Udelay, where γ2 ∈ [0,1]

For example, agent 1 might prefer more weight on loss probability than normalized average delay compared to agent 2 who weighs normalized average delay more than loss probability. Let agent 1 choose γ1 = 0.9, and agent 2 choose γ2 = 0.1. Due to the convexity properties of the loss probability function and the normalized average delay function, the resultant multi-objective utility function is decreasing convex with respect to bandwidth and buffer, respectively. Under the equilibrium condition, the equilibrium prices for the resources have the familiar prroperty that the ratio of prices is equal to the ratio oof marginal utilities with respect to the resources, for each agent. Using the resource constraints c1 + c2 = C and b1 + b2 = B, we can obtain the Pareto surface. To compute a specific Pareto allocation one uses the following parameters: agent 1 and agent 2 have the same traffic arrival rate λ1 = λ2 = 10 . The performance model is the M/M/1/B model for both agents. Using the tatonnement process, where agents negotiate with the link supplier to buy bandwidth and buffer resources, the process converges to a price equilibrium. The Parreto optimal allocation is split evenly with respect to buffer and bandwidth among the agents. The price of link bandwidth is higher than the price of buffer. 5.3. SERVICE ECONOMY: ARCHITECTURE FOR INTERACTION Consider a large scale distributed information system with many consumers and suppliers. Suppliers are content providers such as web servers, digital library servers, multimedia database and transaction servers. Consumers request for and

Generalized Quality of Service

Internet Economics: Models, Mechanisms and Management 105

access information objects from the various suppliers and pay a certain fee or no fee at all for the services rendered. Consider that third party suppliers provide information about suppliers to consumers in order to let consumers find and choose the right set of suppliers. Access and Dissemination Consumers query third-party providers for information about the suppliers, such as services offered and the cost (price). Likewise, suppliers advertise their services and the costs via the third party providers in order to attract consumers. Consumers prefer an easy and simple way to query for supplier information, and suppliers prefer to advertise information securely and quickly across many regions or domains. For example, consider a user who wishes to view a multimedia object (such as a video movie). The user would like to know about the suppliers of this object, and the cost of retrieval of this object from each supplier. Performance Requirements Users wish to have good response time for their search results once the queries are submitted. However, there is a tradeoff. For more information about services offered, advanced searching mechanisms are needed, but at the cost of increased response time. In other words, users could have preferences over quality of search information and response time. For example, users might want to know the service costs in order to view a specific information object. In large networks, there could be many suppliers of this object, and users may not want to wait forever to know about all the suppliers and their prices. Instead, they would prefer to get as much information as possible within a certain period of time (response time). From the above example, in order to let many consumers find suppliers, a scalable decentralized architecture is needed for information storage, access and updates. Naming of services and service attributes of suppliers becomes a challenging issue when hundreds of suppliers spread across the globe. A simple naming scheme to connect consumers, across the internet, with information about suppliers is essential. The naming scheme must be extensible for new suppliers who come into existence. A name registration mechanism for new suppliers and a de-registration mechanism (automatic) to remove non-existent suppliers is required. In addition, naming must be hierarchical, domain based (physical or spatial domains) for scalability and uniqueness. Inter-operability with respect to naming across domains is an additional challenging issue not covered in this Chapter.

106 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

The format of information storage must be simple enough to handle many consumer requests quickly within and across physical domains. For better functionality and more information, a complex format of information storage is necessary, but at the cost of reduced performance. For example, a consumer, in addition to current service cost, might want to know more information such as the cost of the same service during peak and off-peak hours, the history of a supplier, its services, and its reputation, in order to make a decision. This information has to be gathered when requested. In addition, the storage formats must be interoperable across domains. Performance A good response time is important to make sure consumers get the information they demand about suppliers within a reasonable time period, so that decisionmaking by consumers is done in a timely fashion. In addition, the design of the right architectures for information storage and dissemination is necessary for a large scale market economy to function efficiently. Using the previous example, consumers and suppliers would prefer an efficient architecture to query for and post information. Consumers would prefer good response time in obtaining the information, and suppliers prefer a secure and fast update mechanism to provide up-to-date information about their services. Security in transferring information and updating information at the bulletin boards (name servers) is crucial for efficient market operation and smooth interaction between consumers and suppliers. For this the third party suppliers (naming services) have to provide authentication and authorization services to make sure honest suppliers are the ones updating information about their services CONCLUSION We show some applications of mechanism design and queueing systems to resource management problems in distributed systems and computer networks. These concepts are used to develop effective market based control mechanisms, and to show that the allocation of resources are Pareto optimal. We propose novel methodologies of decentralized control of resources, and pricing of resources based on varying, increasingly complex QoS demands of users. We bring together economic models and performance models of computer systems into one framework to solve problems of resource allocation and efficient QoS provisioning matching large-scale e-commerce applications. The work can be applied to pricing services in ATM networks and (wireless) Integrated Services Internet of the future. We address some of the drawbacks to this form of modelling where several agents have to use market mechanisms to decide where

Generalized Quality of Service

Internet Economics: Models, Mechanisms and Management 107

to obtain service (which supplier?). If the demand for a resource varies substantially over short periods of time, then the actual prices of the resources will also vary causing several side effects such as indefinite migration of consumers between suppliers. This might potentially result in degradation of system performance where the resources are being underutilized due to the bad decisions (caused by poor market mechanisms) made by the users in choosing the suppliers. As in real economies, the resources in a computer system may not easily be substitutable. The future work is to design robust market mechanisms and rationalized pricing schemes which can handle surges in demand and variability, and can give price guarantees to consumers over longer periods of time. Another drawback is that resources in a computer system are indivisible resulting in nonsmooth utility functions which may yield sub-optimal allocations, and potential computational overhead. In addition to models for QoS and pricing in computer networks, we are also working towards designing and building distributed systems using market based mechanisms to provide QoS charge users either in a commercial environment or in a private controlled environment by allocating quotas via fictitious money (charging and accounting) by central administrators. REFERENCES Deng, X., Graham, F.C. (2007). Internet and Network Economics Third International Workshop. WINE San Diego, Berlin, New York: Springer Dodis, Y., Rabin, T. (2007). Cryptography and Game Theory In: Nisan, N., Roughgarden, T., Tardos, E., Vazirani, V.V., (Eds.), Algorithmic Game Theory chapter 8 . (pp. 181-205). Cambridge: Cambridge Univ. Press. [http://dx.doi.org/10.1017/CBO9780511800481.010] Feigenbaum, J., Schapiro, M., Shenker, S. (2007). Distributed Algorithmic Mechanism Design In: Nisan, N., Roughgarden, T., Tardos, E., Vazirani, V.V., (Eds.), Algorithmic Game Theory chapter 14 . (pp. 363-384). Cambridge: Cambridge Univ. Press. [http://dx.doi.org/10.1017/CBO9780511800481.016] Ferguson, D.F., Nikolaou, C., Sairamesh, J., Yemini, Y. (1995). Economic Models for Allocating Resesources in Computer Systems In: Clearwater, S., (Ed.), Market-Based Control : A Paradigm for Distributed Resource Allocation. Singapore: World Scientific. Gottinger, H.W. (2009). Strategic Economics for Network Industries. New York: NovaScience. Kleinrock, L. (1996). Queueing Networks. (Vol. Vol. 2). New York: John Wiley and Sons. Kleinrock, L., Gail, R. (1996). Queueing Systems: Problems and Solutions.. New York: John Wiley and Sons. Macias, M., Smith, G., Rana, O.F., Guitart, J., Torres, J. (2010). Enforcing Service Level Agreements using an Economically Enhanced Resource Manager In: Neumann, D., Baker, M., Rana, O.F., Altmann, J., (Eds.), Economic Models and Algorithms for Distributed Systems, 109-125.Basel: Birkhaeuser. Mohammed, A.B., Altmann, J., Hwang, J. (2010). Cloud Computing Value Chains: Understanding Business and Value Creation in the Cloud In: Neumann, D., Baker, M., Rana, O.F., Altmann, J., (Eds.), Economic Models and Algorithms for Distributed Systems (pp. 187-208). Basel: Birkhaeuser. Myerson, R.B. (2006). Fundamental Theory of Institutions: A Lecture in Honor of Leo Hurwicz. Available at:

108 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

http://home.uchicago.edu/~rmyerson/hurwicz.pdf Neumann, D., Baker, M., Altmann, J., Rana, O.F. (2010). Economic Models and Algorithms for Distributed Systems.. Basel: Birkhaeuser. [http://dx.doi.org/10.1007/978-3-7643-8899-7] Nisan, N., Ronen, A. (2001). Algorithmic Mechanism Design. Games Econ. Behav., 35, 166-196. [http://dx.doi.org/10.1006/game.1999.0790] Nisan, N., Roughgarden, T., Tardos, E., Vazirani, V.V. (2007). Algorithmic Game Theory. Cambridge: Cambridge Univ. Press. [http://dx.doi.org/10.1017/CBO9780511800481] Ross, S. (1970). Applied Probability Models with Optimization Applications. New York: Dover. Zhang, J., Cohen, R. (2007). Towards more effective emarketplaces: A novel incentive mechanism Proc. 6th Joint Conferenceon Autonomous Agents and Multiagent Systems (AAMAS’07).

Internet Economics: Models, Mechanisms and Management, 2017, 109-124

109

CHAPTER 6

Network Platforms “The platform business model is powered by a new set of factors that determine value creation and competitive advantage. These factors are rapidly changing how entire industries operate.” S. Paul Choudary, Platform Scale, 2015 Abstract: A new property of Internet enabled communication and computation opens up a new business model of ‘plug and play’ that is intrinsically linked to the interactive social and commercial use of websites through the World Wide Web (WWW), i.e. two-sided or multi-sided platforms. We analyze a model of platform operations that involves a sequential decision process very much alike a dynamic programming algorithm (DPA) for selecting functionalities of the platform. This could serve as a simple approximation procedure for building the optimal design of a platform. The platform business in a vertically integrated supply chain as well as toward product development through platforms rather than pipelines is a good example of facilitating ‘increasing returns mechanisms’ (IRM), as one can follow the evolution of the Amazon platform in a commercial context but also on the growth of social media like Facebook and LinkedIn. Such a business growth would not have been facilitated without a dedicated, universal, easily accessible and low cost network economy provided by the Internet. From an economic and business perspective, the network platform business utilizes direct or indirect network effects to attract customers and facilitate network economies, and determine how to switch from a ‘pipeline model’ (product line model)to a ‘platform model’. Some business and social network platforms (eBay, Facebook) have been able to harness network effects to fuel truly continuous growth. On the other hand, platform businesses, even without fixed costs or economies of scale, need to acquire critical mass through intensity and scale after they are launched to survive and grow. We may observe ‘tipping point’ effects generated through ‘chicken and egg’ problems in platform building processes.

Keywords: Consumer Utility, Cross-Externalities, Direct/Indirect Network Effects, Dynamic Programming Algorithm, Increasing Returns Mechanism (IRM), Java Specifications, Multi-sided Platforms, Multimedia Services, Platform Business, Platform Economics, Platform Scale, Platform Utility, Social Network Platform, Software Platforms, Supply Chain, Virtual Reality (VR), Web Services, World Wide Web. Hans W. Gottinger All rights reserved-© 2017 Bentham Science Publishers

110 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

6.1. INTRODUCTION Network platforms are a major ingredient of network economies, in fact, they are one of their distinctive features. They provide a wide-ranging applicability and use in multi-sided businesses from financial services (credit cards), healthcare to video games. In particular, they have an increasingly important proliferation in media industries or social networks. Companies such as Google, Ebay or Amazon could not operate as such, or much less efficiently, if it were not for software platforms. The financial service industry used software platforms to leverage (subprime) loans and commercial papers with world-wide exposure – allowing multi-sided business (buyers/sellers). From a consumer’s viewpoint, software platforms are the ‘virtual reality’(VR) equivalent of physical shopping malls. YouTube, Facebook,Twitter are social platforms while digital business evolves through market-place platforms like Amazon and eBay. In terms of product provision, platforms vs. pipelines function through ‘plug and play’. They benefit from superior marginal economics of scaling supply plus various network effects. As in Chapter 7, covering the Internet of Things (IoT), IoTs use various data platforms, technical platform designs are essential features for optimizing IoT generated industrial systems. The much heralded ‘blockchain revolution’ (Tapscott and Tapscott, 2015) would be a platform on a trusted network. From an aggregate or macro perspective, platform economics helps to organize an emerging ‘sharing economy’ which could become the constitutional part of a network economy (Sundararajan,2016). It appears that new digital businesses shift from a pipeline to a platform model with implications to increasing ‘total factor productivity’ that benefits the growth of the national economy.Economies which fall behind in this shift will lose the digital innovation race (Chap. 9). The network platform business utilizes direct or indirect network effects to attract customers and facilitate network economies. How to switch from a ‘pipeline model’ (product line model) to a ‘platform model’? Some business and social network platforms (eBay, Facebook) have been able to harness network effects to fuel truly continuous growth. On the other hand, platform businesses, even without fixed costs or economies of scale, need to acquire critical mass when they are launched even to survive (Evans and Schmalensee, 2010), and we may observe ‘tipping point’ effects or failure of ‘critical mass’ generated through ‘chicken and egg’ problems in platform building processes. They show that in the case of direct network effects, the basic problem is that the level of participation on the platform affects the quality of the products offered to participants, and if the quality is too low, participation falls, which might reduce quality further, to a downward spiraling effect. In the case of indirect network effects participation by each customer group affects the quality of the product experienced by the other group and the dynamics may set off a similar upward or downward spiral.

Network Platforms

Internet Economics: Models, Mechanisms and Management 111

What is the rationale to design Internet platforms? A platform typically provides a set of capabilities through built-in application interfaces (APIs), modules, tools, etc., which make it easier for developers to innovate new applications and services of interest/value to consumers. This, however, comes at a cost to the platform, and this cost grows with the number of features offered. The question for the platform provider is then to determine the number of features that make profits. The issue is here how to activate positive direct, indirect and crossexternalities such as ‘bandwagon effects’ on platform building (Majumdar et al., 2005, Chaps. 3,4). With platform design capabilities, limited applications tend to be more complex, therefore, the number of application developed for the platform is made limited. This makes the platform less attractive to consumers and lowers revenues. On the other hand, a scalable and diversifiable platform is expensive to build, but the cost may be offset by facilitating the development of more applications, therefore attracting more consumers. This trade-off arises in many environments and properly assessing it can have farreaching consequences for sustainable business success or failure. For example, many attribute initial success to the Internet for its simple and transparent design principles. However, as it matures and transforms from a ‘physical’ network platform to a broader ecosystem of software and web services, the question is how new scalable features can be successfully integrated. (Choudary, 2015). The platform business has proliferated over a decade, ranging from established pioneering companies, i.e. from Google, Amazon, Apple, FaceBook, Twitter, Alibaba to strongly growing firms in the services and retail sectors, to Uber, Lyft, Airbnb and eBay, to those creating platform businesses including IBM, Intel, Microsoft, SAP, Salesforce and plenty of startups worldwide in the network/ software business. More and more established firms across a wide industrial spectrum such as WalMart, Nike, GE, Siemens, ABB, Disney, CocaCola and Hilton among many others are adopting a platform model embracing their supply chain. One focus of this chapter is to explore the decision problem faced by a platform provider seeking to select the level of functionality or features the platform should offer. From an architectural perspective, the creation of the worldwide web as a network hub turned into an interactive system as a two-sided market place facilitates the transformation of the Internet from a physical network to an ecosystem of software. The network is the platform and seeks to ‘connect’ users to services. The present day, Internet offers examples of network and operating system platforms whose success largely comes from their ability to connect users and service/application developers.

112 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

Services are offered by developers and asked for by the consumers that rely on the platform and its features. A feature-rich platform facilitates service development, which yields more services. This in turn attracts more users and benefits the platform (Evans, Schmalensee and Hagiu, 2006; Evans and Schmalensee, 2005). A platform provider attracts developers and consumers by creating value that entices them to join the platform. This ‘value’ depends on a number of factors, such as the subscription fees to join it, the cost of developing applications for it, and externalities that affect the value that either developers or customers derive from joining the platform. In a two-sided market, both sides of the market derives cross-externality benefits from the presence of the other, i.e., consumers benefit from more applications offered by developers, and conversely developers benefit from being able to target their applications to more consumers. This fits many software products and services, where platform and applications share a common technology. In general, the construction proceeds on a model’s applicability to platforms that are software ecosystems, e.g., cloud computing, web services, operation systems, etc. In the order of real applications, we observe PC platforms, video game platforms, personal digital assistant (PDA) platforms, smartphone (mobile) platforms and digital media platforms. In principle, such a network platform design would proceed and could be solved using a sequential decision process for the platform to select the level of functionality to offer. Thus we consider platform launch as a decision process rather than an event only as it was considered in the previous literature (Caillaud and Julien, 2003). In the first stage, the platform provider chooses the number of features to build into the platform. Given this choice for a number of features, participation prices (fees) for the two market sides are chosen in the second stage. Compatible capacity levels of consumers and developers are simultaneously realized in a third stage. This sequential decision process is then solved in the reverse order – like a dynamic programming (DP) framework. Capacity levels for users and developers are first computed for a given choice of participation prices and number of built-in features. Next, given a choice for the number of built-in features, ‘optimal’ participation prices are computed based on the scale levels of the previous step. The results characterize the platform’s profit for any given number of built-in features. This is then used to find the ‘optimal’ number of features that maximizes the platform’s profit.

Network Platforms

Internet Economics: Models, Mechanisms and Management 113

This chapter is organized as follows: Section 6.2 sketches a simple applicable two-sided market model. Section 6.3 outlines the solution methodology for this work. Section 6.4 presents the analysis and explores the impact of different factors on the level of platform functionalities. Section 6.5 reviews prior works and positions this work in the extensive internet literature. Section 6.6 concludes the chapter with remarks on future work. 6.2. TWO-SIDED PLATFORMS A platform provider attracts developers, vendors and consumers by creating value that entices them to join the platform. This ‘value’ depends on a number of factors, such as the platform’s intrinsic value, the subscription fees to join in, the cost of developing applications for it, and the (direct and indirect) network externalities that affect the value that either developers or customers derive from joining the platform. When modeling a platform as a two-sided market, externalities are usually distinguished as same-side externalities and cross-side externalities. Same-side externalities arise in each side of the market from the presence of other users and can be positive or negative (Katz and Shapiro, 1994, Shapiro and Varian, 1999). Cross-side externalities measure benefits that one side of the market derives from the other. These are usually positive, i.e., consumers benefit from more applications offered by developers, and conversely developers benefit from being able to target their applications to more consumers. The adoption of the platform by either developers or consumers depends on the overall value they derive from it. As commonly done, we measure this value through a utility function that incorporates the different factors that contribute to it. Similarly, the impact of the decisions that the platform provider makes, i.e., pricing and selection of the platform’s scope and scale, are also reflected through the platform provider’s utility function. The utility functions for the platform, the developers, and the consumers are described in rudimentary terms. Next we briefly review a number of assumptions we make in the model and their implications on its applicability. Assumptions and Implications Under normal circumstances, we could assume that developers generate revenue from advertisements and not from consumers purchasing the applications, i.e., application downloads are free and transaction costs for both developers and consumers are negligible, equally reverse the transaction costs between the parties.

114 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

This is reasonable in many settings where applications are offered for free and the bulk of the developer’s revenue comes from location based and personalized advertising (Schwartz. 2010). The advertising revenue generated by an application is also assumed to be linear in the number of users of the applications. Furthermore, all applications make use of the same set of platform features, and scale and scope embedded in these features can be built by either the platform or the developers themselves. Platform Utility The goal of the platform provider is, for simplicity, to maximize its profit which depends on the revenue it generates from the two sides of the market and the cost of the features it has decided to embed in the platform. We use xc and nd to denote the fraction of a large population of Nc and Nd consumers or developers, respectively, who join the platform. As in Bakos and Katsamakas (2008), the platform charges flat fees of pc and nd to the consumers and to the developers, respectively. These fees may be incurred as a monthly membership fee for consumers and as a licensing or certification fee for application developers. The revenue for the platform is, therefore, pc xc + bd nd. The set of platform features of potential benefits to application developers is assumed known to the platform provider. Embedding more desirable choices (features) in the platform incurs a greater cost, and we denote as C(F) the cumulative cost of incorporating F. We assume that the set of possible features is large. Hence, when mapped on to an interval [0, maxF ], they result in a differentiable, monotonically increasing function of C(F) for F ∈ [0, max F] . An integrability constraint on F is, therefore, not considered explicitly. In Section 6.4 we discuss specific, real-world examples that illustrate different possible behaviors for C(F), i.e., concave or convex. The profit (utility) of a platform with F built-in features and fees of pc and bd is given by: U(p) = pc xc + bd nd − C(F)

(1)

Together with similar expressions for the utility of consumers and application developers will guide the real-valued decisions d(F) of how many features to embed in the platform and how to price it.

Network Platforms

Internet Economics: Models, Mechanisms and Management 115

Developing applications incurs a certain cost, which depends on the level of support provided by the platform (d(F) and number of features). A platform of few and limited features will usually have lower subscription fees, which may partly offset its correspondingly higher development costs. The revenues generated by an application depend on the number of users the platform has managed to attract, and grow in proportion to that number. Equation (2) captures the combined effect of these factors on the developers’ utility. U(d) = αxc − bd - (K(F) + τϕ)

(2)

The first component of Equation (2) represents the application revenues generated from the xc consumers that joined the platform (the factor α is a normalization constant that can also be interpreted as the marginal value that a consumer generates for the developer). Those revenues are in the form of advertising revenues, as is commonly assumed in many two-sided markets. For example, the online service iLike has developed a free application for Facebook (the platform) that based on a user profile allows her to play clips of music she may like, and collect revenues from both advertising and referrals to iTunes or Ticketmaster. The second component of Equation (2) is the flat-fee, bd, a developer pays the platform, e.g., to be certified. This charging model is used by many platforms. The last component of Equation (2) reflects development costs, which as alluded to earlier depends on the number F of features embedded in the platform. This is captured by the function K(F) that is a differentiable, monotonically decreasing function of F [0, maxF]. As for C(F), there are illustrative, real-world examples associated with different behaviors for K(F). In particular, K(F) can be convex or concave depending on whether the marginal reduction in development costs is increasing or decreasing as the platforms add more features. For a given F, K(F) is the same for all developers. Hence, it can be interpreted as the base cost of developing applications when the platform includes F built-in features. This assumes that developers have similar expertise in developing applications, e.g., software engineers draw from a similar skill base. Developers can, however, be expected to exhibit heterogeneity in their overall development costs, e.g., because of different fixed costs, overhead, benefit levels, etc. This is captured in the factor τϕ of Equation (2), where ϕ is uniformly distributed on a unit interval (Yoo et al., 2002). The value of ϕ for an individual developer is private information, but the distribution is known to all. Consumer Utility The value that consumers derive from joining a platform depends on the subscription fees charged, and the number of applications and services accessible

116 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

through the platform. Consumers are typically heterogeneous, and this heterogeneity can manifest itself in how they value the platform, access to applications and services (cross-side externality), or both. For simplicity and analytical tractability, we focus on a model where heterogeneity is present only in how users value access to applications. The consumer utility function is then of the form: U(c) = θβnd − pc

(3)

The first component, θβnd, captures the cross-side externality benefits that consumers enjoy from accessing applications available on the platform. These benefits grow with the number of developers, nd, creating applications for the platform, e.g., the many (say) iPhone developers being responsible for the large number of applications available on it, which contributes to its attractiveness. The assumption of linear growth in nd is consistent with the previous literature (Bakos and Katsamakas, 2008). The factor β denotes the marginal externality benefit associated with each developer. The term θ ∈ [0, 1] is a random variable that accounts for heterogeneity in how users value these externality benefits. A user who uses many applications will have a higher θ value, and thus derive higher externality benefits for a given number of application developers. The value of θ for individual users is private information, but its distribution across users is known. We make the common assumption that θ is uniformly distributed in [0, 1]. The last element of Equation (3) is the price pc, which is a flat membership fee paid to the platform provider. An integrated analytical treatment of this decision procedure is due to Sen (2011). 6.3. REVIEW OF PLATFORM ECONOMICS This review focuses on two streams of information systems and network economics literature, namely, (i) Two-sided markets and (ii) Platform intermediaries in e-commerce markets. The model shares a key structural element with the two-sided markets literature in that we consider a platform provider who facilitates interaction between two interdependent customer groups (i.e., App developers and consumers). We also address the topics of platform pricing and customer adoption decisions which have received considerable attention in many of earlier works (Armstrong, 2006; Weyl, 2010). Another stream of literature mentioned above focus on the question of how can a platform invest in impacting the cross-side network effects in an e-commerce

Network Platforms

Internet Economics: Models, Mechanisms and Management 117

market so as to increase its profits in a supply chain. Platform formation in targeted media industries creates the selective role of gatekeeping in value generation for both sides (Tonon, 2010) . A key difference from these previous works is that the platform provider in the present model does not directly alter the network effects, instead it invests in adding platform functionalities that reduce the development costs for application developers. Such a scenario is typical in many software ecosystems where the trade-offs lie in the costs borne by the platform in adding functionalities and the benefits these bring to the developers. It also opens up an interesting question that we investigate, namely, what factors influence whether a platform has scale and scope. Two-sided markets: Two-sided markets are made of two interdependent groups of customers (e.g., sellers and buyers) who benefit from each other’s participation. A platform intermediary facilitates interactions between these two customer groups and generates its revenue by charging them a price for joining the platform. The adoption of the platform by the two customers groups and the volume of interaction between them depend not only the prices set by the platform provider but also on the price structure (Rochet and Tirole, 2006). They provide definition and examples of two-sided platforms along with extensive literature survey on the topic. Two-sided models have also been used recently to analyze net-neutrality issues (Economides and Tag, 2007). Several works have also focused on pricing strategies for two-sided platform (Hagiu, 2006; Parker and VanAlstyne, 2005). While our work builds on the existing literature, we use the two-sided platform model to consider issues that were not the focus of these earlier works. We consider software based platforms that bring application developers and consumers together, and focus on the question of features design. This environment is particularly relevant in the context of today’s Internet as it evolves from a physical infrastructure to a software ecosystem. The success of the Amazon Web Services platform, Facebook platform etc., bear witness to the progress made in that direction. Although the Internet started out with a simple grid design, its gradual evolution raises new questions as to which functionality is desirable in this new environment. The model of the two-sided platform we develop is aimed towards creating an analytical framework that explores such questions. Platform intermediaries in e-commerce markets: The literature in electronic intermediaries is also related to this work. Electronic market intermediaries lower search costs for buyers and increases price competition among sellers (Bakos, 1991). While much of the work in this area have analyzed the role of electronic intermediaries, some have focused on the impact of infrastructural investments of the intermediary platform on its cross-network externalities. Balos and

118 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

Katsamakas (2008) show that it is optimal for an intermediary to invest in network externalities asymmetrically to maximize the network benefits for one market side. Following up, we consider scenarios where the platform provider does not have the means to directly impact the cross-externalities. Instead, the platform can invest in features that make application development easier for developers, and thus indirectly influence the customer adoption levels and pricing. Such scenarios are typical for most web services and social network platforms where the level of investment decisions determine how the costs and benefits are shared by the platform and its developers. One focus is different from previous works, e.g. the interest in analyzing the factors that influence the platform’s decision regarding a fitting design. The results would contribute to the growing literature on e-commerce intermediary investments and platform design. Platform enterprises rely on direct or indirect network effects to attract customers, face demand-side constraints when launched that distinguish themselves from conventional pipeline product companies. As the emergence of such entgerprises show us, such as Amazon’s marketplace, eBay’s auction and Facebook’s social networking platform (SNP), some platforms have been able to harness network effects to leverage extraordinary growth on an annual basis. On the other hand, we equally observe why even without fixed costs or economies of scale platform enterprises typically need to attain critical mass when they are launched even to survive. The preponderance of tipping of network advantages in two-sided markets is building a consistent lead that is hard to reverse. In the case of direct network effects, the basic problem is that the level of participation on the platform affects the quality of the product it offers to participants, and if quality gets to low, participation falls,which reduces quality further, and participation converges toward zero. In the case of indirect network effects, being of more widespread concern, participation by each customer group affects the quality of the product experienced by the other group and, though the dynamics are less straightforward, participation levels below critical mass will settle in a similar downward spiral. In a ‘chicken and egg’ situation participation adjusts more rapidly or slowly upward or downward toward an equilibrium state. Of particular interest is the study of competition and competition policies of existing platforms and with new entrants to the platform business. For example, in view of market design and platform competition we observe two groups of agents to interact via a platform. An agent’s utility from using the platform increases in the number of agents on the other side of the market. This incurs some cost since his net utility depends positively on the size of the other user group, but negatively on the fee paid for access to the platform. But both parties depend on the utilities derived from joining the platform. The platform’s profit is then the sum of the revenues extracted from both parties, net of costs. Because of their growing importance and

Network Platforms

Internet Economics: Models, Mechanisms and Management 119

the diversity of their strategies, multi-sided platforms are increasingly subject of antitrust intervention in competition policies as covered in Gottinger (2016, Chap. 3). 6.4. PLATFORM OPERATIONS By presenting how platform utility functions introduced in Sec. 6.2 combine in the platform provider decision process, we introduce examples that illustrate possible combinations of the cost functions C(F) and K(F). In all examples, there is an inherent ordering of the features the platform provider is considering offering, i.e., from basic features to more advanced ones, with the latter building on the former. The examples differ in the relative cost of more advanced features compared to basic ones, and in how useful each additional feature is to application developers. Here we sketch the main combinations of interest between the costs of platform features and their benefits to consumers and application developers. From many early platforms and dedicated applications (Evans, 2003) to largescale industrial platforms as technically unified and described by Sen (2011). 1. Amazon Web Services Platform Amazon Web Services (AWS) is a cloud computing platform that offers functionalities which third-party developers can use to create services for clients (consumers). These functionalities or features include Amazon EC2 (computation), SimpleDB (database), Amazon S3 (storage), FPS (flexible payment), CloudFront (content delivery), MTurk (Internet marketplace), etc. Consumers and developers of services and applications on the AWS platform enjoy cross-side externality benefits from joining the platform, for which they pay subscription fees (AWS, 2017). The introduction of features on the AWS platform proceeded in two steps. Amazon introduced a set of core features (EC2, FPS, SimpleDB, etc.) that offered basic capabilities such as computation, database, and payment for its AWS platform. Additional features (e.g., SQS, SNS, DevPay, etc.) that built on these core capabilities were subsequently introduced. Adding each feature to the AWS platform came at a cost. Using API complexity as a proxy for the platform’s development cost together with data from (Garnatt, 2010), it can be observed that capabilities such as EC2, FPS, and SimpleDB came at a higher cost than that of follow-on enhancements such as SNS and DevPay. From this data, one can infer that the AWS platform has a feature development cost function, C(F), that is a concave increasing function of F.

120 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

Conversely, the benefits of each feature can be estimated based on its ‘popularity’ among developers, i.e., presuming that more useful features are more likely to be used by developers. Using again Garnatt (2010), we see that most core features are significantly more popular than subsequent enhancement features. In other words, the features that were the most costly to develop and incorporate in the platform were also the most useful to developers; at least based on how often developers took advantage of them. As a result, one can conclude that while the development cost function C(F) of the AWS platform is a concave increasing function, the benefits that developers derive from those features, as captured by the function K(F), is a convex decreasing function, i.e., the more expensive initial features are also the most useful. 2. IP Multimedia Systems (IMS) Platform The IMS platform is meant to facilitate the development of new integrated multimedia applications and services. Both applications developers and subscribers (consumers of services) pay a fee to the IMS platform. The platform offers a number of built-in capabilities such as a registration mechanism, colocation of multiple IMS services, quality of service, etc. These capabilities are exposed to developers through APIs using Java specifications (JSRs), JSR (2010). The development cost function C(F) of the IMS platform is a concave increasing function of the number F of features (JSRs) it offers. On the developer side, application development costs are high when only lowlevel APIs are available. This is mainly because of the greater technical knowledge and programming consistency they require from developers. As APIs that hide many of the platform’s low-level intricacies are made available, development costs decrease rapidly. In other words, the function K(F) that captures development costs as a function of the number F of features (APIs) that the platform offers is a concave decreasing function, i.e., low-level APIs have little effect on developers costs, while higher-level ones deliver significant benefits. 3. Social Network Platform With location-based services (LBS) support: A social network platform such as Facebook provides application developers access to basic capabilities, e.g., APIs to access the users’ social graph, database of user interests, affiliations, etc. However, it also offers more sophisticated capabilities such as realtime updates and location-based services (LBS). These have enabled the rapid growth of

Network Platforms

Internet Economics: Models, Mechanisms and Management 121

applications that offer personalized services, e.g., Facebook’s Recommendation and Places (Reed, 2010). Adding this level of sophistication to the platform is, however, technically challenging. It calls for integrating capabilities such as spatial database management, location tracking, real time generation of cryptographic data, all of which are significantly more complex than the basic features at the core of a social network platform, e.g., access to the underlying social graph or to a user database. In other words, the function C(F) that captures the cost of adding new (sophisticated) capabilities to a social network platform such as Facebook is a convex increasing function of F. On the other hand, the benefits to application developers of those advanced features can be very high. For example, in the absence of LBS support from the platform, developers would need to build this capability into their application, e.g., by interfacing to the GPS service built into the user’s mobile device, when available. That those development costs are high is readily seen from the growth in the number of applications that rely on location information once LBS became available. Specifically, in spite of the large revenue potential of location-based services, there were relatively few applications that used location information before LBS became readily accessible to application developers (Göx, 2002). In other words, the substantial decrease in development costs that this produced, enabled many more developers to offer such applications. As a result, it can be argued that a social network platform such as Facebook is an environment where while sophisticated features are expensive to add, they are the ones that deliver the most benefits (reduction in development costs) to application developers. CONCLUSION This chapter develops a model to explore when a platform carries designs in scale and scope .The question is formulated using a two-sided market model in which the platform is the market and service developers and consumers are the two sides of the market. Consumers, developers and the platform provider have utility functions that account for externality benefits, prices, and costs. The platform provider seeks to identify how many features to include in the platform to maximize its profit. This is formulated as a three stage sequential decision process, for which the solution is characterized. The solution reveals the impact of cross-externalities and confirms the benefits of asymmetric pricing. More importantly, it shows that the platform choice is highly

122 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

sensitive to how additional features affect the costs of the platform and service developers. This unfortunately establishes that minor changes in either costs can yield drastically different solutions. This is illustrated through numerical examples, which point to the significant challenge of answering the question of optimal platform design in practice. In spite of the limitations of its results, the work provides initial insight and a possible methodology for tackling the complex question of (software) platform design. There are many directions in which the work’s initial results can be extended. Empirical validations are obviously at the forefront, and exploring if this can be done for one of the examples of Section 6.4 is of interest. Some modeling extensions are also worth pursuing. One involves allowing developers to use different subsets (bundles) of features, each assigned a different price. Platforms generate value by bringing together two interdependent groups of customers, the developers and the consumers, to provide functionalities to foster service innovation and user adoption. In doing so, we considered a framework of a two-sided market, with the platform as the market and the developers and consumers as the two sides of this market. We introduced utility functions for the two sides and the platform and incorporated in them the impact of network externalities, prices, and functionality development costs. The investigation revealed a number of interesting properties depending on how the cost of features to the platform and the benefits application developers derive from them relate to each other. In particular, it showed that the ratio of the marginal increase in platform provider’s cost to the marginal decrease in application developer’s cost from an increase in the platform’s functionalities play an important role in determining the optimal functionality level. Besides contributing to the growing economics literature on two-sided markets. A platform moving up in scale and in a variety of features and interactively producing value could be a better economic design and business model than the one that produces values through a pipe and moves the product to consumers. Related to the questions of new service deployment and adoption is the issue of the right design for the underlying network platform (e.g., Internet). Platforms generate value by bringing together two interdependent groups of customers, the developers and the consumers, and provide capabilities to foster service innovation and user adoption. REFERENCES Armstrong, M. (2006). Competition in Two-Sided Markets. Rand J. Econ., 37(4), 668-691.

Network Platforms

Internet Economics: Models, Mechanisms and Management 123

[http://dx.doi.org/10.1111/j.1756-2171.2006.tb00037.x] AWS. (2017). Overview of Amazon Web Services (AWS Whitepaper). Amazon. Bakos, Y. (1991). Reducing buyer search costs: Implications for electronic marketplaces. Manage. Inf. Syst. Q., 15(3), 295-310. [http://dx.doi.org/10.2307/249641] Bakos, Y., Katsamakas, E. (2008). Design and ownership of two-sided networks: Implications for internet platforms. J. Manage. Inf. Syst., 25(2), 171-202. [http://dx.doi.org/10.2753/MIS0742-1222250208] Choudary, S.P. (2015). Platform Scale. Boston: Platform Thinking Labs Pte.Ltd. Economides, N., Tag, J. (2007). Net neutrality on the internet: A two-sided market analysis. Working Paper no. 07-14, NET Institute, New York University. Evans, D. (2003). Some Empirical Aspects of Multi-Sided Platform Industries. Rev. Netw. Econ., 2, 191-209. [http://dx.doi.org/10.2202/1446-9022.1026] Evans, D.S., Hagiu, A., Schmalensee, R. (2006). Invisible Engines: How Software Platforms Drive Innovation and Transform Industries.. Cambridge, Ma.: MIT Press. Evans, D.S., Schmalensee, R. (2010). Failure to Launch: Critical Mass in Platform Businesses. Aspects of Multi-Sided Platforms, Conference in Competition in High-Tech Markets, Univ. of Milano, Bicocca. [http://dx.doi.org/10.2202/1446-9022.1256] Garnatt, M. (2010). AWS by the numbers. Available at: http://www.elastician.com/2010/06/aws-bynumbers.html Göx, R.F. (2002). Capacity planning and pricing under uncertainty. J. Manage. Account. Res., 14(1), 59-78. [http://dx.doi.org/10.2308/jmar.2002.14.1.59] Gottinger, H.W. (2016). Networks, Competition, Innovation and Industrial Growth. New York: Nova Science. Hagiu, A. (2006). Pricing and commitment by two-sided platforms. Rand J. Econ., 37(3), 720-737. [http://dx.doi.org/10.1111/j.1756-2171.2006.tb00039.x] Hagiu, A. (2009). Multi-Sided Platforms: From Microfoundations to Design and Ex-pansion Strategies. Harvard Business School Strategy Unit Working Paper No. 09-115. JSR. (2010). IMS Communication Enablers (ICE). Available at: http://jcp.org/en/jsr/detail?id=325 Katz, M.L., Shapiro, C. (1985). Network externalities, competition, and compatibility. Am. Econ. Rev., 75(3), 424-440. Katz, M.L., Shapiro, C. (1994). Systems competition and network effects. J. Econ. Perspect., 8(2), 93-115. [http://dx.doi.org/10.1257/jep.8.2.93] Majumdar, S., Vogelsang, I., Cave, M. (2005). Technology Evolution and the Internet. (Vol. 2). Amsterdam: North Holland.Handbook of Telecommunications Economics Parker, G.G., Van Alstyne, M.W. (2005). Two-sided network effects: A theory of information product design. Manage. Sci., 51(10), 1494-1504. [http://dx.doi.org/10.1287/mnsc.1050.0400] Reed, R. (2010). Facebook places: The de facto platform for lbs. Available at: http://momentfeed.com/ 2010/11/facebook-places-platform-lbs/ Rochet, J.C., Tirole, J. (2003). Platform Competition in Two-Sided Markets. J. Eur. Econ. Assoc., 1, 9901029. [http://dx.doi.org/10.1162/154247603322493212] Rochet, J.C., Tirole, J. (2006). Two-sided markets: a progress report. Rand J. Econ., 37(3), 645-667.

124 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

[http://dx.doi.org/10.1111/j.1756-2171.2006.tb00036.x] Schwartz, E.I. (2010). Behind the surge in mobile advertising. Available at: http://www.technology review.com/business/26648/ Sen, S. (2011). On the Economic Viability of Network Systems and Architecture Ph.D. Diss., Electrical and Systems Engineering, The Univ. of Pennsylvania, UMI, Ann Arbor, Mich. Shapiro, C., Varian, H.R. (1999). Information Rules: A Strategic Guide to the Network Economy. Harvard Business School Press. Sundararajan, A. (2016). The Sharing Economy. Cambridge, MA: MIT Press. Tapscott, D., Tapscott, A. (2016). Blockchain Revolution. New York: Penguin. Tonon, J-C. (2010). The Role of Gatekeeping in the Music Industry: Why Bad Artists Might Prefer SelfPromotion IMPRS-CI, Munich, and ICE, Munich School of Management, Univ. of Munich (LMU). Weyl, E.G. (2010). A Price Theory of Two-Sided Platforms. Am. Econ. Rev., 100(3), 1642-1672. [http://dx.doi.org/10.1257/aer.100.4.1642] Yoo, B., Choudhary, V., Mukhopadhyay, T. (2002). A model of neutral B2B intermediaries. J. Manage. Inf. Syst., 19(3), 43-68. [http://dx.doi.org/10.1080/07421222.2002.11045734]

Internet Economics: Models, Mechanisms and Management, 2017, 125-143

125

CHAPTER 7

The Internet of Things and the Industrial Internet The Internet of Things is “the first real evolution of the Internet – a leap that will lead to revolutionary applications that have the potential to dramatically improve the way people live, learn, work and entertain themselves” Cisco Systems Futurist Dave Evans (2011) Abstract: The Internet in the form of expansion of things, or everything, the Internet of Things (IoT) shows how Internet could lead to a radical dimensional and combinatorial type expansion. The Internet of Things (IoT) embraces connectivity up to a scale of aggregate capacity and complexity. It could evolve into a network where not only each node would be a computational device but by itself, it would also be an intelligent computing device that could replace human supervision and control through Artificial Intelligence (AI). The Internet of Things (IoT) as being embedded in the ‘Industrial Internet’ is a new paradigm shift that comprehensively affects computers and networking technology. This is also recently referred to the buzzword ‘Industry 4.0’. This technology is going to increase the utilization followed by Bandwidth of the Internet. More and more (intelligent) devices in this network are connected to the Internet through various combinations of sensor networks and interactive machine learning.

Keywords: Artificial Intelligence (AI), Bandwidth of the Internet, CEBIT, Complexity, Hannover Industrial Fair, Industry 4.0, Industrial Internet, International Telecommunication Union (ITU), M2M Communication, Quality of Service (QoS), Radio Frequency Identification (RFID), Security of Things, Semantic Web, Supply Chain, Web 2.0, Web of Things (WoT), Wireless Sensor Network (WSN). 7.1. INTRODUCTION Over a few decades, the Internet has been in a constant state of evolution. The early days of the Internet were characterized by the World Wide Web, a network of linked Hypertext Markup Language (HTML) documents that resided on the top of the Internet architecture. This network of static HTML pages gradually evolved in to what is referred to as Web 2.0, in which two-way communication became common, which enabled user participation, collaboration and interaction. Web 2.0 technologies include social networking services - technologies that have become Hans W. Gottinger All rights reserved-© 2017 Bentham Science Publishers

126 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

essential to modern social interaction as well as for global business. While Web 2.0 currently dominates the Internet, network engineers have been working towards another goal, commonly referred to as the Semantic Web and sometimes referred to as Web 3.0. The goal of the Semantic Web is to mark up web content in a way that makes it understandable by machines, allowing machines and search engines to behave more intelligently. Marking up web content in standardized formats would allow machines to process and share data on their own, without the need for human mediation. Alongside developments in Internet technologies, technologies in Sensor Networks and Near Field Communication (NFC) using Radio Frequency Identity (RFID) tags have also been evolved. Convergence of these two technologies, i.e. the Internet and Sensor Networks, is leading to new possibilities and visions. The possibility of a framework that would allow direct machine-to-machine communication over the Internet has led researchers to envision the benefits of bringing more machines online and allowing them to participate in the web as a vast network of autonomous, self-organizing devices. This vision has produced a paradigm being referred to as the Internet of Things (IoT). The Internet of Things (IoT) as being embedded in the ‘Industrial Internet’ (Evans and Annunziata, GE, 2012), or the Industrial Internet of Things (IIoT) is a new paradigm shift that comprehensively affects computers and networking technology. This is also recently referred to as the buzzword ‘Industry 4.0’. This technology is going to increase the utilization followed by Bandwidth of the Internet. More and more (intelligent) devices in this network are connected to the Internet through various combinations of sensor networks. For example, RFID tags, NFC and Bluetooth devices can be used to communicate in this Internet-like network. Devices at home, irrespective of their size, are given active or passive RFID tag, NFC chip, and devices like an air conditioner are attached to a transceiver to communicate with a local server. This local server updates the sensor output data into the Internet through the local server. The data from Internet is accessed from a desktop, mobile phone or any device that is connected to the Internet. One of the main advantages of this kind of networking architecture involving home appliances and devices is that we can have control over our devices throughout a global reach where communication is possible. The potential arises on nearly every scale as we could imagine with smart cars (autonomous selfdriving vehicles), smart planes, smart homes and smart cities. IoT can be implemented in business or industrial environments based on the requirement enabled technically by machine-to-machine (M2M) communication (Alpaydin, 2016).

The Internet of Things

Internet Economics: Models, Mechanisms and Management 127

There are many devices connected to the Internet. For example, to generate a print job we had to give all the instructions manually, but now we have printers that can be connected wirelessly through the Wi-Fi home network. The printing job can be done from anywhere and on any device such as a computer or a smartphone. In the same way, all the appliances and devices at home can be controlled remotely. It is achieved by arranging the devices and appliances in the smart home in a network and connecting them to a decision-making circuit. It is a centralized network because the decision-making circuit takes a decision on the status of the devices in the network. In an IoT environment, the controlling and communication of devices involve much sensor networks with input and output ports. 7.2. BACKGROUND RESEARCH ON INTERNET OF THINGS IoT means connecting things to the Internet, it can be an object or a device. The communication of devices on the Internet-like network can be one sided or two sided in terms of networking. We can say the communication can be a simplex kind of communication or duplex communication. The Internet has been using different types of technologies like http, www, and com (commercial), the communication between the devices has evolved as Internet technologies were developed. The new web development on the web are Web 2 and Web 3 technologies. In these technologies, engineers are designing the web server so that the machine can understand the instructions directly given by the user. This kind of technology will help the wireless sensor devices to communicate with the web directly, making them decision-making circuits (when given with a proper algorithm to make a decision and act accordingly). For instance, consider an active RFID chip equipped with a relevant sensor; this RFID label usually gathers digital output from the powered sensor in the chip and sends that data to the server but does not take any action. However, when we give commands in computer language, the machine can understand the result due to which it can choose an appropriate action from the given set of actions. In this case, we are talking about machine-to-machine (M2M) communication where the machine communicates within themselves and performs an appropriate action. Industrial Driving Forces in the IoT Context According to IoT development, IoT business evolves through several stages from expediting logistics to tele-operation and tele-presence as a result of the interaction forces of both market pull and technology push (Gottinger, 2006). This development roadmap indicates that the progress in relevant technology will continuously contribute to the development of IoT, while the commercialization in the market place is another key issue relating to the progress of IoT. Therefore,

128 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

we argue that IoT business is promoted by industrial driving forces of both technology push and market pull. First, technology push is viewed as a new invention being pushed through the development of related technologies. Rapid development of IoT technology brings enormous potential to firms in IoT business. For example, many important technologies such as cloud computing, RFID technology, and sensor network technology have promoted the development of IoT business to a new level (Whitmore et al., 2014). Second, market pull is defined as an innovative force being developed by firms in response to an identified and/or potential market on a given acceptable security level. Enormous market demand in IoT business provides unprecedented market opportunities to firms. For example, IoT applications in home appliances, healthcare, and automobiles have been increased by new market demands like household demand for home automation, patient demand for customized service, and customer demand for intelligent vehicles. The International Telecommunication Union (ITU) for instance now defines the Internet of Things as “a global infrastructure for the Information Society, enabling advanced services by interconnecting (physical and virtual) things based on, existing and evolving, interoperable information and communication technologies” (ITU, 2013). At the same time, a multitude of alternative definitions has been proposed. Some of these definitions exhibit the connectivity scale on the things which become connected in the IoT. Other definitions focus on Internetrelated aspects of the IoT, such as Internet protocols and network technology. And a third type centers on semantic challenges in the IoT relating to, e.g., the storage, search and organization of large volumes of information (Atzori et al., 2010). While there is no universal definition for the IoT, the core concept is that every day, objects can be equipped with identifying, sensing, networking and processing capabilities that will allow them to communicate with one another and with other devices and services over the Internet to achieve some useful objective. The core single concepts underlying the IoT are not new. For years, technologies such as RFID and sensor networks have been used in industrial and manufacturing contexts for tracking large-ticket items such as cranes and livestock. The idea of direct machine-to-machine communication is also not that new, as it is basic to the idea of the Internet in which clients, servers and routers communicate with each other. What the IoT represents is an evolution of the use of these existing technologies in terms of the number and kinds of devices as well as the interconnection of networks of these devices across the Internet. For example, most devices currently on the Internet were originally designed to be part of the Internet and have integrated processing, storage and network capabilities. These devices include servers, desktops laptops, tablets and smart phones. What the IoT

The Internet of Things

Internet Economics: Models, Mechanisms and Management 129

proposes is to attach technology to everyday devices, such as audio/video receivers, smoke detectors, home appliances, etc. and making them online, even if they were not initially designed with this capability in mind. The other major evolutionary change promised by the IoT, is the integration of networks that contain these devices, making each device directly accessible through the Internet. For example, RFID has been used for years to track products through certain parts of the supply chain. However, once the product left the shelf of a retail outlet, the manufacturer’s ability to track the object was lost. Likewise, consumers were unable to gain access to the lifecycle information of products they purchased. By giving each product a unique identifier and making its data available through the web, the IoT promises to enable product traceability throughout the entire product lifecycle. The emerging wirelessly sensory technologies have significantly extended the sensory capabilities of devices and therefore the original concept of IoT hence is extending to ambient intelligence and autonomous control. To date, a number of technologies are involved in IoT, such as wireless sensor networks (WSNs), barcodes, intelligent sensing, RFID, NFC, low energy wireless communications, cloud computing network depending on the standardization. Depending on various technologies for the implementation, the definition of the IoT varies. However, the fundamental of IoT implies that objects in an IoT can be identified uniquely in the virtual representations. Within an IoT, all things are able to exchange data and if needed, process data according to predefined schemes. 7.3. SPECIFIC TECHNOLOGIES AND USAGE At the core of the idea of the Internet of Things is the notion that everyday ‘things’ such as vehicles, refrigerators, medical equipment, and general consumer goods will be equipped with tracking and sensing capabilities. When this vision is fully actualized, ‘things’ will also contain more sophisticated processing and networking capabilities that will enable these smart objects to understand their environments, interact with people and become smart. Like any information system, the IoT will rely on a combination of hardware, software and architectures, and through software more and more Artificial Intelligence (AI) is being built in that makes the ‘things’ to act autonomous and serve as intelligent decision-making agents. (In the recent 2017 CeBIT Fair, Hannover, IBM displayed a BMW in which an AI engineered IBM Watson IoT was embedded for autonomous driving). Some of the hardware upon which the IoT is being built already exists and is currently in industrial use. Critical hardware infrastructure includes: RFID, NFC and Sensor Networks Radio-Frequency Identification (RFID) is a short range communication technology where an RFID tag

130 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

communicates with an RFID reader via radio-frequency electromagnetic fields. Tags may contain different forms of data, but the data form most commonly used for IoT applications is the Electronic Product Code (EPC). An EPC is a universally unique identifier for an object. These unique identifiers ensure that NFC objects tracked with RFID tags have individual identities in the IoT. RFID is not a new technology designed specifically for the IoT. RFID’s usefulness in terms of tracking objects has been well established. The technology has applications in the areas of logistics and supply chain management, aviation, food safety, retailing, public utilities and others. The use of RFID has been mandated by organizations such as Wal-Mart, the U.S. Department of Defense, and others. However, the tracking capabilities offered by RFID are generally understood to be a precursor to the Internet of Things (Ngai et al. 2008) and the benefits of RFID can be extended by making their data. A newer technology that builds on the RFID standard is Near Field Communication (NFC). NFC is a short-range communication standard where devices are able to engage in radio communication with one another when touched together or brought into close proximity to one another. Each NFC tag contains a Unique Identification (UID) that is associated with the tag. The NFC technology is frequently integrated into smartphones which are able to exchange data with one another when brought together. NFC devices are also able to make connections with passive, unpowered NFC tags that are attached to objects. One common use for NFC is in smart posters. Smart posters contain readable NFC tags that transmit data to the user’s smart phone which reads the data from the tag. Sensor networks are devices that monitor characteristics of the environment or other objects such as temperature, humidity, movement, and quantity. When multiple sensors are used together and interact, they are referred to as a wireless sensor network (WSN). Wireless sensor networks contain the sensors themselves and may also contain gateways that collect data from the sensors and pass it on to a server. While sensors ‘sense’ the state of an environment or object, actuators perform actions to affect the environment or object in some way. Actuators can affect the environment by emitting sound, light, radio waves or even smells. These capabilities are one way that IoT objects can communicate with people. Actuators are frequently used in combination with sensors to produce sensor-actuator networks. One example of the use of actuators in such a network would be the use of a sensor to detect the presence of carbon monoxide in a room and the use of an actuator to produce a loud noise alerting people to the detection of the harmful gas. Thus, the combination of sensors and actuators can enable objects to

The Internet of Things

Internet Economics: Models, Mechanisms and Management 131

simultaneously be aware of their environment and interact with people, both goals of the IoT. In practical industrial usage pattern IoT presently embraces RFID-enabled identification and tracking technologies. The RFID system has been widely applied in logistics, such as package tracking, supply chain management, healthcare, monitoring and maintenance applications. A RFID system could provide sufficient real-time information about things in IoT, which are very useful to manufacturers, distributors, and retailers. For example, RFID application in supply chain management can improve inventory management. Some identified advantages include reduced labor cost, simplified business processes, and improved efficiency. Five years ago, it was reported that 3 percent of EU based companies are using RFID (Kranenburg and Anzelmo, 2011) but with a strong growth potential. In RFID-based applications, 56 percent of firms go for access control, 29 percent for supply chain, 25 percent for freeway tolls, 24 percent for security control, 21 percent for product control, and 15 percent for asset management. Hardware devices involve much diversified specifications in terms of communication, computation, memory, and data storage capacity, or transmission capacities. An IoT application consists of many types of devices. All types of hardware devices should be well organized through the network and be accessible via available communication. Typically, devices can be organized by gateways for the communication purpose over the Internet. IoT can be an aggregation of heterogeneous networks, such as WSNs, wireless mesh networks, mobile networks, and Wireless Locally Area Network (WLAN). These networks help the things in fulfilling complex activities such as decisionmaking, computation, and data exchange. In addition, the reliable communication between gateway and things is essential to make a centralized decision with respect to IoT. The gateway is capable of running the complicate optimization algorithm locally by exploiting its network knowledge. The computational complexity is shifted from things to the gateway; the global optimal route and parameter values for the gateway can be obtained. This is feasible since the size of the gateway domain is in the order of a few of tens in comparison with the sizes of things. Hardware capabilities and the communication requirements vary from one device type to another. The things in IoT can have very different capabilities for computation, memory, power, or communication. For instance, a cellular phone or a tablet has much better communication and computation capabilities than a

132 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

special-purpose electronic product such as a heart rate monitor watch. Similarly, things can have very different requirements of Quality of Service (QoS), in particular, in the aspects of delay, energy consumption, and reliability. For example, minimizing the energy use for communication/computation purposes is a major constraint for the battery-powered devices without efficient energy harvesting techniques; this energy constraint is not critical for the devices with power supply connection. 7.4. BREADTH OF APPLICATION AREAS IoT is an interconnected network of smart objects and devices. Objects in the world of the Internet of Things, mostly referred to as smart objects, and are augmented to have RFIDs, NFCs, microprocessors or sensors built in or attached to them. Networked objects create a continuous data stream, which can be connected to offer services to users. The Internet of Things affects everyday lives of people all around us. Users get information from objects, and have novel ways of interacting with them. One can swipe (or wave) a smart card to process payment while taking a bus, train or subway. One crosses a bridge, and the tollbooth automatically deducts money from the owners account. One can look at an object (e.g. with Google Goggles or with augmented reality), and one can find out all that he wants to know about the object. RFIDs are used in stores for supply chain management. Parking meters update their status, so that drivers can know the vacancy in a parking lot. Smart meters in homes give updated information about electricity consumption; smart devices communicate with the appliance to find out its electricity consumption. Shoes with embedded accelerometers keep track of the pace and distance covered while running. Bio sensors and wearable medical gadgets keep a record of the health conditions and parameters in a body. The Internet of Things semantically means “a world-wide network of interconnected objects uniquely addressable, based on standard communication protocols” (Greengard, 2015). This implies a huge number of (heterogeneous) objects involved in the process. The vision is a new era of data production where “humans may become the minority as generators and receivers of traffic and will be dwarfed by those prompted by the networking of everyday objects” (ITU, 2013). The IoT is a convergence of various visions, evolving together based on the stakeholders’ focus of development. Atzori et al. (2010) identifies three classes of visions within the research community of the IoT: Things-oriented, Internetoriented and Semantic-oriented. The Things oriented vision focuses on the Things in the Internet of Things. Perspectives include Near Field Communications (NFC), Wireless Sensor and Actuator Networks (WSAN), microprocessors,

The Internet of Things

Internet Economics: Models, Mechanisms and Management 133

embedded systems, 3D and barcodes together with RFID. The Internet oriented vision is concerned with the protocols for communication networks. Some protocols are the Internet Protocol, IPv6 and the Web of Things paradigm. The third vision within the IoT is semantic oriented. The semantic oriented vision focuses on the organization, storage, retrieval, representation, interconnection, and search of the immense data coming in from interconnected objects (Toma et al., 2009). The objects are not necessarily addressed not by their unique identifiers, but by associating meaning (semantics) to their addressing and identification mechanisms. The domain of the application areas for the IoT is limited only by imagination at this point. For a thorough discussion of the common application areas see (Atzori et al. 2010; Miorandi et al. 2012). Based on a broad review of literature, the applications categories can be sub-classified into the following application domains: smart infrastructure, healthcare, supply chains/ logistics, and social applications. Smart Infrastructure Integrating smart objects into physical infrastructure can improve flexibility, reliability and efficiency in infrastructure operation. These benefits can reduce cost and manpower requirements as well as enhance safety. A fully designed and implemented smart (electric) grid (SG) would be a scaled-up IoT with a big data (BD) dimension in cloud computing (CC) subject to security challenges. A SG would not only cover cost-effective and efficient cloud but also secure services applying techniques of data analytics (Li and Liu, 2012). IoT technologies are also being used inside homes, offices and cities. Homes and buildings are being equipped with sensors and actuators that track utility consumption, monitor and control building infrastructure such as lights and conduct surveillance to meet security needs (Appendix). On a broader scale, IoT technologies can be employed to make cities more efficient. The goal of smart cities is to leverage the IoT to improve the lives of citizens by improving traffic control, monitoring the availability of parking spaces, evaluating air quality and even providing notification when trash containers are full (Schaffers et al. 2011; Vicini et al. 2012). Healthcare IoT is proposed to improve the quality of human life by automating some of the basic tasks that humans must perform. In that sense, monitoring and decision making can be moved from the human side to the machine side. One of the main

134 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

applications of IoT in healthcare is in assisted living scenarios. Sensors can be placed on health monitoring equipment used by patients. The information collected by these sensors is made available on the Internet to doctors, family members and other interested parties in order to improve treatment and responsiveness (Dohr et al. 2010). Additionally, IoT devices can be used to monitor a patient’s current medicines and evaluate the risk of new medications in terms of allergic reactions and adverse interactions (Jara et al. 2010). Supply Chains/Logistics RFID and sensor networks already have long established roles in supply chains. Sensors have long been used in assembly lines in manufacturing facilities and RFID is frequently used to track products through the part of the supply chain controlled by a specific enterprise. While the use of these technologies in supply chains is not new, the pervasiveness and ubiquity promised by the IoT will enable the use of these technologies across organizational and geographic boundaries. Specifically, the IoT can further improve logistics and supply chain efficiency by providing information that is more detailed and up-to-date (Flügel and Gehrmann 2009). 7.5. SECURITY OF THINGS For IoT, security and privacy are two important challenges. To integrate the devices of sensing layers as intrinsic parts of the IoT, effective security technology is essential to ensure security and privacy protection in various activities such as personal activities, business processes, transportations, and information protection (Tan et al. 2010; Wang et al. 2012; Xing et al. 2013). The applications of IoT might be affected by pervasive threats such as RFID tags attacks and data leakage. A number of security issues could be approached beyond a basic cryptographic structure both through software tools, decentralized random control and through artificial intelligence techniques providing any component of the IoT with an inherent security control. What is essentially needed though not yet available is some universal safe-proof security design that comprises all these elements. “The Internet wasn’t designed with security in mind and, in today’s world, security experts are playing a game of cat and and mouse with cybercrooks and hackers. As every new threat and breach occurs, security teams scramble to plug the dike. This has led to a mélange of tools, approaches and techniques- none of which solves the problem alone” (Greengard, 2015, Chap.6). In RFID systems, a number of security schemes and authentication protocols have been proposed to cope with security threats. Recently, in their IoT Security

The Internet of Things

Internet Economics: Models, Mechanisms and Management 135

Architecture under ‘Threat Intelligence’ IBM invokes their AI based Watson program for integrated security assessment. IoT devices are typically wireless and may be located in public places. Wireless communication in today’s Internet is typically made more secure through encryption. Encryption is also seen as key to ensuring information security in the IoT. However, many IoT devices are not currently powerful enough to support robust encryption. To enable encryption on the IoT, algorithms need to be made more efficient and less energy- consuming, and efficient key distribution schemes are needed (Yan and Wen, 2012). In addition to encryption, identity management is an important component of any security model and unique identifiers are essential to IoT devices. These identifiers may be used to establish personal identities at financial institutions, identify illegal activity and other functions. Thus, ensuring that smart objects are who they say they are is essential to IoT success (Mahalle et al., 2010; Roman et al., 2011). Privacy As more and more objects become traceable through IoT, threats to personal privacy become more serious. In addition to securing data to make sure that it doesn’t fall into the wrong hands, issues of data ownership need to be addressed in order to ensure that users feel comfortable participating in the IoT. Thus, the ownership of data collected from smart objects must be clearly established. The data owner must be assured that the data will not be used without his/her consent, particularly when the data will be shared. Privacy policies can be one approach to ensuring the privacy of information. Smart objects and reading devices in the IoT can each be equipped with privacy policies. When the object and reader come into contact, they can each check the other’s privacy policy for compatibility before communicating (Roman et al. 2011). 7.6. ECONOMIC BENEFITS In a survey on IoT and its ‘economic energy’ Fleisch (2010, 14) addresses one of the core economic impacts of IoT: “The IOT, with its technologies to automate the bridging of the last mile between the Internet and the physical world, dissolves the transaction costs that are caused by real world-virtual world media breaks. A real world-virtual world media break occurs when a piece of information is transferred from one carrier medium, e.g., a bar code, to another, e.g., a data base that serves a warehouse management

136 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

system. When things become computers, these media breaks, along with their attached costs fade away.” In specific IoT applications, the costs of interactive objects in production or service industries would not carry only physical (material) or labor costs but also coordination or regulatory costs, the costs of transaction. Depending on the industry, its scale, scope and value generation, they could be substantial and even dominate the others. Thus cost avoidance or substantial reduction would have multiple benefits for any network industry concerned, nationally and across borders. For the companies in that industry would benefit in industrial competitiveness in an international context. On the other hand, they could even reverse or at least diminish past and current flows of trade in manufactured goods from emerging to advanced economies. This equally well applies to supply chains in industrial economies. The dynamics of relationship among supply chain partners is viewed in terms of their respective innovation competence. It is emphasized that varying power arrangements (through information asymmetry) in a supply chain leads to different implications for investments in innovation by buyer and supplier (Gottinger, 2015). Williamson's (1995, 1996) transaction cost approach provides a conceptual grounding for understanding the fundamental basis on which relationship between buyer and supplier takes place. With multiple firms constituting a supply chain, investments by supply chain partners have implications that transcend the traditional cost minimization or revenue/profit maximization objectives. In present dynamic environments, firms are investing in risky innovations and pursue associated strategies to gain first-mover advantage. In high technology industries more firms strategically decide to enter a collaborative relationship. In a joint product development context, many firms outsource the manufacturing process of components which would be used in the final product. At times, this outsourcing goes beyond just the manufacturing of a fully specified component to allowing and expecting the supplier to build resource competence through active innovation. With a decrease of transaction costs through IoT we could also envision a decrease in quality of control costs through IoT induced network industries increasing the value chain of its producers lifting their productivity and their profitability and boosting industrial growth. The GE Report on the Industrial Internet by Evans and Annunziata (2012) foresees those benefits in particular, in the aviation, transportation, electric power and distribution (‘smart grids’), the retail and healthcare industries facilitated

The Internet of Things

Internet Economics: Models, Mechanisms and Management 137

through complementary enabling elements in intelligent machines, advanced analytics and networked people at work. Those industries in particular are bound to favor ‘increasing returns mechanisms’ that generally speaking complementarize technologies, business operations and strategies and form the core engine of network-centered industrial growth. (Gottinger, 2016, Chaps. 4, 7). On a more negative note, the IoT, in some industries, may contribute to replace jobs and even may not create new ones beyond the commonly observed process of automation. In the mid to longer term the net effects of jobs replacement may well be negative according to MIT economists Brynjolfsson and McAfee (2011). As with all other economic activities network externalities could well have positive economic impacts but could well be negative as in the case of security breaches, cyber piracy and loss of privacy potentially inducing irreversible economic damages that would be even out of proportion with a universal scalable and scaling effect of IoT. As in the economics treatment of accident law this requires assessment of a delicate balance of tradeoffs and compensation. Summarizing we can say that IoT promises multiple economic benefits such as: 1. Lowering transaction costs for various industries and the entire economy, 2. Improving quality of products, services for consumers, and quality of life in environmental protection and energy provision, 3. Enabling substantial industrial performance in a broad set of value generating industries and therefore inducing industrial and economic growth and well being though without creating new jobs in some industries. 7.7. FUTURE DIRECTIONS Since the IoT has not yet been fully realized – conceptually and in practical implementations, it might seem precocious to forecast the future directions, the extent of industrial broadening and deepening of IoT. Yet, the manufacturing sector together with 3D Printing and supply side chaining will be universally affected. As regards the Industrial Internet it has been a major theme on the World Economic Forum’s (WEF) 2016 Annual Meeting at Davos, also an all comprehensive showcase at the CEBIT and the Hannover Industrial Fair (2016) with companies such as Siemens, ABB, Intel, IBM, Huawei, ZTE, Bosch, Kuka in partnership with software firms showing their projects, and exposing over 100 cases of real-world implementations of Industry 4.0. Future visions of the IoT will affect its current development and must therefore be considered.

138 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

One future vision for the IoT is the Web of Things. The Web of Things proposes the use of web standards to fully integrate smart objects into the World Wide Web (WWW). Using web technologies can make it easier for developers to build applications using smart objects and existing web protocols can more easily enable the interoperability and communication of different devices. A mashup is a Web 2.0 concept where an application uses data and functionality from a variety of web resources. Some researchers proposing the Web of Things model suggest building on the mashup paradigm, except this time applying it to physical devices instead of applications (Guinard et al., 2011). With the generality of the Web, the WoT and IoT are on the edge of experiencing vast progress. Today, we are one step closer to this vision due to latest advances in web services, identification technologies, convergence services, wireless networks, which make communication capabilities and processing power available in increasingly smaller packages. Obviously, the Internet is evolving into the so-called ‘Web of Things’ (WoT), an environment where everyday devices such as traffic lights, sidewalks, buildings, and commodities are recognizable, identifiable, addressable, and even controllable via the Internet. Certainly, starting from an Internet of nearly one thousand million computers, the Web now turns to be an Internet of nearly 100 billion of things or devices presaging transition from an IoT to a WoT. Thus, in a WoT platform the workload is put at the extreme level and scalability is a compulsory requirement. This does not refer to any technology or any network structure, but only to the idea of interconnecting devices as well as interconnecting computers with the Internet. Use of the Web as the platform hosting and exposing connected devices can be explained by multiple technological and economic benefits, a few of which include deployment, high availability and versatility, use of standardized communication protocols and the ecosystem created thanks to Web 2.0 paradigm. A thing or object becomes Internet- enabled if it is associated with networking ability, which peculiarly identifies it on the Internet. Recently, devices or objects such as electric meters, access cards, street lights, and sensors are already accessed and networked on to the Internet. A thing or object becomes Webenabled when it is augmented with a Web server so that it can exploit its functional and non-functional abilities on the Web through http. Researchers have long ago successfully embedded tiny Web servers on resource-constrained things, making Web-enabled things or devices a reality. Certainly, there is a domain for Representational State Transfer (REST) in the area of Web services. The advances in REST based web service structure are the abstraction of physical things as services on the Web. This trend gives rise to the possibilities of wrapping things in the physical world as Web services.

The Internet of Things

Internet Economics: Models, Mechanisms and Management 139

SUMMARY AND CONCLUSION This chapter reported on the current state of IoT research by challenges that threaten IoT diffusion, presenting open research questions and future directions, and compiling a comprehensive reference list to assist researchers. We proposed a classification scheme with six major categories: technology, applications, challenges, business models, future directions and overview/survey. The IoT holds the promise of improving people’s lives through both automation and augmentation. The capabilities offered by the IoT can save people and organizations time and money as well as help improve decision making and outcomes in a wide range of application areas. The IoT builds on existing technologies such as RFID and Wireless Sensor Networks along with standards and protocols to support machine-to-machine communication such as those envisioned for the semantic web. One question that remains is whether or not the IoT is to be an enduring technology, whether it will fail to materialize, or whether it is a stepping stone to another paradigm. Only time will ultimately answer that question. However, by bringing existing technologies together in a novel way, the IoT has the potential to reshape our world. APPENDIX: SMART HOME SKELETON DESIGN – AN ILLUSTRATIVE EXAMPLE A home becomes smart if the access and control of the devices and utilities in the home start taking a decision (we are not talking here about Artificial Intelligence). These devices start making the decision if a proper set of instructions is given to them, and only if Machine to machine communication is happening. Machine to machine learning is possible if sensor networks control these devices. Technologies, protocols, networks like RFID, NFC, Zigbee, Z-wave, Wi-Fi, Wimax, Bluetooth and various sensors (temperature, PIR, smoke, and humidity) can make a decision-making circuit. The features below illustratively explain the communication process. Smart Home Network We can see small circuits (sensors and smart circuits) fixed on all the devices and appliances, these sensors and decision-making circuits communicate with access gateway. The access gateway is connected to a local router that provides Internet connection through which the data is transferred to the Internet. Smart homes have the following features:

140 Internet Economics: Models, Mechanisms and Management

1. 2. 3. 4. 5.

Hans W. Gottinger

Decision making circuits. Power efficient devices. Remote access. Security. User friendly.

1. Decision Making Circuits

Decision-making circuits are nothing but the sensors with a combination of programmable memory device and transmitter (wired or wireless) which make decisions depending on the input from the sensors. The sensor data is converted into machine understandable format and given to the memory unit that compares the predefined set of instructions and performs an appropriate action. The performed action and the status of the sensor are transmitted to the gateway so that the user can know the situation of that particular device. 2. Power Efficient Appliances

High-efficiency appliances are used to get the same output as the regular appliance but consume minimal power. Appliances like high-intensity LEDs and power saving appliances are installed to make smart utilization of energy without compromising on comfort levels. 3. Remote Access

The main purpose of smart homes is to achieve control over appliances remotely. Most of the times we see, people forget to switch off the appliances while rushing to work due to which they have to pay higher electricity bills and frequent breakdown of appliance’s due to excessive usage. The appliances, when linked to a universal controlare accessible from anywhere remotely with our smartphones and computers, the user can switch them as he/she wants and monitor their home. 4. Security

We build our home the way we want, and we also want it to stay the same way, so safety is also the primary concern in a smart home. For safety from fire water and high temperature, appropriate sensors should be installed to alert us in cases of fire breakout or water leakage. Our smart home can also be vulnerable to robbery. To prevent our house from getting robbed, we can install cameras that record the footage in the cloud. Also to get informed on any unauthorized entries and suspicious behavior, infrared alarm system (it could be a PIR sensor or burglar alarm system) can be arranged in the fencing and entrances.

The Internet of Things

Internet Economics: Models, Mechanisms and Management 141

5. User Friendly

All the members living in the house may not be aware of the latest technology (elderly people), so the operation of these appliances should be easily understandable for people of all age groups. A simple system is designed so that user can troubleshoot on its self. The interface of control should also be as simple and friendly as possible. For illustrative purposes, we have designed a model to demonstrate Internet of Things for remote monitoring and controlling of appliances remotely through the Internet using a web interface. The objective of this model is to provide nonphysical control over appliances and to give an update on the temperature and humidity levels at home through the web interface from anywhere in the world. This kind of a model helps the physically challenged and the aged persons to control appliances as it might be difficult for them to move to the switch every time they want to control it. Moreover, the proposed model is cost effective and power efficient. The concept of Internet of Things can be further improved by creating a proper interface and designing a specific user-friendly gateway to communicate with the Internet. Available goods and quantity can be known in a single scan when all the objects and goods are given an RFID tag. In a similar way misplaced lost and found objects can be tracked using this concept by tracking them using appropriate sensors fixed to them. A mobile application can be created to control these appliances from our smart phones. By arranging cameras in business places, every camera can be operated and monitored from anywhere by logging into the account. Similarly, this technology can be implemented in various applications for easy accessibility. REFERENCES Alpaydin, E. (2016). Machine Learning, The New AI. Cambridge, Ma.: MIT Press. Atzori, L., Iera, A., Morabito, G. (2010). The Internet of Things: A survey. Comput. Netw., 54(15), 27872805. [http://dx.doi.org/10.1016/j.comnet.2010.05.010] Brynjolfsson and McAfee A. (2011). The Race against the Machine. Lexington, MA: Digital Frontier Press. Dohr, A., Modre-Opsrian, R., Drobics, M., Hayn, D., Schreier, G. (2010). The Internet of Things for ambient assisted living Proceedings of the Seventh International Conference on Information Technology: New Generations (ITNG). [http://dx.doi.org/10.1109/ITNG.2010.104] Evans, P.C., Annunziata, M. (2012). Industrial Internet, General Electric (GE) Report, New York. Fleisch, E. (2013). What is the Internet of things?. Available at: http://www.im.Ethz.ch/education/HS10/ AUTOIDLABS-WP-BIZAPP-53.pdf Flügel, C., Gehrmann, V. (2009). Scientific Workshop 4: Intelligent objects for the Internet of Things:

142 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

Internet of Things – application of sensor networks in logistics In: Gerhäuser, H., Hupp, J., Efstratiou, C., Heppner, J., (Eds.), Constructing ambient intelligence, communications in computer and information science [http://dx.doi.org/10.1007/978-3-642-10607-1_4] Giusto, G.M., Iera, A. (2010). The Internet of Things. Berlin: Springer. [http://dx.doi.org/10.1007/978-1-4419-1674-7] Gottinger, H.W. (2006). Innovation, Technology and Hypercompetition. London: Routledge. Gottinger, H.W. (2015). Supply-Chain Coopetition. International Journal of Business and Economics Research, 4(2), 67-71. [http://dx.doi.org/10.11648/j.ijber.20150402.16] Gottinger, H.W. (2016). Networks, Competition, Innovation and Industrial Growth. New York: NovaScience. Greengard, C. (2015). The Internet of Things.. Cambridge, Ma: MIT Press. Guinard, D., Trifa, V., Mattern, F., Wilde, E. (2011). From the Internet of Things to the Web of Things: Resource-oriented architecture and best Practices In: Uckelmann, D., Harrison, M., Michahelles, F., (Eds.), Architecting the Internet of Things (pp. 97-129). Berlin: Springer. [http://dx.doi.org/10.1007/978-3-642-19157-2_5] Haller, S., Karnouskos, S., Schroth, C. (2009). The Internet of Things in an enterprise context In: Domingue, J., Dieter, F., Paolo, T., (Eds.), Future internet – FIS 2008, Lecture Notes in Computer Science (LNCS) (Vol. 5468, pp. 14-28). Berlin: Springer. [http://dx.doi.org/10.1007/978-3-642-00985-3_2] ITU. (2013). The Internet of Things, International Telecommunication Union (ITU), Internet Report. Available at: http://www.itu.int/dms_pub/itu-s/opb/pol/S-POL-IR.IT-2005-SUM-PDF-E.pdf Kranenburg, R., Anzelmo, E. (2011). The Internet of Things 1st Berlin Symposium on Internet and Society, 243-259. Li, L., Liu, J. (2012). An efficient and flexible web services-based multidisciplinary design optimisation framework for complex engi-neering systems. Enterprise Inf. Syst., 6 (3), 345-371. [http://dx.doi.org/10.1080/17517575.2011.651627] Mahalle, P., Babar, S., Prasad, N.R., Prasad, R. (2010). Meghanathan, N. (2010). Identity management framework towards Internet of Things (IoT): Roadmap and key challenges Recent trends in network security and applications, communications in computer and information science (pp. 430-439). Berlin: Springer. Miorandi, D., Sicari, S., De Pellegrini, F., Chlamtac, I. (2012). Internet of things: vision, applications and research challenges. Ad Hoc Netw., 10(7), 1497-1516. [http://dx.doi.org/10.1016/j.adhoc.2012.02.016] Ngai, E. (2008). RFID research: an academic literature review (1995–2005) and future research directions. Int. J. Prod. Econ., 112(2), 510-520. [http://dx.doi.org/10.1016/j.ijpe.2007.05.004] Roman, R., Najera, P., Lopez, J. (2011). Securing the Internet of Things. IEEE Computers, 44 (9), 51-58. [http://dx.doi.org/10.1109/MC.2011.291] Schaffers, H., Komninos, N., Pallot, M., Trousse, B., Nilsson, M., Oliveira, A. (2011). Smart cities and the future internet: Towards cooperation frameworks for open innovation In: Domingue, J., (Ed.), The future Internet, Lecture Notes in Computer Science (Vol. 6656, pp. 431-446). Berlin: Springer. Tan, L., Wang, N. (2010). Future Internet: The Internet of Things Proceedings of the 3rd International Conference on Advanced Computer Theory and Engineering (ICACTE). Toma, I., Simperl, E., Hench, G. (2009). A joint roadmap for Semantic technologies and the Internet of Things. Technology, 1-5. Vicini, S., Sanna, A., Bellini, S. (2012). Uckelmann, D., Scholz-Reiter, B., Rügge, I., Hong, B., Rizzi, A. (2012). A living lab for Internet of Things vending machines The Impact of Virtual, Remote, and real

The Internet of Things

Internet Economics: Models, Mechanisms and Management 143

Logistics Labs. Commun. Comput. Inf. Sci. (Vol. 282, pp. 35-43). Berlin: Springer. [http://dx.doi.org/10.1007/978-3-642-28816-6_4] Wang, S., Li, L., Wang, K., Jones, J.D. (2012). e-Business systems integration: a systems perspective. Inf. Technol. Manage., 13(4), 233-249. [http://dx.doi.org/10.1007/s10799-012-0119-8] Whitmore, A., Agarwal, A., Da Xu, L. (2015). The Internet of Things—A survey of topics and trends. Inf. Syst. Front., 17, 261-274. [http://dx.doi.org/10.1007/s10796-014-9489-2] Williamson, O.E. (1995). Organization Theory. New York: Oxford Univ. Press. Williamson, O.E. (1996). The Mechanisms of Governance. New York: Oxford Univ. Press. Xing, Y., Li, L., Bi, Z., Wilamowska-Korsak, M., Zhang, L. (2013). Operations research (OR) in service industries: a comprehensive review. Syst. Res. Behav. Sci., 30(3), 300-353. [http://dx.doi.org/10.1002/sres.2185] Yan, T., Wen, Q. (2011). Building the Internet of Things using a mobile RFID security protocol based on information technology. Adv. Comput. Sci. Intell. Syst. Environ, 104, 143-149. [http://dx.doi.org/10.1007/978-3-642-23777-5_24]

144

Internet Economics: Models, Mechanisms and Management, 2017, 144-166

CHAPTER 8

The Internet, Data Analytics and Big Data “Be an expensive complement (stats knowhow) to something that’s getting cheaper” Hal Varian, Chief Economist, Google, 2014 Abstract: For observing a huge online data generation through an expanded Internet (IoT and the Industrial Internet) one can use the flux of information for monitoring, predicting, controlling and decision making. During the time of applying methods of statistical inference and statistical decisions some 70 years ago, information derived from data collection was considered costly. Models were built where information was linked to payoff relevance of a decision making criterion (utility or payoff function), therefore statistical information was handled to satisfy these criteria. What really is subsumed under ‘Big Data’ (BD) qualifies under a few main characteristics: (i) BD is primarily network generated on a large scale by volume, variety and velocity and comprises large amounts of information at the enterprise or public level, in the categories of terabytes (1012 bytes), petabytes (1015) and beyond of online data. (ii) BD consists of a variety and diversity of data types and formats, many of them are dynamic, unstructured or semi-structured and are hard to handle by conventional statistical methods. (iii) BD is generated by disparate sources as in interactive application through IoT from wireless devices, sensors, streaming communication generated by machine-to-machine interactions. The traditional way of formatting information from transactional systems to make them available for ‘statistical processing’ does not work in a situation where data arrived in huge volumes from diverse sources, and where even the formats could be changed.

Keywords: Big Data, Business Intelligence, Cloud Computing, Complexity, Cybersecurity, Cyber Physical System (CPS), Data Analytics, Data Mining, Exploratory Data Analysis (EDA), GPS Systems, Hadoop, IBM, IDC, Industry 4.0, M2M Communication, MapReduce, McKinsey, Predictive Analytics, Statistical Decisions, Statistical Inference. 8.1. INTRODUCTION As outlined in the previous chapters, the Internet, the IoT and the Industrial Internet, in particular, are about data, huge online data generation and using the flux of information for monitoring, predicting, controlling and decision making. During the time of forming methods of statistical inference and statistical * Contributory author Dr. S. Sedkaoui, Montpellier, France Hans W. Gottinger All rights reserved-© 2017 Bentham Science Publishers

The Internet, Data Analytics

Internet Economics: Models, Mechanisms and Management 145

decisions over 70 years ago, information derived from data collection was considered costly. Models were built where information was linked to payoff relevance of a decision making criterion (utility or payoff function), therefore statistical information was handled to satisfy these criteria. Now as masses of online data through the IoT are produced at relatively low costs, all these data could be quickly aggregated for business or government decisions. Statisticians have coined a term, ‘value of perfect information’, which is set up to integrate data points, collection and analysis through statistical inferential models, exploratory data analysis (EDA) or through statistical decision models (Piegorsch, 2015). In the conventional statistical days, achieving this goal is quite challenging to gather all the data for perfect information. What really is subsumed under ‘Big Data’ (BD) qualifies under a few main characteristics: i. BD is primarily network generated on a large scale volume, by variety and velocity and comprises large amounts of information at the enterprise and public level, in the categories of terabytes (1013 bytes), petabytes (1015) and beyond online data. ii. BD consists of a variety and diversity of data types and formats, many of them are dynamic, unstructured or semi-structured and are hard to handle by conventional statistical methods. iii. BD is generated by disparate sources as in interactive application through IoT from wireless devices, sensors, streaming communication generated by machine-to-machine interactions. The traditional way of formatting information from transactional systems to make them available for ‘statistical processing’ does not work in a situation where data arrive in huge volumes from diverse sources, and where even the formats could be changed. For many types of such situations, in view of insights from computational imperatives, Google developers codified a style of programming they called ‘MapReduce’ that used to be quite effective, processing large amounts of data and yet being able to express a wide range of algorithms. This straightforward architecture was expanded to Hadoop that in 2008-through the Apache Software Foundation (ASF), was promoted as an open source software project. Here are a few application areas of using Hadoop for diverse data-linked purposes (O’Reilly, 2012): Financial institutions are using Hadoop as a critical part of their security architecture to predict phishing behavior and payments fraud in real time and minimize their impact. They hold on to data for longer periods and run more detailed analytics and forensics.

146 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

An online advertising company provides real-time trading technology to its users and relies on Hadoop to store and analyze petabytes worth of data. 90 billion real time ad auctions are processed each day on their Hadoop distribution. A digital marketing intelligence provider uses Hadoop to process over 1.7 trillion internet and mobile records per month providing syndicated and custom digital marketing intelligence. While one major aspect of big data is the computational handling of network induced data; another is the proper application of data analytic and statistical tools for large scale use in business and commercial contexts. Cukier and Mayer-Schoenberger(2013a; 2013b,29) see a paradigmatic change in the statistical handling of large data: “Using great volumes of information … requires three profound changes in how we approach data. The first is to collect and use a lot of data rather than settle for small amounts or samples as statisticians have done for well over a century. The second is to shed our preference for highly curated and pristine data and accept messiness: in an increasing number of situations, a bit of inaccuracy can be tolerated, because the benefits of using vastly more data of variable quality outweigh the costs of using smaller amounts of very exact data. Third, in many instances, we will need to give up our quest to discover the cause of things, in return for accepting correlations. With big data, instead of trying to understand precisely why an engine breaks down or why a drug’s side effect disappears, researchers can instead collect and analyze massive quantities of information about such events and everything that is associated with them, looking for patterns that might help predict future occurrences. Bid data helps answer what, not why, and often that’s good enough.” Though we consider quite a few challenges in handling large data of diverse nature and sources we do not see the dichotomy proposed, rather we see the issues in terms of complementarity not substitution. A case in point is the following example which stands for many diverse computational-statistical paradigms. With massive production of online data, Google, in 2009,was able to predict the timely spread of influenza through simple statistical correlation (Foster et al., 2017, Chap. 1). However, four years later, using the same type of procedures, Google’s model for another flu outbreak was off the mark, overstating the spread of flu-like illnesses by a factor of two. After all, it shows that a theory-free, data-rich model even with massive provision of online data is not good enough to make accurate or even good predictions. It would require tailored statistical methods and data quality control to superimpose on large data streams to make sense of the data and use them for statistical inference and decisions. More frequently than not also

The Internet, Data Analytics

Internet Economics: Models, Mechanisms and Management 147

good theoretic insights and models of the subject discipline would be helpful to identify the ‘payoff relevance’ of data for predictive purposes (Harford, 2014). This explains the formation of handling data through scalable tools as developed in econometrics, psychometrics, technometrics etc. that have been established before and during the generation of big online data. On the interface of Internet generated Big Data and proper statistical treatment H. Varian(2014) from Google suggests a targeted range of statistical ‘tricks’ for data analysis involving classification and regression analysis, variable selection, econometrics and machine learning. The fast computational generation of online data could also activate built-in intelligent mechanisms of classifying, categorizing, visualizing, aggregating diverse data types and providing data analytic tools for statistical metrics. Artificial intelligence methods could help in this regard (Gale, 1986; Gottinger, 1993; Gottinger and Weimann, 1990). A way to fine tune large numbers of raw data with respect to making them treatable for optimally targeted statistically processing on predictive and decision making purposes would be to link them to statistical expert systems. These expert systems may work endogeneously on newly generated online data and formats to answer more sophisticated and more specific statistical queries – beyond large scale correlative number crunching. This explosion of data volumes will gradually increase as the Internet of Things (IoT) develops. Some data analysts suggest that we are entering the ‘Industrial Revolution of Data’ where the amount of data will be generated not only by people and companies but also by machines and interactive devices. These machines generate data a lot faster than people can, and their production rates will grow exponentially. According to the IDC group, the amount of world data will expand to 40 zetabytes (1021) by 2020. Emerging markets and machine-generated data will account for increasing proportions of it. This group notes that machinegenerated data is a key factor behind this expansion, increasing from 11 percent of the digital universe in 2005 to more than 40 percent in 2020. In this context, the Internet of Things (IoT) paradigm relies on a world of interconnected objects (Atzori et al., 2010), able to communicate between each other and collect data about their context. The Gartner group predicts up to 26 billions of things connected to the Internet by 2020 (Middleton et al., 2013). This is a typical example of Big Data collection and analysis as it addresses the four Vs: large Volume of Various data, collected with a high Velocity to define application with added value. The trend is part of an environment quite popular lately: the proliferation of web pages, image and video applications, social

148 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

networks, mobile devices, apps, sensors, and so on, able to generate, according to IBM, more than 2.5 quintillion bytes per day, to the extent that 90% of the world’s data have been created over the few past years (see Cukier,K. and V. Mayer-Schoenberger (2013a,b); Dietrich et al., (2014); Foster et al., (2017)). The biggest challenge of the zetabytes age will not be storing all that data, it will be figuring out how to make sense of it. Big data deals with unconventional, unstructured databases, which can reach petabytes, exabytes or zetabytes, and require specific treatments for their needs, either in terms of storage or processing/display. A study carried out by McKinsey Global Institute (2013) showed that companies that adopted advanced data analysis tools attain more productivity and better profit margins than their competitors. IDC (2011) describes big data technologies as a new generation of technologies and architectures, designed to economically extract value from very large volumes of a wide variety of data, by enabling high-velocity capture, discovery, and/or analysis. The big data technology stack includes: (i) Infrastructure, such as storage systems, servers, and datacenter networking infrastructure (ii) Data organization and management software, (iii) Analytics and discovery software, (iv) Decision support and automation software, (v) Services including business consulting, business process outsourcing, IT outsourcing, IT project-based services, and IT support and training related to big data implementations. As the amount of data being collected continues to grow, more and more companies have started building BD repositories to store, aggregate and extract meaning from their data. Volumes of data are expanding rapidly, and effectively harnessing the data volumes generated by organizations today brings both significant gains as well as challenges. Thanks to cloud computing and the socialization of the Internet, petabytes of data are created daily online and much of this information has an intrinsic business value if it can be captured and analyzed (Yeluri and Castro-Leon, 2014). Big data brings along with it some huge analytical challenges and a plethora of problem solving tools that for adequate coverage would require a diversified book rather than a single chapter. The type of analysis to be done on this huge amount of data requires a large number of advanced skills. Moreover the type of analysis which is needed to be done on the data depends highly on the results to be obtained through ‘decision making’. This can be done to (i) incorporate massive data volumes in analysis or (ii) determine upfront which big data is relevant. So, we have two technical entities to manage together. First, there is big data for massive amounts of data. Second, there is advanced analytics, which is actually a

The Internet, Data Analytics

Internet Economics: Models, Mechanisms and Management 149

collection of different tool types, including those based on predictive analytics, data mining, statistics, clustering, data visualization, text analytics, artificial intelligence, and so on (Shroff, 2013; Siegel, 2016). The potential value in the massive data flows, typical of big data can only be extracted under the condition that appropriate measures are taken for adapting society, business, and technical solutions to meet prospective new needs. In this chapter, we give a broad overview of the challenges that need to be addressed in order to build capabilities in big data analytics. We sketch some initiatives that aim to draw computer science and statistics closer together, with particular reference to big data problems. It shows that the big data revolution and Internet of Things have been a driving force for the deployment of commensurate business plans and operations. The objective of this chapter is to discuss what specifically the role of ‘data’ is for companies and to show that the challenges around the era of ‘data revolution’ focus on data uses. It is linked to the rise of the intangible economy that mobilizes knowledge and highlights the importance of data. In order to discuss this more deeply the following sections will provide insights on the state-of-the-art of big data analytics. 8.2. BIG DATA DIMENSIONS Since the advent of IT and the Internet, the amount of data stored in digital form has been rapidly growing. An increasing number of data silos are created across the world, which means this growth will further proliferate. Today ‘data’ comes from everywhere: geolocation sensors, smartphones, social networking, web services and so on. Thus currently, not only is the quantity of digitally stored data is much larger, but the type of data is also tremendously diversified, due to various new technologies (Sedkaoui and Monino, 2016). Individuals are putting more and more publicly available data on the web. Many companies collect information from their clients and their respective behavior. As such, many industrial and commercial processes are being controlled by computers. The results of medical tests are also being retained for analysis. Financial institutions, companies, and health service providers, administrations generate large quantities of data through their interactions with suppliers, patients, customers, and employees. Beyond those interactions, large volumes of data are created through Internet searches, social networks, GPS systems, and stock market transactions. Many of these companies’ datasets are within the petabytes range but, soon they could reach exabytes or even zetabytes. In 1999, Wal-Mart (one of America’s most important retail chains) had a database of 1,000 terabytes (that is, 1,000,000 gigabytes of data). In 2004, Wal-Mart claimed to have the largest data warehouse

150 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

with 500 terabytes storages. By 2012, this quantity had grown to over 2.5 petabytes (2.5 million gigabytes) of data. In 2008, eBay storage amounted to 8 petabytes. Three years later, the Yahoo warehouse totaled 170 petabytes. The Internet Age has triggered a boom in information research. Companies are flooded by the wealth of data that results from simple Internet browsing. In other words, they are forced to purchase pertinent information to develop high added value strategies that allows them to succeed in the face of incessant changes in their business environment. Industrial strategies now rely strongly on the capacity of companies to access strategic information to better navigate their environment. This information can, thus, become the source of new knowledge (knowledge pyramid) (see Ackoff, 1989). As long-term economic growth depends on the accumulation of knowledge and the ability to introduce new products, processes, services, and business and organizational models, business competitiveness is determined by the company’s ability to organize a space that is beneficial to science, technology and innovation. The problem is that most current technologies and software try to overcome the challenges that the ‘Vs’ (volume,velocity,variety) raise. One of these is Apache Hadoop, which is open source software that its main goal is to handle large amounts of data in a reasonable time. What Hadoop does is dividing data across a multiple systems infrastructure in order to be processed. Also, Hadoop creates a map of the content that is scattered so it can be easily found and accessed. The exploration of large amounts of data enables the launch of new products and services, new processes, and even new business models. By making data speak each company will have access to a better understanding of the context and its environment. BD is therefore part of ‘open innovation’ approach that encourages organizations to collaborate with other actors outside of the traditional company boundaries, on either strategic or unexpected topics, in what is deemed ‘open innovation outsidein’. It also requires reflecting on the value creation from projects and produced data, in which the company has no plans to capitalize directly into its core activity of ‘open innovation inside-out’. Beyond the huge volume, it is the diversity of data sources that gives BD its full scope. But the development of new products and services and their adaptation to new uses are facilitated by the mixing of large data sets. BD marks a major turning point in the use of data and is a powerful vehicle for growth and profitability. A comprehensive understanding of a company’s data, its potential and the processing methods can be a new vector for performance.

The Internet, Data Analytics

Internet Economics: Models, Mechanisms and Management 151

8.3. THE 3 VS: ISSUES AND CHALLENGES Large data contains the value, but there are still many difficulties and challenges in the use of BD technologies. The larger the scale of data, the more difficult the processing and storage. The analysis of BD involves multiple distinct phases which include data acquisition and recording, information extraction and cleaning, data integration, aggregation and representation, query processing, data modeling and analysis and interpretation. Each of these phases introduces challenges. Heterogeneity, scale, timeliness, complexity, quality, security and privacy are certain challenges of BD analytics. The problems start right away during data acquisition, when the data deluge requires us to make decisions, currently in an ad hoc manner, about what data to keep and what to discard, and how to store what we keep reliably with the right data. Heterogeneity Implement application of classification or regression in a BD framework poses difficult specific questions. The first hurdle is the heterogeneity. BD is often obtained by aggregating different sources of very different nature of data. We may have to deal simultaneously with numerical, categorical data, but also with text, preference data, browsing histories, historical purchase on e-commerce websites, social media data, analyzed by using methods of natural language processing, being fused with sales data to determine the effect of advertising on consumer sentiment about a product and behaviors of purchase. A majority of business problems faced by the enterprise are not BD problems. They are distributed data problems with information, data, value and analysis hugely distributed across heterogeneous locations, technology and sources. In the BD era, the large sample size enables us to better understand heterogeneity, shedding light toward studies such as exploring the association between certain covariates and rare outcomes and understanding why certain treatments benefit a subpopulation and harm another subpopulation. Data can be both structured and unstructured. They are highly dynamic and does not have particular format. It may exist in the form of email attachments, images, pdf documents, medical records, graphics, video, audio etc. and they cannot be stored in row/column format as structured data. Transforming this data to structured format for later analysis is a major challenge in BD analytics. However, machine analysis algorithms expect homogeneous data, and cannot understand

152 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

nuance. In consequence, data must be carefully structured as a first step in data analysis. Scale Managing large and rapidly increasing volumes of data has been a challenging issue for many decades. In the past, this challenge was mitigated by processors getting faster, following Moore’s law, to provide us with the resources needed to cope with increasing volumes of data. The difficulties of BD analysis derive from its large scale as well as the presence of mixed data based on different patterns or rules (heterogeneous mixture data) in the collected and stored data (heterogeneous mixture data issue). Especially, in the case of complicated heterogeneous mixture data, the data has not only several patterns and rules but characteristically, the properties of the patterns vary greatly. Timeliness As the size of the data sets to be processed increases, it will take more time to analyze. In some situations results of the analysis is required immediately. So we need to develop partial results in advance so that a small amount of incremental computation with new data can be used to arrive at a quick determination. In BD the realization time to information is critical to extract value from various data sources, including mobile devices, radio frequency identification, the web and a growing list of automated sensory technologies. Complexity Due to the development of technology, modern datasets are evolving not only in term of size or volume but also in terms of ‘complexity’. The amount of data that is traveling across the internet today, not only that is large, but is complex as well. Industrial companies are faced with a rapidly increasing mountain of data and devices that is growing in both quantity and complexity. Complexity measures the degree of interconnectedness and interdependence in BD structures such that a small change in one or a few elements can yield very large changes or a small change that ripple across or cascade through the system and substantially affect its behavior, or no change at all (Katal et al., 2013). Traditional software tools are not enough for managing the increasing volumes of data. Data analysis, organization, retrieval and modeling are also challenges due to scalability and complexity of data that needs to be analyzed. Quality BD processing requires an investment in computing architecture to store, manage,

The Internet, Data Analytics

Internet Economics: Models, Mechanisms and Management 153

analyze, and visualize an enormous amount of data. It is the indispensable raw material of one of the new century’s most important activities. But it is important to be prudent in our analysis and predictions because a lot of data is not yet ‘the right data’. There is, therefore, underlying difficulty behind BD, since more data is not necessarily better data. Data uncertainty refers to the level of reliability associated with certain types of data. Striving for high data quality is an important BD requirement and challenge, but even the best data cleansing methods cannot remove the inherent unpredictability of some data, like a customer’s actual future buying decisions. Security and Privacy In addition, security (which will be more developed in the next section) and privacy are becoming very urgent BD aspects that need to be tackled (Agrawal et al., 2011). The vast majority of data comes from the many devices and machines reporting to each other and to those running them. From the assembly line at the manufacturing plant to the passenger jet in flight, millions of bytes of data are generated and then analyzed. Some of captured data is personal information, and as such, both cutting-edge security and responsible stewardship models must be used to make sure this information is safe and correctly used. However, many technical challenges must be addressed before this potential can be realized fully. The challenges include not just the obvious issues of scale, but also heterogeneity, lack of structure, error-handling, timeliness, provenance, and visualization, at all stages of the analysis driving from data acquisition to result interpretation. Privacy is also a big concern, especially when considering that the linking of databases can disclose information that was meant to remain anonymous. The advance in BD analytics brought us tools extract and correlates this data which would make data violation much easier. That makes developing the BD applications a must without forgetting the needs of privacy principles and recommendations. Once the implementation of the first data warehouse in the 1990s the question of the quality of the data was a major issue. In the US, the theorem ‘garbage in, garbage out’ was immediately widespread. So there is nothing new about this description: only data quality will help produce an event, a forecast or strategic information and define an action lever. The reconciliation of internal and external data has always been a challenge. It is possible to obtain better results by making

154 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

better use of available data. When researchers encounter a set of data, they need to understand not only the limits of the available set of data, but also the limits of the questions that it can respond to, as well as the range of possible appropriate interpretations. The ultimate goal is not only to collect, combine, or process all data, but also to increase its value and efficiency. This means that we must evolve from big data to smart data, since the effectiveness of companies’ strategies now depends on the quality of data. Indeed, companies must not rely on the size of their data – it is not useful unless it is applied in an intelligent manner. Therefore, the volume of data is of little importance, since internal data must be combined with external data in order for a company to obtain the most out of its data. To make the most out of BD, the issue is not limited to the “simple” technical issues of collection, storage and processing speed. The use of BD requires rethinking the process of collecting, processing and the management of data. It’s the “analysis” that will be applied to data which will justify BD, not the collection of data itself. What is truly necessary are excellent analytic skills, a capacity to understand and manipulate large sets of data, and the capacity to interpret and apply the results. The need to analyze and use enormous amounts of data more efficiently drives companies towards ‘data science’ in the hope of unlocking the power of BD. 8.4. BIG DATA ANALYTICS AND SECURITY CHALLENGE Data analysis, when it is not preceded by the word ‘Big’, refers to the development and sharing of useful and effective models. For the most part, it uses a variety of methods from different research fields, like statistics, data mining, visual analysis, etc. It caters to a wide range of applications, including data summarization, classification, prediction, correlation, etc. In the 1970s and 1980s, computers could process information in batch processing, but its operations were constrained and too costly. Only large firms could hope to analyze data with them. They started to work on data organization by designing database management systems (DBMSs), in particular on relational databases. Data processing and analysis, in the present day, are brought together under the notion of Business Intelligence (BI), due especially to computers’ increased processing capabilities (Shroff, 2013). Currently, one of the innovations that make it possible to share and store large volumes of data is Cloud Computing (CC). The ‘Cloud’ allows access to shared computing resources through an on-demand telecommunication network or selfservice modules. The cloud transforms storage infrastructure and computing power into services through the intermediary of companies that possess servers and rent out their capacities. This approach makes it possible to share costs and to provide greater data storage and processing flexibility for users.

The Internet, Data Analytics

Internet Economics: Models, Mechanisms and Management 155

Companies must first understand the potential value creation of connected devices and big data in their markets. Porter and Heppelmann (2015) identifies four key capabilities of connected objects combined to big data: ●







Monitoring: Sensors placed on the objects provide information about their environment and the conditions of use and objects operation. The use of this data can be the source of new services, such as preventive medicine. These data can also be used indirectly to better consider the design of future objects, better segment the market and prices, or provide more effective customer service. Control: Smart, connected products can be controlled through software embedded within them or that resides in the cloud. Users have an unprecedented ability to tailor product function and personalize interactions. Remote control of products increases employee safety and can reduce the number of employees needed. Optimization: Algorithms and analytics can optimize product operation, capacity utilization, and predictive maintenance. Autonomy: Access to monitoring data, remote control, and optimization algorithms allow for product autonomy. This enables autonomous operation, self-coordination, and self-diagnosis.

Indeed, when computers were first introduced in organizations, it was in an attempt to tame the data. Businesses gather large amounts of data. However, they were unable to use this data to generate useful information in a timely manner which fragmenting their vision and understanding of their business. As they were unable to conduct an analysis of large volumes of data, and their real-time processing. The BD analytics extends beyond traditional data management capabilities; it can deal with the volume, velocity, and variety. This is the all important difference between both. Big Data can manage it’s ‘3Vs’, focusing on the acquisition of large-scale data and promoting a quick turnaround for a deeper analysis. Pioneers In this field have led the way: Google, Facebook, Amazon and others. Data is at the heart of their business models. The American company ‘Harrah’s’ has made progress in sales of 8 to 10 percent by analyzing customer segmentation data, while Amazon stated that 30 percent of its turnover came from its engine analytical recommendations (McKinsey, 2011, 2013). Such examples, and many others, share common principles: extreme digitalization of their process leads to extensive use of data to experiment with new business models, beyond their original boundaries. The new analytical power is seen as an opportunity to invent and explore new methods which are able to detect correlations between the quantities of available data. BD allows for recovering data hitherto unknown but useful for inspiring new ideas that can lead to new discoveries and thus fuel the cycle of research an

156 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

innovation. The applications of BD are numerous and are a main factor in strengthening the capacity for innovation within companies by playing the two dynamics of exploration and processing. Innovation will certainly stem from combinations and processes that were not originally thought of. The Danish company ‘Vestas Wind Systems’, one of the most influential wind turbine manufacturers in the world, uses ‘IBM Big Data analytics’ and ‘IBM Systems’ solutions to decide the location of wind turbines within a few hours by crossing varied data such as weather data, satellite images, etc. In the BD universe, companies seek to unlock the potential of data in order to generate value. The keystone of BD exploitation is to leverage the existing datasets to create new information, enriching the business value chain. The advent of BD is upsetting conventional management practices and invites the organization to adopt new visions allowing it to have ‘strategic information from this data’ likely to generate more value. In other words, we must get the message that data in itself is not power; it is its using that gives power, and more one gives an exchange of data and information, more one receives (Martinet and Marti, 2001). It is the use of data that empowers decision-making. Being increasingly aware of the importance of data and information, companies are pressing to rethink the way to ‘manage’, to enrich and to benefit from them. Faced with this challenge companies have sought to go further by automating their strategic decision-making, on the basis of precise indicators from ‘Big Analytics’. This causes two main challenges: ●



Big data contains invisible models, which must be viewed using tools and analytical techniques. The knowledge gained should be used at the right time in the right context and with the right approach. To capture, manage, combine, secure, and always take advantage of a huge amount of data is much more complicated than the simple data storage problem.

They are also impatient to find new ways to process that data and make more intelligent decisions, which will result in better client service, improved process efficiency, and better strategic results. There are many technical challenges that must be addressed to realize the full potential of big data. Covington (2015) provides a comprehensive discussion of such challenges based on the notion of data analysis pipeline: ●



Data Acquisition and Recording: It is critical to capture the context into which data has been generated, to be able to filter out non relevant data and to compress data, to automatically generate metadata supporting rich data description and to track and record provenance. Information Extraction and Cleaning: Data may have to be transformed in order

The Internet, Data Analytics







Internet Economics: Models, Mechanisms and Management 157

to extract information from it and express this information in a form that is suitable for analysis. Data may also be of poor quality and/or uncertain. Data cleaning and data quality verification are thus critical. Data Integration, Aggregation and Representation: Data can be very heterogeneous and may have different metadata. Data integration, even in more conventional cases, requires huge human efforts. Novel approaches that can improve the automation of data integration are critical as manual approaches will not scale to what is required for big data. Also different data aggregation and representation strategies may be needed for different data analysis tasks. Query Processing, and Analysis: Methods suitable for big data need to be able to deal with noisy, dynamic, heterogeneous, untrustworthy data and data characterized by complex relations. However despite these difficulties, big data even if noisy and uncertain can be more valuable for identifying more reliable hidden patterns and knowledge compared to tiny samples of good data. Also the (often redundant) relationships existing among data can represent an opportunity for cross-checking data and thus improve data trustworthiness. Supporting query processing and data analysis requires scalable mining algorithms and powerful computing infrastructures. Interpretation: Analysis results extracted from big data needs to be interpreted by decision makers and this may require the users to be able to analyze the assumptions at each stage of data processing and possibly re-tracing the analysis. Rich provenance is critical in this respect (Salmin et al., 2012).

The analysis of a larger amount of data in real time is likely to improve and accelerate decisions in multiple sectors, from finance to health, both including research. The considerable increase in the volume and diversity of digital data generated, coupled with Big Data technologies, offer significant opportunities for value creation. This value cannot be reduced to simply what we can solve or improve, but rather it knows what the new potential discoveries are that may arise from cross-exchanges and correlations. Big data and analytics can reap huge benefits to both individuals and organizations. 8.5. BIG DATA ANALYTICS FOR INDUSTRY 4.0 In today’s competitive business environment, companies are facing challenges in dealing with BD issues of rapid decision-making for improved productivity. The problem is that most enterprises and technology vendors have yet to explore the possibilities of an expanded Internet and are not operationally or organizationally ready. In addition, we have much more powerful methods of capturing, transferring, storing, and evaluating large amounts of data. The technological advances have an effect not only on the volume of data but also on its analysis and practical application (Analytics).

158 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

The complementarity with approaches like Cyber Physical Systems (CPS), cloud technologies, BD and future networks like 5G is highly evident. A series of announcements, from the acquisition of Nest Labs by Google for $3.2 billion, to Samsung Gear and health-related wearables to the development of Smart Home features into Apple’s iOS, have made IoT an increasingly tangible business opportunity. As outlined in Chapter 7 IoT architecture as conceptualized embraces three types of devices: ● ●



Objects directly connected to the Internet; M2M that defines the communication between machines and the access to information system without human intervention via Bluetooth, RFID, NFC, 4G or Wi-Fi for example; Communicating terminals (smart connected devices) such as tablets or smartphones.

The IoT is generally viewed as an internet- like structure where physical objects have network connectivity allowing them to communicate on-line. Powerful analytics tools can then be used to process the information gathered in large sets of structured and unstructured data. On a high level, the IoT space is broken down into three main areas: data collection, data transport, and data analysis. From an industry point of view, BD is going to play an important role in Industry 4.0. An important component of Industry 4.0 is the fusion of the physical world and the virtual world (Hermann et al., 2015). This fusion is made possible by cyberphysical systems. Smart factories constitute a key feature of Industry 4.0 (Kagermann et al., 2013). A smart factory can be defined as a factory where CPS communicate over the IoT and help people and machines in the execution of their tasks (Hermann et al., 2015). The following is some sub-processes for a smart factory (Beckhoff Automation, 2013): ● ● ● ●

M2M communication via IoT. Consistent communication from the sensor to the cloud. Integration of robotics and innovative drive technologies. RFID as the basis for parts tracking and intelligent products.

For example, object tagging and Internet-to-object communication is vital for real-time data capturing and accessibility. Cloud computing can offer computing and storage power for digitally enhanced production or manufacturing. The Industrial Internet draws together fields such as M2M communication, machine learning, and BD analytics to collect data from machines, analyze it, and use it to adjust operations.

The Internet, Data Analytics

Internet Economics: Models, Mechanisms and Management 159

BD, CPS, mobile computing, cloud computing, and Internet of the Things (IoT), etc. have great impacts on Industry 4.0. Cybersecurity in manufacturing, BD Analytics for heterogeneous data in manufacturing, and BD Analytics in real-time processing of sensor data or stream data generated in the manufacturing environment and so on. For industrials, the digital challenge is supporting the development of three new value creation fields: the production of connected things to the end consumers, production of connected equipment to industrials and the development of related markets data analysis. This last point is strategic because it is not only a better understanding of the uses of its clients and to focus resources on their specific needs. This data also materialize the border between Customer Relations Management (CRM) and know-how of the company. This data can also be collected and analyzed using BD techniques to identify and solve small but ongoing problems. The rise in quality plays an important role in reducing the first anticipated result of the analytics on the Industry 4.0 in enterprises is the integration between client and supplier networks along the value creation chain that binds the different stakeholders, their added value and expertise to compete in the manufacture of a given product. Productivity can also increase through various Industry 4.0 effects. By using advanced analytics in predictive maintenance programs. For example, the factory in Amberg, Bavaria is the demonstration of the integrated offering ‘Software for Industry’ of Siemens. The Siemens electronics plant produces custom Programmable Logic Controls (PLCs) in a state-of-the-art ‘smart factory’ where product management, manufacturing and automation systems are integrated. Intelligent machines coordinate the production and distribution of 950 products with more than 50,000 different variants, for which roughly 10,000 materials are sourced from 250 suppliers. By linking intelligent machines with data-rich components and workers, innovation cycles can be shortened, productivity raised and quality improved: the Amberg plant now records only 12 defects per million (versus 500 in 1989) and has a 99 percent reliability rate. Also, in Airbus manufacturing operators must follow tens of thousands of steps, with resulting high costs if one is missed or goes wrong. The current process involves more than a thousand different tightening tools. However Airbus is launching the development of network-enabled handheld tightening tools. For a particular task, employees can be quickly directed to the right tool which automatically ‘knows’ the next step and sets the correct calibration for the specific part that the employee wants to tighten. The smart tools can also record the operation to ensure quality control and eliminate manual logging.

160 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

Another thing is noticeable on the very organization of the company’s value chain. This value chain articulates the functions that contribute to the production, procurement of raw materials to finished product delivery, through the design and all associated manufacturing process and underpinned by the supply chain. Structuring that value chain, information flows are organized in a ‘pyramid of automation’. At the head of this pyramid we have the ERP system which is related to the production management software in which production information back from the workshop. In the traditional value chain, communication is limited between what concerns the core business and core activities for the upstream part, composed of the network of subcontractors and partners, and downstream part consisting of customer network. Embracing Industry 4.0 triggers innovation and productivity, spur growth, increase safety, and improve operations. To capitalize on its potential, companies must put data analytics at the center of their strategy. But, they need to establish clear guidelines for data integrity and security, as digital ecosystems can only function efficiently if all parties involved can trust in the security of their data and communication. The analysis of BD is not only a matter of solving computational problems, even if those working on BD in industry primarily come from the natural sciences or computational fields. Rather, expertly analyzing BD also requires thoughtful measurement, careful research design, and the creative deployment of statistical techniques. 8.6. STATISTICAL AND COMPUTATIONAL NEEDS FOR BIG DATA Data now stream from daily life thanks to technological advances, and BD has indeed become a big deal (Shaw, 2014). Data comes from multiple sources, which makes it difficult to collect, transform, and analyze. In this way, the term ‘variety’ involves several different issues. First of all, data – especially in an industrial environment – can be presented in several different ways, such as texts, functions, curves, images, and graphs, or a combination of these elements. On the other hand, this data shows great variety, which often reflects the complexity of the studied phenomenon. So data complexity is growing with the increase of its quantity its velocity and diversification of its types and sources. The increase in data produced by companies, individuals, scientists and public officials, coupled with the development of IT tools, offers new analytical perspectives. Faced with this volume and diversification, it is essential to develop techniques to make best use of all of these stocks in order to extract the maximum amount of information. So data analyzed is no longer necessarily structured in the same way as in previous analysis, but can now be text, images, multimedia content, digital traces, connected objects, etc. the rise of BD reflects the growing awareness of the

The Internet, Data Analytics

Internet Economics: Models, Mechanisms and Management 161

‘power’ behind data, and of the need to enhance gathering, exploitation, sharing and processing. It is not just about the quantity and speed of production of these data, the real revolution lies in what can be created by combining and analyzing these flows. For example, what we understand by Variety, they usually mean heterogeneity of data types, representation, and semantic interpretation. By Velocity we mean both the rate at which data arrive and the time in which it must be acted upon. The most important asset of large volumes of data has to do with the fact that they make it possible to apply knowledge and create considerable value. Combined with advanced analysis methods, they can provide new explanations for several phenomena. There are two ways to transforms data into a valuable contribution to a company (Sedkaoui and Monino, 2016): ●



Transforming data into information is one of the stages of data value production, which is exploited in order to obtain useful information and to successfully carry out company strategies. This automatically involves database information in company decision making processes; Transforming data into products or processes adds value to companies. This is produced when data analysis must be implemented in the physical world.

While these three Vs are important, this short list fails to include additional important requirements such as privacy and usability. BD are data on a massive scale in terms of volume, intensity, and complexity that exceed the capacity of standard software tools. They present opportunities as well as challenges to statisticians. The challenge in the statistical setting is that the analysis of subsets of data may present different statistical properties than the overall dataset. For example, confidence intervals based on subsets of data will generally be wider than confidence intervals based on the original data; thus, care must be taken that the overall divide-and-conquer procedure yields a correctly calibrated interval (Jordan, 2013). The continuous increase in measurement capabilities and development of new data uses make it increasingly necessary to have statistical tools to summarize and extract information from data. However, the nature of modern data (greatest dimension, diverse types, mass of data) does not authorize the use of most conventional statistical methods (tests, regression, classification). Indeed, these methods are not adapted to these specific conditions of application and in particular suffer from the scourge of dimension. When the dimensions of problems (n, p) are reasonable and that of model assumptions (linearity) and distributions are checked, or in other words, usually, when the sample or residue are assumed to follow laws putting themselves in the

162 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

form of an exponential family (Gaussian, binomial), modeling statistical techniques drawn from the general linear model are optimal (maximum likelihood) and, especially in the case of samples small size, it seems difficult to do much better. As soon as the distributional assumptions are not verified, as the assumed relationships between variables or variable to be modeled are not linear or when the volume of data (BD) is important, other computational methods compete with traditional statistical tools. Other statistical dimensions of BD concern: high dimensionality and non-Gaussian data. For example, queueing time is usually non-negative, distribution of wealth can be heavy tailed, and responses to questionnaires usually discrete. The fields of computer science and statistics have undergone mostly separate evolutions during their respective histories. It is applying statistical methods to BD where the two areas meet. This is changing, due in part to the phenomenon of BD involving digital exploration in the oil industry, exploding web mining, and massive use of GPS trails of vehicle fleets, smart buildings, clad sensors, 3D . The applications and problems require more thorough reflection on the data structures. With the rapidly increasing richness and constraining power of data, there is an urgent need to develop accurate and efficient data analysis and inference techniques. Concerning statistical methods the literature summarizes the change in two points: ●



The new approaches are on the crossroads of IT tools and Statistics: we talk about Machine Learning where algorithms generate models on large amounts of data. Machine Learning dated from the1960s: its refocussed development is due to the fact that these techniques work especially well on high amounts of information.

The applied statistics and machine learning community have been quite concerned with identifying ways to cross-validate predictions produced by these techniques, and avoid simply capitalizing on chance by overfitting their data (James et al., 2014). But, it’s necessary to point that there are two computational barriers for BD analysis: the first concerns the data that can be too big to hold in a computer’s memory; while the second is related to the computing task that can take too long to wait for the results. These barriers can be approached either with newly developed statistical methodologies and/or computational methodologies (Wang et al., 2015).

The Internet, Data Analytics

Internet Economics: Models, Mechanisms and Management 163

From an IT point of view, knowledge of Hadoop is highly useful. It allows the creation of distributed applications and “scalable” on thousands of nodes to manage petabytes of data. The principle is to split and parallelize (distribute) data batch tasks to linearly reduce the computation time (scalable) depending on the number of nodes. Hadoop becomes the mining web reference tool for e-commerce. From a statistical point of view the new challenge is both the functional representation of bases of construction and relevant models to address and take into account the complex data structures: geolocation on graphs, real-time signal, 3D images, and sequences. Every problem, especially industrial, requires a specific approach after a search in conventional engineering development. In the case of data streams, the decision support becomes adaptive or sequential. The computational tools that often are associated with the analysis of BD also can help scholars who are designing experiments or making causal inferences from observational data. Besides the aforementioned advantages, the heterogeneity of BD also poses significant challenges to statistical inference. The model is also changing; reason why data were collected to feed statistical models but now models reinventing or adapting to best exploit available data. Processing BD, in turn, puts demand on computational frameworks and models that need to be fault tolerant, flexible and light weight; example by supporting iterative and stream computing, as well as local processing of data. Computing and storage solutions form basis for advanced data analysis, including machine learning and statistical modeling. For example, Imai and Ratkovic (2013) extended variable selection methods to estimate treatment-effect heterogeneity, whereas Green and Holger (2012) used Bayesian additive regression trees to capture systematic heterogeneity in treatment effects. The scalability of statistical methods also poses a major challenge. When data becomes big, the possible number of simultaneous hypotheses, as well as data points, can be on the order of millions (American Statistical Association, 2015). Data sets derived from BD sources are not necessarily random samples of the target population. BD introduce unique computational and statistical challenges, including scalability and storage bottleneck, noise accumulation, spurious correlation and measurement errors. These challenges are distinguished and require new computational and statistical paradigm. In order to confront the challenges mentioned above statistical methods will need to be modernized. More application is needed to overcome the methodological difficulties impeding the exploitation of BD sources.

164 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

BD are characterized by high dimensionality and large sample size. These two features raise three unique challenges: (i) high dimensionality brings noise accumulation, spurious correlations and incidental homogeneity; (ii) high dimensionality combined with large sample size creates issues such as heavy computational cost and algorithmic instability; (iii) the massive samples in BD are typically aggregated from multiple sources at different time points using different technologies (Fan et al., 2014). The volume of data generated and stored by enterprises made a new step. As already indicated in previous sections, this new step creates new approaches for both architectures databases, the parallelization of computations as for algorithms and methods used. While the parallel and distributed architectures present new capabilities for storage and manipulation of data, from an inferential point of view, it is unclear how the current statistical methodology can be transported to the paradigm of BD. Cutting-edge data management, querying, and analysis techniques in computer science must be linked with fundamental approaches in statistics and machine learning to create data systems that are flexible, responsive, and predictive. Computer-science techniques need to incorporate more statistical approaches, while statistical techniques need to develop approaches for trading off statistical power and computational complexity. CONCLUSION One major challenge of the rapid advance in science and technology is that the advance of computing resources still lags far behind the exponential growth of database. Offering to businesses and decision-makers, unprecedented opportunities arise to tackle much larger and more complex BD challenges. The opportunity, however, has not yet been fully utilized, because effective and efficient statistical and computing tools for analyzing super-large dataset are still lacking. To handle the challenges of BD, we need new statistical thinking and computational methods. The collaboration between statistics and computer science is needed to control runtimes that will maintain the statistical procedures usable on large-scale data while ensuring good statistical properties. Recently some statistical methods have been adapted to process BD, like linear regression models, clustering methods and bootstrapping schemes. But, why are classical statistical methods often unsuited for BD purposes? We can link this to a lack of flexibility in existing methods, but also to the assumptions that are typically made for mathematical convenience, and the particular way of drawing inference from data.

The Internet, Data Analytics

Internet Economics: Models, Mechanisms and Management 165

For example, many traditional methods that perform well for moderate sample size do not scale to massive data. Similarly, many statistical methods that perform well for low-dimensional data are facing significant challenges in analyzing highdimensional data. In terms of statistical methods, dimension reduction and variable selection play pivotal roles in analyzing high-dimensional data. New statistical procedures with these issues in mind are crucially needed. In this chapter, we discussed one way in which development of new computational and statistical methods can contribute to the challenges introduced by BD. The point of this chapter was to show how BD can benefit from the mixture of two fields: ‘computer science’ and ‘statistics’, making the inquiry from data a more successful endeavor, rather than dwelling on theoretical issues of dubious value for industrial use. REFERENCES Ackoff, R.L. (1989). From data to wisdom. J. Appl. Syst. Anal., 15, 3-9. Agrawal, D., Das, S., El Abbadi, A. (2011). Big data and cloud computing: current state and future opportunities, New York, NY, 530-533. American Statistical Association. (2015). Discovery with Data: Leveraging Statistics with Computer Science to Transform Science and Society. Available at: http://www.amstat.org/policy/pdfs/BigDataStatistics June2014.pdf Atzori, L., Iera, A., Morabit, G. (2010). The Internet of Things: A survey. Comput. Netw., 54, 2787-2805. [http://dx.doi.org/10.1016/j.comnet.2010.05.010] Beckhoff Automation. (2013). PC-based Control – The technological foundation for Industrie 4.0, Technical Report. Covington, D. (2015). Analytics. London: Data Science, Data Analysis and Predictive Analytics for Business. Cukier, K., Mayer-Schonberger, V. (2013a). Big Data: A Revolution That Will Transform How We Live, Work and Think.. Boston, Ma: Houghton Mifflin Harcourt. Cukier, K., Mayer-Schoenberger, V. (2013b). The Rise of Big Data. Foreign Aff., 92(3), 28-40. Dietrich, B.L., Plachy, E.C., Norton, M.F. (2014). Analytics across the Enterprise, How IBM realizes Business Value from Big Data and Analytics. New York: IBM Press. Fan, J., Han, F., Liu, H. (2014). Challenges of Big Data Analysis. Natl. Sci. Rev., 1(2), 293-314. [http://dx.doi.org/10.1093/nsr/nwt032] [PMID: 25419469] Foster, I., Ghani, R., Jarmin, R.S., Kreuer, F., Lane, J. (2017). Big Data and Social Science.. Boca Raton, FL: CRC Press. Green, D.P., Holger, L.K. (2012). Modeling Heterogeneous Treatment Effects in Survey Experiments with Bayesian Additive Regression Trees. Public Opin. Q., 76(3), 491-511. [http://dx.doi.org/10.1093/poq/nfs036] Gale, W.A. (1986). Artificial Intelligence and Statistics. Reading, Ma.: Addison Wesley. Gottinger, H.W., Weimann, H.P. (1990). Artificial Intelligence: A Tool for Industry and Management. New York, London: Ellis Horwood. Gottinger, H.W. (1993). Statistical Expert Systems. Encyclopedia of Computer Science and Technology. (Vol. Vol. 27, pp. 337-349). New York: Marcel Dekker.

166 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

Harford, T. (2014). Big Data: A Big Mistake. Significance, 11(5), 14-19. [http://dx.doi.org/10.1111/j.1740-9713.2014.00778.x] Hermann, M., Pentek, T., Otto, P. (2015). Design Principles for Industry 4.0 Scenarios: A Literature Review Working Paper, TU Dortmund University, Germany. IDC. (2011). World’s data will grow by 50X in next decade IDC Study. Imai, K., Ratkovic, M. (2013). Estimating Treatment Effect Heterogeneity in Randomized Program Evaluation. Ann. Appl. Stat., 7(1), 443-470. [http://dx.doi.org/10.1214/12-AOAS593] James, G., Witten, D., Hastie, T., Tibshirani, R. (2014). An introduction to statistical learning with applications in R.. New York: Springer. Jordan, M.I. (2013). On statistics, computation and scalability. Bernoulli, 19(4), 1378-1390. [http://dx.doi.org/10.3150/12-BEJSP17] Kagermann, H., Wahlster, W., Helbig, J. (2013). Recommendations for implementing the strategic initiative Industrie 4.0 Final Report of the Industrie 4.0 Working Group, National Academy of Science and Engineering, Germany. Katal, A., Wazid, M., Goudar, R.H. (2013). Big Data: Issues, Challenges, Tools and Good Practices. IEEE Spectr., 404-409. Martinet, B., Marti, Y-M. (2001). L’intelligence économique: Comment donner de la valeur concurrentielle à l’information.. Paris: Editions d’Organisation. McKinsey. (2011). Internet matters: The Net’s sweeping impact on growth, jobs, and prosperity. McKinsey. (2013). Big Data, Analytics and the Future of Marketing and Sales.. New York: McKinsey. Middleton, P., Kjeldsen, P., Tully, J. (2013). Forecast: The internet of things, worldwide Gartner. O’Reilly Media. (2012). Big Data Now. Piegorsch, W.W. (2015). Statistical Data Analytics. New York: Wiley. Porter, M., Heppelmann, J.-E. (2015). How smart, connected products are transforming Competition Harvard Business Review. Salmin, S., Bertino, E. (2012). A Comprehensive Model for Provenance Proceedings of the First International Workshop on Modeling Data-Intensive Computing (MoDIC 2012), Florence, ItalyBerlin: Springer. Sedkaoui, S., Monino, J-L. (2016). Big data, Open Data and Data Development. New York: ISTE-Wiley. Shaw, J. (2014). Why big data is a big deal. Harv. Mag. Shroff, G. (2013). The Intelligent Web, Search, Smart Algorithms and Big Data. Oxford: Oxford Univ. Press. Siegel, E. (2016). Predictive Analytics. New York: Wiley. Varian, H. (2014). Big Data: New Tricks for Econometrics. J. Econ. Perspect., 28(2), 3-28. [http://dx.doi.org/10.1257/jep.28.2.3] Wang, X., Guo, F., Heller, K.A., Dunson, D.B. (2015). Parallelizing MCMC with Random Partition Trees. Yeluri, R., Castro-Leon, E. (2014). Building the Infrastructure for Cloud Security. New York: Apress Open (Springer Business).

Internet Economics: Models, Mechanisms and Management, 2017, 167-189

167

CHAPTER 9

Internet, Innovation and Macroeconomics “If I were to look for adjectives to describe this second (‘Internet’) economy, I’d say it is vast, silent, connected, unseen, and autonomous (meaning that human beings may design it but are not directly involved in running it)” W. Brian Arthur, The Second Economy, McKinsey Quarterly, Oct. 2011 Abstract: From Internet induced economic effects in micro-economic structures, i.e., on enterprise and industry levels, we now address network effects on a macro scale involving productivity, growth, and the business cycle. The network effect is strongly facilitated by computerization and information technologies (ITs). Ubiquitous computerization and digitalization increasingly pervade many sectors of the economy, and communication by network technologies through the Internet (the network is the computer) as a strong catalyst. Eventually, through this synergy, most sectors of the economy will be impacted by network effects. Thus, networking and computerization have far-reaching impacts on the pace and path of the economy but they could also make the economy more vulnerable to economic shocks and security breaches. We address three important issues: networks and productivity, endogeneous growth and increasing returns. The relationship between technology and productivity used for the United States on the economy or sector level, found little evidence of a relationship in the 1980s. Capital IT investment between 1977 and 1989 rose several hundred per cent but was barely reflected in a rise in output per worker. There was this famous saying by Nobel prizewinning economist Robert Solow (1987): “You can see the computer age everywhere except in productivity statistics”. In the 1990s, such a positive relationship on the firm level was established empirically. On a short-term basis: one-year difference in IT investments vs one-year difference in firm productivity should be benchmarked by benefits equal to costs. However, benefits are supposed to rise by a factor of 2–8 in forecasting future benefits through productivity growth.

Keywords: Catchup Race, DARPA, E-commerce, Economic transformation, Endogeneous Growth, General Purpose Technologies (GPT), Increasing Returns, Information Goods, Information markets, Minitel, Multi-Factor Productivity (MFP), Productivity Paradox, SABRE, Total Factor Productivity(TFP).

Hans W. Gottinger All rights reserved-© 2017 Bentham Science Publishers

168 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

9.1. INTRODUCTION While the effects of Internet on micro structures, that is, on industry levels, have been our primary concern in the previous chapters, we now address network effects on a macro scale involving productivity, growth, and the business cycle. The network effect is here strongly facilitated through computerization and information technologies (ITs or ICTs). Ubiquitous computerization pervades many sectors of the economy, and communication by network technologies such as the Internet (the network is the ‘computer’) is a strong catalyst. Eventually, through this synergy, most sectors of the economy will be impacted by network effects and ‘increasing returns’ (Arthur, 2011). Thus, networking and computerization have far-reaching impacts on the pace and path of the economy but they could also make the economy more vulnerable to economic shocks. In the World Bank Development Report (2016) on digitization and development the economics part identifies several advantages of the digital economy through the Internet: 1. Facilitating widespread sectoral increasing returns mechanisms (IRM) with corresponding technological entrepreneurship. 2. Advancing significant productivity increase and competition across industrial sectors. 3. More scalable trade and capital utilization. 4. Expansion of a more product/service /platform differentiated economy. We will address three important issues: networks and productivity, endogeneous growth and increasing returns. Early on, Brynjolfsson and Hitt (2000) provided some examples on some productivity- enhancing activities on the enterprise level which aggregate to impacting total factor productivity on a macro scale, e.g. (i) computerization of ordering along the supply chain, (ii) Internet-based procurement system, and (iii) computer-based supply chain integration. There are many other examples of pervasive internet applications that trickle down to aggregate productivity increases. The relationship between technology and productivity used for the United States, on the economy or sector level, found little evidence of a relationship in the 1980s. The investment between 1977 and 1989 rose several hundred percent but was barely reflected in a rise in output per worker. There was this famous saying by Nobel prize-winning economist Robert Solow (1987): “You can see the computer age everywhere except in productivity statistics” though somewhat modified weakened a few years later (Uchitelle, 2000). But even 30 years later, a similar dictum is echoed for the US in a Wall Street Journal column (Ip, 2016).

Internet, Innovation

Internet Economics: Models, Mechanisms and Management 169

In the 1990s, a positive relationship on the firm level was established empirically. On a short-term basis: one-year difference in IT investments vs one-year difference in firm productivity should be benchmarked by benefits equal to costs. However, benefits are supposed to rise by a cumulative factor in forecasting future benefits (in productivity growth). Economic techniques focus on the relatively observable aspects of investment, such as price and quantity of computer hardware in the economy, but neglect intangible investments in developing complementary new products, services, markets, and business processes. The recent World Bank Development Report (2016, 51) clearly states three key factors for growth on the macro economy: (i) Inclusion – through international trade, (ii)efficiency – through capital utilization and (iii) innovation − through competition. Current statistics typically treat the accumulation of intangible capital assets, new production systems, and new skills as expenses rather than as investments. This leads to lower levels of measured outputs in periods of net capital accumulation. Output statistics miss many of the gains of IT brought to consumers such as variety, speed, and convenience. For instance, US productivity figures used not to take account of quality changes, in particular, in services industries: (a) financial services sector (Automatic Teller Machines), (b) health care (diagnosis, medical decision making), and (c) legal services (online information and legal advice). 9.2. BASICS OF NETWORK ECONOMY If we mirror a real economy through Internet activities and interactions, we arrive at a network economy that consists of positive production and consumption specific network externalities (Gottinger, 2016). That is, in a typical network, the utility of each consumer is likely to increase by new entry (network nodes) of others. The collective utility (welfare) through entry of each new consumer exceeds the private utility of adding another node to the network. Depending on the initial network structure, adding a new node to an n-node network would create at least 2n new products. With such a network expansion, we would be able to increase economies of scale as well as economies of scope which originate from positive network externalities. Network goods are not automatically complementary to each other. For this, it needs compatibility and coordination among network components. In technological networks, compatibility originates from technical standards, or one creates compatibility through bridging technologies. In industrial situations, positive network effects are triggered at critical tipping points under dynamic competition (Varian and Shapiro, 1999). Network economies likely thrive under monopolistic and oligopolistic competition rather than in market forms closer to atomistic

170 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

competition when production follows marginal cost pricing. For a monopolist, in this context, there is an incentive to reduce output at given capacity if his marginal revenue is lower than price. However, more often it is in the interest of the monopolist to expand network size for utilizing positive network effects, in particular if his marginal revenue is positive and his marginal costs of additional products (e.g., software) are tending closer to zero. It is expected that in dynamic network markets the latter incentive dominates. In those cases, monopolistic or oligopolistic markets can generate positive welfare effects in the economy. In such circumstances, there should be differentiations in the assessment of competitive rules and anti-trust judgements. The Internet use hinders price discrimination in promoting price competition since it increases market transparency for liberalized and globalized markets. The Internet reinforces an already existing monopolistic tendency in network markets so that firms with larger market shares benefit most from market expansion. Network markets bear the property that the strong become stronger and the weak become weaker. In markets for software, for example, a particular software is worth the most if it allows most of the compatible applications. Structural conditions on network markets tend to lead to oligopolistic markets but with strongly dynamic competitive features. In order to gain and maintain leadership in such dynamic markets, there often is a technological race between rivals (Gottinger, 2006) which can take heated forms on innovation. Since the product portfolio is not intrinsically limited, there are increased incentives of innovative firms to enter the market, and therefore to contribute to a broader expansion. 9.3. ECONOMIC TRANSFORMATION The story of the revolution in IT is a story of technology and a story of innovations in business organization and practice. The network economy is as much a story about changes in business organization, market structures, government regulations, and human experience as it is about new technology. Information technology builds the capabilities to process and distribute digital data to multiply the scale and speed with which thought and information can be applied. Thought and information can be applied to almost everything across the board. Computer chips, lasers, broadband Internet, and software are the key components of the technology that drive the network economy. The networking of the economy through processes of production, distribution, transportation and communication all started out in the second half of the nineteenth and continued much in the twentiest century, all driven by the ‘dynamics of industrial capitalism’. It shows the evolutionary path of industrial growth that characterizes modern western societies (Chandler, 1990).

Internet, Innovation

Internet Economics: Models, Mechanisms and Management 171

Productivity growth and technological change in the core sectors of the information-processing extension has been immense. Skeptics can see this productivity explosion as just another example of a ‘leading sector’– an explosion of invention and innovation that revolutionizes productivity in a broad industrial cross-section of the economy, now commonly understood as general purpose technologies (GPT). There have been many such leading sectors in the past – air transport in the 1960s, television in the 1950s, automobiles in the 1920s, organic chemicals in the 1890s, electricity and railroads in the 1870s. Yet, they did not only change the standard dynamics of economic growth, they defined it. But what we are experiencing is not just a decade-long boom as technology opens up new possibilities in a leading sector of economic growth. We are experiencing something deeper and broader. Semiconductors, computers, and communications do constitute a large leading sector, but this is not the whole story (Gershenfeld, 1999). Some technological changes do not just amplify productivity in one sector but enhance productivity in all economic sectors. They open new possibilities for economic organization across the board. They change what can be done and how it can be done across a wide range of industries. And they require changes in ideas about property and control, in the way that the government regulates the economy, in order for these new possibilities to be realized. Innovative users began to discover how they could employ computer in new ways. For example, IBM/American Airlines used computers to create its SABRE automated reservations system (Campbell-Kelly, 2003). The insurance industry first automated its traditional process, its back office applications of sorting and classifying. But insurance companies then began to create customized insurance products. The user cycle became one of the first learning about the capabilities of computers in the course of automating established processes, and then applying that learning to generate innovative applications. As computing power has grown, computer-aided aided product design from airplanes built without wind-tunnels to pharmaceuticals designed at the molecular level for particular applications has become possible. A major function of a computer is that of a ‘what-if ’ machine. The computer creates models of ‘what-if ’: that is, what would happen if the airplane, the molecule, the business, or the document were to be built up in a particular way. It thus enables an amount and a degree of experimentation in the virtual world that would be prohibitively expensive in resources and time in the real world. What does it mean to say that computing is becoming pervasive? The new production and distribution processes that pervasive computing makes possible are visible at the checkout counter, the petrol station, and in transportation services. The most important part of pervasive computing is the ‘computers that we do not see’ (Brynolfsson and MacAfee, 2011).They become embedded in

172 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

traditional products and alter the way such products operate. In automobiles, antilock brakes, air bags, engine self-diagnosis and adjustment and self-driving are performed by embedded microprocessors that sense, compute, and adjust. The level of automotive performance in systems from brakes to emission control is vastly greater today than it was a generation ago because of embedded microprocessors, and it extends to the choice of transportation modes through the emergence of ‘intelligent transportation systems’(ITS). Now much more than before we see the computer as a network (the ‘network is the computer’) and more recently, the network containing ‘things’ as computing devices in a network (IoT). As the cost of communication bandwidth dropped, it became possible to link individual sensing, computing, and storage units. Today it is well taken that an automated teller machine (ATM) verifies the bank balance we hold in a bank in a distant city. The key point is not that rapid transmission has become technically feasible, but that the costs of data communication are dropping so far and fast as to make the wide use of the network for data transmission economically feasible for nearly every use imaginable. With the early data use of data networks it was once again leading-edge users who created new applications in their pursuit of competitive advantage. The experimental nature of the Internet in the beginning, its evolution to a full-scale global data and communication network is a case in point. Networking began either as private corporate networks, or, as in the case of the French Minitel, as a public network with defined and limited services. And data communications networks began their exponential expansion as experimenting users found new useful applications and configurations. But few foresaw the true long-run potential of high-speed data networking until the TCP/IP protocol and the World-Wide Web (WWW) revealed the potential benefit of linking networks to networks. Every computer became a window into the world’s data store. And as the network grew it became more and more clear that the value of the network to everyone grew as well – a principle that is now termed Metcalfe’s law. The build-up of the Internet has been so rapid in large part because the Internet could be initially run over the existing voice telecommunications system. Even before the new technologies designed from the ground up to manage data communications emerged the global Internet had already established its reach. The first generation Internet has grown exponentially world-wide during the last decade of the century (Odlyzko, 1998). Some of the elements of the next generation of data networks are already evident. First, for consumers and small businesses one dramatic advance will be broadband to the home to create high-bandwidth (Ferguson, 2002). The acceleration in speed will change the kinds of tasks that can be accomplished over

Internet, Innovation

Internet Economics: Models, Mechanisms and Management 173

the Internet. Second, wireless voice networks will be even more extensively deployed as the wired phone network. Widely diffused wireless data networks will set off another round of experimentation and learning. Third, the capacity and cost of the very backbone of the network will evolve dramatically over the next years, bringing new architectures, lower costs, ongoing experimentation and new applications. Technological advance and diffusion clouds the view of how the growth of the network will transform business organization and business competition. How will the entire economy be linked into information processing and data communications? The prospects are still grossly uncertain because the growth of the network promises to transform the whole society. Traditional businesses that act as intermediaries – like stockbrokers and travel agents − will be irrevocably altered. Traditional products like automobiles are already being marketed in new ways. Stores will not disappear, but the mix of stores and what stores do will change. New ways of reaching customers will in turn drive new ways of organizing production and delivering goods to consumers. Today we continue to see strategic experiments in the form of new companies trying to exploit the Web and established companies trying to defend their positions. But we do not know which of these experiments in corporate information and network strategy will be successful. We can, however, see which strategies have been successful in the recent past. Consider the retail sector, and take the consumer-goods distributor Wal-Mart. Wal-Mart starting as a conventional retail distributor, has evolved into a very efficient network retailer by successfully solving problems of control and distribution, by adopting consistently and on a large-scale basis modern information technology. So Wal-Mart is among the top firms that excel in the use of information and network technology related to output and revenue. Some of those benefits may not come forward as rapidly as expected. Economic historian Paul David (1990) points out that it took nearly half a century for business users to figure out the possibilities for increased efficiency through factory reorganization opened up by the electric motor. Finding the most valued uses for the next wave of computer and communications technologies will probably not take as long, but it will take time. 9.4. ASSESSING THE TRANSFORMATION Why, skeptics ask, are boosters of information processing and communications so sure that this current transformation is more important than the leading sectors we have seen in the past? Is the claim justified that the impact on our material welfare through the Internet is so much greater than the impact of the automobile, or of penicillin and other antibiotics, or of network television? R. J. Gordon (2000,

174 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

2004) is one of the early and leading skeptics, and it is appropriate to discuss the line of his main arguments. Even Gordon admits the pervasive impact of IT: “A revival in productivity growth … is impressive in comparison with the American historical record.. .” He argues, based on econometric studies, that ‘spillover effects on multi- factor productivity in the non-computer economy are absent or slightly negative’. However, Gordon looks at the computer primarily as a ‘stand alone’ IT, he only considers 12 percent of US GDP as generated by the computer/semiconductor industry, but neglects the indirect and cross network effects on the economy. In particular, the computerized network industry and the Internet that comprises about 30 percent of the US economy (most of the professional services, media, telecoms), will have a significantly larger impact in the ‘catch-up’ economies. There also appear to be problems with measurements as suggested by Gordon: ●

● ● ●

the comparative statistical basis is too small (only 5 years in the 1990s) which ignores that the innovation effect is cumulative depending on critical mass and spillover into several sectors. neglect of synergy of IT and network economy (the ‘network is the computer’). neglect of intangible effects (quality, speed, new products). neglect of laggardness of IT effects, not considered in econometric estimation.

According to Gordon (2000, 2004): “The second industrial revolution took place from 1860 to 1900 and was triggered through invention of electricity, internal combustion engine, the telephone, and led to increased productivity growth from 1913 to 1972”. What is not mentioned is that productivity growth did not increase by the emergence of those technologies themselves but by the fact that they were embedded in products or even created an own network industry. A question mark is placed on: Does Information Technology and the Internet sustain a Third Industrial Revolution? And the New Economy is certainly more about computers than pharmaceuticals. Why this artificial distinction? We can venture the thesis: Progress in Science is heavily carried through computers (Wolfram, 2002). In most cases of recent progress in science we could put forward the thesis that all science (now) is (in a way) computer science. Examples: Human Genome projects to disclose DNA classification, new drug designs, system products such as the heart pacemaker, as there are many new discoveries in astronomy, physics, biology, medicine, with new technology

Internet, Innovation

Internet Economics: Models, Mechanisms and Management 175

platforms which could snowball into new network industries. Increasing returns industries are at work in emerging and maturing biotechnology, medical instruments and pharmaceuticals (Gottinger et al., 2010). This is not as yet reflected fully in the measurements, because it takes time to have this synergy come out in hard numbers. Thus computers/communications is a cross-sectional technology impacting almost every sector of the economy, horizontally (industry and firm specific) as well as vertically (downstream and upstream, i.e. computers in basic research, basic IT research in non-computer industries, etc.). The roadmap of this industrial revolution is new computing technology plus networking expanding to traditional existing industries and crafts. Compared to one of the five big innovations in the Second Industrial Revolution, how do they compare with the Internet today? How far goes the information revolution? In terms of welfare impacts for the population concerned, Gordon asks, “is for the population of Houston air conditioning more important than the Internet”, or for the population of Minneapolis, “is indoor plumbing more important than the Internet?”. Gordon (2000, 2004): “The cost of computing has dropped exponentially, the cost of thinking is what it always was. I cannot type or think any faster than I did with my first 1983 personal computer that contained 1/100th of memory and operated 1/60th of the speed of my present model”. This looks like a scientist who has everything in his head, and gets his ideas out where it does not make a difference what vintage the computer is and whether networked or not. Look at how one produces a ‘paper’, or launches a new product. Thinking is not an island, in fact, many ideas originate through a network. Criticality, size of network and speed matters, even if one finds out by speedy ways that somebody else had the idea before (because then he can realize his opportunity costs faster and pursue a more productive idea). Suppose a paper is incremental knowledge, as a new product, then its faster production and diffusion is higher productivity! Increasing network size and speed (say through broadband fiber optic lines etc.) could even make more than an order-of-magnitude difference. Gordon (2000, 2004): “ . . . the main productivity gains of computers have already been achieved”.

176 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

Here, one of the bigger weaknesses of the Gordon study is the presumption of projecting the past and existing functionality of present IT (computers) into the future. Gordon (2000, 2004): “Much of computer usage is through competition, and thus “burned up” on an more aggregate level which does not show up in aggregate productivity (zero sum game) while (paradoxically?) on a micro level the rate of return on investment in computers substantially exceed other investments”. This is an interesting argument, but does it stick? Yes, but it is only a one-shot effect, therefore, only a blip in the time series on productivity rates. Gordon (2000, 2004): “The Internet creates no new products, they are already available”. Again these are strong words. If we describe goods and services through commodity characteristics, then speed and higher quality (less ‘externality bound’) could be important characteristics defining a new product. If the Internet allows me to do things electronically which otherwise would need physical transportation, then it would be a partial or complete substitute (with zero mileage costs and almost zero pollution) for a fossil-fuel-type transportation device (Park and Roone, 2002). And there are network effects, increasing returns fostered though network complementarity and foremost artificial intelligence (AI) that favors productivity increase by leaps and bounds. In Gordon’s (2016, Chap. 17) most recent update work he basically still keeps his skepticism toward what he calls the ‘techno-optimists’ by “arguing that the main benefits of digitalization for productivity growth have already occurred during the temporary productivity growth revival of 1996-2004, (in the US). As a summary we could subsume that Gordon would not put the Internet as a GPT or transformational innovation on a a par with electricity and the internal combustion engine to explain major growth processes for the US economy (as a proxy for a leading advanced economy). 9.5. THE PRODUCTIVITY PARADOX But if this wave of technological innovation is so important, why has it not yet had a more powerful impact on our picture of the overall economy?. Nobelist Solow (1987) shared the general view that computers-and-communications held a potential for economic revolution. Yet when he looked at the aggregate overall measurements of the state of the economy, he saw slow productivity growth.

Internet, Innovation

Internet Economics: Models, Mechanisms and Management 177

After having made a detailed accounting of the computer’s contribution to productivity statistics, Triplett (1998) of the Brookings Institution responded: “You don’t see computers in the productivity statistics yet, but wait a bit and you will.” The fourteen years from the date generally accepted as the beginning of the productivity ‘slowdown’, the oil crisis years 1973–87, with Solow’s observations, had seen measured output per hour worked in the non-farm business sector of the US economy grow at a pace of only 1.1 per cent per year. By contrast, the fourteen years before 1973 had seen measured output per hour worked grow at a pace of 2.8 per cent per year. Even after Solow had asked his question the productivity performance worsened, between 1987 and 1995 measured output per hour worked for the US non-farm business sector grew at only 0.8 per cent per year (Council of Economic Advisers, 1999). This ‘productivity paradox’ was sharpened because at the microeconomic level economists and business analysts had no problem finding that investments in high technology had enormous productivity benefits. Typical rates of return on investments in computers and networks amounted to more than 50 per cent per year. Firms that invested heavily in IT and transformed their internal structures so that they could use their new technological capabilities flourished in the 1980s and 1990s – and their lagging competitors did not (Brynjolfsson and Hitt, 2000). Attempts have been put forward to resolve the paradox. One comes from the economic historian Paul David (1990) arguing that it takes considerable time for an economy to restructure itself to take full advantage of the potential opened up by revolutionary technology. For example, David observed that it took forty years for the American economy to realize the productivity potential of the dynamo. Electric power became a reality in the 1880s. But it was not until the 1920s that there had been enough experimentation and use of electricity-based technologies for businesses to learn how to use electric power effectively, and for the inventions of Edison, Westinghouse and Siemens to pay off in big leaps in industrial sector productivity. Another resolution of the productivity paradox stems from the fact that while technologists look at the leading edge of technologies national income accountants see changes reflected in their aggregate data only when a critical mass has been building up. With a considerable lag in invention and innovation dates and its economic impacts there have been observations in economic history with innovations having taken place in 1760 that only led to a marked acceleration of economic growth in the 1840s and 1850s when industrial technology diffused more widely.

178 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

One interesting explanation is provided by Sichel (1997), as to the lag of productivity growth, for example, as the lack of critical mass. În the 1970s and 1980s computers were simply too small a share of total investment in total GDP to expect to see a strong contribution to aggregate economic growth – even if for each investing company the rate of return on IT investments was very high. This observation is superimposed by findings that manufacturing sectors that use computers most intensively showed a gain in relative productivity growth in the 1990s, although the most computer-intensive service sectors do not (McGuckin and Stiroh, 2001). This measured weak productivity growth of the services needs to be reconciled with the massive IT investments and high expectations of the new economy. However, as Sichel (1997) points out, what was true in the 1980s was no longer true in the 1990s: then investments in IT were more than half of total investment. From 1995, productivity growth has been accelerating, and computers and communications are the prime source for this recent acceleration in American economic growth. For OECD countries there have been comprehensive recent investigations supporting this view (OECD, 2000). Still, there continue to be conflicting arguments as to the extent of IT’s contribution to productivity and economic growth. The most comprehensive explanations are Onliner and Sichel (1994) and Jorgensen and Stiroh (1995,1999). For both the growth accounting equation is calculated as: dt Y = scdt Kc + snc dt Knc + sLdt L + dt π (*)

where dtY = dY/dt, the rate of growth of output, dtKc, dtKnc, and dtL are rates of growth of the inputs – Kc computer capital (or computer capital services), Knc non-computer capital (services) and L, labor, si is the share of input i, and dt π the growth of multifactor productivity. This equation says that the rate of growth of output (dtY) equals the share-weighted growth in inputs (for example, scdtKc is the rate of growth of computer capital, weighted by the share of computer capital in total cost), plus the rate of growth of multifactor productivity. Jorgenson and Stiroh (1995) estimate the share of capital services provided by computer equipment capital, using the capital accounting framework developed by Jorgenson (1990); Onliner and Sichel (1994) use computer equipment’s income share. The results of both papers are highly consistent. Computer equipment made a relatively small contribution to economic growth, even during the period of the 1980s when computer technology became so widely diffused throughout the economy. In the growth accounting framework of equation (*), even very rapid rates of input growth, and the growth of computing equipment has been rapid indeed, make only relatively small contributions to growth when the

Internet, Innovation

Internet Economics: Models, Mechanisms and Management 179

share of this equipment is small. In these calculations, computer equipment still accounts for only around 2 percent or less of the physical capital stock, and under 2 percent of capital services. Onliner and Sichel (1994) enlarge the definition of computers to encompass all of information-processing equipment and also computing software and computerusing labor. The result remains unchanged. On any of these definitions the shares remain small and so does the growth contribution of IT. To check the reasonableness of their results, Onliner and Sichel (1994) simulate results for the assumption that computers earn supernormal returns, as equation (*) implies that computers earn the same return as earned on other capital equipment. Romer (1986), Brynjolfsson and Yang (1996) all argued or implied that computers yield higher returns than investment in other capital. The alternative simulations raise the contribution of computing equipment to growth (from 0.2 to 0.3 or 0.4), but all of them confront the same problem: the share of computing equipment is simply too small for any reasonable return to computer investment to result in large contribution to economic growth. This could change with universal digitization. Growth-accounting exercises calculate the computer’s contribution to growth, not its contribution to multifactor productivity. As equation (*) shows, multifactor productivity’s contribution to economic growth is separate from the contribution of any input, including the input of computers. If one interprets the productivity paradox as applying to multifactor productivity, growth accounting exercises do not shed very much light on it. In summary, computers make a small contribution to growth because they account for only a small share of capital input. Does the same small share suggest that they likewise cannot have an impact on productivity? Not so. The paradox remains a subject for discussion for other reasons to follow. One reason is that computer impact is less observable in sectors of the economy whose output is poorly measured. Griliches (1994) noted that more than 70 per cent of private sector US computer investment was concentrated in whole- sale and retail trade, financial insurance, and real estate and services. These are exactly the sectors of the economy where output is least well measured, and where in some cases even the concept of output is not well defined (finance, insurance, consulting services). That there are serious measurement problems in all these areas is well established. It is also the case that services account for a large part of output in advanced economies. Services that directly affect the calculation of GDP are those in personal consumption expenditures (PCE) and in net exports

180 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

(and of course the output of the entire government sector is notoriously mismeasured). Thus, services make up a large proportion of the aggregate productivity ratio and they are poorly measured. Of course, services include many that have probably not benefited appreciably from output-enhancing productivity improvements caused by computers. The other position that the real computer impact is delayed and needs a critical mass is argued by David (1990). He has drawn an analogy between the diffusion of electricity and computers linking electricity and computers because both ‘form the nodal elements’ of networks and ‘occupy key positions in a web of strongly complementary technical relationships’. Because of their network parallels, David predicts that computer diffusion and the effects of computers on productivity will follow the same protracted course as electricity. More than four decades have passed since the introduction of the commercial computer. Since then, in about forty years the price of computing power has declined more than two thousand fold. No remotely comparable price decreases accompanied the dawning of the electrical age. David reports that electricity prices only began to fall in the fourth decade of electric power; and although Nordhaus (1997) estimates that the per lumen price of lighting dropped by more than 85 percent between 1883 and 1920, two-thirds of that is attributable to improved efficiency of the light bulb, rather than to electric-power generation. Because their price histories are so different, the diffusions of electric power and computing power have fundamentally different – not similar – patterns. In the computer diffusion process, the initial applications supplanted older technologies for computing. Water and steam power long survived the introduction of electricity; but old, pre-computer age devices for doing calculations disappeared long ago. Brynjolfsson and McAfee (2011, Chap. 2) argue that computerized networks belong to those general purpose technologies (GPTs) that linearly improve over a first part of the chessboard but in its further expansion dramatically explode nonlinearly in the second part of the chessboard. Clearly, in a review of the recent literature, ICTs act as GPTs and spread values to many sectors (Cardona et al., 2013). Therefore it breaks barriers of performance through combinatorial innovation. This is a model of technology evolution that in view of an increasing returns mechanism (IRM) or a Moore Law type projection could materialize but it is not clear yet at which point we have arrived. GPTs not only act upon horizontal applications across industries but also in a time-directed way in inducing applications in emerging industries. Thus like electricity, the ‘computer as a

Internet, Innovation

Internet Economics: Models, Mechanisms and Management 181

network’ induces future industries through combinatorial innovations toward increasing returns industries (IRI). If performance growth would have the shape of a hockey stick we could envision dramatic advancements. Just to cite one example. In a contest on autonomous car driving in the 2004 DARPA Grand Challenge, a driverless vehicle to navigate a 150-mile route couldn’t even make it eight miles into the course and it took hours to achieve even that. In contrast, only six years later, Google announced officially that it had a fleet of autonomous cars that had driven more than 1,000 miles on American roads without any human involvement, and more than 140,000 miles with only minor inputs from the person behind the wheels. In only six years a sea-side change in effectiveness of IoT technologies, also an overwhelming success of big data processing. With this preview in mind it would be no surprise if we enter a new age of broad and deep technological advances comparable to advancements predicted for the second chess board. In view of spreading the ‘network intelligence’ over even larger leading sectors of the economy and see bottom up microprocesses interact to an aggregate macro innovation impact we could be confident to expect a significant productivity boost in the decades ahead as compiled by a growing list of taxonomic industrial changes (Tapscott, 2015). 9.6. GROWTH PROCESSES If we look for growth in a network centered industry we may start to single out growth processes for individual companies in that industry. A natural way for a firm to grow is to engage in organic growth through obtaining larger market shares even in a mature or declining industry. This may be achieved through indigeneous innovation on a broader front: technology, organization, services. Second, strategic alliances on major business projects as well as mergers and acquisitions (M&As) can lead companies to new growth opportunities. For example, many well-known companies in the IT business, from Cisco Systems, Google, Intel, Oracle have been consistently growing through M&As even if some have been left under their growth potential. If acquisitions were driven through innovations and entry into new or complementary fields they were targeted at new opportunities and markets. If innovations in-house or through acquisitions led into new complementary markets the resulting growth potential could be characterized as network centered growth, for example through the ‘increasing returns’ paradigm. The Network Economy is formed through an ever-emerging and interacting set of increasing returns industries (IRIs); it is a high-intensity, technology driven racing, dynamic entrepreneurship, and focused risk-taking through (free) venture capital markets endogenized by societal and institutional support.

182 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

Racing behavior on technological positions among firms in high technology industries, as exemplified by the globally operating telecommunications, computer industries but also biotech industries, produce spillover benefits in terms of increasing returns and widespread productivity gains. Due to relentless competition among technological leaders the network effects result in significant advantages in the value added to this industry, as value networks, contribute to faster growth of GDP, and through a flexible labor market, also to employment growth. This constitutes a new paradigm in economic thinking through network economies and is a major gauge to compare the wealth-creating power of major economic regions in the global context. The trajectories of technological evolution certainly seem to suggest that firms from one frontier cannot simply jump to another trajectory. Witness, in this regard, the gradual process necessary for a firm in the catch-up race to approach those in the frontier race. There appears to be a frontier ‘lock-in’, in that once a company is part of a race, the group of rivals within that same race are the ones whose actions influence that company’s strategy the most. Advancing technological capability is a cumulative process. The ability to advance to a given level of technical capability appears to be a function of existing technical capability. Given this path dependence, the question remains: why do some firms apparently choose a path of technological evolution that is less rapid than others? Two sets of possible explanations could be derived from our case analysis, which need not be mutually exclusive. The first explanation lingers primarily on the expensive nature of R&D in industries like telecommunications and computers which rely on novel discovery for their advancement. Firms choosing the catchup race will gain access to a particular technical level later than those choosing the frontier, but will do so at a lower cost (Gottinger and Goosen, Chap. 4). The evolution of a cross section of high technology industries reflects repetitive strategic interactions between companies in a continuous quest to dominate the industry or at least to improve its competitive position through company level and industry level technological evolution. We can observe several racing patterns across industries, each of which is the result of a subset of firms jockeying for a position either as a race leader or for a position near the leader constituting a leadership club. The identification and interpretation of the races relies on the fact that different firms take very different technological paths to target a superior performance level with the reward of increasing market shares, maintaining higher productivity and profitability. In a Schumpeterian framework such races cannot be interpreted in a free-riding situation where one firm expands resources in advancing the state of technology and the others follow closely behind

Internet, Innovation

Internet Economics: Models, Mechanisms and Management 183

(Gottinger, 2006) such spillover interpretations are suspect when products are in the domain of high complexity, of high risk in succeeding, and different firms typically adopt different procedural and architectural approaches. 9.7. THE GLOBAL NETWORK ECONOMY A strong companion of the network economy is deregulation, liberalization, and privatization. We take the case of telecommunication as a prototype of a network industry. Once telecommunications deregulation began, deregulation in air travel and finance (banking, brokering, insurance) freed major companies in those industries to experiment with new applications of computing and telecommunications. The newly competitive environment in their own industries gave them every possible incentive to try to gain advantage on their competitors. Subsequently, other companies had to imitate the innovators, or leapfrog them by developing still newer applications – usually in conjunction with IT producers. Deregulation thus created eager experimenters willing to try new applications developed by IT firms. In the US deregulated companies in key industries eagerly sought new competitive tools and experimented with new organizational forms and new products and services. In Japan, at least in the early stage, telecommunications remained a regulated monopoly, and so established companies and entrepreneurial groups could not experiment in how they used telecommunications; they could only do what NTT (the monopoly PSTN company) had planned for them to do. Japanese banks and brokerages remained regulated, so there was little competitive pressure to seek out new IT applications to transform their businesses. Both the possibilities of experimenting with new uses for telecommunications and the supply of lead users eager to do that experimenting were absent. Policies assuring competitive markets, deregulation, and competition policy, were essential. In the very beginning it was antitrust that moved the infant transistor technology into the competitive realm. The then-monopoly AT&T was forced to make the new technology – invented at Bell Labs – available to all. Had the transistor remained under monopoly control as the AT&T’s development review process tried to think of uses for it, recent industrial history would be fundamentally (but unknowably different). It is highly likely that innovation and diffusion would have proceeded more slowly and narrowly, resulting in both fewer innovative and fewer innovative competitive technology firms. New companies such as Fairchild, Intel, AMD, National Semiconductor took the new technology and led the process of innovation and risk taking that has brought prices down further and faster, and performance up further and faster, than with any other major new technology in history.

184 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

Most of the large integrated companies that dominated vacuum-tube consumer electronics – RCA, Philco, Sylvania – were soon bypassed by the new companies that were born and grew with new technology. New (and some reborn) technology producers working closely with innovative users created new applications, many of which were not initially obvious: from automobile door locks, anti- skidding brakes, through medical imaging and surgical equipment, to Automatic Teller Machines, personal computers, and gene sequencers. The creation of competitive markets in telecommunications was not an easy policy to establish or to implement. Deregulation and competition drove the rapid build-up of private networks in America and the rapid private take-up of Internet access. As we move to next generation Internet, wireless networks, and high-speed Internet access in homes and small businesses, the question reposes itself: how to sustain competition and in which parts of the market? (Gottinger, 2016, Chap.5). The network economy is an ‘idea economy’. Certainly economic growth has always been engineered by ideas. Given the increasing pace of innovation that rests on ever new ideas and the importance of information-based products that are easily diffused in perfect form over digital networks, the question of intellectual property becomes central. Ideas and ‘information goods’ have particular characteristics that distinguish them from ordinary goods. These include (1) marginal costs of reproduction and distribution that approach zero, (2) problems of transparency, and (3) non-rival possession. The emergence of and transition to a digitized network economy opens a broad range of issues. The network economy creates new products and transforms established industries with new ways. It transforms storage, search, diffusion, and indeed generation of ideas and information. It reopens issues such as rights to copy because previously copies generally involved degraded quality or substantial cost. Summarizing, the Internet-based economy creates new opportunities and paradigms. As laid out in previous chapters the transaction-based Internet is e-commerce. It has a very large regional and often global reach, it creates an enhanced form of competitiveness in competitive markets, eases market entry (but accelerates market exit), it provides a very fast mechanism which is highly cost efficient for incremental costs, it is highly able to induce significant increases in productivity through the entire business process such as (i) just-in-time (JIT) delivery by data exchange, integration of production and sales processes of different companies, (ii) in many cases no inventory is necessary because of manufacturing on demand

Internet, Innovation

Internet Economics: Models, Mechanisms and Management 185

(disintermediation), (iii) lowering costs by planned product obsolescence, thus no price dumping, (iv) lower sunk capital, higher profitability, and greater reach. These advantages exist in a stable or cyclical growth environment of the industry (as part of the network economy) but, for instance, JIT production based on effective demand only constitutes partial (incomplete) information, and can turn to a big disadvantage in a network economy along the supply chain because a cyclical downturn can reinforce itself in a cumulative, steep decline. Summing up the Productivity Advantage of IT Industries: ●





● ●





IT is pervasive in all business activities (disaggregation of value chain), as in Orders-Marketing-Billing-Logistics. It allows to concentrate on certain activities and outsource others ® Higher Specialization with Lower Capital Costs. Significant Increase of Market Transparency, Reduction of Transaction Costs and Barriers to Entry ® Intensification of Competition. IT is truly global. More product information available all the time, sales around the clock and around the globe. Speeding up information and innovation, reducing product launch time. Database management: Saving of Individual Customer Profiles. Customer Relations Management (CRM) Business-to-Business (B2B) online ordering cuts procurement costs, by finding cheaper suppliers and reducing errors in orders. Much lower distribution costs in electronic delivery (financial services, software, music).

Intangible Assets Current statistics typically treat the accumulation of intangible capital assets, new production systems, and new skills as expenses rather than as investments. This leads to lower levels of measured outputs in periods of net capital accumulation. Output statistics miss many of the gains of IT brought to consumers such as variety, speed, and convenience (Brynjolfsson and McAfee, 2011, 2014). Even though the most visible developments in digital markets recently have been in business-to-consumer (B2C) markets, the biggest economic changes are likely to be in the business-to-business (B2B) part of the value chain. B2B electronic commerce has been around longer than B2C commerce with the introduction of technologies such as electronic data interchange (EDI). However, now that the B2C part of the value chain is becoming totally digital, it is increasingly easy to integrate the whole value chain so that consumers become an important player in all steps of value creation. The most immediate impact of

186 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

this change will be in logistics and supply chain management. Logistics and supply chain management practices are changing to include customers in the value chain. Traditional logistics issues address the moving of physical items along the value chain so that a product is available at a retailer when a consumer wants to purchase it. Information Markets The ability of the Internet to deliver a good, and not just create a transaction that requires fulfillment via some other channel, may be one of the key factors to enhance productivity increases across major sectors. Information goods have unique properties including marginal reproduction costs that are close to, if not exactly, zero. Therefore, the pricing strategies must change as well to reflect the new economics. For instance, some of the financial information freely made available on the Web today by online brokerage companies was sold through proprietary networks for hundreds of dollars per month just a few years ago. Software, another information good, is also enjoying a new economic model of ‘open source’ where the source code that comprises the good is made freely available to use and improve on the design. Information goods may be most affected by integrating consumers in the value chain. Instead of an information product being created ex ante for consumers to purchase, information products can be dynamically rendered based upon the wishes of the consumer. Not only will this enable (Internet) retailers to price discriminate, it can also help change the number of product offerings to a very large set of products. While there may be mass customization of physical products once the consumer is included in the value chain, information products can be customized to individual consumers at almost no additional costs. Digital information goods also raise interesting pricing opportunities. Traditional rules of thumb such as ‘price equals marginal cost’ or using a standard mark-up over cost are not very useful in this environment. Instead value-oriented strategies are likely to be more effective (Shapiro and Varian, 1999). At the same time, the special characteristics of digital goods combined with the Internet open up new opportunities including disaggregation of previously aggregated content such as newspaper or journal articles and/or massive aggregation items, such those sold by America Online (Bakos and Brynjolfsson, 1999). CONCLUSION Traditional economics views economic growth as a result of input accumulation

Internet, Innovation

Internet Economics: Models, Mechanisms and Management 187

and technical progress in a world of roughly constant returns to scale. While there is some debate about how to measure inputs and how to define technical progress, there is a consensus that much of economic growth involves tradeoffs, for example, increasing capital means investment, savings, and foregone consumption, while increasing labor input requires education expenditure and fore- gone leisure. Any unexplained growth is labelled the contribution of total factor productivity (TFP), also known as the Solow residual, which reflects technical progress, spillovers, improved efficiency, scale economies, etc. The network or new economy view takes the idea of TFP very seriously and argues that important sectors benefit from increasing returns, externalities, standards, and network economies. In a network economy, more creates more. For example, the value of new IT products like an internet connection, digital interaction, or a software program increases when others invest and obtain campatible equipment (Tapscott, 2015). Thus investment of firm A improves the productivity and value of firm B’s investment. This type of production spillover allows for an on-going growth that can quickly outpace traditional explanations. Related ideas include non-linear dynamics growth after a critical mass is reached, virtuous circles of positive feedback in fast-developing industrial sectors, and falling prices and increased quality via technology and scale economies. These ideas are not entirely new to conventional economics but in network economies they gain greater importance as the driving force of economy-wide growth. Proponents point to commonplace observations about hardware and software use as evidence of increasing return and network effects. Since modern networks are typically high-tech, for example, communications systems like the Internet show many characteristics of increasing returns, the IT sector is seen as the vital force in the new economy. Globalization also plays a role by expanding the scope of markets and allowing critical sectors to grow and achieve the necessary size of scale effects. Finally, the very nature of the new digitized economy is seen as inherently different from the old industrial economy due to obvious physical production and pricing differences between information and physical products or commodities. REFERENCES Arthur, W.B. (2011). The Second Economy McKinsey Quarterly. Bakos, Y., Brynjolfsson, E. (1999). Bundling Information Goods, Pricing, Profits and Efficiency. Manage. Sci., 45(12), 1613-1630. [http://dx.doi.org/10.1287/mnsc.45.12.1613] Brynjolfsson, E., Hitt, L. (2000). Beyond Computation: Information Technology, Organizational Transformation and Business Performance. J. Econ. Perspect., 14(94), 23-48.

188 Internet Economics: Models, Mechanisms and Management

Hans W. Gottinger

[http://dx.doi.org/10.1257/jep.14.4.23] Brynjolfsson, E., Yang, S. (1996). Information Technology and Productivity; A Review of the Literature. Adv. Comput., 43, 179-214. [http://dx.doi.org/10.1016/S0065-2458(08)60644-0] Brynjolfsson, E., McAfee, A. (2011). Race against the Machine. Lexington, Ma.: Digital Frontier Press. Brynjolfsson, E., McAfee, A. (2014). The Second Machine Age, Cambridge,Ma: MIT Press Council of Economic Advisers (1999). Economic Report of the President. Washington, DC: GPO. Cardona, M., Kretschmer, T., Strobel, T. (2013). ICT and Productivity: conclusions from the empirical literature. Inf. Econ. Policy, 25, 109-125. [http://dx.doi.org/10.1016/j.infoecopol.2012.12.002] Chandler, A.D. (1990). Scale and Scope, The Dynamics of Industrial Capitalism. Cambridge, Ma.: Harvard Univ. Press. David, P.A. (1990). The Dynamo and the Computer: An Historical Perspective on the Productivity Paradox. Am. Econ. Rev., 80, 315-348. [PP]. Ferguson, Ch. (2002). The United States Broadband Problem: Analysis and Policy Recommendations Working Paper, Washington, DC: Brookings Institution. Gershenfeld, N. (1999). When Things start to Think. New York: Henry Holt. Gordon, R. (2000). Does the ‘New Economy’ Measure up to the Great Inventions in the Past?. [http://dx.doi.org/10.3386/w7833] Gordon, R. (2016). The Rise and Fall of American Growth. Princeton: Princeton Univ. Press. [http://dx.doi.org/10.1515/9781400873302] Gottinger, H.W. (2006). Innovation, Technology and Hypercompetition. London: Routledge. Gottinger, H.W., Umali, C., Floether, F. (2010). Strategic Alliances in Biotechnology and Pharmaceuticals. New York: NovaScience. Gottinger, H.W., Goosen, M.F. (2012). Strategies of Economic Growth and Catch-Up. New York: NovaScience. Gottinger, H.W. (2016). Networks, Competition, Innovation and Industrial Growth. New York: NovaScience. Griliches, Z. (1994). Productivity, R&D and the Data Constraint. Am. Econ. Rev., 84(1), 1-23. Ip, G. (2016). The Economy’s Hidden Problem: We’re Out of Big Ideas The Wall Street Journal. Jorgensen, D.W. (1990). Berndt, E., Triplett, J. (1990). Productivity and Economic Growth Fifty Years of Economic Measurement. Chicago: University of Chicago Press. Jorgensen, D.W., Stiroh, K.J. (1995). Computers and Growth. Econ. Innov. New Technol., 3, 295-316. [http://dx.doi.org/10.1080/10438599500000008] Jorgensen, D.W. (1999). Computers and Growth. Am. Econ. Rev., 89(2), 109-115. [PP]. McGuckin, R.H., Stiroh, K.J. (2001). Do Computers Make Output Harder to Measure? J. Technol. Transf., 2, 295-321. [http://dx.doi.org/10.1023/A:1011170416813] Nordhaus, W.D. (1997). Bresnahan, T.F., Gordon, R.J. (1997). Do Real-Output and Real-Wage Measures Capture Reality? The History of Light Suggests Not The Economics of New Goods, National Bureau of Economic Research, Studies in Income and Wealth 58. Chicago: University of Chicago Press. Odlyzko, A. (2001). Internet Traffic Growth: Sources and Implications Univ. of Minnesota, Minneapolis. Available at: http://www.dtc.umn.edu/odlyzko OECD. (2000). A New Economy?: The Changing Role of Innovation and Information Economy in Growth. Paris: OECD.

Internet, Innovation

Internet Economics: Models, Mechanisms and Management 189

Onliner, St.D., Sichel, D.E. (1994). Computers and Output Growth Revisited; How Big is the Puzzle? Brookings Pap. Econ. Act., 2, 273-334. [http://dx.doi.org/10.2307/2534658] Onliner, St.D., Sichel, D.E. (2000). The Resurgence of Growth in the Late 1990s: Is Information Technology the Story? J. Econ. Perspect., 14, 3-22. [http://dx.doi.org/10.1257/jep.14.4.3] Park, J., Roome, N. (2002). The Ecology of the Economy. London: Greenleaf Publishers. Romer, P. (1986). Increasing Returns and Long-Run Growth. J. Polit. Econ., 94(5), 1002-1037. [http://dx.doi.org/10.1086/261420] Shapiro, C., Varian, H.R. (1999). Information Rules, A Strategic Guide to the Information Economy. Cambridge, Ma.: Harvard Business School Press. Sichel, D.E. (1997). The Computer Revolution; An Economic Perspective, Washington. DC: The Brookings Institution. Solow, R.E. (1987). We’d better watch out New York Times Book Review. Tapscott, D. (2015). The Digital Economy. New York: McGraw Hill. Triplett, J.E. (1999). Economic Statistics, the New Economy, and the Productivity Slowdown. Bus. Econ., 34(2), 13-17. Uchitelle, L. (2000). Economic View: Productivity finally shows the impact of Computers New York Times. Wolfram, St (2002). A New Kind of Science, Champaign,Il.: Wolfram Media.

190 Internet Economics: Models, Mechanisms and Management

ABBREVIATIONS AI Artificial Intelligence AMD Algorithmic Mechanism Design AOL America Online API Application Programming Interface ASF Apache Software Foundation ATM Asynchronous Transfer Mode AWS Amazon Web Services B-ISDN Broadband-Integrated Services Digital Networks BBL Burstiness Budget Line BD Big Data BGP Border Gateway Protocol BI Business Intelligence CAM Computer Aided Manufacturing CC Cloud Computing CDN Content Delivery Network CLS Connectionless Servers CPS Cyber Physical System CPU Central Processing Unit CRM Customer Relations Management DAMD Distributed Algorithmic Mechanism Design DARPA Defense Advanced Research Projects Agency DBMS Data Base Management System DSL Digital Subscriber Line DP Dynamic Programming EDA Exploratory Data Analysis EC Electronic Commerce ECN Explicit Congestion Notification EDI Electronic Data Interchange EPC Electronic Product Code ERP Enterprise Resource Program FCC Federal Communication Commission FIFO First-in first-out

Hans W. Gottinger

Abbreviations

Internet Economics: Models, Mechanisms and Management 191

4G/5G Fourth/Fifth Generation Digital Standard FTP File Transfer Protocol FTTH Fiber-to-the-Home GPS Geographic Positioning System GPT General Purpose Technology HTML Hypertext Markup Language ICT Internet Communication Technology IoT Internet of Things IMS IP Multimedia Systems IPv4,6 Internet Protocol Version 4,6 IRI Increasing Returns Industry IRM Increasing Returns Mechanism ISN Integrated Services Network ISP Internet Service Provider ITS Intelligent Transportation System ITU International Telecommunications Union JIT Just In Time JSR Java Specification LAN Local Area Network LBS Location Based Services MAS Multi-agent Systems Mbps Mega bits per second MPC Multiparty Computation M2M Machine-to-Machine Communication NFC Near Field Communication NC Network Computing OECD Organisation of Economic Cooperation and Development PCE Personal Consumption Expenditure PDA Personal Digital Assistant PLC Programmable Logic Control PMP Paris Metro Pricing PSTN Public Switched Telephone Network QoS Quality of Service REST Representational State Transfer RFID Radio Frequency Identification

192 Internet Economics: Models, Mechanisms and Management SG Smart Grid SLA Service Level Arrangement SNP Social Network Platform TCP/IP Transmission Control Protocol/Internet Protocol TELNET Telecommunications Network TFP Total Factor Productivity UID Unique Identification Code VOIP Voice over Internet Protocol WB Wireless Broadband WEF World Economic Forum WLAN Wireless Locally Area Network WSAN Wireless Sensor and Actuator Network WSN Wireless Sensor Network WWW World Wide Web

+DQV:*RWWLQJHU

GLOSSARY

Internet Economics: Models, Mechanisms and Management 193

GLOSSARY Before launching into the main text of this book, we have found it pertinent to recall the definitions of some key concepts. Needless to say, the following list is not exhaustive. Key Terms

Definitions

Algorithmic Mechanism Design (AMD)

Algorithmic mechanism design (AMD) lies at the intersection of economic game theory and computer science. It combines ideas such as utility maximization and mechanism design from economics, rationality and Nash equilibrium from game theory, with such concepts as complexity and algorithm design from discrete mathematics and theoretical computer science.

Amazon Web Services (AWS)

Amazon Web Services (AWS) is a comprehensive, evolving cloud computing platform provided by Amazon.com. Web services are sometimes called cloud services or remote computing services. The first AWS offerings were launched in 2006 to provide online services for websites and client-side applications.

Analytics

Analytics has emerged as a catch-all term for a variety of different business intelligence (BI)- and application-related initiatives. For some, it is the process of analyzing information from a particular domain, such as website analytics. For others, it is applying the breadth of BI capabilities to a specific content area (for example, sales, service, supply chain and so on). In particular, BI vendors use the “analytics” moniker to differentiate their products from the competition. Increasingly, “analytics” is used to describe statistical and mathematical data analysis that clusters, segments, scores and predicts what scenarios are most likely to happen. Whatever the use cases, “analytics” has moved deeper into the business vernacular. Analytics has garnered a burgeoning interest from business and IT professionals looking to exploit huge mounds of internally generated and externally available data.

Asynchronous Transfer Modes (ATM)

Asynchronous transfer modes (ATM) is a transfer mode for switching and transmission that efficiently and flexibly organizes information into cells; it is asynchronous in the sense that the recurrence of cells depends on the required or instantaneous bit rate. ATM’s powerful flexibility lies in its ability to provide a high-capacity, low-latency switching fabric for all types of information, including data, video, image and voice, that is protocol, speedand distance-independent. ATM supports fixed-length cells 53 bytes in length and virtual data circuits between 45 megabits per second (Mbps) and 622 Mbps. Using statistical multiplexing, cells from many different sources are multiplexed onto a single physical circuit.

Bandwidth-Buffer

In a virtual channel a bandwidth-buffer tradeoff operates in such a way that bandwidth can be traded for buffer space and vice versa to provide the same QoS. If bandwidth is scarce, then a resource pair that uses less bandwidth and more buffer space should be used. Resource pricing is targeted to exploit this tradeoff to achieve efficient utilization of the available resources.

194 Internet Economics: Models, Mechanisms and Management

+DQV:*RWWLQJHU

&ontd.....

Key Terms

Definitions

Big Data

The term Big Data is used when the amount of data that an organization has to manage reaches a critical volume that requires new technological approaches in terms of storage, processing, and usage. Volume, velocity, and variety are usually the three criteria used to qualify a database as “Big Data”.

Border Gateway Protocol (BGP)

Border Gateway Protocol (BGP) is protocol that manages how packets are routed across the internet through the exchange of routing and reachability information between edge routers. BGP directs packets between autonomous systems (AS), networks managed by a single enterprise or service provider.

Bottom-up

In contrast, the bottom-up approach focuses on selecting network technologies and design models first. This can impose a high potential for design failures, because the network will not meet the business or applications’ requirements.

B-ISDN is both a concept and a set of services and developing standards for integrating digital transmission services in a broadband network of fiber optic Broadband Integrated and radio media. B-ISDN will encompass frame relay service for high-speed Services Digital Network data that can be sent in large bursts, the Fiber Distributed-Data Interface (Fiber (B-ISDN) Distributed-Data Interface), and the Synchronous Optical Network (Synchronous Optical Network). B-ISDN will support transmission from 2 Mbps up to much higher, but as yet unspecified, rates. Business intelligence (BI) is an umbrella term that includes the applications, Business Intelligence (BI) infrastructure and tools, and best practices that enable access to and analysis of information to improve and optimize decisions and performance.

Central Processing Unit (CPU)

The component of a computer system that controls the interpretation and execution of instructions. The CPU of a PC consists of a single microprocessor, while the CPU of a more powerful mainframe consists of multiple processing devices, and in some cases, hundreds of them. The term “processor” is often used to refer to a CPU.

Cloud Computing: National Institute of Standards and Technology (NIST)

Definition of Cloud Computing: “Cloud Compuing is a model for enabling convenient, on-demand network access to a shared pool of configurable computing resources(e.g. networks, servers, storage, applications), and services that can be rapidly provisioned and released with minimal management effort or service provider interaction”. Cloud computing: This term designates a set of processes that use computational and/or storage capacities from remote servers connected through a network, usually the Internet. This model allows access to the network on demand. Resources are shared and computational power is conFigured according to requirements.

Content Delivery Networks (CDNs)

Content delivery networks (CDNs) are a types of distributed computing infrastructure, where devices (servers or appliances) reside in multiple points of presence on multihop packet-routing networks, such as the Internet, or on private WANs. A CDN can be used to distribute rich media downloads or streams, deliver software packages and updates, and provide services such as global load balancing, Secure Sockets Layer acceleration and dynamic application acceleration via WAN optimization techniques.

GLOSSARY

Internet Economics: Models, Mechanisms and Management 195

&ontd.....

Key Terms

Definitions

Customer Relationship Management (CRM)

Customer relationship management (CRM) is a business strategy that optimizes revenue and profitability while promoting customer satisfaction and loyalty. CRM technologies enable strategy, and identify and manage customer relationships, in person or virtually. CRM software provides functionality to companies in four segments: sales, marketing, customer service and digital commerce.

Cyber-Physical System (CPS)

Cyber-physical system (CPS) is a mechanism controlled or monitored by computer-based algorithms, tightly integrated with internet and its users. In cyber physical systems, physical and software components are deeply intertwined, each operating on different spatial and temporal scales, exhibiting multiple and distinct behavioral modalities, and interacting with each other in a myriad of ways that change with context.

Cyber Security

Cyber security also known as Computer security or IT security, is the protection of computer systems from the theft or damage to the hardware, software or the information on them, as well as from disruption or misdirection of the services they provide.

Data

This term comprises facts, observations, and raw information. Data itself has little meaning if it is not processed.

Data Analysis

This is a class of statistical methods that makes it possible to process a very large volume of data and identify the most interesting aspects of its structure. Some methods help to extract relations between different sets of data, and thus, draw statistical information that makes it possible describe the most important information contained in the data in the most succinct manner possible. Other techniques make it possible to group data in order to identify its common denominators clearly, and thereby understand them better.

Data Mining

This practice consists of extracting information from data as the objective of drawing knowledge from large quantities of data through automatic or semiautomatic methods. Data mining uses algorithms drawn from disciplines as diverse as statistics, artificial intelligence, and computer science in order to develop models from data; that is, in order to find interesting structures or recurrent themes according to criteria determined beforehand, and to extract the largest possible amount of knowledge useful to companies. It groups together all technologies capable of analyzing database information in order to find useful information and possible significant and useful relationships within the data.

Data Science

It is a new discipline that combines elements of mathematics, statistics, computer science, and data visualization. The objective is to extract information from data sources. In this sense, data science is devoted to database exploration and analysis. This discipline has recently received much attention due to the growing interest in Big Data.

Information

It consists of interpreted data, and has discernible meaning. It is lies in descriptions and answers questions like “Who?” “What?”, “When?”, and “How many?”

196 Internet Economics: Models, Mechanisms and Management

+DQV:*RWWLQJHU

&ontd.....

Key Terms

Definitions

Dial-up Modems

Dial-up modems is a connection that is established using a modem. To make the dial-up connection the modem must be connected to an active phone line that is not in use. When connecting the modem will pick up the phone and dial a number that is attached to another computer. After the connection has been made the computer can check e-mail, browse the Internet, and share files.

Digital Subscriber Lines (DSL)

Digital subscriber lines (DSL) is a family of technology for bringing highbandwidth information to homes and small businesses over ordinary copper telephone lines. DSL is a communications medium used to transfer digital signals over standard telephone lines. Along with cable Internet, DSL is one of the most popular ways ISPs provide broadband Internet access.

Distributed Algorithmic Mechanism Design (DAMD)

Distributed algorithmic mechanism design (DAMD) is an extension of algorithmic mechanism design. DAMD differs from Algorithmic mechanism design since the algorithm is computed in a distributed manner rather than by a central authority. This greatly improves computation time since the burden is shared by all agents within a network.

Economies of Scale

In microeconomics, economies of scale are the cost advantages that enterprises obtain due to size, output, or scale of operation, with cost per unit of output generally decreasing with increasing scale as fixed costs are spread out over more units of output.

Electronic Product Code (EPC) is designed as a universal identifier that Electronic Product Code provides a unique identity for every physical object anywhere in the world, for (EPC) all time. Explicit Congestion Notification (ECN)

is an extension to the Internet Protocol and to the Transmission Control Protocol and is defined in RFC 3168.

Exploratory Data Analysis (EDA)

In statistics, EDA is an approach to analyzing data sets to summarize their main characteristics, often with visual methods.

Federal Communication Commission (FCC)

Federal Communication Commission (FCC) the communications regulator in the United States of America.

Fiber to the Home (FTTH)

Fiber to the home (FTTH) also called “fiber to the premises” (FTTP), is the installation and use of optical fiber from a central point directly to individual buildings such as residences, apartment buildings and businesses to provide unprecedented high-speed Internet access. FTTH dramatically increases the connection speeds available to computer users compared with technologies now used in most places.

File Transfer Protocol (FTP)

A Transmission Control Protocol/Internet Protocol (TCP/IP) standard used to log onto a network, list directories and copy files. That is, it provides authentication of the user and lets users transfer files, list directories, delete and rename files on the foreign host, and perform wild-card transfers.

FIFO and LIFO accounting are methods used in managing inventory and financial matters involving the amount of money a company has tied up within inventory of produced goods, raw materials, parts, components, or feed stocks. First-in First-out (FIFO) They are used to manage assumptions of cost sheet related to inventory, stock repurchases (if purchased at different prices), and various other accounting purposes.

GLOSSARY

Internet Economics: Models, Mechanisms and Management 197

&ontd.....

Key Terms

Definitions

Flat-Rate Pricing

A flat fee, also referred as a linear rate, refers to a pricing structure that charges a single fixed fee for a service, regardless of usage. Less commonly, the term may refer to a rate that does not vary with usage or time of use.

Garbage in, Garbage Out (GIGO)

Garbage in, garbage out (GIGO) in the field of computer science or information and communications technology refers to the fact that computers, since they operate by logical processes, will unquestioningly process unintended, even nonsensical, input data (“garbage in”) and produce undesired, often nonsensical, output (“garbage out”). The principle applies to other fields as well.

Hadoop

Big Data software infrastructure that includes a storage system and a distributed processing tool.

Hypertext Markup Language (HTML)

A document-formatting language derived from the Standard Generalized Markup Language (SGML), predominately used to create Web pages. The user’s browser interprets HTML commands and formats the page layout, fonts and graphics on the screen. One of the more powerful features of HTML is its ability to create hyperlinks that enable the user to navigate between documents and files with a single click. HTTP is also sometimes used for messaging attachments as a way of supporting rich text formatting across product boundaries. HTML that is generated by a program or service is considered “dynamically generated HTML,” which has been confused with the Netscape and Microsoft technology called dynamic HTML (DHTML). These technologies offer client-side mechanisms for enhancing the capabilities of the Web browser and HTML documents. Dynamically generated HTML might contain DHTML, but they are not the same thing.

Industry 4.0

Industry 4.0 is the current trend of automation and data exchange in manufacturing technologies. It includes cyber-physical systems, the Internet of things and cloud computing.

Integrated Services Digital Network (ISDN)

Integrated Services Digital Network (ISDN) a technical standard and design philosophy for digital networks. ISDN provides high-speed, high-bandwidth channels to every subscribers on the public switched telephone network, achieving end-to-end digital functions with standard equipment interface devices. ISDN networks enable a variety of mixed digital transmission services to be accommodated at a single interface.

Intelligent Transportation Systems (ITS)

Intelligent transportation systems (ITS) aim to provide innovative services relating to different modes of transport and traffic management and enable various users to be better informed and make safer, more coordinated, and 'smarter' use of transport networks.

Internet of Thing (IoT)

Internet of thing (IoT) according to Gartner group IoT is the network of physical objects that contain embedded technology to communicate and sense or interact with their internal states or the external environment.

Internet Protocol Version 4 (IPv4)

Internet Protocol version 4 (IPv4) is the fourth version of the Internet Protocol (IP). It is one of the core protocols of standards-based internetworking methods in the Internet, and was the first version deployed for production in the ARPANET in 1983. It still routes most Internet traffic today, despite the ongoing deployment of a successor protocol, IPv6.

198 Internet Economics: Models, Mechanisms and Management

+DQV:*RWWLQJHU

&ontd.....

Key Terms

Definitions

Internet Protocol Version 6 (IPv6)

Internet Protocol version 6 (IPv6) is the next version of Internet Protocol (IP), designed to overcome several key limitations of IP version 4 (IPv4), the most widely used networking protocol. The main benefits of IPv6 are vastly increased address space, integrated security and quality-of-service mechanisms, as well as support for autoconfiguration and mobility. In addition, large network operators may see better routing stability as platforms mature.

Internet Service Providers (ISPs)

A company that provides Internet access to its customers. The majority of ISPs are too small to purchase access directly from the network access point (NAP), and instead buy pieces of bandwidth that are available from larger ISPs. Access to the Internet can be provided either via modem or by direct connection, which offers far higher speeds. Internet service providers are different from online services, although these services sometimes also provide access to the Internet. Online services provide access to exclusive content, databases and online discussion forums that are not available outside the service.

Knowledge

It is a type of know-how that makes it possible to transform information into instructions. Knowledge can either be obtained through transmission from those who possess it, or by extraction from experience.

Local Area Networks (LANs)

Local area networks (LANs) is a computer network that interconnects computers within a limited area such as a residence, school, laboratory, university campus or office building and has its network equipment and interconnects locally managed.

Location-Based Services (LBS)

Services based on the location of a mobile user as determined by using network and/or mobile-device-based technology. Technologies supporting this include cell of origin (also known as cell ID), AOA, time of arrival (TOA), EOTD and GPS or assisted GPS. GPS can be used without network modification but requires mobile devices to support GPS. In WLAN systems, location can be determined by triangulation between several access points. Location data can be used for a variety of services to mobile-device users, including advertisements, billing, information, tracking and safety. See also e911 and GPS.

Machine-to-Machine (M2M)

Machine-to-machine (M2M) communications is used for automated data transmission and measurement between mechanical or electronic devices. The key components of an M2M system are: Field-deployed wireless devices with embedded sensors or RFID-Wireless communication networks with complementary wireline access includes, but is not limited to cellular communication, Wi-Fi, ZigBee, WiMAX, wireless LAN (WLAN), generic DSL (xDSL) and fiber to the x (FTTx).

MapReduce

MapReduce is a programming model or algorithm for the processing of data using a parallel programming implementation and was originally used for academic purposes associated with parallel programming techniques.

GLOSSARY

Internet Economics: Models, Mechanisms and Management 199

&ontd.....

Key Terms

Definitions

MD aims to transfer privately known preferences of the relevant population to an aggregate of social choice that accordingly implement resource allocation Mechanism Design (MD) processes. Algorithmic MD combines concepts of utility maximization and mechanism design from economics, rationality and game theory with such concepts as complexity and algorithm design of computer science

Mergers and Acquisitions (M&A)

Mergers and acquisitions (M&A) is a general term that refers to the consolidation of companies or assets. While there are several types of transactions classified under the notion of M&A, a merger means a combination of two companies to form a new company, while an acquisition is the purchase of one company by another in which no new company is formed.

Multi-Agent System (MAS)

Multi-agent system (MAS) is a computerized system composed of multiple interacting intelligent agents within an environment. Multi-agent systems can be used to solve problems that are difficult or impossible for an individual agent or a monolithic system to solve. Intelligence may include some methodic, functional, procedural approach, algorithmic search or reinforcement learning. Although there is considerable overlap, a multi-agent system is not always the same as an agent-based model (ABM). The goal of an ABM is to search for explanatory insight into the collective behavior of agents (which don't necessarily need to be “intelligent”) obeying simple rules, typically in natural systems, rather than in solving specific practical or engineering problems.

Multimedia Subsystems (IMS)

Next-generation application delivery architecture. In the IMS architecture, applications can be created, controlled and changed, regardless of the kind of network or platform on which they run. IMS promises to bring flexibility, operational effectiveness, openness and standardization to the delivery of applications across fixed and mobile networks.

Multiobjective Utility Maximization (MUM)

Simultaneous optimization of conflicting criteria often under constraints thus permitting tradeoffs.

Near Field Communication (NFC)

Near Field Communication (NFC) is a wireless technology that enables a variety of contactless and proximity-based applications, such as payments, information retrieval, mobile marketing and device pairing. It has an operating range of 10 cm or less using the 13.56MHz frequency band.

Network Nodes

In a communications network, a network node is a connection point that can receive, create, store or send data along distributed network routes. Each network node - whether it's an endpoint for data transmissions or a redistribution point - has either a programmed or engineered capability to recognize, process and forward transmissions to other network nodes.

Open Data

This term refers to the principle according to which public data (that gathered, maintained, and used by government bodies) should be made available to be accessed and reused by citizens and companies.

Open Innovation

It is defined as increased use of information and knowledge sources external to the company, as well as the multiplication of marketing channels for intangible assets with the purpose of accelerating innovation.

200 Internet Economics: Models, Mechanisms and Management

+DQV:*RWWLQJHU

&ontd.....

Key Terms

Definitions

Packet Switching

Packet switching is a digital networking communications method that groups all transmitted data into suitably sized blocks, called packets, which are transmitted via a medium that may be shared by multiple simultaneous communication sessions. Packet switching increases network efficiency, robustness and enables technological convergence of many applications operating on the same network.

Personal Digital Assistant (PDA)

Data-centric handheld computer weighing less than 1 pound that is designed primarily for use with both hands. These devices use an open-market OS supported by third-party applications that can be added into the device by end users. They offer instant on/off capability and synchronization of files with a PC. A PDA may offer WAN support for voice, but these are data-first, voicesecond devices.

Programmable Logic Controls (PLCs)

The fundamental building block of factory and process automation. A specialty purpose computer, including input/output processing and serial communications, used for executing control programs, especially control logic and complex interlock sequences. PLCs can be embedded in machines or process equipment by OEMs, used stand-alone in local control environments or networked in system configurations.

Quality of Service (QoS)

A service standard of a service level arrangement (SLA) that satisfies a best level of communication service under the prevaiiling internet technology (‘best effort packet service’).

Queueing Models (QM)

QMs effectively determine the demand size in view of available supply lines in a network. Typical questions in queueing networks in view of QoS involve bottlenecks or major delays, comparing one network design with another, good set of rules for operating the network, a least-cost network satisfying given demand.

Radio Frequency Identification Tags (RFID)

Radio frequency identification tags (RFID) refers to an automated data collection technology that uses radio frequency waves to transfer data between a reader and a tag to identify, track and locate the tagged item. There are two basic categories of tags used for logistics and transportation: passive and battery-enabled. Passive tags collect the necessary energy from the antenna of the reader, which can be fixed or portable. Battery-enabled tags fall into two major groupings: battery-assisted passive (BAP) technology and active RFID tag technology.

Representational State Transfer (REST)

Is one way of providing interoperability between computer systems on the Internet. REST-compliant web services allow requesting systems to access and manipulate textual representations of web resources using a uniform and predefined set of stateless operations. Other forms of web service exist, which expose their own arbitrary sets of operations such as WSDL and SOAP.

Scalability

Scalability the measure of a system’s ability to increase or decrease in performance and cost in response to changes in application and system processing demands. Enterprises that are growing rapidly should pay special attention to scalability when evaluating hardware and software.

GLOSSARY

Internet Economics: Models, Mechanisms and Management 201

&ontd.....

Key Terms

Definitions

Service Discipline (SD)

In a network the SD must transfer traffic at a given bandwidth by scheduling the cells (fixed size packets in an ATM network) and make sure that it does not exceed the buffer space reserved for each channel. SD must support the provision of quality of service guarantees.

Service Level Arrangements (SLAs)

Service Level Arrangements (SLAs) is defined as an official commitment that prevails between a service provider and the customer. Particular aspects of the service – quality, availability, responsibilities – are agreed between the service provider and the service user. The most common component of SLA is that the services should be provided to the customer as agreed upon in the contract.

Smart Data

The flood of data encountered by ordinary users and economic actors will bring about changes in behavior, as well as the development of new services and value creation. This data must be processed and developed in order to become “Smart Data”. Smart Data is the result of analysis and interpretation of raw data, which makes it possible to effectively draw value from it. It is, therefore, important to know how to work with the existing data in order to create value.

Statistical Inference

Statistical inference is the process of deducing properties of an underlying distribution by analysis of data. Inferential statistical analysis infers properties about a population: this includes testing hypotheses and deriving estimates. The population is assumed to be larger than the observed data set; in other words, the observed data is assumed to be sampled from a larger population.

Top-Down Approach

The top-down design approach simplifies the design process by splitting the design tasks to make it more focused on the design scope and performed in a more controlled manner, which can ultimately help network designers to view network design solutions from a business-driven approach.

Transmission Control Protocol/Internet Protocol (TCP/IP)

A set of protocols covering (approximately) the network and transport layers of the seven-layer Open Systems Interconnection (OSI) network model. TCP/IP was developed during a 15-year period under the auspices of the U.S. Department of Defense. It has achieved de facto standard status, particularly as higher-level layers over Ethernet.

Voice over Internet Protocol (VoIP)

Voice over Internet Protocol (VoIP) is a methodology and group of technologies for the delivery of voice communications and multimedia sessions over Internet Protocol (IP) networks, such as the Internet. VoIP represents a form of communication that allows making phone calls over a broadband internet connection instead of typical analog telephone lines. Basic VoIP access usually allows calling others who are also receiving calls over the internet. Interconnected VoIP services also allow making and receiving calls to and from traditional landline numbers, usually for a service fee.

Web 1.0

This term refers to the part of the Internet that makes it possible to access sites composed of web pages connected by hyperlinks. This Web was created at the beginning of the 1990s. It creates a relationship between an edited site that publishes content or services and Internet users who visit it and who surf from site to site.

202 Internet Economics: Models, Mechanisms and Management

+DQV:*RWWLQJHU

&ontd.....

Key Terms

Definitions

Web 2.0

This term designates the set of techniques, functions, and uses of the World Wide Web that have followed the original format of the Web. It concerns, in particular, interfaces that allow users with little technical training to appropriate new Web functions. Internet users can contribute to information exchanges and interact (share, exchange, etc.) in a simple manner.

Web 3.0

(Also known as the Semantic Web). This is a network that allows machines to understand semantics, which is to say the meaning of information published online. It expands the network of Web pages understandable by humans by adding metadata that is understandable by a machine and that creates links between content and different pages, which in turns allows automatic agents to access the Web in a more intelligent manner and to carry out some tasks in the place of users.

Web of Things (WoT)

Web of Things (WoT) is a term used to describe approaches, software architectural styles and programming patterns that allow real-world objects to be part of the World Wide Web.

Wireless Broadband (WB)

Wireless broadband (WB) is technology that provides high-speed wireless Internet access or computer networking access over a wide area.

Wireless Local-Area Network (WLAN)

Wireless Local-Area Network (WLAN) is a LAN communication technology in which radio, microwave or infrared links take the place of physical cables. The 802.11 family of standards issued by the IEEE provides various specifications covering transmission speeds from 1 Mbps to 54 Mbps. The four main physical-layer standards are 802.11a, 802.11b, 802.11g and 802.11n.

Wireless Sensor Networks (WSNs)

Wireless sensor networks (WSNs) is a group of specialized transducers with a communications infrastructure for monitoring and recording conditions at diverse locations. Commonly monitored parameters are temperature, humidity, pressure, wind direction and speed, illumination intensity, vibration intensity, sound intensity, power-line voltage, chemical concentrations, pollutant levels and vital body functions.

World Wide Web(WWW)

World Wide Web(WWW) is a hypertext-based global information system that was originally developed at the European Laboratory for Particle Physics in Geneva. It is a subset of the Internet, technically defined as the community on the Internet where all documents and resources are formatted using Hypertext Markup Language (HTML). HTML, and the related Hypertext Transport Protocol (HTTP), make it easy to find and view data and documents stored on computers connected to the Internet. HTML creates the links (“hyperlinks”) that enable the user to move among many Web documents with the click of a mouse.

Internet Economics: Models, Mechanisms and Management, 2017, 203-205

203

SUBJECT INDEX

A ABB 111, 137 Admission control 13, 14, 17, 34, 35, 49, 56, 62, 66, 72, 98 Amazon ix, 123, 155, 190, 193 AMD 21, 78, 79, 95, 183, 190, 193 API 119, 190 Auction 8, 9, 37, 75, 82, 84, 118 AWS 119, 120, 123, 190, 193

B Bandwagon effects 27, 54, 111 Bandwidth 1, 2, 22, 23, 25, 33, 34, 40, 58, 59, 61, 62, 67, 68, 76, 86, 93, 96, 125, 126, 172, 193, 200 Big Data 133, 160, 165, 166, 181, 190, 194, 195, 197 Blockchain 110, 124 Border Gateway Protocol 93, 95, 190, 194 Bosch 137 Bridging technologies 169 Burstiness 1, 18, 19, 56, 58, 93, 98, 190

C Circuit switching 5 Cisco Systems 125, 181 Cloud computing 77, 82, 83, 92, 93, 107, 112, 119, 128, 129, 133, 144, 148, 154, 158, 159, 165, 190, 193, 194, 197 Computational complexity 78, 131, 164 Congestion 53, 75, 84, 85, 91, 190, 196

D Data Analytics viii, 133, 144, 160, 166 Data Management Economy 75, 77, 82, 83 Decentralization 26, 31, 34, 54, 77 Distributed computing 2, 26, 194

Distributed System 23, 35, 57, 76, 78, 79, 82 Dynamic Programming 109, 112, 190

E eBay 118, 150 Equilibrium 21, 24, 25, 34, 37, 38, 49, 52, 53, 56, 58, 70, 71, 75, 82, 95, 96, 118, 193 Exploratory Data Analysis 144, 145, 190, 196

F Facebook 115, 117, 118, 120, 121, 123, 155 FCC 3, 190, 196

G Globalization 187 Google 110, 111, 132, 155, 158, 181 Growth accounting 178, 179

H Hadoop 150, 163, 197 Huawei 137

I IBM ix, 111, 129, 135, 137, 144, 148, 156, 165, 171 IMS Platform 120 Incentive compatible 7, 8, 76, 84, 94, 96 Industry 4.0 125, 126, 137, 144, 166, 197 Information goods 167, 184, 186, 187 Intel 111, 137, 181, 183 Internet of Things viii, 2, 110, 132, 147, 149, 165, 166, 191, 197 Internet Pricing 1, 3, 5, 6, 27, 83 ITU 125, 128, 132, 142, 191

Hans W. Gottinger All rights reserved-© 2017 Bentham Science Publishers

204

Internet Economics: Models, Mechanisms and Management

J JIT 184, 185, 191

K Kuka 137

Hans W. Gottinger

Price elasticities 1, 17, 18 Productivity Paradox 167, 176, 177, 179, 188

Q

Link capacity 13, 25, 39, 44, 98, 100, 102, 103

Queueing Model 21, 22, 25, 33, 97, 101 Queueing Service 22, 33 Queueing Systems 21, 23, 54, 63, 74, 97, 106, 107

M

R

L

Market based control 26, 47, 85, 106 Markovian type models 97 Mechanism Design 7, 9, 26, 53, 54, 75, 76, 78, 84, 85, 190, 193, 196, 198 Minitel 2, 167, 172 Multi-factor productivity 167 Multimedia systems 120, 191

N National Semiconductor 183 Network complexity 75, 79 Network externalities 3, 26, 83, 113, 117, 118, 122, 123, 137, 169 Network platforms viii, 109, 110, 118 Network routing 56-58

O OECD 28, 76, 178, 189, 191 Open source 145, 150, 186 Optimality conditions 42 Optimal pricing 14, 55 Oracle 181

P Packet loss 58, 66, 67, 73, 74, 93, 94, 96, 97, 99, 101-103 Packet switching 5, 10, 199 Packet tail distribution 103 Pareto optimal allocations 25, 33, 43, 46, 49, 58, 62, 72 Platform utility 109, 114, 119 Poisson arrivals 41, 63, 67, 97

RCA 184 Resource allocation 1, 8, 9, 16, 18, 19, 37, 38, 42, 47, 52, 54, 57, 58, 63, 70, 82, 86, 92, 95, 98, 102, 106, 107, 198 Response time 21, 23, 32, 35, 36, 51, 52, 57, 60, 91, 93, 94, 105, 106 Revelation of preferences 7, 93, 94 Routing algorithm 56, 58, 65, 66, 71, 72

S Scalability 32, 36, 75, 78, 80, 88, 89, 105, 138, 152, 163, 166, 200 Security 2, 6, 26, 27, 37, 81, 83, 87, 89, 93, 94, 96, 106, 125, 128, 131, 137, 140, 142, 143, 145, 151, 153, 154, 160, 166, 167, 195, 198 Server Economy 35, 50, 61, 67, 72, 75, 79 Service discipline 1, 30, 31, 200 Siemens ix, 111, 137, 159, 177 Social network platform 109, 120, 121, 192 Software ecosystem 117 Software platforms 60, 109, 110, 123 Supply chain 77, 109, 111, 117, 125, 134, 136, 160, 168, 185, 186, 193

T Tail probability 49, 51, 62, 69, 97, 101, 102 Technological entrepreneurship 168 TELNET 21, 29, 192 Tipping point 109, 110

Subject Index

Internet Economics: Models, Mechanisms and Management

Traffic classes 21, 27, 33, 34, 45, 46, 48, 50, 56, 59, 61, 63, 64, 66, 72, 103 Traffic model 18, 99 Tragedy of the Commons 8, 28 Transaction classes 51, 52, 60, 68-70 Two-sided markets 53, 84, 91, 122, 123

U Utility function 25, 37, 45, 46, 56, 74, 81, 82, 104, 113, 116

205

V Vestas 156 Vickrey auction 8, 75, 82 Virtual Reality 109, 110

W WalMart 111 Watson 129, 135 Web 2.0 125, 126, 138, 201 Web of Things 125, 133, 138, 142, 202 Web Services 75, 109, 111, 112, 123, 138, 142, 149, 190, 193, 200