High-Performance Communication Networks [2 ed.] 1558605746

Table of contents :
Front-Matter_2000_High-Performance-Communication-Networks
Copyright_2000_High-Performance-Communication-Networks
The-Morgan-Kaufmann-Series-in-Networ_2000_High-Performance-Communication-Net
Dedication_2000_High-Performance-Communication-Networks
Preface_2000_High-Performance-Communication-Networks
CHAPTER-1---Overview_2000_High-Performance-Communication-Networks
CHAPTER-2---Network-Services-and-Layered-_2000_High-Performance-Communicatio
CHAPTER-3---Packet-Switched-Networ_2000_High-Performance-Communication-Netwo
CHAPTER-4---The-Internet-and-TCP-IP-Ne_2000_High-Performance-Communication-N
CHAPTER-5---Circuit-Switched-Networ_2000_High-Performance-Communication-Netw
CHAPTER-6---Asynchronous-Transfer-M_2000_High-Performance-Communication-Netw
CHAPTER-7---Wireless-Networks_2000_High-Performance-Communication-Networks
CHAPTER-8---Control-of-Networks_2000_High-Performance-Communication-Networks
CHAPTER-9---Control-of-Networks--Mathemati_2000_High-Performance-Communicati
CHAPTER-10---Network-Economics_2000_High-Performance-Communication-Networks
CHAPTER-11---Optical-Networks_2000_High-Performance-Communication-Networks
CHAPTER-12---Switching_2000_High-Performance-Communication-Networks
CHAPTER-13---Toward-a-Global-Multimedia_2000_High-Performance-Communication-
No-title-available_2000_High-Performance-Communication-Networks
No-title-available_2000_High-Performance-Communication-Networks1

Citation preview

High-performance mill

SECOND EDITION

Jean Walrand Pravin Varaiya U n i v e r s i t y of California, Berkeley

Μ

l i

Morgan Kaufmann Publishers San Francisco, California

Sponsonng Editor Jennifer Mann Director of Production and Manufacturing Yonie Overton Production Editor Heather Collins Editorial Assistant Karyn Johnson Cover Design Ross Carron Design Tbxt Design, Composition, and Illustrations Windfall Software Copyeditor Carol Leyba Proofreader Erin Milnes Indexer Ted Laux Printer Courier Corporation Cover credit: © Lockyer, Romilly/The Image Bank Morgan Kaufmann Publishers Editorial and Sales Office 340 Pine Street, Sixth Floor San Francisco, CA 94104-3205 USA Telephone 415 / 392-2665 Facsimile 415 / 982-2665 E-mail [email protected] Web site http: //www. mkp. com © 2000 by Morgan Kaufmann Publishers All rights reserved Printed in the United States of America 03 02 01 00 5 4 3 2 1 No part of this publication may be reproduced, stored in a retrieval system, or transmitted in any form or by any means—electronic, mechanical, photocopying, recording, or otherwise—without the prior written permission of the publisher. Library of Congress Cataloging-in-Publication Data Walrand, Jean. High-performance communication networks / Jean Walrand, Pravin Varaiya. - 2 n d ed. p. cm.—(The Morgan Kaufmann series in networking) Includes bibliographical references. ISBN 1-55860-574-6 1. Computer networks. 2. Multimedia systems. 3. High performance computing. 4. Asynchronous transfer mode. 5. Wireless communications systems. I. Varaiya, P.P. (Pravin Pratap) II. Title. III. Series. TK5105.5 .W353 2000 621.382 l-dc21 99-047341 CIP /

T h e Morgan K a u f m a n n Series i n N e t w o r k i n g Series Editor, David Clark High-Performance Communication J e a n Walrand a n d Pravin Varaiya

Networks,

Computer Networks: A Systems Approach, Larry Peterson a n d Bruce Davie

2e 2e

Internetworking Multimedia J o n Crowcroft, Mark H a n d l e y , Ian W a k e m a n Understanding Networked David G. M e s s e r s c h m i t t

Applications:

A First

Course

Integrated Management of Networked Systems: Concepts, Architectures, and their Operational Application Heinz-Gerd Hegering, Sebastian Abeck, a n d B e r n h a r d N e u m a i r Virtual Pnvate Networks: D e n n i s Fowler

Making

the Right

Connection

Networked Applications: A Guide to the New Computing David G. M e s s e r s c h m i t t

Infrastructure

Modern Cable Television Technology: Video, Voice, and Data Walter Ciciora, J a m e s Farmer, a n d David Large

Communications

Switching in IP Networks: IP Switching, Tag Switching, Bruce S. Davie, Paul Doolan, a n d Yakov Rekhter

and Related

Wide Area Network Robert S. C a h n

Optimization

Design: Concepts

and Tbols for

Optical Networks: A Practical Perspective Rajiv R a m a s w a m i a n d K u m a r Sivarajan Practical Computer Network J a m e s D. McCabe Frame Relay Applications: J a m e s P. C a v a n a g h

Analysis Business

and

Design

and Technology

Case

For a list of forthcoming titles, p l e a s e visit o u r website at http://www. mkp. com/publish/mann/networking. htm

Studies

Technologies

We dedicate this b o o k

to Annie a n d Isabelle a n d Julie, a n d to Ruth

Preface

M u c h h a s c h a n g e d in t h e n e t w o r k i n g world since 1995 w h e n w e w r o t e t h e first edition. O n l i n e " a n d "Web" j o i n e d "Internet" in t h e p o p u l a r vocabulary. Cellular p h o n e s b e c a m e as c o m m o n as t h e t e l e p h o n e . T h e "fast" 56 Kbps m o d e m i n t r o d u c e d w i t h m u c h publicity in 1998 w a s quickly s u r p a s s e d b y t h e megabit-per-second access delivered at h o m e b y cable TV a n d ADSL. A n d at work, 100 Mbps E t h e r n e t c a m e to t h e d e s k t o p . T h e ongoing p r o c e s s of n e t w o r k c o n v e r g e n c e t o d a y is s e e n i n t h e multibillion-dollar acquisitions of cable TV o p e r a t o r s a n d data n e t w o r k s b y t e l e p h o n e c o m p a n i e s . Service providers a n d e q u i p m e n t m a n u f a c t u r e r s are b e g i n n i n g to c o m p e t e in t h e delivery of quality of service or QpS. T h a t c o m p e t i t i o n will s h a p e t h e future of ATM a n d IP. Advances in wireless c o m m u n i c a t i o n p r o m i s e s o o n to b r i n g "anytime, any­ w h e r e " connectivity. Within a d e c a d e optical n e t w o r k i n g will provide orders of m a g n i t u d e i n c r e a s e s in b a n d w i d t h . T h e s e a d v a n c e s will sustain t h e n e t w o r k i n g b o o m of t h e 1990s. T h e social c o n s e q u e n c e s of t h e s e d e v e l o p m e n t s are diffi­ cult to predict, b u t t h e technological t r e n d s are in place. We w r o t e t h e s e c o n d edition to explain t h e s e a n d r e l a t e d technological a d v a n c e s , s o m e u n e x p e c t e d , o t h e r s already e v i d e n t in 1995.

Audience This b o o k is a u n i q u e l y c o m p r e h e n s i v e s t u d y of t h e major c o m m u n i c a t i o n n e t w o r k s : data, t e l e p h o n e , cable TV, a n d wireless. We describe t h e t e c h n o l o g i e s t h a t h e l p create t h e s e n e t w o r k s , explain t h e protocols a n d control m e c h a n i s m s t h a t o p e r a t e t h e m , a n d analyze t h e e c o n o m i c p r i n c i p l e s t h a t r e g u l a t e t h e i r u s e

Preface

xviii

a n d evolution. We w r o t e t h e b o o k for t h e professionals a n d s t u d e n t s w h o w a n t s u c h a c o m p r e h e n s i v e view of n e t w o r k i n g . T h e professionals are t h o s e in i n d u s t r y w h o m u s t evaluate t h e i r decisions in t h e context of t h e w i d e role t h a t n e t w o r k i n g plays in t h e i r organizations. T h e y m a y b e n e t w o r k i n g e n g i n e e r s a n d c o m p u t e r scientists a n d t h e i r m a n ­ agers, corporate n e t w o r k m a n a g e r s a n d administrators, o p e r a t i o n s r e s e a r c h a n d s y s t e m e n g i n e e r s e n g a g e d in n e t w o r k design a n d o p e r a t i o n s or in u p ­ grading n e t w o r k i n g infrastructure. T h e i r decisions often will r e q u i r e a n u n d e r ­ standing of alternative technologies a n d p e r f o r m a n c e evaluation a n d a s e n s e of t h e p a c e a n d direction of i n n o v a t i o n . We believe this b o o k will h e l p gain such an understanding. College seniors a n d g r a d u a t e s t u d e n t s c o m e to n e t w o r k i n g w i t h t r a i n i n g in electric e n g i n e e r i n g , c o m p u t e r science, or o p e r a t i o n s r e s e a r c h . T h e y are at­ tracted to t h e field b e c a u s e of its c a r e e r o p p o r t u n i t i e s or b e c a u s e a familiarity w i t h n e t w o r k i n g is n o w n e c e s s a r y for t h e i r o w n specialization in c o m m u n i ­ cations or software e n g i n e e r i n g or control. T h e y m a y h a v e t a k e n at m o s t o n e or, m o r e likely, n o u n d e r g r a d u a t e course in n e t w o r k i n g . This b o o k will m e e t t h e diverse n e e d s of t h e s e s t u d e n t s a n d give t h e m a wider, m o r e sophisticated appreciation of this exciting field t h a n o t h e r b o o k s w i t h a n a r r o w v i e w of net­ working. T h e distinction b e t w e e n t h e s e two i n t e n d e d a u d i e n c e s is o n l y n o m i n a l . Tbday's professional w a s y e s t e r d a y ' s s t u d e n t , a n d t h e astonishing p a c e of tech­ nical c h a n g e will t o m o r r o w m a k e h i m or h e r a s t u d e n t again. Tb cope well w i t h t h a t p a c e r e q u i r e s a c o m p r e h e n s i v e view, a n d t h a t is w h a t w e b e l i e v e this b o o k offers.

Approach We h a v e c o n d u c t e d r e s e a r c h in n e t w o r k i n g for t w e n t y years. For t h e p a s t fifteen y e a r s w e h a v e taught a n i n t r o d u c t o r y g r a d u a t e course in n e t w o r k i n g to s t u d e n t s from electrical e n g i n e e r i n g , c o m p u t e r science, a n d o p e r a t i o n s r e s e a r c h . For m o r e t h a n t e n y e a r s w e h a v e offered short courses to professionals from t h e t e l e c o m m u n i c a t i o n s i n d u s t r y a n d to m a n a g e r s in charge of n e t w o r k i n g in t h e i r c o m p a n i e s . For t h e past five y e a r s w e h a v e h a d i n t e n s e i n t e r a c t i o n s w i t h industry, as c o n s u l t a n t a n d t e c h n i c a l adviser (PV, JW) a n d as e n t r e p r e n e u r (JW). This e x p e r i e n c e in research, teaching, a n d i n d u s t r y h a s s h a p e d o u r book. In all t h r e e contexts w e find t h e n e e d for c o m p r e h e n s i v e n e s s of coverage a n d m u l t i p l e perspectives. Most b o o k s take a n a r r o w v i e w of t h e subject a n d a p p r o a c h n e t w o r k i n g from a single perspective. Typically, it is identified w i t h t h e I n t e r n e t or ATM

Preface

xix

n e t w o r k s a n d described t h r o u g h t h e associated protocols. Or n e t w o r k s are m o d e l e d as n e t w o r k s of q u e u e s , w h o s e o p e r a t i o n is e x p l a i n e d t h r o u g h r o u t i n g algorithms a n d q u e u i n g analyses. Or n e t w o r k s are d e s c r i b e d t h r o u g h t h e i r e n a b l i n g technology: wireless c o m m u n i c a t i o n , optics, or switching. We p r e s e n t a c o m p r e h e n s i v e study, discussing n e t w o r k s as t h e n e e d arises from t h e basis of first p r i n c i p l e s from c o m m u n i c a t i o n s e n g i n e e r i n g , c o m p u t e r science, o p e r a t i o n s research, a n d e c o n o m i c s . We h a v e m i n i m i z e d t h e u s e of a d v a n c e d c o n c e p t s from t h e s e disciplines. It is o u r h o p e t h a t t h e r e a d e r c a n t h u s gain a greater a p p r e c i a t i o n of t h e s e m u l t i p l e views a n d a d e e p e r u n d e r s t a n d i n g of h o w n e t w o r k s are built, h o w t h e y are used, a n d w h o will p a y for t h e m . We discuss q u e s t i o n s of n e t w o r k p e r f o r m a n c e a n d control in a n intuitive m a n n e r and, in a s e p a r a t e chapter, w e p r e s e n t t h e rigorous m a t h e m a t i c a l a r g u m e n t .

Highlights of the Second Edition In addition to c h a n g e s a n d u p d a t e s w e h a v e m a d e t h r o u g h o u t t h e m a n u s c r i p t , w e w o u l d like to highlight four major e n h a n c e m e n t s in t h e n e w edition. First, in t h e p r e v i o u s edition t h e I n t e r n e t w a s t r e a t e d s i m p l y as a n e x a m p l e of packet-switched n e t w o r k s . T h e r e is n o w a c o m p l e t e s t u d y of t h e I n t e r n e t , including t h e T C P / I P protocol suite a n d t h e a d v a n c e s p r o p o s e d to i m p r o v e its p e r f o r m a n c e or provide quality of service. Second, wireless c o m m u n i c a t i o n , a b s e n t from t h e p r e v i o u s edition, n o w receives a n e x t e n d e d discussion. T h e growing i m p o r t a n c e of wireless tele­ p h o n e access a n d its potential for u s e in data transfer m a n d a t e d its inclusion. Third, rapid a d v a n c e s in t h e last five y e a r s in wave-division m u l t i p l e x i n g a n d wave-selective switching h a v e b r o u g h t forward t h e era of optical n e t w o r k ­ ing. T h e s e a d v a n c e s will e v e n t u a l l y c h a n g e t h e f u n d a m e n t a l s of n e t w o r k de­ sign, operations, a n d e c o n o m i c s , a n d so t h e y are d e s c r i b e d h e r e . Lastly, quality of service (QpS) is likely to b e c o m e a n i m p o r t a n t d i m e n s i o n of c o m p e t i t i o n a m o n g providers. T h e ability to o p e r a t e n e t w o r k s t h a t c a n give QoS g u a r a n t e e s is also k e y to service integration. T h e e c o n o m i c s of QoS a n d t h e m e c h a n i s m s n e e d e d to g u a r a n t e e QoS receive m u c h a t t e n t i o n in t h e n e w edition.

Contents We give a c h a p t e r b y c h a p t e r outline, p o i n t i n g out t h e c h a n g e s in t h e n e w edition. C h a p t e r 1 c o n t a i n s a brief historical a c c o u n t a n d explains t h e p r i n c i p l e s

Preface

of n e t w o r k i n g . Added are r e c e n t e s t i m a t e s of t h e size, growth, a n d t r e n d s in t h e t e l e c o m m u n i c a t i o n s i n d u s t r y . C h a p t e r 2 explains h o w n e t w o r k services are p r o d u c e d b y l a y e r e d a r c h i t e c t u r e s . A n e w section s u m m a r i z e s applications t h a t are driving n e t w o r k i n g d e m a n d . C h a p t e r 3 discusses packet-switched n e t w o r k s u s i n g t h e OSI m o d e l , a n d t h e i m p o r t a n t LAN i m p l e m e n t a t i o n s . T h e 100-Mbps E t h e r n e t , a n d t h e replace­ m e n t of E t h e r n e t h u b s b y intelligent E t h e r n e t switches t h a t c a n c r e a t e virtual local area n e t w o r k s or VLANS, h a v e reorganized e n t e r p r i s e n e t w o r k i n g . De­ scriptions of t h e s e i n n o v a t i o n s a n d gigabit E t h e r n e t s are a d d e d . A unified t r e a t m e n t of t h e I n t e r n e t a n d T C P / I P n e t w o r k s o c c u p i e s Chap­ t e r 4. Advances in I n t e r n e t t e c h n o l o g y in addressing, faster switching, i m p r o v e ­ m e n t s in t h e T C P / I P protocol suite, a n d protocol proposals t h a t s e e k b e t t e r control are discussed. Circuit-switched n e t w o r k s is t h e subject of C h a p t e r 5. SONET c o n t i n u e s to receive e m p h a s i s . T h e m o s t significant addition is t h e discussion of b r o a d b a n d access n e t w o r k s : cable TV a n d ADSL, a n d E u r o p e a n proposals a d v a n c i n g passive optical n e t w o r k s . W i d e s p r e a d d e p l o y m e n t of t h e s e t e c h n o l o g i e s will s p u r c o m m e r c i a l d e v e l o p m e n t of b r o a d b a n d services. C h a p t e r 6 u p d a t e s t h e e x p l a n a t i o n of ATM w i t h i m p o r t a n t r e c e n t work, including i n t e r n e t w o r k i n g protocols MPOA, a n d m o r e detailed specifications of PNNI r o u t i n g a n d UNI signaling. M u c h of this w o r k is focused o n m o r e efficient ATM s u p p o r t of IP. H o w ATM a n d IP will c o m p e t e a n d c o o p e r a t e to provide QpS r e m a i n s u n r e s o l v e d . Wireless access h a s exploded w o r l d w i d e over t h e last five y e a r s . Primarily u s e d for voice a n d short m e s s a g e transfers, wireless c o m m u n i c a t i o n is begin­ n i n g to b e u s e d for data. C h a p t e r 7 explains t h e characteristics of wireless links a n d t h e challenges t h e s e characteristics p o s e for n e t w o r k i n g . T h e discussion explains w h y , u n l i k e t h e c o n v e r g e n c e e x p e r i e n c e d in w i r e l i n e n e t w o r k s , wire­ less n e t w o r k i n g is f r a g m e n t e d a n d w i d e s p r e a d a d o p t i o n of wireless t e c h n o l o g y for data r e m a i n s u n c e r t a i n . C h a p t e r 8 provides a n accessible discussion, a n d C h a p t e r 9 explains t h e m a t h e m a t i c a l derivations, of n e t w o r k p e r f o r m a n c e a n d control. T h e t r e a t m e n t covers circuit-switched, packet-switched, a n d ATM n e t w o r k s . Resource allo­ cation ( b a n d w i d t h a n d priority a s s i g n m e n t ) to a c h i e v e QpS g u a r a n t e e s u s i n g w i n d o w a n d rate control algorithms are discussed t h e r e . T h e t r e a t m e n t of con­ gestion control is novel. C h a p t e r 9, d e v o t e d to e c o n o m i c s , n o w h a s a focus b a s e d o n a f o r m u l a t i o n of d e m a n d for n e t w o r k services. I m p l e m e n t a t i o n of QpS g u a r a n t e e s will r e q u i r e pricing of QpS-differentiated services—a major d e p a r t u r e from t h e c u r r e n t practice of flat-rate tariffs for n e t w o r k access. T h e r e are a n a l y s e s of data a b o u t

h o w u s e r s value service quality in t e r m s of t h e i r willingness to pay. T h e data are o b t a i n e d from a m a r k e t trial at Berkeley t h a t b e g a n in April 1998. Five y e a r s ago, wave-division m u l t i p l e x i n g (WDM) w a s l i m i t e d to labo­ ratory d e m o n s t r a t i o n s . I b d a y , b a c k b o n e optical links are b e i n g u p g r a d e d b y installing WDM e q u i p m e n t . WDM links w i t h 1 terabit p e r s e c o n d s p e e d (equal to t h e traffic carried b y t h e e n t i r e I n t e r n e t today) will b e sold n e x t year. Ad­ v a n c e s in optical r o u t i n g a n d switching in less t h a n t e n y e a r s will c u l m i n a t e in all-optical n e t w o r k s , offering o r d e r s of m a g n i t u d e h i g h e r s p e e d s w i t h a small i n c r e a s e in cost. T h i s could i n a u g u r a t e a n o t h e r r e v o l u t i o n in c o m m u n i c a t i o n s . WDM a n d optical switching are d i s c u s s e d in C h a p t e r 10. T h e t r e a t m e n t of op­ tical links in t h e first edition h a s b e e n abridged. C h a p t e r 11 u p d a t e s t h e discussion o n fast p a c k e t switching to i n c o r p o r a t e m u l t i c a s t i n g a n d s o m e r e c e n t w o r k o n fast table search. C h a p t e r 12 gives a revised version of t h e future of n e t w o r k i n g .

How To Use This Book T h i s b o o k c a n b e u s e d b y i n d u s t r y professionals, or as a text for u n d e r g r a d u a t e or g r a d u a t e s t u d e n t s . Professionals m a y s t u d y a topic as t h e y n e e d it to facili­ tate u n d e r s t a n d i n g of a p a r t i c u l a r d e v e l o p m e n t . A n i n t e r e s t i n g u n d e r g r a d u a t e course c a n b e t a u g h t a r o u n d C h a p t e r s 1 t h r o u g h 3 a n d e i t h e r C h a p t e r s 4 a n d 6, if t h e a u d i e n c e is p r i m a r i l y from c o m p u t e r science, or C h a p t e r s 5 a n d 7, if t h e s t u d e n t s are p r i m a r i l y from electrical e n g i n e e r i n g . We ourselves h a v e u s e d this m a t e r i a l in two w a y s . At Berkeley, w e h a v e t a u g h t a one-semester, 45-hour i n t r o d u c t o r y g r a d u a t e course to s t u d e n t s from electrical e n g i n e e r i n g , c o m p u t e r science, a n d o p e r a t i o n s r e s e a r c h . ( T h e course always attracts s o m e seniors.) S t u d e n t s n e e d n o prior e x p o s u r e to c o m m u ­ nication n e t w o r k s — t h e e m p h a s i s is o n descriptive b r e a d t h t h a t c o n v e y s t h e e x c i t e m e n t of t h e technological a d v a n c e s a n d t h e c h a l l e n g e s p o s e d b y speed, distance, a n d d e m a n d i n g applications. T h r e e or four t i m e s e a c h y e a r w e h a v e t a u g h t a short course to practitioners, b e t w e e n 8 a n d 20 h o u r s long. T h e a i m t h e r e is to provide a n o v e r v i e w of r e c e n t d e v e l o p m e n t s , to d e c i p h e r t r e n d s , a n d to s p e c u l a t e a b o u t o p p o r t u n i t i e s .

Support Materials O u r o w n l e c t u r e s m a k e h e a v y u s e of t h e figures in t h e book. Postscript files of t h e figures are available from t h e Web page for o u r b o o k at http://www.rnkp. com.

Preface

xxii

Each c h a p t e r of t h e b o o k e n d s in p r o b l e m s t h a t test u n d e r s t a n d i n g of t h e m a t e r i a l a n d challenge t h e r e a d e r to u s e t h a t u n d e r s t a n d i n g in situations t h a t m a y arise in practice. We will k e e p a d d i n g to t h e s e p r o b l e m s a n d post t h e m at t h e Web site. A solutions m a n u a l is also available from t h e publisher.

Acknowledgments T h i s b o o k s y n t h e s i z e s t h e different v i e w p o i n t s of n e t w o r k i n g specialists w h o k n o w m o r e a b o u t e a c h view t h a n w e do. Inevitably, errors of fact a n d j u d g m e n t a n d b a l a n c e of t r e a t m e n t h a v e c r e p t into t h e book. We w o u l d b e v e r y grateful to o u r r e a d e r s for b r i n g i n g t h o s e e r r o r s to o u r a t t e n t i o n a n d for providing u s w i t h feedback a b o u t t h e i r e x p e r i e n c e s in l e a r n i n g or t e a c h i n g from this book. We c a n b e r e a c h e d via e-mail at {voir, varaiya}@eecs.berkeley.edu. We will post corrections a n d c o m m e n t s at t h e Web site http://www.mkp.com. I n this s e c o n d edition w e h a v e i n c o r p o r a t e d c o m m e n t s from instructors w h o h a v e u s e d t h e first edition. A n d r e a Goldsmith's c h a p t e r o n wireless c o m m u n i c a t i o n s discusses a v e r y i m p o r t a n t t e c h n o l o g y t h a t w a s entirely m i s s i n g in t h e first edition. We are greatly i n d e b t e d to h e r for t h e excellent discussion of a rapidly evolving field. She c a n b e r e a c h e d at [email protected]. A draft of t h e e n t i r e m a n u s c r i p t for t h e s e c o n d edition w a s r e v i e w e d b y Vijay Bhagavath, AT&T Labs; Scott J o r d a n , N o r t h w e s t e r n University; Ivy Hsu, Nortel Networks; a n d R a m e s h Rao, UC, San Diego. A n t h o n y E p h r e m i d e s , Uni­ versity of Maryland, r e v i e w e d C h a p t e r 7; Kevin Fall, UC, Berkeley, r e v i e w e d C h a p t e r 4; Riad Hartani, Nortel Networks, r e v i e w e d C h a p t e r 6; a n d E y t a n Mediano, MIT Lincoln Laboratory, r e v i e w e d C h a p t e r 10. We are i m m e n s e l y grateful to t h e m for t h e i r criticisms as well as t h e i r suggestions for i m p r o v i n g t h e book. Most of t h o s e suggestions h a v e b e e n incorporated. O u r editor, J e n n i f e r M a n n , p r o v i d e d t h e e n c o u r a g m e n t a n d friendly coax­ ing t h a t w e n e e d e d to start w o r k o n t h e s e c o n d edition a n d to b r i n g it to c o m p l e t i o n . H e r assistant, Karyn J o h n s o n , h e l p e d w i t h logistics a n d w i t h h e r e n t h u s i a s m . Finally, o u r t h a n k s to o u r p r o d u c t i o n editor, H e a t h e r Collins, w h o s o m e h o w m a n a g e d a v e r y tight s c h e d u l e .

1

I

Overview

CHAPTER

i n f o r m a t i o n t e c h n o l o g y is c h a n g i n g t h e world e c o n o m y , society, a n d daily life. T h e c h a n g e p r o c e e d s in waves, originating w i t h t h e n e w technologies, growth, a n d r e s t r u c t u r i n g of t h e affected industries. Next c o m e c h a n g e s in o t h e r b u s i n e s s e s as t h e y absorb t h o s e technologies. N e w b u s i n e s s e s are started, a n d traditional practices are o v e r t h r o w n . Societies a n d g o v e r n m e n t s t r y to a d a p t to shifts in t h e d e m a n d for goods a n d services, i n v e s t m e n t , a n d skilled workers. Millions of u s c h a n g e o u r daily r o u t i n e s in w o r k a n d r e c r e a t i o n a n d t h e w a y w e interact w i t h others, as w e adapt, willingly or u n d e r p r e s s u r e , to t h e n e w o p p o r t u n i t i e s of t h e World Wide Web.

Telecommunications Industry According to t h e I n t e r n a t i o n a l Tele­ c o m m u n i c a t i o n U n i o n (ITU), t h e total t r a d e in t e l e c o m m u n i c a t i o n s e q u i p ­ m e n t a n d services w o r l d w i d e r e a c h e d $1 trillion in 1998. T h e t r a d e is growing at 7% a n n u a l l y , twice t h e rate of t h e world e c o n o m y . I n t e r n a t i o n a l traffic d o u b l e d b e t w e e n 1990 a n d 1996 to 70 billion t e l e p h o n e m i n u t e s . Technological a d v a n c e s in fiber optic c o m m u n i c a t i o n s a n d i n c r e a s e d com­ petition are dramatically r e d u c i n g costs of r a w b a n d w i d t h . According to British Ifelecom, e q u i p m e n t a n d installation cost p e r voice p a t h o n t h e transpacific r o u t e fell from $73,000 in 1975 to $2,000 in 1996, $200 in 1999, a n d is likely to r e a c h a m e r e $5 in 2010. T h e average charge for a t h r e e - m i n u t e peak-rate U.S.-Europe p h o n e call fell in five y e a r s from $4 to $1.50 in 1998 a n d will r e a c h 50Φ b y 2000. D e m a n d for access is insatiable. In 1996, forty-eight million n e w fixed l i n e s w e r e installed b r i n g i n g t h e global installed b a s e to 741 million. T h e

2

Overview

n u m b e r of cellular subscribers g r e w b y 50% in 1997 to r e a c h 200 million in 1998. T h e traffic m i x is changing. T h e n u m b e r of c o m p u t e r s o n t h e I n t e r n e t g r e w from 1 million in early 1993, to 5 million in 1995, 16 million in 1997, a n d over 50 million in 1999. T h e I n t e r n e t Society e s t i m a t e s t h a t t h e n u m b e r of I n t e r n e t u s e r s will r e a c h 300 million b y t h e e n d of 2000. T h a t is 5% of t h e world population. By 2047 t h e world p o p u l a t i o n will r e a c h 11 billion, a n d if 25% b e c o m e s c o n n e c t e d to t h e I n t e r n e t , t h a t is n e a r l y 3 billion p e o p l e . Data traffic in t h e t r a n s a t l a n t i c corridor is doubling e a c h y e a r a n d s u r p a s s e d voice traffic in S e p t e m b e r 1997. By 2000, it m a y a c c o u n t for 75% of all traffic. T h e c h a r a c t e r of t h e 100-year-old t e l e p h o n e i n d u s t r y will c h a n g e as it a t t e m p t s to m e e t t h e shift in d e m a n d . U n d e r p r e s s u r e s from users, suppliers, a n d a worldwide m o v e m e n t for deregulation, t e l e c o m m u n i c a t i o n s m o n o p o l i e s are e n d i n g . T h e 1997 World Tirade Organization a g r e e m e n t o n t e l e c o m m u n i c a t i o n s services, signed b y 69 c o u n t r i e s w i t h m o r e t h a n 90% of all t e l e c o m m u n i c a t i o n s trade, calls for t h e liberalization of n e a r l y all m a r k e t s b y 2000 or soon after. In Europe, t h e o p e n i n g of m o s t m a r k e t s in J a n u a r y 1998 b r o u g h t in its w a k e n e w e n t r a n t s a n d large i n v e s t m e n t s . T h i s w a s foreshadowed b y t h e 1996 T e l e c o m m u n i c a t i o n s Act in t h e U.S. T h e i m p a c t of t h e 1996 Act is still w o r k i n g itself out. AT&T, t h e world's largest long-distance carrier, set a n e w course for itself, w h e n it a c q u i r e d two large cable TV operators, TCI a n d M e d i a O n e , as p a r t of its strategy to b e a n integrated provider of TV, data, a n d p h o n e services. A g r o u p of on­ line service providers is challenging AT&T, urging Congress a n d t h e FCC to e x t e n d t h e Act to i n c l u d e "open access" to cable TV n e t w o r k s . T h e r e p o r t e d $58 billion price tag for MediaOne, a m o u n t i n g to $10,000 for e a c h of its five million c u s t o m e r s , suggests t h a t AT&T is anticipating a n a n n u a l r e v e n u e of at least $2,000 p e r customer. T h e regional t e l e p h o n e c o m p a n i e s are s e e i n g t h e i r m o n o p o l i e s e r o d e b y c o m p e t i t i v e local exchange carriers or CLECs a n d b y cable TV c o m p a n i e s . Several of t h e s e c o m p a n i e s are m e r g i n g (NYNEX/Bell Atlantic, SBC/Pacific B e l l / A m e r i t e c h ) as p a r t of a strategy to e n t e r long-distance m a r k e t s in exchange for t h e i r m o n o p o l y position in local m a r k e t s . A less noticed b u t potentially significant c h a n g e m a y b e initiated b y t h e u s e of ATM switches in place of t h e large a n d expensive time-division circuit switches u s e d in t e l e p h o n e n e t w o r k s e v e r y w h e r e . ATM switches are cheaper, c a n b e d e p l o y e d in s m a l l e r sizes, a n d are m o r e versatile since ATM s u p p o r t s a n y form of traffic. T e l e p h o n e c o m p a n i e s , b u r d e n e d b y t h e i r large installed infrastructure a n d m o n o p o l y heritage, are likely to b e further t h r e a t e n e d b y n e w e n t r a n t s t h a t offer ATM-based services initially catering to n i c h e m a r k e t s .

Overview

3 Yesterday's n i c h e t e l e c o m m u n i c a t i o n s m a r k e t m a y grow into a large m a r ­ k e t tomorrow. Mobile subscribers i n c r e a s e d from 10 million i n 1990 to 200 million i n 1998. WWW applications h a v e t r a n s f o r m e d b u s i n e s s n e t w o r k s into i n t r a n e t s — i n t e r n a l c o r p o r a t e n e t w o r k s b a s e d o n I n t e r n e t protocols. U n k n o w n in 1994, b y 1997 over half of all large c o r p o r a t i o n s h a d i m p l e m e n t e d i n t r a n e t s .

Economy and Society

Advances i n t e l e c o m m u n i c a t i o n s are felt i n t h e

rest of t h e e c o n o m y . Long skeptical a b o u t t h e productivity b e n e f i t s of informa­ tion technology, m a n y e c o n o m i s t s n o w attribute to it a productivity i n c r e a s e of 1%. For t h e $8.5 trillion U.S. e c o n o m y , this a m o u n t s to a n additional $85 billion of "free" o u t p u t e a c h year. I n v e s t m e n t in c o m p u t i n g a n d c o m m u n i c a ­ tions e q u i p m e n t h a s q u a d r u p l e d over t h e last decade, a c c o u n t i n g for 5 3 % of all b u s i n e s s s p e n d i n g o n e q u i p m e n t . T h e r e a r e similar i n c r e a s e s i n s p e n d i n g o n software, consulting, t e c h n i c a l support, a n d training. Some e c o n o m i s t s b e l i e v e e v e n this productivity g r o w t h is u n d e r s t a t e d b e c a u s e g o v e r n m e n t statistics do n o t a d e q u a t e l y define a n d m e a s u r e o u t p u t in t h e fast-growing service sector, i n c l u d i n g b a n k i n g , finance, h e a l t h care, a n d education. While "hot" I n t e r n e t e - c o m m e r c e c o m p a n i e s gain m e d i a attention, it is i n t h e m u n d a n e , b a c k office o p e r a t i o n s of invoicing, p u r c h a s i n g , a n d i n v e n t o r y control t h a t t h e i m p a c t of i n t r a n e t s a n d i n t e r n e t s is m o s t profound. Businessto-business c o m m e r c e over t h e I n t e r n e t is projected to j u m p from $48 billion in 1998 to $1.5 trillion b y 2003, according to Forrester Research, I n c . D u r i n g t h e s a m e period, c o n s u m e r sales over t h e I n t e r n e t will rise from $3.9 billion to $108 billion. Some believe t h a t t o m o r r o w ' s c o r p o r a t i o n s will b e "virtual"—defined n o t b y t h e i r location b u t b y t h e i r ability to a c q u i r e k n o w l e d g e , organize informa­ tion, a n d o r c h e s t r a t e i n d e p e n d e n t contractors a n d s u p p l i e r s worldwide. I n t h e process, b u s i n e s s e s a n d professions a r e d i s a p p e a r i n g . C u s t o m e r s a r e leaving travel a g e n t s a n d o t h e r retailing i n t e r m e d i a r i e s , preferring to m a k e t h e i r reser­ vations a n d p u r c h a s e s on-line. T h e world of w o r k is changing. More t h a n t h r e e - q u a r t e r s of w o r k e r s i n t h e U.S. today a r e "information workers." T h e r e is a global shortage of w o r k e r s i n information m a n a g e m e n t a n d t e c h n o l o g y . S p u r r e d b y h i g h labor costs, U.S. c o m p a n i e s a r e h i r i n g p r o g r a m m e r s a n d e n g i n e e r s in India, E a s t e r n Europe, a n d Russia. We m a y b e w i t n e s s i n g t h e next p h a s e i n w h i c h capital s e a r c h e s for o p p o r t u n i t i e s i n India, Israel, a n d e l s e w h e r e . At t h e s a m e t i m e t h e r e is i n c r e a s e d e m p l o y m e n t i n s e c u r i t y as c o r p o r a t i o n s shift t h e i r d e m a n d for skilled labor in t h e face of c o m p e t i t i v e p r e s s u r e to c h a n g e t h e i r o p e r a t i o n s .

4

Overview

With m o r e t h a n a million pages a d d e d e a c h day, t h e Web is n o w a n infinitely large b u l l e t i n board. Media Metrix e s t i m a t e s t h a t 61 million p e o p l e worldwide visited t h e top 50 Web sites in t h e U.S. in April 1999. T h e Web is c h a n g i n g o u r habits. A m e r i c a O n l i n e ' s 14 million u s e r s are s e n d i n g 15 million e-mails e a c h d a y a n d s p e n d i n g o n average 21 h o u r s on-line e a c h m o n t h . By c o m p a r i s o n , t h e average p e r s o n in t h e U.S. w a t c h e s TV 25 h o u r s p e r week.

Problems and Opportunities Technologies o p e n u p possibilities t h a t c a n create p r o b l e m s . T h e p r o b l e m s suggest o p p o r t u n i t i e s for further advances. T h e r e are n o w several million Web sites. Tb navigate this w o r l d w i d e b u l l e t i n board, s e a r c h e n g i n e s w e r e i n v e n t e d . T h e y crawl t h r o u g h t h e s e sites m a i n ­ taining, for e v e r y word, a list of all Web pages c o n t a i n i n g t h a t word. But this is n o t always satisfactory. A s e a r c h for " m o d e m " in Alta Vista r e t u r n e d 2 mil­ lion pages, a s e a r c h for "DSL m o d e m " r e t u r n e d 800 pages, still too large. This creates t h e o p p o r t u n i t y for s e a r c h facilities t h a t c a n b e t t e r figure o u t w h a t t h e v i e w e r is looking for. Some p e o p l e are t r o u b l e d b y t h e g r o w t h of on-line p o r n o g r a p h y a n d vio­ l e n c e . Technological solutions are b e i n g offered t h a t provide services t h a t rate Web sites a n d m e a n s to p r e v e n t access to u n d e s i r e d sites. T e l e c o m m u n i c a t i o n s e n a b l e s giant m e d i a c o m p a n i e s to m a n u f a c t u r e a n d s p r e a d a worldwide m a s s culture, erasing local b o u n d a r i e s a n d sensibilities. But t h e t e c h n o l o g y also gives o p p o r t u n i t i e s for creative resistance. T h e copying m a c h i n e a n d t h e fax m a c h i n e b r o u g h t o r d i n a r y citizens t h e m e a n s to p u b l i s h a n d p r o p a g a t e t h e i r o w n views. O v e r t h e last 10 y e a r s t e l e c o m m u n i c a t i o n s h a s vastly i n c r e a s e d t h e r e a c h a n d l o w e r e d t h e cost of d i s s e m i n a t i n g n e w s a n d ideas. Geographically isolated, l i k e - m i n d e d individuals are forming small groups to p u r s u e t h e i r c o m m o n i n t e r e s t s a n d to b u i l d n e w cultural islands outside m a s s culture. A m o v e m e n t s u c h as this w o u l d h a v e b e e n impossible a d e c a d e ago. F u r t h e r in t h e future lies t h e possibility t h a t e d u c a t i o n will b e c o m e a truly lifelong process, e n a b l i n g a m u c h fuller d e v e l o p m e n t of o u r potential. It is difficult to predict t h e l o n g - t e r m i m p a c t of t h e i n f o r m a t i o n revolution. But t h e technologies t h a t u n d e r l i e t h e revolution c a n b e u n d e r s t o o d u s i n g t h e l a n g u a g e a n d c o n c e p t s of t h e n e t w o r k engineer, t h e c o m p u t e r scientist, a n d t h e e c o n o m i s t . T h e s e technologies c o m p r i s e a d v a n c e s in c o m p u t e r s , c o m m u ­ nications, signal processing, a n d t h e i r applications in diverse d o m a i n s . T h e w a y t h e s e c o m b i n e to form a d v a n c e s in n e t w o r k i n g form t h e s u b s t a n c e of this b o o k . T h i s b o o k provides a system-level u n d e r s t a n d i n g of t h e n e t w o r k i n g tech­ nologies a n d t h e actual n e t w o r k s t h a t p r o m o t e t h e i n f o r m a t i o n revolution. You will l e a r n t h a t while t h e r e are m a n y details, only a few p r i n c i p l e s are sufficient to grasp t h e field of n e t w o r k i n g . You will t h e n k n o w w h a t q u e s t i o n s to ask in

1.1

History of Communication Networks

o r d e r to c o m p a r e different n e t w o r k s , a n d i n m a n y i n s t a n c e s y o u will k n o w h o w t h o s e q u e s t i o n s s h o u l d b e a n s w e r e d . If y o u w o r k in a n organization, y o u will h a v e t h e basis to j u d g e h o w well different n e t w o r k i n g solutions will m e e t t h e n e e d s of t h e organization, n o w a n d i n t h e future. If y o u are a s t u d e n t , y o u will b e able to critically r e a d m a n y of t h e r e c e n t r e s e a r c h c o n t r i b u t i o n s to net­ working, a n d y o u will gain t h e s e n s e of w h a t directions of r e s e a r c h are likely to b e fruitful. Advances in n e t w o r k i n g t e c h n o l o g y feed on, a n d are c o n s t r a i n e d by, t h e c u r r e n t state of n e t w o r k i n g . Within t h o s e constraints, t h e a d v a n c e s are g u i d e d b y w i d e r technological a n d e c o n o m i c forces. T h i s c h a p t e r p r e s e n t s a h i g h l y abbreviated history of t h e k e y i n n o v a t i o n s in t e l e p h o n e , data, cable TV, a n d wireless n e t w o r k s , a n d t h e p r i n c i p l e s t h a t c a n serve as a c o m p a s s for j u d g i n g w h i c h directions of technological a d v a n c e are likely to b e m o r e successful. By t h e e n d of t h e c h a p t e r y o u will b e a c q u a i n t e d w i t h t h e m a i n c o n t e n d e r s for t h e role of "network of t h e future," a n d y o u will a p p r e c i a t e t h a t n o single n e t w o r k t e c h n o l o g y will b e t h e w i n n e r : t h e future n e t w o r k will i n t e g r a t e all of t h e major n e t w o r k i n g solutions. In section 1.1 w e r e v i e w t h e history of t h e t e l e p h o n e n e t w o r k , c o m p u t e r or data n e t w o r k s , cable television or CATV, a n d wireless n e t w o r k s . In t h e past, t h e s e n e t w o r k s u s e d different t e c h n o l o g i e s t h a t w e r e well s u i t e d to t h e information services t h e y provided. In section 1.2 w e explain t h e four p r i n c i p l e s t h a t u n d e r l i e t h e forces driving t h e i n d u s t r y toward c o n v e r g e n c e a n d l e a d i n g to t h e i n t e r p e n e t r a t i o n of t h e s e networks. In section 1.3 w e discuss s o m e plausible s c e n a r i o s for n e t w o r k s of t h e future. T h e s e future n e t w o r k s will offer services r a n g i n g from t e l e p h o n e to interactive video to high-speed file transfers.

1.1

HISTORY OF COMMUNICATION NETWORKS C o m m u n i c a t i o n n e t w o r k s e n a b l e u s e r s to transfer i n f o r m a t i o n in t h e form of voice, video, electronic mail or e-mail, a n d c o m p u t e r files. Users r e q u e s t t h e c o m m u n i c a t i o n service t h e y n e e d b y m e a n s of s i m p l e p r o c e d u r e s u s i n g a t e l e p h o n e h a n d s e t or cellular p h o n e , set-top TV box, or t h r o u g h applications r u n n i n g o n a h o s t c o m p u t e r s u c h as a PC or w o r k s t a t i o n . I n t h e following sections, w e identify t h e major s t e p s in t h e evolution of c o m m u n i c a t i o n

5

Overview

6

AiL—^ GU RE

Telephone network around 1890. The transmissions are analog, and the switches are manually operated.

n e t w o r k s t h a t provide t h e services u s e r s w a n t . T h e r e w e r e m a n y o t h e r in­ novations, b u t t h e steps e m p h a s i z e d h e r e w e r e decisive.

1Λ Λ

Telephone Networks T h e k e y i n n o v a t i o n s in t e l e p h o n y are circuit switching, digitization, s e p a r a t i o n of call control from voice transfer, optical links, a n d service integration. In 1876, Alexander G r a h a m Bell i n v e n t e d a pair of t e l e p h o n e s . A r o u n d 1890, simple n e t w o r k s c o n n e c t e d t e l e p h o n e s b y m a n u a l l y o p e r a t e d switches. In s u c h a network, as s h o w n in Figure 1.1, t h e signal is analog, as indicated b y t h e letter A o n t h e links. Tb call a n o t h e r t e l e p h o n e , a c u s t o m e r first rings t h e operator a n d provides t h e p h o n e n u m b e r of t h e o t h e r party. T h e operator t h e n d e t e r m i n e s t h e line t h a t goes e i t h e r directly to t h e o t h e r p a r t y or to a n o t h e r operator along a p a t h to t h e o t h e r party. I n t h e latter case, t h e operators talk to e a c h other, decide h o w to h a n d l e t h e call, a n d t h e p r o c e d u r e of c o n s t r u c t i n g t h e p a t h to t h e o t h e r p a r t y c o n t i n u e s , possibly involving o t h e r operators. Eventually, o n e operator rings t h e destination, and, if t h e t e l e p h o n e is picked up, t h e two parties are c o n n e c t e d . T h e parties r e m a i n c o n n e c t e d for t h e d u r a t i o n of t h e conversation a n d are d i s c o n n e c t e d b y t h e operators at t h e e n d of t h e conversation. Note h o w t h e t r a n s m i s s i o n lines are allocated to t h e p h o n e conversation. T h i s is a c c o m p l i s h e d b y circuit switching, w h e r e "circuit" refers to t h e capability of t r a n s m i t t i n g o n e t e l e p h o n e conversation along o n e link. Tb set u p a call, a set of circuits h a s to b e c o n n e c t e d , j o i n i n g t h e two t e l e p h o n e sets. By modifying t h e connections, t h e operators c a n switch t h e circuits. Circuit switching occurs at t h e b e g i n n i n g of a n e w t e l e p h o n e call. Operators w e r e later r e p l a c e d b y m e c h a n i c a l switches and, 100 y e a r s later, b y electronic switches.

1.1

History of Communication Networks

1.2 IMX^LWIJ....!..

Telephone network around 1988. The transmissions are analog (A) or digital ^ The switches are electronic and exchange control information b y using a data network called c o m m o n channel signaling (CCS).

Figure 1.2 illustrates t h e t e l e p h o n e n e t w o r k a r o u n d 1988. O n e major de­ v e l o p m e n t at this stage is t h a t t h e t r a n s m i s s i o n of t h e voice signals b e t w e e n switches is digital, as i n d i c a t e d b y t h e l e t t e r D, i n s t e a d of analog. A n electronic interface in t h e switch c o n v e r t s t h e analog signal traveling o n t h e link from t h e t e l e p h o n e set to t h e switch into a digital signal, called a bit stream. T h e s a m e interface c o n v e r t s t h e digital signal t h a t travels b e t w e e n t h e switches into a n analog signal before s e n d i n g it from t h e switch to t h e telephone. T h e switches t h e m s e l v e s are c o m p u t e r s , w h i c h m a k e s t h e m v e r y flexible. T h i s flexibility allows t h e t e l e p h o n e c o m p a n y to modify c o n n e c t i o n s b y send­ ing specific i n s t r u c t i o n s to t h e c o m p u t e r . Figure 1.2 also s h o w s a n o t h e r major d e v e l o p m e n t — c o m m o n channel signaling (CCS). CCS is a data c o m m u n i c a t i o n n e t w o r k t h a t t h e switches u s e to e x c h a n g e control information. T h i s "conver­ sation" b e t w e e n switches serves t h e s a m e function as t h e c o n v e r s a t i o n t h a t took place b e t w e e n operators in t h e m a n u a l n e t w o r k . T h u s CCS s e p a r a t e s t h e functions of call control from t h e transfer of voice. C o m b i n e d w i t h t h e flexi­ ble c o m p u t e r i z e d switches, this s e p a r a t i o n of function facilitates n e w services s u c h as call waiting, call forwarding, a n d call back. In c u r r e n t t e l e p h o n e n e t w o r k s , t h e bit s t r e a m s in t h e t r u n k s (lines con­ n e c t i n g switches) a n d access links (lines c o n n e c t i n g subscriber t e l e p h o n e s to t h e switch) are organized in t h e digital signal (DS) h i e r a r c h y . T h e links t h e m s e l v e s — t h e "hardware"—are called digital carrier systems. T h i n k capacity is divided into a h i e r a r c h y of logical c h a n n e l s . In N o r t h A m e r i c a t h e s e chan­ nels, listed in Tkble 1.1, are called DS-1, . . . , DS-4 a n d h a v e r a t e s r a n g i n g from 1.544 to 274.176 Mbps (megabits p e r s e c o n d ) . T h e b a s i c u n i t is set b y t h e DS-0 c h a n n e l , w h i c h carries 64 Kbps (kilobits p e r s e c o n d ) a n d a c c o m m o d a t e s o n e voice circuit. Larger-capacity c h a n n e l s m u l t i p l e x several voice c h a n n e l s . T h e

Overview

8

11 TABLE

No. of voice circuits

Rate in Mbps

Medium

Signal

T-l paired cable

DS-1

24

1.5

T-l C paired cable

DS-1C

48

3.1

North America

Japan

Europe

1.5

2.0

T-2 paired cable

DS-2

96

6.3

6.3

8.4

T-3 coax, radio, fiber

DS-3

672

45.0

34.0

32.0

Coax, waveguide, radio, fiber

DS-4

4032

274.0

Digital carrier systems. This is the hierarchy of digital signals that the telephone network uses. Note that the bit rate of a DS-1 signal is greater than 24 times the rate of a voice signal (64 Kbps) because of the additional framing bits required.

rates in J a p a n a n d E u r o p e are different. T h e m o s t c o m m o n c h a n n e l s are DS-1 a n d DS-3. Observe in t h e table t h a t t h e rates are n o t m u l t i p l e s of o n e a n o t h e r : t h e DS-1 signal carries 24 DS-0 c h a n n e l s , b u t its rate is m o r e t h a n 24 t i m e s 64 Kbps. T h e additional bits are u s e d to a c c o m m o d a t e DS-0 c h a n n e l s w i t h r a t e s t h a t deviate from t h e n o m i n a l 64 Kbps b e c a u s e t h e signals are g e n e r a t e d u s i n g clocks t h a t are n o t perfectly s y n c h r o n i z e d . Since t h e 1980s t h e t r a n s m i s s i o n links of t h e t e l e p h o n e n e t w o r k h a v e b e e n c h a n g i n g to t h e SONET, or S y n c h r o n o u s Optical Network, standard. SONET rates are a r r a n g e d in t h e STS ( S y n c h r o n o u s Transfer Signal) h i e r a r c h y s h o w n in T&ble 1.2. In N o r t h A m e r i c a a n d J a p a n t h e basic SONET signal, STS-1, h a s a rate of 51.840 Mbps. (In E u r o p e t h e basic signal is STS-3 a n d h a s a rate of 155.52 Mbps. T h e h i e r a r c h y is called S y n c h r o n o u s Digital Hierarchy, or SDH.) T h e m o s t c o m m o n links in t h e b a c k b o n e today are OC-3, OC-12, a n d OC-48. OC-192 links are n o w c o m i n g into service. Wave-division m u l t i p l e x i n g n o w e n a b l e s a single optical fiber to c a r r y 100 OC-192 links for a n aggregate rate of 1 terabit p e r second. Two differences are i m m e d i a t e l y a p p a r e n t w h e n c o m p a r i n g t h e STS a n d DS hierarchies. First, SONET signals h a v e m u c h h i g h e r bit rates, t h a n k s to t h e m u c h h i g h e r rates t h a t optical links c a n s u p p o r t c o m p a r e d w i t h t h e c o p p e r links of t h e c u r r e n t n e t w o r k . Second, t h e STS-n rate is exactly η t i m e s t h e STS-1 rate. Because all clocks in a SONET n e t w o r k are s y n c h r o n i z e d to t h e s a m e m a s t e r clock, it is possible to c o m p o s e a n STS-n signal b y m u l t i p l e x i n g

1.1

History of Communication Networks

1.2 riimiiiiiHii

iiiii

9

Carrier

Signal

OC-1

STS-1

51.840

OC-3

STS-3

155.520

OC-9

STS-9

466.560

OC-12

STS-12

622.080

Rate in Mbps

OC-18

STS-18

933.120

OC-24

STS-24

1244.160

OC-36

STS-36

1866.240

OC-48

STS-48

2488.320

OC-192

STS-192

9853.280

SONET rates. The rates of multiplexed STS-1 signals are exact multiples; no additional framing bits are used. exactly η STS-1 signals. As a result, m u l t i p l e x i n g a n d d e m u l t i p l e x i n g e q u i p m e n t for STS signals is less c o m p l e x t h a n for DS signals. T h e last major i n n o v a t i o n in t e l e p h o n y is t h e i n t e g r a t i o n of voice a n d data signals t h r o u g h t h e i n t r o d u c t i o n of t h e Integrated Services Digital Network (ISDN), illustrated in Figure 1.3. T h e ISDN b a s i c access offered to a c u s t o m e r consists of two Β c h a n n e l s a n d o n e D c h a n n e l (both Β a n d D c h a n n e l s are dig­ ital). Each Β c h a n n e l is a bidirectional, or full-duplex, c h a n n e l at 64 Kbps. O n e Β c h a n n e l c a n c a r r y e i t h e r a circuit-switched c o n n e c t i o n , a packet-switched

1.3 FIGURE

Integrated Services Digital Network (ISDN). The basic access provides two bidirectional 64-Kbps links and one 16-Kbps link. These links can be used to transmit voice or data.

Overview

t r a n s m i s s i o n service (described below), or a p e r m a n e n t digital c o n n e c t i o n . T h e D c h a n n e l carries a 16-Kbps packet-switched service. ISDN m a k e s avail­ able to subscribers t h e digital t r a n s m i s s i o n facilities t h a t w e r e p r e v i o u s l y u s e d b e t w e e n t h e switches of t h e n e t w o r k , t h u s e x t e n d i n g t h e digital t r a n s m i s s i o n all t h e w a y to t h e users. Applications of t h e ISDN services i n c l u d e c o m p u t e r com­ m u n i c a t i o n , high-speed facsimiles, r e m o t e m o n i t o r i n g of buildings, videotex, a n d low bit rate v i d e o p h o n e s . With ISDN, t h e t e l e p h o n e s y s t e m is t r a n s f o r m e d into a n e t w o r k t h a t c a n transfer i n f o r m a t i o n in m a n y forms, if at m o d e s t s p e e d s . A n e w t e c h n o l o g y for t r a n s m i s s i o n of data over u n t w i s t e d or twisted pair cables for distances u p to 4,000 m will displace ISDN. T h e t e c h n o l o g y u s e s existing t e l e p h o n e subscriber l i n e s in a m a n n e r similar to ISDN. A l t h o u g h t h e t e l e p h o n e voice c h a n n e l is l i m i t e d to a b a n d w i d t h of 3 kHz, t h e twisted pair cable itself w h i c h c o n n e c t s to t h e central office h a s a b a n d w i d t h of m o r e t h a n 1 MHz, l i m i t e d b y signal a t t e n u a t i o n a n d noise. A s y m m e t r i c Digital Subscriber Line (ADSL) service n o w offered b y t h e t e l e p h o n e c o m p a n i e s c a n provide u p to 1.5 Mbps or m o r e d o w n s t r e a m (to t h e h o m e ) a n d u p to 1.5 Mbps u p s t r e a m (from t h e h o m e ) , in addition to r e g u l a r analog t e l e p h o n e service. It is e s t i m a t e d t h a t 66% of subscriber loops in t h e U n i t e d States c a n s u p p o r t t h e ADSL t e c h n o l o g y .

1Λ 2

Computer Networks T h i s section discusses t h e following k e y i n n o v a t i o n s in c o m p u t e r or data net­ works: organization of data in packets, p a c k e t switching, t h e I n t e r n e t Protocol hierarchy, m u l t i p l e access m e t h o d s , a n d service integration. We b e g i n o u r historical sketch in 1969 w i t h t h e RS-232-C s t a n d a r d for t h e serial port of c o m p u t e r devices, illustrated in Figure 1.4. T h i s s t a n d a r d is for low bit rate t r a n s m i s s i o n s ( u p to 38 Kbps) over short distances (less t h a n 30 m ) . Serial t r a n s m i s s i o n p r o c e e d s o n e c h a r a c t e r at a t i m e . T h e c o m p u t e r devices e n c o d e e a c h c h a r a c t e r into s e v e n bits, to w h i c h t h e y c a n add a p a r i t y bit for e r r o r detection, a n d successive c h a r a c t e r s are s e p a r a t e d b y s o m e t i m e interval. W h e n t h e receiver d e t e c t s t h e b e g i n n i n g of a n e w character, it starts a clock t h a t t i m e s t h e s u b s e q u e n t bits. Both bit rate a n d distance m u s t b e k e p t small, b e c a u s e t r a n s m i s s i o n s take place over u n t w i s t e d wires, w h i c h c a n i n t r o d u c e errors d u e to cross-talk. Cross-talk b e c o m e s m o r e severe as t h e rate a n d t h e distance increase. A serial link is often u s e d to attach a c o m p u t e r to a modem. A m o d e m , or modulator-demodulator, t r a n s m i t s data b y c o n v e r t i n g bits into t o n e s t h a t c a n b e t r a n s p o r t e d b y t h e t e l e p h o n e n e t w o r k as if t h e y w e r e voice signals. T h e receiving m o d e m t h e n c o n v e r t s t h e s e t o n e s b a c k into bits, t h u s e n a b l i n g two c o m p u t e r s w i t h c o m p a t i b l e m o d e m s to c o m m u n i c a t e over t h e t e l e p h o n e

1.1

History of Communication Networks

1.4 FIGURE

The RS-232-C standard for the serial line specifies the transfer of one 8-bit character at a time, separated b y time intervals. The speed and distance of the serial line are limited.

n e t w o r k as if t h e y w e r e directly c o n n e c t e d b y a serial link. In 1999 m o s t m o d e m s r u n at a s p e e d of 28,800 b p s . M o d e m s c o n f o r m i n g to t h e n e w V.90 s t a n d a r d c a n t r a n s m i t 56,000 b p s in t h e d o w n s t r e a m direction. Figure 1.5 illustrates t h e synchronous transmission s t a n d a r d s i n t r o d u c e d in t h e 1970s to increase t h e t r a n s m i s s i o n rate a n d t h e u s a b l e l e n g t h of trans­ m i s s i o n links. T h e s e s t a n d a r d s are k n o w n as SDLC ( S y n c h r o n o u s Data Link Control). A n u m b e r of s t a n d a r d s are b a s e d o n SDLC, i n c l u d i n g HDLC (HighLevel Data Link Control), LAPB (Link Access P r o c e d u r e B), LAPD, a n d LAPM. T h e m a i n idea of SDLC is to avoid t h e t i m e w a s t e d b y RS-232-C c a u s e d b y gaps b e t w e e n successive characters. Tb e l i m i n a t e t h a t lost time, SDLC g r o u p s m a n y data bits into packets. A p a c k e t is a s e q u e n c e of bits p r e c e d e d b y a special bit p a t t e r n called t h e header a n d followed b y a n o t h e r special bit p a t t e r n called t h e trailer. T h e n u m b e r of bits in a p a c k e t m a y b e fixed or variable. T h e receiver is s y n c h r o n i z e d b y a p r e a m b l e c o n t a i n e d in t h e h e a d e r (H) of t h e p a c k e t a n d b y a self-synchronizing code t h a t c o n t a i n s t h e t i m i n g informa­ tion in addition to t h e data. Moreover, SDLC u s e s a n error-detection code called t h e cyclic redundancy check, or CRC, t h a t is m o r e efficient a n d m o r e powerful

I Η I Data

1.5 GURE

| CRC |

The Synchronous Data Link Control and related standards transmit long packets of bits. The header (H) contains the preamble that starts the receiver clock, which is kept in phase b y the self-synchronizing encoding of the bits. The receiver uses the cyclic r e d u n d a n c y check (CRC) bits to verify that the packet is correctly received.

Overview

1JL„... GURE

Store-and-forward transmissions proceed b y sending the packet successively along links from the source to the destination. The packet h e a d e r specifies the source and destination addresses (A and E, for example) of the packet. When it receives a packet, a computer checks a routing table to find out on which link it should next send the packet.

t h a n t h e single parity bit of RS-232-C. Two c o m p u t e r s , t h e n , c a n e x c h a n g e in­ formation over a t r a n s m i s s i o n link u s i n g e i t h e r RS-232-C or SDLC. But w h a t if m a n y c o m p u t e r s are to b e i n t e r c o n n e c t e d ? In t h e early 1960s, c o m m u n i c a t i o n e n g i n e e r s p r o p o s e d t h e store-and-forward packet-switching m e t h o d illustrated in Figure 1.6. This figure s h o w s c o m p u t e r s c o n n e c t e d b y point-to-point links. Tb s e n d a p a c k e t to c o m p u t e r E, c o m p u t e r A p u t s t h e source a d d r e s s A a n d t h e des­ t i n a t i o n address Ε into t h e p a c k e t h e a d e r a n d s e n d s t h e p a c k e t to c o m p u t e r B. W h e n Β gets t h e p a c k e t from A, it r e a d s t h e d e s t i n a t i o n a d d r e s s a n d deter­ m i n e s t h a t it m u s t forward t h e p a c k e t to D. W h e n D gets t h e packet, it r e a d s t h e destination address a n d forwards t h e p a c k e t to E. In this s c h e m e , w h e n a n o d e receives a packet, it m u s t first store it, t h e n forward it to a n o t h e r n o d e (if n e c e s s a r y ) . H e n c e t h e n a m e store-and-forward given to this switching m e t h o d . W h e n c o m p u t e r s u s e store-and-forward p a c k e t switching, t h e y u s e a given link only w h e n t h e y s e n d a packet. As a result, t h e s a m e links c a n b e u s e d efficiently b y a large n u m b e r of i n t e r m i t t e n t t r a n s m i s s i o n s . T h i s m e t h o d for sharing a link a m o n g t r a n s m i s s i o n s is called stanstical multiplexing. Statistical multiplexing contrasts w i t h time-division multiplexing-based circuit switching, w h i c h r e s e r v e s circuits for t h e d u r a t i o n of t h e c o n v e r s a t i o n e v e n t h o u g h t h e parties c o n n e c t e d b y t h e circuit m a y n o t t r a n s m i t c o n t i n u o u s l y .

1.1

History of Communication Networks

Starting in t h e late 1960s, t h e U.S. D e p a r t m e n t of Defense A d v a n c e d Re­ s e a r c h Projects A g e n c y (DARPA) b e g a n p r o m o t i n g t h e d e v e l o p m e n t of packetswitched n e t w o r k s . T h e resulting n e t w o r k , ARPANET, b e g a n o p e r a t i o n s in 1969 b y c o n n e c t i n g four c o m p u t e r s . T h e r u l e s of operations, or protocols, ARPANET u s e d w e r e p u b l i s h e d in t h e o p e n literature. By i m p l e m e n t i n g t h e s e protocols, e n g i n e e r s in m a n y r e s e a r c h a n d e d u c a t i o n a l institutions a t t a c h e d t h e i r com­ p u t e r s to t h e ARPANET. T h r o u g h t h e ARPANET protocols, e n g i n e e r s a g r e e d o n a single p a c k e t format s t a n d a r d a n d a c o m m o n a d d r e s s i n g s c h e m e . T h i s allowed n e t w o r k s t h a t c o n f o r m e d to this p a c k e t format to b e easily i n t e r c o n n e c t e d . T h e benefits of i n t e r c o n n e c t i v i t y soon b e c a m e obvious, a n d t h e ARPANET evolved into t h e I n t e r n e t , w h i c h today is u s e d to i n t e r c o n n e c t a large n u m b e r of c o m p u t e r s a n d local area n e t w o r k s t h r o u g h o u t t h e world. In 1983, only 500 "host c o m p u t e r s " h a d I n t e r n e t access. In 1999, t h e r e w e r e 50 million s u c h c o m p u t e r s b e i n g u s e d b y 300 million people. T h e single p a c k e t format of t h e I n t e r n e t Protocol offered t w o advantages. O n t h e o n e h a n d , t h e format could b e s u p p o r t e d b y a v a r i e t y of physical net­ works, including local area n e t w o r k s (LANs) s u c h as E t h e r n e t a n d t o k e n ring, as well as b y point-to-point links. O n t h e o t h e r h a n d , e n g i n e e r s a n d c o m p u t e r sci­ entists could d e v e l o p c o m m u n i c a t i o n s applications a s s u m i n g t h a t data w o u l d b e t r a n s p o r t e d in p a c k e t s of a s t a n d a r d i z e d format. T h e ARPANET implicitly i m p l e m e n t e d a t h r e e - l a y e r e d a r c h i t e c t u r e consisting of (1) t h e physical net­ w o r k t h a t transfers bits, (2) g r o u p s of data e n c a p s u l a t e d into p a c k e t s w i t h a c o m m o n format a n d a d d r e s s i n g s c h e m e , a n d (3) applications t h a t a s s u m e transfer of p a c k e t s w i t h n o regard to t h e u n d e r l y i n g physical n e t w o r k . T h i s implicit l a y e r e d a r c h i t e c t u r e w a s s u b s e q u e n t l y e l a b o r a t e d a n d formalized in t h e O p e n Systems I n t e r c o n n e c t i o n , or OSI, m o d e l . In t h e late 1960s a n d early 1970s, e n g i n e e r s p r o p o s e d a n e w m e t h o d for c o n n e c t i n g c o m p u t e r s . This m e t h o d is called multiple access. It d r a m a t i c a l l y r e d u c e d t h e cost of i n t e r c o n n e c t i n g n e a r b y c o m p u t e r s in a LAN as well as t h e cost of access to a w i d e area n e t w o r k (WAN). Figure 1.7 illustrates a p o p u l a r i m p l e m e n t a t i o n of m u l t i p l e access called Ethernet. In a n E t h e r n e t n e t w o r k , c o m p u t e r s are a t t a c h e d to a c o m m o n coaxial cable via a n interface t h a t today consists of a small c h i p set m o u n t e d o n t h e m a i n b o a r d . W h e n c o m p u t e r A w a n t s to s e n d a p a c k e t to c o m p u t e r E, it p u t s t h e source a d d r e s s A a n d t h e d e s t i n a t i o n a d d r e s s Ε into t h e p a c k e t h e a d e r a n d t r a n s m i t s t h e p a c k e t o n t h e cable. All t h e c o m p u t e r s r e a d t h e packet, b u t only t h e c o m p u t e r w i t h t h e d e s t i n a t i o n a d d r e s s i n d i c a t e d o n t h e p a c k e t copies it. T h e original E t h e r n e t t r a n s m i s s i o n rate w a s 10 Mbps; n o w 100-Mbps a n d 1000Mbps E t h e r n e t are available.

Overview

In t h e early 1980s, IBM d e v e l o p e d a n o t h e r m u l t i p l e access m e t h o d , called token ring, illustrated in Figure 1.8. W h e n n e t w o r k e d as a t o k e n ring, c o m p u t e r s are a t t a c h e d b y point-to-point links in a unidirectional ring configuration u s i n g t o k e n ring interface b o a r d s . W h e n t h e c o m p u t e r s h a v e n o i n f o r m a t i o n to transmit, t h e interfaces pass a token a r o u n d t h e ring. T h e interface b o a r d s b e t w e e n t h e c o m p u t e r s a n d t h e n e t w o r k are configured so t h a t t h e y p u t b a c k o n t h e ring w h a t e v e r information t h e y receive w i t h a delay e q u a l to a few bit t r a n s m i s s i o n t i m e s . T h i s e n a b l e s t h e t o k e n to circulate v e r y fast a r o u n d t h e ring. Suppose c o m p u t e r A w a n t s to s e n d a p a c k e t to c o m p u t e r E. C o m p u t e r A p u t s t h e source a d d r e s s A a n d t h e d e s t i n a t i o n a d d r e s s Ε into t h e p a c k e t h e a d e r a n d gives t h e p a c k e t to its interface, w h i c h t h e n waits for t h e t o k e n . As s o o n as it gets t h e token, t h e interface of c o m p u t e r A t r a n s m i t s its packet, i n s t e a d of forwarding t h e token. T h e o t h e r c o m p u t e r s k e e p forwarding t h e p a c k e t t h e y receive while m a k i n g a c o p y for t h e m s e l v e s . In particular, t h e interface of c o m p u t e r Ε copies t h e p a c k e t d e s t i n e d to it. T h e o t h e r interfaces discard t h e i r copy of t h e p a c k e t w h e n t h e y find out t h a t it is n o t for t h e m . W h e n A receives t h e last bit of its o w n packet, after t h e p a c k e t h a s traveled a r o u n d t h e ring, it p u t s t h e t o k e n b a c k o n t h e ring. W h a t is i m p o r t a n t is t h a t t h e c o m p u t e r s get to t r a n s m i t in t u r n , w h e n t h e y get t h e token. H a r d w a r e is available for t o k e n ring n e t w o r k s at 4 Mbps a n d at 16 Mbps. T h e m a x i m u m t i m e a c o m p u t e r waits before it gets to t r a n s m i t in a t o k e n ring or E t h e r n e t n e t w o r k is small e n o u g h for m a n y applications b u t too large for

1.1

History of Communication Networks

1.8 FIGURE

Tbken ring. The computers share a ring. Access is regulated b y a token-passing protocol.

interactive audio or video applications. Also, t h e t r a n s m i s s i o n rate of t o k e n ring (4 or 16 Mbps) or 10-Mbps E t h e r n e t n e t w o r k s is too slow for s o m e m u l t i m e d i a applications. T h e s e t w o l i m i t a t i o n s l e d e n g i n e e r s in t h e late 1980s to d e v e l o p a n e w n e t w o r k called Fiber Distributed Data Interface (FDDI), illustrated in Figure 1.9. FDDI n e t w o r k s u s e optical fibers to t r a n s m i t at 100 Mbps; access to t h e c h a n n e l is r e g u l a t e d b y a t i m e d - t o k e n m e c h a n i s m . T h i s m e c h a n i s m is similar to t h e access control of a t o k e n ring n e t w o r k except t h a t w i t h FDDI, t h e arrivals of t o k e n s are t i m e d to a s s u r e t h a t t h e y are r e t r a n s m i t t e d w i t h i n a fixed t i m e .

1.9 FIGURE

Fiber Distributed Data Interface (FDDI). A token-passing protocol is used to share the ring. The computers time their holding of the token. This network guarantees that every computer gets to transmit within an agreed-on time.

Overview

1.10 FIGURE

Asynchronous Transfer Mode (ATM) network. The network transports information in 53-byte cells. Tbtal throughput of this network is m u c h larger than that of FDDI or of a 100-Mbps Ethernet.

T h e high s p e e d of FDDI m a k e s it suitable for n e t w o r k i n g w o r k s t a t i o n s w i t h instruction rates of a few h u n d r e d Mips (millions of i n s t r u c t i o n s p e r s e c o n d ) . Because it c a n g u a r a n t e e a t o k e n rotation time, FDDI c a n offer i n t e g r a t e d services for applications t h a t c o m b i n e audio a n d video w i t h data. Faster LAN a n d WAN n e t w o r k s are today b e i n g deployed, u s i n g Asynchro­ nous Transfer Mode (ATM). With ATM, a c o m p u t e r t r a n s m i t s i n f o r m a t i o n at rates b e t w e e n 25 Mbps a n d 2.5 Gbps (gigabits p e r s e c o n d ) in p a c k e t s of 53 b y t e s (1 b y t e = 8 bits). T h e s e fixed-size packets, called cells, c a n b e switched rapidly b y ATM switches. Figure 1.10 illustrates a n ATM LAN n e t w o r k w i t h o n e switch. T h e h e a d e r c o n t a i n s a virtual circuit a d d r e s s or VCI, i n s t e a d of a source a n d destination address. With t h e a p p r o p r i a t e control software, n e t w o r k e n g i n e e r s c a n c o n n e c t m a n y ATM switches t o g e t h e r to build large n e t w o r k s . Moreover, t h e links b e ­ t w e e n ATM switches c a n b e long optical fibers. Using this technology, t h e n , c o m p a n i e s c a n build a worldwide n e t w o r k . In a n ATM n e t w o r k , data is trans­ ferred from source to d e s t i n a t i o n over a fixed route, j u s t like in a t e l e p h o n e c o n n e c t i o n . Unlike t e l e p h o n e n e t w o r k s , however, a n ATM c o n n e c t i o n is n o t allocated a fixed b a n d w i d t h . T h e ATM n e t w o r k d e t e r m i n e s h o w m u c h b a n d ­ w i d t h to allocate so t h a t information is t r a n s p o r t e d w i t h v e r y low loss rate or delay, as r e q u i r e d b y t h e application. T h u s this t e c h n o l o g y is well s u i t e d for building large i n t e g r a t e d services n e t w o r k s . Figure 1.11 s u m m a r i z e s t h e i n c r e a s e in s p e e d of data n e t w o r k s . O v e r t h e 30 y e a r s since 1970, s p e e d h a s i n c r e a s e d b y six orders of m a g n i t u d e , from 10 Kbps to 10 Gbps.

1.1

History of Communication Networks

ATM •

10G Γ1 G

ATM

FDDI

•

100 Μ 10M

•

Ethernet

1 Μ 100 Κ

Token ring •

•

•

Ethernet

•

Ethernet

SMDS • Frame Relay

RS-232-C •

10 Κ 65 1.11 FIGURE

1Λ 3

70

75

80

85

90

95

00

The speed of data networks increased b y six orders of magnitude b e t w e e n 1970 and 1999.

Cable Television Networks Cable television, originally k n o w n as Community Antenna Television or CATV, w a s i n t r o d u c e d in t h e 1940s in a r e a s t h a t could n o t receive t h e TV signal w i t h o u t obstruction. T h e solution w a s to place a n a n t e n n a o n t o p of a large utility pole a n d locally s h a r e t h e signal. CATV, as w e n o w k n o w it, w a s c r e a t e d w h e n t h e signal from o n e m a s t e r a n t e n n a w a s distributed over a large area u s i n g coaxial cable a n d amplifiers. T h e k e y i n n o v a t i o n s i n cable TV are optical feeder links, digital c o m p r e s s i o n t e c h n i q u e s , a n d service integration. Cable television s y s t e m s deliver television signals to m o r e t h a n 50% of U.S. h o u s e h o l d s a n d could s e r v e close to 90% w i t h relatively m i n o r additional infrastructure. C h a n g e s in CATV n e t w o r k s are d e p i c t e d in Figure 1.12. Tbday, CATV u s e s frequency-division m u l t i p l e x i n g to t r a n s m i t u p to 69 analog TV c h a n n e l s , e a c h 4.5-MHz wide. T r a n s m i s s i o n is over coaxial cables a r r a n g e d as a u n i d i r e c t i o n a l tree, w i t h w i d e b a n d amplifiers u s e d to c o m p e n s a t e for t h e a t t e n u a t i o n of t h e cable signal (top p a n e l in Figure 1.12). T h e n u m b e r of TV c h a n n e l s is l i m i t e d b y t h e b a n d w i d t h of coaxial cables. T h e s p a n of a CATV n e t w o r k is l i m i t e d b y t h e noise power, w h i c h i n c r e a s e s as m o r e amplifiers are a d d e d to c o m p e n s a t e for t h e signal p o w e r loss d u r i n g propagation. T h e n e x t major step in CATV is to utilize optical fibers to t r a n s m i t t h e TV signals over l o n g e r distances ( m i d d l e p a n e l in Figure 1.12). Fibers h a v e

Overview

H —

1

—

1

— L - a

Amplifier

1

< A,,.,........ FIGURE

Coax Head station

Coax Fiber

O/E

γ

Video data voice J

u

I Data^

Coax Fiber


1 is t h e gain a n d η is noise (refer to Figure 1.16). Suppose t h e p o w e r of t h e CATV signal at t h e b e g i n n i n g of t h e distribution n e t w o r k is S, a n d s u p p o s e t h e signal goes t h r o u g h a net­ w o r k t h a t is Κ sections long. At t h e e n d of this n e t w o r k t h e signal Ζ is a s u m of two t e r m s , Z = X + N, w h e r e X is t h e CATV signal p o w e r a n d Ν is t h e noise power. For p r o p e r reception, w e n e e d (1) m i n i m u m CATV signal s t r e n g t h X >XQ a n d (2) a m i n i m u m signal-to-noise ratio X/N > RQ. H O W

y = ax

1.16 ^^ΆΜ^

ζ = yy + η

A section of a CATV distribution network consists of a length of cable, which attenuates the signal, followed by an amplifier, which boosts the input signal and adds noise.

1.6

Problems

1.17 nniMiMi.

35

A finite alphabet is encoded into binary words b y associating each letter with £ binary rp-^g ^ gj fixed-length encoding; the

a

Q

a

t r e e

t r e e

o

n

6

v e s

a

tree on the right gives a variable-length encoding.

long c a n t h e n e t w o r k b e ? We c a n i n c r e a s e t h e n e t w o r k l e n g t h b y increas­ ing t h e i n p u t signal p o w e r S. We c a n also r e p l a c e t h e cable b y a n optical fiber w h o s e a is m u c h larger t h a n t h a t of cable, b u t for w h i c h t h e i n p u t signal level S is m u c h smaller. Discuss t h e trade-off b e t w e e n i n c r e a s i n g S a n d i n c r e a s i n g a. 5. Suppose s e q u e n c e s of letters from a finite a l p h a b e t are to b e t r a n s m i t t e d over a b i n a r y c o m m u n i c a t i o n c h a n n e l . A s s u m e t h a t t h e r e are 2 l e t t e r s in t h e alphabet, so t h a t e a c h letter c a n b e e n c o d e d into a n η-bit word, as in t h e left p a n e l of Figure 1.17. A n m-letter s e q u e n c e is t h u s e n c o d e d into a b i n a r y s e q u e n c e of l e n g t h mn. A variable-length e n c o d i n g is obtained, as in t h e right p a n e l , b y associating e a c h l e t t e r w i t h a leaf of t h e tree. T h u s t h e letters / a n d g are e n c o d e d into 4-bit words, h is e n c o d e d into a 2-bit word, a n d t h e rest into 3-bit words. A variable-length e n c o d i n g is p r e f e r r e d if s o m e of t h e letters are u s e d m o r e f r e q u e n t l y t h a n others. Suppose, in t h e e x a m p l e above, t h e l e t t e r ft is u s e d w i t h probability 0.5, a n d t h e r e m a i n i n g s e v e n letters are u s e d w i t h probability 1/14. Show t h a t t h e average l e n g t h of t h e b i n a r y e n c o d i n g o n t h e right is s m a l l e r t h a n t h e average l e n g t h o n t h e left. If t h e r e are Ν letters in t h e a l p h a b e t a n d t h e y o c c u r w i t h probabilities p(l),... ,p(N), t h e n t h e m i n i m a l l e n g t h e n c o d i n g r e q u i r e s Η bits p e r letter, w h e r e Η is t h e e n t r o p y : n

H = - ] £ p ( f c ) log p(fc). 2

k

Show t h a t if all letters are e q u a l l y likely, t h e m i n i m a l l e n g t h e n c o d i n g r e q u i r e s l o g Ν bits p e r letter. 2

Overview

6. A customized CD c o m p a n y m i g h t o p e r a t e as follows. All its m u s i c w o u l d b e available b y m e a n s of c e n t r a l servers. C u s t o m e r s w o u l d c o m e into a n y b r a n c h store a n d select t h e songs t h e y w a n t . Bit s t r e a m s r e p r e s e n t i n g those songs w o u l d b e t r a n s p o r t e d over t h e n e t w o r k from t h e s e r v e r s a n d w r i t t e n onto a CD. T h e c o m p a n y w o u l d save m o n e y since it w o u l d h a v e n o i n v e n t o r y . Is t h e s c h e m e feasible? As w e h a v e seen, a 70-minute CD stores a b o u t 700 MB. H o w large a c o m m u n i c a t i o n b a n d w i d t h s h o u l d b e provided b e t w e e n s e r v e r a n d store so t h a t a c u s t o m e r w o u l d n o t n e e d to wait for m o r e t h a n 10 m i n u t e s ? If a 45-Mbps link h a s a m o n t h l y r e n t of $20,000, a n d a CD sells for $20, a n d if a 100% m a r k u p is n e e d e d for o t h e r costs, is this a profitable idea? H o w m u c h s h o u l d t h e r e n t b e to m a k e t h e s c h e m e profitable? Can y o u t h i n k of o t h e r b u s i n e s s o p p o r t u n i t i e s w h e r e c o m m u n i c a t i o n substitutes for i n v e n t o r y ? 7. Sketch t h e data n e t w o r k in y o u r c a m p u s or c o m p a n y . H o w m a n y h o s t s are there, a n d h o w large is t h e u s e r p o p u l a t i o n ? W h a t is t h e s p e e d of t h e access link to t h e I n t e r n e t ? H o w do y o u gain access to t h e I n t e r n e t ? H o w m u c h does h o m e access to t h e I n t e r n e t cost? 8. Sketch h o w y o u r t e l e p h o n e is c o n n e c t e d to t h e c e n t r a l office of y o u r t e l e p h o n e c o m p a n y . H o w large is t h e switch located in t h a t c e n t r a l office? H o w c a n y o u gain access to different long-distance p h o n e c o m p a n i e s u s i n g t h e s a m e local t e l e p h o n e c o m p a n y ? 9. Sketch t h e CATV n e t w o r k in y o u r city. It is likely t h a t t h e city h a s g r a n t e d this n e t w o r k a franchise, m e a n i n g t h a t it h a s t h e exclusive right to o p e r a t e in y o u r city. W h a t a r g u m e n t s c a n y o u m u s t e r for a n d against s u c h a franchise? 10. Besides CATV a n d b r o a d c a s t TV, w h a t o t h e r m e a n s do y o u h a v e to receive video p r o g r a m s ? H o w w o u l d y o u characterize t h e differences b e t w e e n t h e s e alternative s o u r c e s of s u p p l y from t h e c u s t o m e r ' s p o i n t of view? 11. Is t h e r e a n e w s p a p e r in y o u r area t h a t y o u c a n r e a d on-line, for e x a m p l e o n t h e World Wide Web? If t h e r e is, c o m p a r e t h a t service w i t h t h e p r i n t e d version of t h e s a m e n e w s p a p e r . 12. Recent p u r c h a s e s of CATV c o m p a n i e s suggest t h a t t h e y are v a l u e d at $3,000 p e r customer, so t h a t if a CATV c o m p a n y h a s o n e million c u s t o m e r s , its m a r k e t value is a b o u t $3 billion. Tb obtain s u c h a m a r k e t value, t h e CATV c o m p a n y m u s t m a k e profits of a b o u t $300 p e r c u s t o m e r p e r year. H o w does t h e c o m p a n y m a k e s u c h profits? Look at t h e a n n u a l r e p o r t s of s o m e CATV c o m p a n i e s in y o u r library.

1.6

Problems

37

13. W h a t is t h e r e v e n u e p e r residential a n d b u s i n e s s c u s t o m e r of y o u r local phone company? 14. C o m p a r e two proposals to distribute m o v i e s . A s s u m e t h a t in b o t h s c h e m e s 100 video c h a n n e l s are available to distribute m o v i e s to 300 h o u s e h o l d s . In t h e first s c h e m e , w h i c h u s e s t h e existing CATV n e t w o r k , t h e CATV o p e r a t o r finds o u t t h e 25 m o s t p o p u l a r m o v i e s a n d plays t h e m c o n t i n u o u s l y starting e v e r y 15 m i n u t e s . In t h e s e c o n d s c h e m e , h o u s e h o l d s o r d e r t h e m o v i e s t h e y w a n t b y u s i n g a control c h a n n e l . T h i s s c h e m e r e q u i r e s costly modification of t h e existing CATV n e t w o r k to create t h e control c h a n n e l a n d additional t e r m i n a l e q u i p m e n t t h a t u s e r s m u s t e m p l o y to register t h e i r d e m a n d . H o u s e h o l d s m u s t p a y for w a t c h i n g a m o v i e . Build a m o d e l for h o u s e ­ h o l d d e m a n d t h a t c a n b e u s e d to predict w h i c h s c h e m e will m a k e t h e g r e a t e r profit. H o w w o u l d y o u go a b o u t collecting data to e s t i m a t e or vali­ date y o u r m o d e l ? W h a t is y o u r guess as to w h i c h s c h e m e is m o r e likely to b e profitable a n d w h y ? 15. O n e c o m m o n a p p r o a c h to public regulation of a m o n o p o l i s t i c t e l e p h o n e c o m p a n y is called rate of return regulation, in w h i c h a c o m p a n y ' s profit is l i m i t e d to ρ χ V, w h e r e ρ is a "fair" rate of r e t u r n (say, 15%) a n d V is t h e value of t h e c o m p a n y ' s assets. (V c a n b e m e a s u r e d as t h e d e p r e c i a t e d value of t h e c o m p a n y ' s past i n v e s t m e n t s . ) It h a s b e e n suggested t h a t rate of r e t u r n regulation e n c o u r a g e s t h e t e l e p h o n e c o m p a n i e s to invest in m o r e capital e q u i p m e n t a n d h i r e fewer w o r k e r s t h a n is o p t i m u m from society's viewpoint. Can y o u argue in favor of or in opposition to this suggestion?

Network Services and Layered Architectures

a n y n e t w o r k s provide t r a n s p o r t a t i o n services. T h e postal s y s t e m , w h o s e services i n c l u d e t h e transfer of letters a n d parcels, is a familiar example. T h e postal s y s t e m ' s services are differentiated b y quality: t h e r e is registered mail, o v e r n i g h t delivery, surface mail, third-class mail. T h e s e services are built from basic t r a n s p o r t a t i o n services, s u c h as t r u c k or rail or air t r a n s p o r t . T h e postal s y s t e m u s e s t h e s e basic services to create t h e m o r e sophisticated services t h a t its c u s t o m e r s p u r c h a s e . If y o u mail a letter, t h e s y s t e m selects a r o u t e over w h i c h to s e n d it; p u t s t h e l e t t e r w i t h o t h e r s going o n t h e s a m e r o u t e in a larger container; ships t h e c o n t a i n e r u s i n g air transport, say; transfers t h e c o n t a i n e r to a post office n e a r t h e d e s t i n a t i o n u s i n g t r u c k t r a n s p o r t ; a n d finally b r i n g s t h e letter to t h e d e s t i n a t i o n u s i n g y e t a n o t h e r service, n a m e l y h a n d delivery b y t h e letter carrier. Characteristics of t h e basic services a n d s y s t e m p e r f o r m a n c e are s u m m a ­ rized a n d m e a s u r e d b y a few p a r a m e t e r s : t h e v o l u m e of l e t t e r s t h a t c a n b e h a n d l e d p e r d a y or p e r hour, t h e s p e e d of delivery, t h e fraction of l e t t e r s t h a t are lost. Tb a considerable e x t e n t t h e s e characteristics are d e t e r m i n e d b y t h e capabilities of t h e postal s y s t e m "hardware": t h e n u m b e r of trucks, t r a i n cars, a n d a i r p l a n e s t h e postal s y s t e m h a s ; t h e i r s p e e d s ; t h e r o u t e s t h e y c a n use; a n d so on. T h e s e h a r d w a r e - d e t e r m i n e d capabilities are t h e n m a n a g e d a n d controlled b y t h e postal s y s t e m ' s "intelligence"—embodied in h a r d w a r e or soft­ w a r e s y s t e m s or postal s y s t e m e m p l o y e e s — t o p r o d u c e t h e m o r e sophisticated services offered to c u s t o m e r s . Naturally, t h e characteristics of t h e u n d e r l y i n g basic services limit t h e r a n g e a n d quality of t h e sophisticated services t h a t c a n b e provided. For example, if t h e postal service u s e d o n l y railway a n d t r u c k

Network Services and Layered Architectures

transport, it w o u l d n o t b e able to offer o v e r n i g h t delivery. Over time, postal services of different c o u n t r i e s h a v e b e e n interconnected. Lastly, t h r o u g h service integration, t h e postal s y s t e m u s e s its existing r e s o u r c e s to offer n e w services: y o u can "wire" m o n e y a n d telegrams, y o u c a n mail a letter a n d s e n d a fax, y o u can o p e n a post office box a n d a postal savings account, a n d so on. We will find t h a t t h e k e y c o n c e p t s i n t r o d u c e d in this s i m p l e d e s c r i p t i o n of t h e postal s y s t e m carry over w i t h a p p r o p r i a t e r e i n t e r p r e t a t i o n to t h e case of c o m m u n i c a t i o n n e t w o r k s . T h o s e c o n c e p t s are u s e r services a n d service qual­ ity, basic or b e a r e r services a n d t h e i r characteristics, t h e u n d e r l y i n g "hardware," a n d m a n a g e m e n t "intelligence." We h a v e already e n c o u n t e r e d t h e c o n c e p t s of n e t w o r k i n t e r c o n n e c t i o n a n d service integration in section 1.2.4. T h e services provided b y c o m m u n i c a t i o n n e t w o r k s e n a b l e u s e r s to ex­ c h a n g e information. T h e r e is a wide r a n g e of services: u s e r s c a n talk to e a c h o t h e r over a t e l e p h o n e n e t w o r k , transfer data over a c o m p u t e r n e t w o r k , a n d w a t c h a video p r o g r a m over a cable TV n e t w o r k . Since u s e r s i n t e r a c t w i t h t h e n e t w o r k t h r o u g h s o m e t e r m i n a l device ( t e l e p h o n e , computer, TV controller), it is s o m e t i m e s m o r e a p p r o p r i a t e to say t h a t t h e n e t w o r k services are u s e d ( c o n s u m e d ) b y u s e r applications (processes r u n n i n g o n t h e t e r m i n a l device). E n g i n e e r s build a n e t w o r k b y i n t e r c o n n e c t i n g two t y p e s of "hardware" or n e t w o r k e l e m e n t s : t r a n s m i s s i o n links a n d switches. Links transfer strings of bits from o n e location to another. Switches are c o m p u t e r s t h a t store, route, a n d m a n i p u l a t e t h o s e bit strings. This h a r d w a r e s u p p o r t s t h e n e t w o r k ' s bearer services: t h e transfer of bit strings, in a few s t a n d a r d formats, from o n e source or u s e r to o n e or m o r e n e t w o r k destinations. Bearer service p e r f o r m a n c e characteristics are s u m m a r i z e d in a few p a r a m e t e r s : t h e acceptable formats; t h e connectivity a n d selection of r o u t e s from source to destination; a n d t h e speed, delay, errors, a n d so forth, of t h e bit string. A n e t w o r k can effectively s u p p o r t a particular u s e r application o n l y if its b e a r e r services h a v e t h e requisite characteristics. Tb s u p p o r t voice con­ versations, for example, t h e end-to-end delay s h o u l d n o t b e m o r e t h a n 200 milliseconds (ms), say. Tb s u p p o r t data transfer t h e error rate s h o u l d n o t b e m o r e t h a n 10~ , say, a n d so on. T h e m o s t d e m a n d i n g applications, s u c h as t h e transfer of X rays w i t h a fidelity a n d s p e e d t h a t m a k e t h e m acceptable for u s e b y radiologists for diagnosis a n d interactive videoconferencing, r e q u i r e a high-performance network. More sophisticated services are built from less sophisticated o n e s in a l a y e r e d h i e r a r c h y or architecture. Each layer adds functionality. Each l a y e r c o m p r i s e s specific r u l e s of control a n d m a n a g e m e n t , i m p l e m e n t e d in software or h a r d w a r e . T h e s e r u l e s a n d t h e software t h a t i m p l e m e n t s t h e m are called protocols. T h e protocols reside in t h e switches a n d h o s t c o m p u t e r s . S o m e sets 4

2.1

Applications

41

of protocols h a v e b e c o m e standardized, a n d i n t e r c o n n e c t i o n of n e t w o r k s is facilitated if t h e y c o n f o r m to t h e s a m e standard. T h i s c h a p t e r i n t r o d u c e s t h e c o n c e p t s of h o w n e t w o r k s function a n d t h e logical l a y e r s u s e d to divide n e t w o r k s into s m a l l e r s u b s e t s of functionality. By t h e e n d of this c h a p t e r y o u will u n d e r s t a n d t h e m e c h a n i s m s u s e d to i m p l e m e n t n e t w o r k functions, i n c l u d i n g error-control s c h e m e s s u c h as t h e A l t e r n a t i n g Bit a n d Go Back Ν protocols a n d flow a n d c o n g e s t i o n control. We start b y discussing a few c o m m o n l y u s e d applications in section 2.1. In section 2.2 w e e x a m i n e t h e c o m m u n i c a t i o n traffic g e n e r a t e d b y u s e r applica­ tions. We c o m m e n t o n t h e i n f o r m a t i o n t h a t different t y p e s of u s e r applications exchange. In particular, w e explain w h y different applications r e q u i r e n e t w o r k services w i t h different characteristics. In section 2.3 w e describe n e t w o r k services a n d discuss t h e i r characteris­ tics. T h e s e characteristics m a t c h t h o s e r e q u i r e d b y u s e r applications. In section 2.4 w e identify h i g h - p e r f o r m a n c e n e t w o r k s in t e r m s of t h e i r scalability, t h e i r ability to s u p p o r t d e m a n d i n g a n d diverse u s e r applications, a n d t h e ease w i t h w h i c h t h e y c a n b e c o n n e c t e d to o t h e r n e t w o r k s . In section 2.5 w e discuss n e t w o r k e l e m e n t s a n d explain h o w t h e p r o p e r t i e s of t h e s e e l e m e n t s affect t h e characteristics of t h e services t h a t t h e y i m p l e m e n t . We explain t h e m e c h a n i s m s t h a t t h e n e t w o r k e l e m e n t s u s e to i m p l e m e n t services in section 2.6. In section 2.7 w e i n t r o d u c e t h e l a y e r e d organization of n e t w o r k o p e r a t i o n s . We explain t h a t t h e m a i n a d v a n t a g e s of a l a y e r e d organization are m o d u l a r i t y a n d standardization. In section 2.8 w e p r e s e n t t h e O p e n Data N e t w o r k (ODN) m o d e l . T h i s m o d e l provides a useful f r a m e w o r k for describing h o w n e t w o r k s are organized. T h e O p e n Systems I n t e r c o n n e c t i o n or OSI r e f e r e n c e m o d e l , u s e d in data n e t w o r k s , is p r e s e n t e d in C h a p t e r 3. Section 2.9 discusses h o w t h e various c o m p o n e n t s t h a t w e e x p l a i n e d in t h e c h a p t e r are a s s e m b l e d in c o m m o n n e t w o r k a r c h i t e c t u r e s . Section 2.10 discusses t h e n e t w o r k b o t t l e n e c k s a n d e x a m i n e s t h e t e c h n o ­ logical i m p r o v e m e n t s t h a t are r e q u i r e d to i m p r o v e t h e n e t w o r k s .

2.1

APPLICATIONS Tb m a k e t h e discussion of i n f o r m a t i o n transfers m o r e concrete, w e start b y discussing a few c o m m o n l y u s e d n e t w o r k applications. We explain h o w t h e s e applications w o r k a n d describe s o m e of t h e i r characteristics.

Network Services and Layered Architectures

2.1.1

World Wide Web T h e World Wide Web is a distributed application t h a t e n a b l e s y o u to navigate t h r o u g h a set of h y p e r l i n k e d d o c u m e n t s , called Web pages. Each Web page m a y contain text, pictures, audio clips, video clips, a n d possibly links. A link specifies e i t h e r a location in t h e s a m e d o c u m e n t or a n o t h e r Web page location a n d n a m e . T h e location m a y b e a n o t h e r file in t h e s a m e c o m p u t e r or in a n o t h e r c o m p u t e r a t t a c h e d to t h e I n t e r n e t . W h e n y o u click o n a link, t h e application a r r a n g e s to display t h e n e w location in t h e d o c u m e n t or to transfer a n d display t h e n e w Web page. Accordingly, w h e n y o u b r o w s e t h e Web, y o u initiate a s e q u e n c e of file transfers. T h e size of a Web page typically r a n g e s from a few kilobytes to a few h u n d r e d kilobytes. If t h e link is to a video clip or to a large file, t h e n t h e transfer m a y b e of a few m e g a b y t e s . As y o u m a y h a v e experienced, s o m e Web pages take a long t i m e to appear. T h i s d e l a y is n o t difficult to u n d e r s t a n d . For instance, if t h e r e q u e s t e d Web page c o r r e s p o n d s to 100 KB a n d if t h e transfer rate is limited to 8 Kbps, t h e n t h e transfer takes a b o u t two m i n u t e s . T h e transfer rate is l i m i t e d n o t o n l y b y y o u r m o d e m b u t also b y o t h e r c o n n e c t i o n s t h a t s h a r e s o m e critical links of t h e n e t w o r k w i t h y o u r o w n transfer. We expect Web page transfers to b e error-free. Accordingly, t r a n s m i s s i o n errors s h o u l d b e corrected.

2.1.2

Audio or Video Streams Streaming audio a n d video applications e n a b l e y o u to listen to or view a p r o g r a m as it is b e i n g transferred. M a n y radio stations t r a n s m i t "live" o n t h e I n t e r n e t , a n d s o m e live feeds are also available from TV stations. T h e s e s t r e a m i n g applications g e n e r a t e s t r e a m s of p a c k e t s t h a t t h e n e t w o r k delivers from t h e source to t h e destination. D u r i n g t h e i r travel across t h e n e t w o r k , t h e p a c k e t s suffer variable delays a n d s o m e p a c k e t s get d r o p p e d . T h e d e s t i n a t i o n buffers a few packets before it starts "playing" t h e m b a c k as a n audio or video s t r e a m . T h e buffering absorbs t h e delay fluctuations. Note that, w i t h t h e buffering, all t h e p a c k e t s face a d e l a y t h a t is at least e q u a l to t h e m a x i m u m value of t h e delay across t h e n e t w o r k . Indeed, to play b a c k t h e p a c k e t s at t h e s a m e c o n s t a n t intervals as t h e y e n t e r t h e n e t w o r k , all t h e p a c k e t s m u s t h a v e t h e s a m e total delay. C o n s e q u e n t l y , t h e faster packets m u s t b e d e l a y e d so t h a t t h e y h a v e t h e s a m e delay as t h e slowest one. Since t h e s e applications are o n e ­ w a y transmissions, t h e fixed delay is u n i m p o r t a n t . T h e t r a n s m i s s i o n rate of s u c h applications d e p e n d s o n t h e quality of t h e p r o g r a m . T h e rate of a n audio t r a n s m i s s i o n typically r a n g e s from 8 Kbps to 30 Kbps. A video t r a n s m i s s i o n h a s

2.1

Applications

43

a rate b e t w e e n 40 Kbps a n d 80 Kbps. S o m e t r a n s m i s s i o n errors are acceptable for audio a n d video. Such errors are p e r c e i v e d as noise or c o r r u p t i o n of t h e pictures.

2.13

Voice over Packets and Videoconferences I n e x p e n s i v e video c a m e r a s a n d a u d i o devices are available to set u p t e l e p h o n e calls or v i d e o c o n f e r e n c e s b e t w e e n PCs. For conversations, t h e o n e - w a y delay is b a r e l y noticeable if it is less t h a n 100 m s . Beyond 350 m s , t h e d e l a y m a k e s t h e c o n v e r s a t i o n u n p l e a s a n t . T h e small delay r e q u i r e m e n t of voice i m p l i e s t h a t t h e voice s a m p l e s m u s t b e p l a c e d in small packets. I b a p p r e c i a t e this implication, a s s u m e t h a t t h e voice is e n c o d e d into a 64 Kbps bit s t r e a m . Say t h a t w e place t h e voice bits into p a c k e t s of 1,600 bits. It t a k e s 1,600/64,000 = 25 m s to collect t h e bits t h a t fill u p a packet. C o n s e q u e n t l y , t h e packetization i n t r o d u c e s a 25-ms delay t h a t a d d s to t h e m a x i m u m d e l a y across t h e n e t w o r k . Distortions c a u s e d b y t r a n s m i s s i o n e r r o r s are preferable to t h e excessive delays t h a t w o u l d b e r e q u i r e d to correct t h e e r r o r s b y r e t r a n s m i t t i n g t h e e r r o n e o u s packets.

2.1.4

Networked Games M a n y n e t w o r k e d g a m e s are p l a y e d across t h e I n t e r n e t , s u c h as b o a r d a n d card g a m e s a n d faster action g a m e s . W h e n playing s u c h a g a m e , c o m p u t e r s exchange short c o m m a n d s . T h e acceptable delays d e p e n d o n t h e g a m e . T h e r e q u i r e d reliability of t h e transfer also d e p e n d s o n t h e g a m e .

2.1.5

Client/Server M a n y n e t w o r k e d applications, from databases to distributed c a l e n d a r s to s h a r e d file servers, are organized according to a c l i e n t / s e r v e r m o d e l . In a c l i e n t / s e r v e r application, a s e r v e r is d e s i g n e d to a n s w e r q u e r i e s from clients. Typically, t h e client s e n d s a q u e r y to t h e s e r v e r a n d waits for t h e reply. W h e n t h e reply arrives, t h e client r e s u m e s t h e execution of its p r o g r a m . T h e client m u s t b e able to d e t e c t a n d react to s e r v e r or n e t w o r k failures. A s e r v e r m u s t b e able to h a n d l e r e q u e s t s t h a t arrive from m a n y clients. A c o m m o n proce­ d u r e for m e e t i n g this objective is to m a k e t h e s e r v e r stateless. T h i s t e r m m e a n s t h a t t h e server s h o u l d n o t h a v e to r e m e m b e r a n y i n f o r m a t i o n a b o u t p r e v i o u s

Network Services and Layered Architectures

queries. In practice, t h e server is rarely stateless. However, application design­ ers a t t e m p t to limit t h e a m o u n t of information t h a t t h e s e r v e r m u s t r e m e m b e r . T h e n e t w o r k r e q u i r e m e n t s d e p e n d o n t h e acceptable r e s p o n s e t i m e of t h e application.

2.2

1

TRAFFIC CHARACTERIZATION AND QUALITY OF SERVICE As t h e e x a m p l e s in t h e p r e v i o u s section show, u s e r s e x c h a n g e information t h r o u g h applications. In this section, w e e x a m i n e t h e characteristics of t h e information transfers of different applications. T h e s e characteristics describe t h e traffic t h a t t h e applications g e n e r a t e as well as t h e acceptable delays a n d losses b y t h e n e t w o r k in delivering t h a t traffic. T h e information t h a t applications g e n e r a t e c a n take m a n y forms: text, voice, audio, data, graphics, pictures, a n i m a t i o n s , a n d videos. Moreover, t h e information transfer m a y b e one-way, two-way, broadcast, or m u l t i p o i n t . T h e information exchanged c a n b e analog or digital. CATV n e t w o r k s , for example, deliver analog video signals to television sets. T h e t e l e p h o n e net­ w o r k t r a n s m i t s analog or digital voice signals b e t w e e n t e l e p h o n e s . C o m p u t e r n e t w o r k s transfer bit files or bit s t r e a m s r e p r e s e n t i n g text, data, still images, a n d audio or video signals. Most n e t w o r k s t r a n s m i t analog signals b y first con­ verting t h e m into bit streams, as w e explained in section 1.2.1. In this c h a p t e r we limit discussion to digital t r a n s m i s s i o n of information, so u s e r applications e v e n t u a l l y r e q u i r e t h e c o m m u n i c a t i o n n e t w o r k to t r a n s m i t bit files or bit s t r e a m s . We call t h e s e bit files or bit s t r e a m s t h e traffic g e n e r a t e d b y t h e application. In o r d e r to s u p p o r t a u s e r application, t h e n e t w o r k m u s t b e able to t r a n s p o r t in a satisfactory m a n n e r t h e traffic t h a t t h e application g e n e r a t e s . Tkble 2.1 p r e s e n t s s o m e characteristics a b o u t t h e traffic g e n e r a t e d b y c o m m o n forms of information. Notice t h a t t h e bit s t r e a m s g e n e r a t e d b y a video signal c a n v a r y greatly d e p e n d i n g o n t h e c o m p r e s s i o n s c h e m e used. W h e n a page of text is e n c o d e d as a string of ASCII characters, it p r o d u c e s only a 2-Kilobyte (KB) string; w h e n t h a t page is digitized into pixels a n d c o m p r e s s e d as in facsimile, it p r o d u c e s a 50-KB string. T h e LalfeX file for this book, i n c l u d i n g figures, c o m p r e s s e s into a 1-MB file. A high-quality digitization of a color picture (similar quality to a good color laser p r i n t e r ) g e n e r a t e s a 33.5-MB string; a low-quality digitization of a black-and-white picture g e n e r a t e s o n l y a 0.5-MB string.

2.2

Traffic Characterization and Quality of Service

2.1

45

Information form

Traffic type

Size or Rate

Voice

CBR

64 Kbps

Video

CBR

64 Kbps, 1.5 Mbps

VBR

Mean 6 Mbps, peak 24 Mbps

Ttext

ASCII

2 KB/page

Fax

50 KB/page

Picture

600 dots/in, 256 colors, 8.5 χ 11 in

33.5 MB

70 dots/in, b/w, 8.5 χ 11 in

0.5 MB

Characteristics of traffic for some c o m m o n forms of information.

TABLE

We classify all traffic into t h r e e t y p e s . A u s e r application c a n g e n e r a t e a c o n s t a n t bit rate (CBR) s t r e a m , a variable bit rate (VBR) s t r e a m , or a s e q u e n c e of m e s s a g e s w i t h different t e m p o r a l characteristics. We briefly describe e a c h t y p e of traffic, a n d t h e n c o n s i d e r s o m e examples.

2.2.1

Constant Bit Rate Tb t r a n s m i t a voice signal, t h e t e l e p h o n e n e t w o r k e q u i p m e n t first c o n v e r t s it into a s t r e a m of bits w i t h a c o n s t a n t rate of 64 Kbps (see section 1.2.1). S o m e v i d e o - c o m p r e s s i o n s t a n d a r d s c o n v e r t a video signal into a bit s t r e a m w i t h a c o n s t a n t bit rate (CBR). For instance, MPEG1 is a s t a n d a r d for c o m p r e s s i n g video into a c o n s t a n t bit rate s t r e a m . T h e rate of t h e c o m p r e s s e d bit s t r e a m d e p e n d s o n t h e p a r a m e t e r s selected for t h e c o m p r e s s i o n algorithm, s u c h as t h e size of t h e video window, t h e n u m b e r of frames p e r second, a n d t h e n u m b e r of q u a n t i z a t i o n levels. MPEG1 p r o d u c e s a p o o r quality video at 1.15 M b p s a n d a good quality at 3 Mbps. Voice signals h a v e a rate t h a t r a n g e s from a b o u t 4 Kbps w h e n h e a v i l y c o m p r e s s e d a n d l o w quality to 64 Kbps. Audio signals r a n g e in r a t e from 8 Kbps to a b o u t 1.3 Mbps for CD quality. For t h e voice or video application to b e of a n acceptable quality, t h e n e t w o r k m u s t t r a n s m i t t h e bit s t r e a m w i t h a short d e l a y a n d c o r r u p t at m o s t a small fraction of t h e bits. (This fraction is called t h e bit error rate or BER.) T h e end-to-end delay s h o u l d b e less t h a n 200 m s for real-time video a n d voice conversations, since p e o p l e find larger delay u n c o m f o r t a b l e . T h a t de­ lay c a n b e a few s e c o n d s for non-real-time interactive applications s u c h as

Network Services and Layered Architectures

interactive video a n d information o n d e m a n d . T h e delay is n o t critical for n o n interactive applications s u c h as distribution of video or audio p r o g r a m s . T h e m a x i m u m acceptable BER is a b o u t 1 0 ~ for audio a n d video trans­ mission, in t h e a b s e n c e of c o m p r e s s i o n . W h e n a n a u d i o a n d video signal is c o m p r e s s e d , however, a n error in t h e c o m p r e s s e d signal will c a u s e a s e q u e n c e of errors in t h e u n c o m p r e s s e d signal. Therefore, t h e tolerable e r r o r rate for t r a n s m i s s i o n of c o m p r e s s e d signals is m u c h less t h a n 1 0 ~ . 4

4

2.22

Variable Bit Rate Some signal-compression t e c h n i q u e s c o n v e r t a signal into a bit s t r e a m t h a t h a s a variable bit rate (VBR). For instance, MPEG2 is a family of s t a n d a r d s for s u c h variable bit rate c o m p r e s s i o n of video signals. T h e bit rate is larger w h e n t h e s c e n e s of t h e c o m p r e s s e d m o v i e are fast m o v i n g t h a n w h e n t h e y are slow m o v i n g . DBS (Direct Broadcast Satellite) u s e s MPEG2 w i t h a n average rate of 4 Mbps. Tb specify t h e characteristics of a VBR s t r e a m , t h e n e t w o r k e n g i n e e r specifies t h e average bit rate a n d a d e s c r i p t i o n of t h e fluctuations of t h a t bit rate. We s t u d y s u c h descriptions in C h a p t e r 6. T h e acceptable delay a n d BER of t h e s e applications are similar to t h o s e of CBR applications.

2.2.3

Messages M a n y u s e r applications o n a n e t w o r k are i m p l e m e n t e d b y p r o c e s s e s t h a t exchange m e s s a g e s . (For p r e s e n t p u r p o s e s , a message is a variable-length bit string.) For instance, w h e n b r o w s i n g t h e Web, a u s e r s e n d s a s e r v e r r e q u e s t s for Web pages, a u d i o or video clips, or d o c u m e n t s . T h e s e r v e r replies b y s e n d i n g t h e r e q u e s t e d records to t h e user. As a n o t h e r example, a distributed c o m p u t i n g application g e n e r a t e s r e m o t e p r o c e d u r e calls for r e m o t e m a c h i n e s , w h i c h t h e n r e t u r n t h e results of t h e execution of t h e p r o c e d u r e s . T h e m e s s a g e traffic g e n e r a t e d b y various u s e r applications c a n h a v e a w i d e r a n g e of characteristics. S o m e applications, s u c h as e-mail, g e n e r a t e isolated m e s s a g e s . O t h e r applications, s u c h as a distributed c o m p u t a t i o n , g e n e r a t e long s t r e a m s of m e s s a g e s . T h e rate of m e s s a g e s c a n v a r y greatly across applications a n d devices. Tb describe t h e a m o u n t of traffic g e n e r a t e d b y a n application t h a t p r o d u c e s a s t r e a m of messages, t h e n e t w o r k e n g i n e e r m a y specify t h e average traffic rate a n d s o m e m e a s u r e of t h e fluctuations of t h a t rate, in a w a y similar to t h e case of a VBR specification.

23

Network Services

47 T h e n e t w o r k m u s t transfer t h e m e s s a g e s w i t h a n acceptable delay, a n d it c a n c o r r u p t o n l y a small fraction of t h e m e s s a g e s . Typical acceptable values of delays are 200 m s for real-time applications, a few s e c o n d s for interactive services, a n d m a n y s e c o n d s for n o n i n t e r a c t i v e services s u c h as e-mail. T h e acceptable fraction of m e s s a g e s t h a t c a n b e c o r r u p t e d r a n g e s from 1 0 ~ for data t r a n s m i s s i o n s to m u c h larger v a l u e s for noncritical applications s u c h as j u n k mail distribution. A m o n g applications t h a t e x c h a n g e s e q u e n c e s of m e s s a g e s , w e c a n distin­ guish those applications t h a t expect t h e m e s s a g e s to r e a c h t h e d e s t i n a t i o n in t h e correct o r d e r a n d t h o s e t h a t do n o t care a b o u t t h e order. 8

2.2.4

Other Requirements In this b o o k w e will c o n s i d e r o n l y t h e d e l a y a n d loss r e q u i r e m e n t s t h a t ap­ plications i m p o s e o n t h e n e t w o r k . We s h o u l d r e m e m b e r , however, t h a t o t h e r r e q u i r e m e n t s m a y b e i m p o r t a n t . We m e n t i o n h e r e reliability a n d security. W h e n o n e or m o r e links or switches fail, t h e n e t w o r k m a y b e u n a b l e to provide a c o n n e c t i o n b e t w e e n s o u r c e a n d d e s t i n a t i o n u n t i l t h o s e failures are repaired. Reliability refers to t h e f r e q u e n c y a n d d u r a t i o n of s u c h failures. Some applications (e.g., control of electric p o w e r plants, hospital life s u p p o r t s y s t e m s , critical b a n k i n g operations) d e m a n d e x t r e m e l y reliable n e t w o r k o p e r a t i o n . Typically, w e w a n t to b e able to provide h i g h e r reliability b e t w e e n a few d e s i g n a t e d source-destination pairs. H i g h e r reliability is a c h i e v e d b y providing m u l t i p l e disjoint r o u t e s b e t w e e n t h e d e s i g n a t e d n o d e pairs. Recall t h a t in a multiple-access n e t w o r k s u c h as E t h e r n e t , e v e r y c o m p u t e r "hears" e v e r y p a c k e t t h a t is t r a n s m i t t e d . T h e case of w i r e l e s s p h o n e t r a n s m i s ­ sion is similar. In t h e s e n e t w o r k s , to g u a r a n t e e privacy of t r a n s m i s s i o n s , it will b e n e c e s s a r y to e n c r y p t t h o s e t r a n s m i s s i o n s . More generally, security is con­ c e r n e d w i t h m e a s u r e s t h a t c a n b e t a k e n to p r e v e n t u n a u t h o r i z e d access to data or information transfer. As m o n e y a n d o t h e r assets take o n a n electronic form t h a t c a n b e t r a n s m i t t e d over a n e t w o r k , issues of s e c u r i t y will b e c o m e m o r e pressing.

2.3

NETWORK SERVICES We h a v e j u s t s e e n t h a t u s e r applications exchange bit s t r e a m s or m e s s a g e s w i t h widely different traffic characteristics. T h e s e applications expect t h e n e t w o r k to deliver t h e bit s t r e a m s or m e s s a g e s w i t h i n a specific d e l a y a n d to c o r r u p t o n l y a small fraction of t h e bits or m e s s a g e s .

Network Services and Layered Architectures

N e t w o r k e n g i n e e r s distinguish two t y p e s of information transfer services: c o n n e c t i o n l e s s a n d c o n n e c t i o n - o r i e n t e d . We explain t h a t i m p o r t a n t distinction next.

2.3.1

Connection-Oriented Service W h e n a n e t w o r k i m p l e m e n t s a c o n n e c t i o n - o r i e n t e d service, it delivers m e s ­ sages from t h e source to t h e d e s t i n a t i o n in t h e correct order. T h u s , t h e data transfer in a c o n n e c t i o n - o r i e n t e d service a p p e a r s to take place over a dedi­ cated t r a n s m i s s i o n line, except for t h e variability in t h e t r a n s m i s s i o n delay of different packets. A c o n n e c t i o n - o r i e n t e d service is r e q u i r e d b y u s e r applica­ tions t h a t expect reliable a n d o r d e r e d t r a n s m i s s i o n s of m e s s a g e s . A CBR or VBR bit s t r e a m is delivered b y a c o n n e c t i o n - o r i e n t e d service. A c o n n e c t i o n - o r i e n t e d service involves t h r e e p h a s e s : a c o n n e c t i o n s e t u p phase, a data transfer phase, a n d a c o n n e c t i o n t e a r d o w n p h a s e . T h e quality of service (QpS) in s o m e c o n n e c t i o n - o r i e n t e d n e t w o r k services specifies w h e t h e r t h e t r a n s m i s s i o n is e r r o r free a n d m a y assign s o m e prior­ ity level to p a c k e t s w i t h t h e u n d e r s t a n d i n g t h a t t h e n e t w o r k will a t t e m p t to t r a n s m i t high-priority p a c k e t s before low-priority packets. T h u s t h e delays are likely to b e s m a l l e r in a high-priority c o n n e c t i o n - o r i e n t e d service t h a n in a low-priority c o n n e c t i o n - o r i e n t e d service. Some n e t w o r k s p e r m i t a m o r e detailed QpS specification. T h a t specifica­ tion i n c l u d e s t h e delay, delay jitter, a n d p a c k e t e r r o r rate. It also i n c l u d e s a description of t h e a m o u n t of traffic t h a t t h e service c a n t r a n s p o r t . T h e s e m o r e detailed specifications are r e q u i r e d b y real-time a n d interactive applications, as w e explained in section 2.2. W h e n r e q u e s t i n g a c o n n e c t i o n - o r i e n t e d service in t h e s e n e t w o r k s , at t h e c o n n e c t i o n s e t u p t i m e t h e u s e r specifies t h e QpS t h a t is r e q u i r e d b y t h e u s e r ap­ plication. T h e n e t w o r k c a n t h e n d e t e r m i n e w h e t h e r it h a s sufficient r e s o u r c e s available to h a n d l e t h a t c o n n e c t i o n w i t h t h e r e q u e s t e d QoS, a n d t h e n e t w o r k c a n t h e n set aside t h e s e r e s o u r c e s for t h e c o n n e c t i o n . In o r d e r to u n d e r t a k e t h e s e tasks, t h e n e t w o r k retains "state" information a b o u t existing c o n n e c t i o n s . H o w t h e s e tasks c a n b e carried o u t is discussed in C h a p t e r s 8 a n d 9.

2.3.2

Connectionless Service W h e n it i m p l e m e n t s a c o n n e c t i o n l e s s service, t h e n e t w o r k transfers e a c h packet of data to t h e d e s t i n a t i o n o n e at a time, i n d e p e n d e n t l y of t h e o t h e r

2.4

High-Performance Networks • ^ -

• —

mmmmmmm

y

ι

,

packets. Unlike t h e case w i t h c o n n e c t i o n - o r i e n t e d services, t h e n e t w o r k h a s n o state information to d e t e r m i n e w h e t h e r a p a c k e t is p a r t of a s t r e a m of o t h e r packets. In particular, t h e n e t w o r k h a s n o k n o w l e d g e of t h e a m o u n t of traffic t h a t will b e s e n t b y t h e user. C o n s e q u e n t l y , t h e n e t w o r k c a n n o t set aside r e s o u r c e s t h a t w o u l d b e n e e d e d to a c h i e v e a specific quality of service. Because of this l i m i t e d information, o n l y a restricted r a n g e of service qual­ ity c a n b e offered. Typical QpS p a r a m e t e r s i n c l u d e a b o u n d o n t h e m a x i m u m p a c k e t size a n d service priority: a higher-priority p a c k e t is t r a n s m i t t e d before a lower-priority packet. C o n n e c t i o n l e s s service is c h a r a c t e r i z e d b y t h e average or typical d e l a y a n d a specification of t h e w a y t h e service h a n d l e s errors. S o m e c o n n e c t i o n l e s s services do n o t indicate w h e n t h e y fail to deliver a m e s s a g e . O t h e r services a c k n o w l e d g e t h e correct delivery of m e s s a g e s . Typically, lower layers provide o n l y c o n n e c t i o n l e s s service b u t h i g h e r l a y e r s m a y a d d functionality to provide c o n n e c t i o n - o r i e n t e d services.

2.4

HIGH-PERFORMANCE NETWORKS

2.4.1

Traffic Increase T h e rate of traffic o n t h e I n t e r n e t h a s b e e n i n c r e a s i n g rapidly. T h i s i n c r e a s e is c a u s e d b y n e w applications, m o s t l y t h e WWW a n d e-mail, t h a t h a v e m a d e t h e n e t w o r k attractive to m a n y users. T h e v o l u m e of traffic t h a t e a c h u s e r g e n e r a t e s o n t h e I n t e r n e t is increasing. E-mail applications let u s e r s attach p i c t u r e s or o t h e r large files. Web p a g e s are increasingly r i c h in a n i m a t i o n s , audio, a n d video clips. I n e x p e n s i v e digital c a m e r a s let u s e r s e x c h a n g e p h o t o g r a p h s over the Internet. T h e d e m a n d is i n c r e a s i n g w i t h t h e s u p p l y of n e t w o r k b a n d w i d t h . As t h e n e t w o r k gets faster, m o r e u s e r s take a d v a n t a g e of applications t h a t r e q u i r e t h a t i n c r e a s e d speed. Clearly t h e t r e n d is to i n c r e a s e t h e t r a n s m i s s i o n r a t e s of t h e n e t w o r k links. Low-speed m o d e m s (33 Kbps) are giving w a y to ISDN links (64 Kbps or 128 Kbps) or to digital subscriber l o o p s (384 Kbps or faster) or cable m o d e m s (a few Mbps). T h e long links ( t e n s or h u n d r e d s of k m ) are b e i n g u p g r a d e d from 1.5 Mbps to 45 Mbps to 155 Mbps to 622 Mbps to 2.4 Gbps a n d 9.6 Gbps. Multiples of t h e s e large r a t e s are m a d e possible b y optical m u l t i p l e x i n g m e t h o d s .

49

Network Services and Layered Architectures

Tb k e e p u p w i t h t h e increasing link speed, t h e n e t w o r k switches are also getting faster. This i n c r e a s e of t r a n s m i s s i o n rates is a n i m p o r t a n t c o m p o n e n t of w h a t w e call h i g h - p e r f o r m a n c e n e t w o r k i n g .

2.4.2

High-Performance We define a h i g h - p e r f o r m a n c e n e t w o r k (HPN) as a c o m m u n i c a t i o n n e t w o r k t h a t s u p p o r t s a large variety of u s e r applications a n d t h a t is scalable. In o r d e r to s u p p o r t m a n y applications, t h e n e t w o r k m u s t b e able to transfer u s e r traffic at high s p e e d a n d w i t h low delay. It m u s t b e able to allocate r e s o u r c e s in w a y s t h a t m a t c h t h e application r e q u i r e m e n t s . N e t w o r k organization a n d m a n a g e m e n t m u s t b e flexible so t h a t n e w applications c a n b e s u p p o r t e d as t h e n e e d arises. Tb i m p l e m e n t a h i g h - p e r f o r m a n c e n e t w o r k , a n u m b e r of critical bottle­ n e c k s m u s t b e addressed. T h e s e b o t t l e n e c k s arise at all l a y e r s of t h e n e t w o r k operations. We e x a m i n e t h e s e b o t t l e n e c k s in section 2.10 after w e explain t h e key mechanisms that networks implement. A scalable n e t w o r k c a n a c c o m m o d a t e growing n u m b e r s of u s e r s w i t h o u t degradation in p e r f o r m a n c e . G r o w t h is usually a c c o m m o d a t e d b y i n t e r c o n ­ n e c t i n g distinct n e t w o r k s . T h e n e t w o r k m u s t b e able to provide c o n n e c t i v i t y over a growing span. T h e s p a n m a y b e expressed in distance, n u m b e r of links, or n u m b e r of s u b n e t w o r k s . A n H P N t h a t s u p p o r t s a wide r a n g e of c u r r e n t a n d future applications a n d t h a t can a c c o m m o d a t e g r o w t h is built a n d m a n a g e d differently from n e t w o r k s t h a t are d e s i g n e d for a specific application or u s e r p o p u l a t i o n . At o n e extreme, p h o n e n e t w o r k s h a v e m a n y n o d e s a n d o p e r a t e over v e r y large distances, b u t u s e r s c a n transfer data only at low s p e e d s . ( T h u s p h o n e n e t w o r k s are scalable, b u t s u p p o r t l i m i t e d applications.) At t h e o t h e r e x t r e m e , a c o m p u t e r b a c k p l a n e b u s o p e r a t e s at h i g h s p e e d s b u t c o n n e c t s o n l y a small n u m b e r of devices v e r y close to e a c h other. Local a r e a n e t w o r k s or LANs (e.g., E t h e r n e t ) c a n transfer data b e t w e e n t e n s of n o d e s at m o d e r a t e s p e e d s (10 Mbps) over m o d e r a t e distances (1 k m ) . More r e c e n t LANs s u p p o r t s p e e d s of 1 Gbps. Wide area n e t w o r k s or WANs, s u c h as X.25 n e t w o r k s a n d t h e I n t e r n e t , c o n n e c t h u n d r e d s of n o d e s over h u n d r e d s of kilometers, b u t t h e y o p e r a t e at l i m i t e d s p e e d s of a few Mbps or less. Metropolitan area n e t w o r k s (MANs—e.g., DQDB, FDDI, SMDS) r u n at a h i g h e r s p e e d a n d c o n n e c t u s e r s s e p a r a t e d b y a b o u t 100 k m . F r a m e Relay n e t w o r k s are s t r e a m l i n e d versions of X.25 n e t w o r k s t h a t c a n o p e r a t e at high speed. T h e "backbone" t e l e p h o n e n e t w o r k is a n HPN: it c o m p r i s e s t h e switches a n d links or t r u n k s c o n n e c t i n g t h e m , b u t excludes t h e low-speed links c o n n e c t i n g u s e r t e l e p h o n e s to t h e switches.

2.5

Network Elements

T h e precise v a l u e s of t h e u s e r transfer rate, t h e acceptable delay, t h e n e t w o r k span, a n d t h e n u m b e r of u s e r s t h a t characterize a n H P N are s o m e w h a t arbitrary. We h a v e in m i n d a u s e r rate t h a t exceeds 100 Mbps, delays o n t h e o r d e r of 100 ms, a s p a n of at least 100 m, a n d a n u m b e r of u s e r s t h a t c a n exceed 100. What is essential for a n e t w o r k to qualify as a n H P N is for it to b e able to s u p p o r t d e m a n d i n g services s u c h as interactive MPEG video a n d LAN i n t e r c o n n e c t i o n s a m o n g m a n y users.

2.5

NETWORK ELEMENTS A c o m m u n i c a t i o n n e t w o r k is a collection of n e t w o r k e l e m e n t s i n t e r c o n n e c t e d a n d m a n a g e d to s u p p o r t t h e transfer of i n f o r m a t i o n from a u s e r at o n e n e t w o r k location or n o d e to a u s e r at a n o t h e r n o d e . In this section w e discuss t h e t w o principal n e t w o r k e l e m e n t s , a n d w e e x a m i n e h o w t h e p r o p e r t i e s of t h e s e e l e m e n t s affect t h e characteristics of t h e services t h a t t h e y i m p l e m e n t .

2.5.1

Principal Network Elements T h e principal n e t w o r k e l e m e n t s are ( t r a n s m i s s i o n ) links a n d switches. A link transfers a s t r e a m of bits from o n e e n d to t h e o t h e r at a c e r t a i n r a t e w i t h a g i v e n b i t e r r o r rate a n d a fixed p r o p a g a t i o n t i m e . Links are u n i d i r e c t i o n a l . T h e m o s t i m p o r t a n t links are optical fiber, c o p p e r coaxial cable, a n d m i c r o w a v e or radio "wireless" links. Optical fiber a n d c o p p e r l i n k s are u s u a l l y point-topoint links, w h e r e a s radio links are u s u a l l y b r o a d c a s t links. We s t u d y links in C h a p t e r 11, focusing o n optical links, w h i c h are essential for h i g h - s p e e d networks. Several i n c o m i n g a n d outgoing links t e r m i n a t e at a switch, w h i c h is a device t h a t transfers bits from its i n c o m i n g links to its outgoing links. W h e n e v e r t h e rate of i n c o m i n g bits exceeds t h a t of outgoing bits, t h e excess bits are buffered at t h e switch. We s t u d y switches in C h a p t e r 12, focusing o n switches t h a t c a n h a n d l e high-speed traffic. W h e n w e view a n e t w o r k as a n i n t e r c o n n e c t i o n of n e t w o r k e l e m e n t s , w e m a y r e p r e s e n t t h e n e t w o r k as in t h e g r a p h of Figure 2.1. I n t h e graph, edges d e n o t e links, large circles d e n o t e switches (includingbuffers), a n d small circles d e n o t e u s e r n o d e s w h e r e bits are g e n e r a t e d or c o n s u m e d . We will u s e t h e n a m e s user, source, destination, station, a n d node as n e a r s y n o n y m s .

Network Services and Layered Architectures

Switch User

2.1 FIGURE

The principal network elements are transmission links and switches. Transmission links connect user nodes and switches.

A n o t h e r v e r y i m p o r t a n t view of a n i n t e r c o n n e c t i o n of n e t w o r k e l e m e n t s is provided b y t h e q u e u i n g n e t w o r k m o d e l . Consider a switch w i t h η i n c o m i n g links a n d m outgoing links, as in t h e left of Figure 2.2. T h e d i a g r a m o n t h e right is t h e q u e u i n g m o d e l . Each i n c o m i n g link's receiver w r i t e s into its i n p u t buffer, a n d e a c h outgoing link's t r a n s m i t t e r r e a d s from its o u t p u t buffer. T h e switch transfers bits or p a c k e t s from i n p u t buffers to t h e a p p r o p r i a t e o u t p u t buffers. This q u e u i n g n e t w o r k m o d e l of switches a n d links is u s e d to describe a n d evaluate n e t w o r k p e r f o r m a n c e . For example, t h e q u e u i n g d e l a y e n c o u n t e r e d b y a p a c k e t in a n o u t p u t buffer is p r o p o r t i o n a l to t h e n u m b e r of p a c k e t s in t h e buffer in front of it. If p a c k e t s arrive into a full buffer, t h e r e will b e p a c k e t loss.

Input buffers 2.2 FIGURE

Output buffers

The queuing network model is used for performance analysis. Each switch has a buffer corresponding to each incoming and outgoing link, which is serviced at the link rate.

2.5

Network Elements

It is difficult to calculate q u e u i n g d e l a y a n d p a c k e t loss. We describe several a p p r o a c h e s in C h a p t e r 8.

2.5.2

Network Elements and Service Characteristics A p a c k e t g e n e r a t e d b y a source travels over o n e link, gets buffered at a switch, is t h e n r o u t e d to a n o t h e r link, a n d so on, until it arrives at its destination. T h e delay p a c k e t s e x p e r i e n c e t h r o u g h a n e t w o r k d e p e n d s o n t h e e l e m e n t s t h a t constitute t h e n e t w o r k , t h e traffic t h a t goes t h r o u g h t h e s e e l e m e n t s , a n d t h e w a y t h e n e t w o r k is o p e r a t e d . T h e detailed analysis of t h e d e l a y t h r o u g h a n e t w o r k is r a t h e r involved. For now, n o t e t h a t w e c a n d e c o m p o s e t h e total d e l a y into four c o m p o n e n t s : total delay = TRANS + PROP + Q D + PROC.

(2.1)

In (2.1) TRANS is t h e t i m e r e q u i r e d to t r a n s m i t a packet, so TRANS = (packet s i z e ) / ( t r a n s m i s s i o n s p e e d ) .

(2.2)

For example, for a 10,000-bit p a c k e t a n d a t r a n s m i s s i o n s p e e d of 1 Mbps, TRANS is 10 m s . PROP is t h e signal p r o p a g a t i o n time, so

P

R

O

p

=

(distance from source to d e s t i n a t i o n ) 7-Γ-, r—r-. τ ( s p e e d of electrical or optical signal). 1

(2.3)

Propagation t i m e for a n electrical or optical signal is b e t w e e n 3.3 a n d 5 mi­ c r o s e c o n d s p e r k i l o m e t e r ( ^ s / k m ) . QD is t h e q u e u i n g d e l a y in t h e switch. It occurs w h e n e v e r t h e bit rate of t h e traffic c o m i n g into t h e switch exceeds t h e capacity of t h e outgoing link. T h e excess bits are q u e u e d in t h e switch buffers. In contrast w i t h t h e o t h e r two s o u r c e s of delay, q u e u i n g d e l a y is significantly affected b y t h e n e t w o r k control policy. Finally, PROC is t h e p r o c e s s i n g t i m e r e q u i r e d b y t h e n e t w o r k switches. We will a s s u m e t h a t this p r o c e s s i n g t i m e is negligible. Suppose as a r o u g h r u l e of t h u m b t h a t t h e n e t w o r k is controlled so t h a t e a c h p a c k e t c o m i n g into t h e switch h a s to wait o n average for four p r e v i o u s p a c k e t s to b e t r a n s m i t t e d . T h e n t h e average q u e u i n g d e l a y is four t i m e s t h e t r a n s m i s s i o n delay, so total delay = 5 χ TRANS + PROP.

(2.4)

Network Services and Layered Architectures

By c o m b i n i n g t h e expressions (2.2) t h r o u g h (2.4), w e see h o w t h e delay de­ p e n d s o n t h e t r a n s m i s s i o n rate a n d l e n g t h of t h e links a n d o n t h e l e n g t h or size of m e s s a g e s .

2.5.3

Examples We illustrate h o w w e c a n u s e t h e p r e c e d i n g analysis in t h r e e examples. Consider a potential t e l e c o m m u t i n g application in w h i c h a c o m p a n y e m ­ p l o y e e r e s p o n d s to c u s t o m e r billing i n q u i r i e s from a t e r m i n a l at h o m e . W h e n a c u s t o m e r calls, a n a u t o m a t i c call-transfer service transfers t h e call to t h e e m ­ ployee's h o m e . T h e e m p l o y e e t h e n r e q u e s t s t h e c u s t o m e r ' s record from t h e c o m p a n y ' s database c o m p u t e r . T h e r e c o r d is transferred b y t h e n e t w o r k a n d displayed o n t h e t e r m i n a l . T h e e m p l o y e e t h e n a n s w e r s t h e c u s t o m e r ' s q u e s ­ tions a n d u p d a t e s t h e record as n e c e s s a r y . For t h e r e s p o n s e to b e satisfactory to t h e customer, t h e retrieval of t h e record a n d its display s h o u l d n o t take m o r e t h a n o n e second, say. A s c r e e n full of text takes a b o u t 16,000 bits, so t h e net­ w o r k r e s p o n s e will b e satisfactory provided t h e t r a n s m i s s i o n s p e e d of Β b p s is such that total delay = 5 χ 16,000 χ B~ + PROP < 1. l

In this example, w e m a y neglect PROP, w h i c h is a few /zs. We see t h e n t h a t this application n e e d s a t r a n s m i s s i o n s p e e d of 80 Kbps. T h i s rate is s u p p o r t e d b y a DSL service t h a t costs a b o u t $50 p e r m o n t h (1999 prices). C o m p a r i n g this cost w i t h t h e savings to t h e e m p l o y e e a n d e m p l o y e r resulting from fewer w o r k c o m m u t e trips, less office a n d p a r k i n g space, a n d m o r e flexible w o r k schedules, w e can i m a g i n e a large potential d e m a n d for c o m m u n i c a t i o n services n e e d e d to s u p p o r t t e l e c o m m u t i n g . (It is e s t i m a t e d t h a t a t e l e c o m m u t e r w o r k i n g at h o m e 1 to 2 days p e r w e e k c a n save m o s t c o m p a n i e s $6,000 to $12,000 a y e a r b e c a u s e of i n c r e a s e d productivity, lower staff turnover, a n d r e d u c e d office space. T h e a n n u a l cost p e r t e l e c o m m u t e r is e s t i m a t e d to b e u p to $10,000 for e q u i p m e n t , space at h o m e , a n d p h o n e services.) As a n o t h e r t e l e c o m m u t i n g example, consider a design e n g i n e e r or archi­ tect using a workstation at h o m e . Most of t h e m a t e r i a l n e e d e d for w o r k is stored in t h e workstation's local disk. However, from t i m e to time, t h e e n g i n e e r n e e d s to retrieve or replace a large file (e.g., several bit m a p s ) stored in a file s e r v e r at t h e workplace. Such a file is a b o u t 3 MB long, a n d if t h e w o r k s t a t i o n is con­ n e c t e d b y a 1-Mbps link, t h e retrieval w o u l d take a b o u t 24 seconds. T h i s m a y b e a d e q u a t e if s u c h t r a n s m i s s i o n s are infrequent. However, if 10 t r a n s m i s s i o n s

2.5

Network Elements

2.3 FIGURE

Delay as a function of transmission rate for a U.S.-wide network with 10 nodes ^ ^ h node of 1 to 100 packets. Delay for the circuit-switched connection equals the propagation time.

a n c

a

u e u e

a t

e a c

e v e r y m i n u t e are n e e d e d , this m a y r e q u i r e a 4-Mbps link, w h i c h at c u r r e n t costs m a y m a k e this t e l e c o m m u t i n g application u n e c o n o m i c a l . (At t h e work­ place, t h e e n g i n e e r ' s w o r k s t a t i o n a n d file server are c o n n e c t e d b y a relatively i n e x p e n s i v e 10-Mbps local a r e a n e t w o r k s u c h as E t h e r n e t . ) T h e third e x a m p l e provides a m o r e g e n e r a l illustration of t h e r e q u i r e m e n t s t h a t applications place o n t h e delay a n d s p e e d of b e a r e r services. Figure 2.3 c o m p a r e s t h r e e b e a r e r services t h a t t r a n s p o r t data across t h e U n i t e d States t h r o u g h 10 switches at various c o m b i n a t i o n s of s p e e d a n d delay. T h e first service is i m p l e m e n t e d b y a circuit-switched n e t w o r k u s i n g w i r e d (optical or c o p p e r coaxial) links. T h e end-to-end delay e q u a l s t h e p r o p a g a t i o n delay, PROP, i n d e p e n d e n t of t h e s p e e d . T h i s d e l a y is a b o u t 25 m s (5,000 k m χ 5 / x s / k m ) . This service is s u m m a r i z e d b y t h e horizontal line labeled "Wired circuit-switched." T h e s e c o n d a n d third services are i m p l e m e n t e d b y a packet-switched n e t w o r k u s i n g 1-kilobit (Kb) a n d 100-Kb sized packets, respectively. If at e a c h switch, a p a c k e t e n c o u n t e r s b e t w e e n 0 a n d 100 p a c k e t s waiting in q u e u e , t h e n t h e q u e u i n g delay ( t h r o u g h t h e 10 switches) is b e t w e e n 0 a n d 1,000 p a c k e t t r a n s m i s s i o n t i m e s . T h e t r a n s m i s s i o n time, TRANS, is 1,000/bit rate for t h e

Network Services and Layered Architectures

smaller a n d 1 0 / b i t rate for t h e larger packet. So for t h e s e c o n d service t h e total delay is 5

PROP < total delay < PROP + 1 0 / b i t rate, 6

w h i c h is t h e lower s h a d e d region in t h e figure. For t h e third service t h e total delay D is PROP < total delay < PROP + 1 0 / b i t rate, 8

w h i c h is t h e u p p e r s h a d e d region. We n o w c o n s i d e r t h r e e applications: a conversation, a n interactive ex­ change, a n d a transfer of a voice or video file. T h e c o n v e r s a t i o n c a n tolerate a d e l a y of 250 m s , t h e interactive exchange c a n a c c e p t a few s e c o n d s of delay, a n d t h e file transfer c a n tolerate a large delay. T h e s e b o u n d s are s h o w n in t h e figure. Finally, d e p e n d i n g o n t h e c o m p r e s s i o n t e c h n i q u e used, voice will gen­ erate traffic at a rate b e t w e e n 8 a n d 64 Kbps, video b e t w e e n 64 Kbps a n d 10 Mbps, a n d HDTV will g e n e r a t e h i g h e r - s p e e d traffic. W h e n w e c o m p a r e t h e delay a n d s p e e d r e q u i r e m e n t s of t h e application w i t h t h e characteristics of t h e t h r e e services, w e see t h a t at s p e e d s above 10 Mbps t h e r e is little difference b e t w e e n circuit-switched c o n n e c t i o n s a n d a 1Kb packet-switched service. With 100-Kb packets, h i g h e r s p e e d is n e e d e d . For example, at T 3 s p e e d of 45 M b p s t h e d e l a y of a 100-Kb p a c k e t service is indis­ tinguishable from a circuit-switched c o n n e c t i o n , p r o v i d e d q u e u e s are s h o r t e r t h a n 100 packets. For interactive a n d transfer applications, t h e a d v a n t a g e of a circuit-switched c o n n e c t i o n over t h e packet-switched services similarly disap­ p e a r s at h i g h e r speeds.

2.6

BASIC NETWORK MECHANISMS A n e t w o r k ' s b e a r e r services c o m p r i s e t h e end-to-end t r a n s p o r t of bit s t r e a m s , in specific formats, over a set of routes. T h e s e services are differentiated b y quality: speed, delay, errors. T h e y are p r o d u c e d u s i n g five basic m e c h a n i s m s : multiplexing, switching, error control, flow control, congestion control, a n d r e s o u r c e allocation. We discuss t h o s e m e c h a n i s m s in this section. As w e n o t e d in section 1.2.2, t h e r e are e c o n o m i e s of scale in t r a n s m i s s i o n . T h a t is, t h e cost of a t r a n s m i s s i o n link c o n n e c t i n g two n o d e s does n o t in­ crease in p r o p o r t i o n to its capacity. Individual u s e r s located at t h o s e t w o points, however, typically n e e d a small t r a n s m i s s i o n b a n d w i d t h for short d u r a t i o n s .

2.6

Basic Network Mechanisms

Multiplexing c o m b i n e s data s t r e a m s of m a n y s u c h u s e r s into o n e large b a n d ­ w i d t h s t r e a m for long d u r a t i o n s . Users t h e r e b y c a n s h a r e in t h e scale e c o n o m i e s of t r a n s m i s s i o n . However, u s e r s are n o t c o n c e n t r a t e d in a few locations; t h e y are geographically dispersed. Switching allows u s to b r i n g t o g e t h e r t h e data s t r e a m s of t h e s e d i s p e r s e d u s e r s . (In essence, t h e c o m m u n i c a t i o n n e t w o r k in­ d u s t r y m a k e s profits b y installing large t r a n s m i s s i o n capacity a n d b y "renting" this capacity to u s e r s in s m a l l e r a m o u n t s u s i n g m u l t i p l e x i n g a n d switching.) In t h e left side of Figure 2.4 w e see a n e t w o r k w i t h n o m u l t i p l e x i n g or switching. Each pair of p h o n e s is c o n n e c t e d b y a d e d i c a t e d link of capacity 64 Kbps n e e d e d to c a r r y o n e voice conversation. I n t h i s n e t w o r k , t h e average n u m b e r of links p e r p h o n e , a n d h e n c e t h e average cost, i n c r e a s e s w i t h t h e n u m b e r of n o d e s . Since a t e l e p h o n e is e n g a g e d in at m o s t o n e c o n v e r s a t i o n at a n y time, link utilization, t h a t is, t h e fraction of t i m e t h e link is b u s y , d e c r e a s e s as t h e n u m b e r of n o d e s i n c r e a s e s . T h u s t h e cost p e r p h o n e grows w i t h t h e size of t h e n e t w o r k . This n e t w o r k w a s t e s r e s o u r c e s . T h e n e t w o r k o n t h e right e m p l o y s multiplexing a n d switching. Each p h o n e is c o n n e c t e d b y a d e d i c a t e d access l i n k to o n e of t w o local switches, w h i c h are c o n n e c t e d b y a single link called a t r u n k . (In t e l e p h o n e n e t w o r k s , a switch is located in a central office; a link b e t w e e n two switches is called a trunk; a link b e t w e e n a subscriber t e l e p h o n e a n d a switch is called a n access line or subscriber loop.) T h e access line capacity is 64 Kbps, a n d t h e t r u n k capacity is a m u l t i p l e of 64 Kbps. (In this e x a m p l e it m a y b e 2 χ 64 Kbps.) D e p e n d i n g on its capacity, a t r u n k c a n c a r r y several voice c o n v e r s a t i o n s s i m u l t a n e o u s l y b y multiplexing. A c o n v e r s a t i o n b e t w e e n two p h o n e s will oc­ c u p y only two access lines—if b o t h p h o n e s are c o n n e c t e d to t h e s a m e local switch—or, in addition, it will o c c u p y a fraction of t h e t r u n k capacity. It is t h e

Network Services and Layered Architectures

task of t h e switch to d e t e r m i n e w h e t h e r a call originating from a local p h o n e is d e s t i n e d for a n o t h e r local p h o n e or for a p h o n e c o n n e c t e d to t h e r e m o t e switch. In this network, t h e r e is o n e access line p e r p h o n e , b u t t h e switch a n d t r u n k capacities grow m u c h less rapidly t h a n t h e n u m b e r of p h o n e s . As a result, t h e average cost of t h e n e t w o r k p e r u s e r d e c r e a s e s w i t h t h e n u m b e r of p h o n e s . This decreasing cost s t r u c t u r e of c o m m u n i c a t i o n n e t w o r k s is m a d e possible b y multiplexing a n d switching. A n o t h e r i m p o r t a n t m e c h a n i s m is error control. All t r a n s m i s s i o n links oc­ casionally c o r r u p t t h e m e s s a g e s t h e y t r a n s m i t . A l t h o u g h carefully d e s i g n e d a n d m a i n t a i n e d links h a v e a v e r y small bit e r r o r rate (e.g., 1 0 ~ ) , e v e n t h e rare errors in t h e t r a n s m i s s i o n s m a y n o t b e acceptable. Moreover, in addition to t r a n s m i s s i o n errors, a m e s s a g e m a y fail to r e a c h its d e s t i n a t i o n b e c a u s e it arrived at a switch w h o s e buffer w a s full a n d so t h e m e s s a g e w a s discarded. It is therefore i m p o r t a n t for t h e n e t w o r k to control s u c h errors. We explain two m e t h o d s t h a t n e t w o r k s u s e to control errors later in this section. A source m u s t p r e v e n t o v e r w h e l m i n g t h e receiver b y s e n d i n g m e s s a g e s faster t h a n t h e receiver c a n store or process t h e m . Flow control is a m e c h a n i s m t h a t e n a b l e s t h e receiver to p a c e t h e t r a n s m i s s i o n s of t h e source. T h e end-to-end delay is t h e s u m of a fixed p r o p a g a t i o n d e l a y a n d a variable q u e u i n g delay. In o r d e r to k e e p this delay w i t h i n a n acceptable range, t h e rate at w h i c h packets e n t e r t h e n e t w o r k m u s t b e controlled. Congestion control is t h e g e n e r i c n a m e for a set of m e c h a n i s m s d e s i g n e d to limit t h e rate or n u m b e r of packets i n t r o d u c e d into t h e n e t w o r k b y a source or a switch. If t h e c o n g e s t i o n control m e c h a n i s m does not function properly, a n excessive n u m b e r of p a c k e t s m a y a c c u m u l a t e in t h e switch buffers causing u n a c c e p t a b l e delay or loss. We h a v e s e e n t h a t s o m e applications r e q u i r e t h e n e t w o r k to provide a m i n i m u m b a n d w i d t h to e n s u r e acceptable p e r f o r m a n c e . For example, a voice conversation r e q u i r e s 64 Kbps a n d MPEG1 r e q u i r e s 1.5 Mbps. A variable bit rate application will r e q u i r e a g u a r a n t e e d c o m b i n a t i o n of m i n i m u m b a n d w i d t h a n d buffers. Because n e t w o r k resources—link b a n d w i d t h a n d switch buffers—are s h a r e d b y m a n y applications at t h e s a m e time, resource allocation m e c h a n i s m s m u s t b e designed to e n s u r e t h a t e a c h application receives t h e n e c e s s a r y re­ sources to m a i n t a i n its quality of service. Multiplexing is carried out in switches or specialized h a r d w a r e . Switching a n d r e s o u r c e allocation are i m p l e m e n t e d in switches. Error control a n d flow control are i m p l e m e n t e d in switches a n d h o s t c o m p u t e r s . 12

2.6.1

Multiplexing We n o w explain t h r e e i m p o r t a n t t e c h n i q u e s for multiplexing Ν i n c o m i n g c h a n n e l s onto o n e outgoing c h a n n e l . T h e s e are time-division multiplexing,

2.6

Basic Network Mechanisms

statistical multiplexing, a n d frequency-division multiplexing. For o u r p r e s e n t p u r p o s e s a channel is a c o m m u n i c a t i o n l i n k of fixed capacity m e a s u r e d , say, in bits p e r s e c o n d (bps). T h a t is, e a c h s e c o n d of t i m e o n t h e c h a n n e l is divided into a n u m b e r of intervals called hit times, t h e n u m b e r b e i n g e q u a l to t h e c h a n n e l capacity. For example, in a 1-Mbps c h a n n e l , a bit t i m e is 1 μδ long. Each bit t i m e m a y b e o c c u p i e d b y a n i n f o r m a t i o n or data bit, or it m a y b e e m p t y . We also say, accordingly, t h a t t h e c h a n n e l is busy or idle at t h a t t i m e . T h e average data rate divided b y c h a n n e l capacity, or t h e fraction of t i m e t h e c h a n n e l is b u s y , is t h e c h a n n e l utilization. Utilization is e q u a l to 100% o n l y if t h e i n s t a n t a n e o u s data rate is always e q u a l to t h e c h a n n e l capacity. Multiplexing is t h e process b y w h i c h i n f o r m a t i o n bits from Ν i n c o m i n g c h a n n e l s are transferred into bit t i m e s o n o n e outgoing c h a n n e l . Demultiplex­ ing is t h e reverse process: t h e i n f o r m a t i o n bits o n o n e i n c o m i n g m u l t i p l e x e d c h a n n e l are s e p a r a t e d a n d t r a n s f e r r e d o n t o Ν outgoing c h a n n e l s . T h e multi­ plexed outgoing c h a n n e l c o n t a i n s s o m e extra bits (in addition to t h e i n c o m i n g c h a n n e l s ' data bits) t h a t t h e d e m u l t i p l e x e r u s e s to d e t e r m i n e w h i c h data bits b e l o n g to w h i c h i n c o m i n g c h a n n e l . Time-division multiplexing or T D M is illustrated in Figure 2.5. In TDM, t h e capacity of t h e outgoing c h a n n e l is divided into Ν logical c h a n n e l s , a n d data in e a c h of Ν i n c o m i n g c h a n n e l s is p l a c e d in a d e s i g n a t e d outgoing logical c h a n n e l . This is a c h i e v e d as follows. T i m e o n t h e outgoing c h a n n e l is divided into fixedl e n g t h intervals called frames. F r a m e s are d e l i m i t e d b y a special bit s e q u e n c e called a framing pattern, n o t s h o w n in t h e figure. T i m e in e a c h frame is further subdivided into Ν fixed-length intervals called slots (this n a m e is n o t always u s e d ) . T h u s e a c h frame consists of a s e q u e n c e of slots: slot 1, slot 2, . . . , slot N.

1

25 FIGURE

2

Ν

\

2

Ν

When a communication link is shared by time-division multiplexing, time is divided into frames. Each frame is divided into time slots that are allocated in a fixed order to the different incoming channels.

Network Services and Layered Architectures

(A slot is usually 1 bit or 1 b y t e wide.) A logical c h a n n e l o c c u p i e s e v e r y N t h slot. T h e r e are t h u s Ν logical c h a n n e l s . T h e first logical c h a n n e l o c c u p i e s slots 1, Ν + 1, 2N + 1 , . . . ; t h e s e c o n d o c c u p i e s slots 2, Ν + 2, 2N + 2 , . . . ; a n d so on. T h e multiplexer o p e r a t e s as follows. T h e data bits in e a c h i n c o m i n g c h a n ­ n e l are r e a d into a s e p a r a t e FIFO (first in, first out) buffer. T h e m u l t i p l e x e r reads this buffer in s e q u e n c e for a n a m o u n t of t i m e e q u a l to t h e c o r r e s p o n d i n g slot t i m e : buffer 1 is r e a d into slot 1, buffer 2 is r e a d into slot 2. (If t h e r e are n o t e n o u g h bits in a buffer, t h e c o r r e s p o n d i n g slot r e m a i n s partially e m p t y . ) T h e bit s t r e a m of t h e outgoing c h a n n e l is easily demultiplexed: t h e d e m u l t i p l e x e r d e t e c t s t h e framing p a t t e r n from w h i c h it d e t e r m i n e s t h e b e g i n n i n g of e a c h frame, a n d t h e n e a c h slot. T D M is easy to i m p l e m e n t . T h e o v e r h e a d d u e to extra framing bits is small. However, t h e utilization of t h e outgoing c h a n n e l m a y vary a great deal d e p e n d i n g o n t h e b u r s t i n e s s of t h e i n c o m i n g data s t r e a m s . Tb see this, observe t h a t t h e capacity of t h e outgoing c h a n n e l m u s t b e as large as t h e s u m of t h e capacities of t h e Ν i n c o m i n g c h a n n e l s . As a result, t h e utilization of t h e outgoing c h a n n e l will b e low or h i g h accordingly as t h e utilization of t h e i n c o m i n g c h a n n e l s is low or high. T D M leads to h i g h utilization if t h e i n c o m i n g data is n o t b u r s t y . T h u s T D M is ideal for c o n s t a n t bit rate traffic. Statistical multiplexing or SM, illustrated in Figure 2.6, is m o s t effective in t h e case of b u r s t y i n p u t data. As in TDM, t h e data bits in e a c h i n c o m i n g c h a n n e l are r e a d into s e p a r a t e FIFOs. T h e multiplexer r e a d s e a c h buffer in t u r n u n t i l t h e buffer e m p t i e s . (It is c u s t o m a r y to call t h e data r e a d in o n e t u r n a data packet.) In T D M e a c h FIFO is r e a d for a fixed a m o u n t of t i m e — o n e slot—and

2.6

In statistical multiplexing, the multiplexer visits the incoming channel buffers in some order. The multiplexer empties a buffer before moving to the next one. The buffer contents are tagged to indicate their incoming channel. An idle channel does not waste transmission time.

2.6

Basic Network Mechanisms

so e a c h i n c o m i n g c h a n n e l is allocated a fixed fraction of t h e outgoing c h a n n e l capacity, i n d e p e n d e n t of t h e data rate o n t h a t c h a n n e l . By contrast, in SM, t h e capacity allocated to e a c h i n c o m i n g c h a n n e l varies w i t h t i m e , d e p e n d i n g o n its i n s t a n t a n e o u s data rate: t h e h i g h e r t h e rate, t h e larger t h e capacity allocated to it at t h a t t i m e . As a result, t h e capacity of t h e outgoing c h a n n e l n e e d s to b e only as large as t h e s u m of t h e average data r a t e s of t h e i n c o m i n g c h a n n e l , which, for b u r s t y traffic, m a y b e m u c h s m a l l e r t h a n t h e s u m of t h e p e a k data rates. H e n c e t h e capacity of t h e outgoing c h a n n e l m a y b e s m a l l e r t h a n t h e s u m of t h e i n c o m i n g c h a n n e l capacities. We call t h e ratio of t h e total i n c o m i n g capacity to t h e total outgoing capacity t h e multiple?cing gain. T h i s gain is u n i t y for T D M b u t c a n b e m u c h larger for SM. Figure 2.7 illustrates t h e possible m u l t i p l e x i n g gains for SM relative to T D M a n d FDM for various applications. As w e h a v e seen, in SM t h e size of p a c k e t s r e a d from e a c h FIFO c a n v a r y across c h a n n e l s a n d over t i m e w i t h i n e a c h c h a n n e l . Therefore, t h e d e m u l ­ tiplexer c a n n o t sort t h e p a c k e t s b e l o n g i n g to different c h a n n e l s m e r e l y from t h e i r position w i t h i n a frame. C o n s e q u e n t l y , additional bits, w h i c h delimit e a c h p a c k e t a n d identify t h e c o r r e s p o n d i n g i n c o m i n g c h a n n e l or source, m u s t b e a d d e d to e a c h packet. T h e resulting o v e r h e a d is significantly larger t h a n u n ­ d e r TDM. It also b e c o m e s m o r e difficult to i m p l e m e n t t h e m u l t i p l e x e r ( w h i c h m u s t n o w a d d t h e p a c k e t d e l i m i t e r a n d c h a n n e l or s o u r c e identifier) a n d t h e

Number of connections (SM)

CBR video

2.7 FIGURE

for given link rates

Voice

VBR video

TCP

Statistical multiplexing can achieve m u c h higher multiplexing gain relative to TDM and FDM, especially for bursty traffic.

Network Services and Layered Architectures

d e m u l t i p l e x e r ( w h i c h m u s t locate a n d d e c o d e t h o s e bit p a t t e r n s ) . T h e s e in­ creases in complexity a n d o v e r h e a d m u s t b e b a l a n c e d against h i g h utilization in t h e face of b u r s t y data to d e t e r m i n e w h e t h e r SM or T D M is m o r e efficient. In general, c o n s t a n t bit rate traffic, s u c h as voice, fixed-rate video, a n d control a n d s e n s o r signals, are b e t t e r h a n d l e d b y TDM, w h e r e a s m e s s a g e traffic s u c h as database t r a n s a c t i o n s a n d variable bit rate video are b e t t e r h a n d l e d b y SM. Not surprisingly, t e l e p h o n e n e t w o r k s u s e TDM, w h e r e a s c o m p u t e r c o m m u n i ­ cation n e t w o r k s u s e SM. In s o m e cases, o n e m a y w a n t to treat different p a c k e t s t r e a m s differently. T h a t is, i n s t e a d of t r a n s m i t t i n g t h e p a c k e t s in t h e i r o r d e r of arrival at t h e statistical multiplexer, s o m e o t h e r strategy m a y b e preferable. A n u m b e r of s c h e d u l i n g strategies h a v e b e e n designed. A priority strategy always t r a n s m i t s from t h e first n o n e m p t y q u e u e in a fixed order. A w e i g h t e d round-robin strategy multiplexes k q u e u e s b y first serving M ( l ) p a c k e t s from q u e u e 1, t h e n Mil) from q u e u e 2, a n d so o n until it serves M(k) p a c k e t s from q u e u e k a n d t h e n r e p e a t s t h e cycle. If a n y q u e u e i r u n s o u t of p a c k e t s before M(z) p a c k e t s are t r a n s m i t t e d from t h e q u e u e , t h e n t h e t r a n s m i t t e r c o n t i n u e s w i t h t h e n e x t q u e u e . T h e objective of t h e s e strategies a n d t h e i r variations is to provide a differentiated service to different classes of applications. Frequency-division multiplexing or FDM is illustrated in Figure 2.8. T h e f r e q u e n c y b a n d of t h e outgoing c h a n n e l is divided into distinct fixed b a n d s , o n e for e a c h i n c o m i n g c h a n n e l . T h e signal in e a c h i n c o m i n g c h a n n e l is m o d u l a t e d to fit into its assigned b a n d . T h e signal o n t h e outgoing c h a n n e l is s i m p l y t h e s u m of t h e s e m o d u l a t e d signals. T h u s t h e b a n d w i d t h of t h e outgoing c h a n n e l m u s t b e greater t h a n or e q u a l to t h e s u m of t h e b a n d w i d t h s of t h e i n c o m i n g c h a n n e l . (In this sense, FDM is similar to TDM.) For demultiplexing, t h e FDM signal is p a s s e d t h r o u g h a n a p p r o p r i a t e filter. By d e m o d u l a t i n g t h e filter output, w e obtain t h e a p p r o p r i a t e i n p u t signal. FDM is u s e d in AM a n d FM radio a n d TV b r o a d c a s t as well as in CATV distribution. It is also u s e d in cellular radio. FDM is m o r e flexible t h a n t h e o t h e r two multiplexing s c h e m e s . I n d e e d , o n e i n c o m i n g c h a n n e l m a y c o n t a i n a n analog signal, a n d a n o t h e r m a y b e digital. FDM h a s two disadvantages. First, it is wasteful of b a n d w i d t h since t h e f r e q u e n c y b a n d s assigned to a n i n c o m i n g c h a n n e l m u s t b e s e p a r a t e d b y a "guard b a n d " from t h e o t h e r c h a n n e l s . Second, if t h e t r a n s m i s s i o n link exhibits significant nonlinearities, t h e r e will b e "cross-talk" a m o n g t h e different signals, leading to errors. We h a v e u s e d b a n d w i d t h (or size of a link's f r e q u e n c y b a n d in Hz) a n d s p e e d or bit rate of a link (in b p s ) i n t e r c h a n g e a b l y , b u t t h a t is n o t strictly cor­ rect. A (digital) bit s t r e a m is c o n v e r t e d into a n analog signal b y a m o d u l a t i o n s c h e m e like phase-shift k e y i n g (PSK). T h e f r e q u e n c y s p e c t r u m of t h e result-

2.6

Basic Network Mechanisms

Modulators

Channel I


24. T h e n e t w o r k e n g i n e e r s h o u l d p r o p o s e a w i n d o w size e q u a l to 24 in this e x a m p l e . Note t h a t Go Back Ν w i t h Ν = 1 is exactly ABP. T r a n s m i s s i o n e r r o r s r e d u c e t h e efficiency of Go Back N. T h e analysis of t h e efficiency w h e n e r r o r s o c c u r is m o r e complicated.

Selective Repeat Protocol T h e final protocol t h a t w e explain is t h e Selective Repeat Protocol (SRP). SRP is similar to Go Back Ν b u t r e t r a n s m i t s o n l y p a c k e t s t h a t w e r e n o t correctly re­ ceived. A w i n d o w size 2N is a g r e e d on, a n d t h e p a c k e t s a n d a c k n o w l e d g m e n t s are n u m b e r e d m o d u l o 2N. O n c e again, t h e s e n d e r waits for t h e a c k n o w l ­ e d g m e n t of p a c k e t η before s e n d i n g p a c k e t η + Ν ( m o d u l o 2 N ) . T h e s e n d e r r e t r a n s m i t s p a c k e t s t h a t h a v e n o t b e e n a c k n o w l e d g e d w i t h i n a t i m e o u t after t h e i r t r a n s m i s s i o n , a n d t h e r e c e i v e r a c k n o w l e d g e s all t h e correct packets. T h e s e n d e r a n d t h e r e c e i v e r m u s t b o t h b e able to store u p to Ν packets. Figure 2.23 s h o w s typical s e q u e n c e s of e v e n t s w h e n t h e n o d e s u s e SRP (with Ν = 4).

2.6.4

Flow Control Consider a source t h a t is s e n d i n g p a c k e t s to a receiver. T h e r e c e i v e r sets aside s o m e m e m o r y space to store t h e i n c o m i n g p a c k e t s u n t i l it c a n p r o c e s s t h e m , possibly b y displaying t h e m or storing t h e m o n a h a r d disk. W h e n t h e m e m o r y gets full, t h e r e c e i v e r s h o u l d notify t h e s o u r c e to stop t r a n s m i t t i n g . W h e n t h e source u s e s a w i n d o w m e c h a n i s m , t h e r e c e i v e r c a n indicate, in t h e a c k n o w l e d g m e n t s , t h e m a x i m u m w i n d o w size t h a t it c a n store. T h e source

Network Services and Layered Architectures

t h e n u s e s t h a t value as a limit to its w i n d o w size. As t h e receiver buffer b e c o m e s full, this m a x i m u m value decreases, w h i c h slows d o w n t h e source a n d p r e v e n t s overflowing t h e receiver m e m o r y .

2.6.5

Congestion Control T h e bit s t r e a m s of m a n y s o u r c e s get m u l t i p l e x e d a n d r e a d into a switch buffer. If t h e sources are u n r e g u l a t e d , occasionally t h e buffer will get filled up, causing long delays and, possibly, buffer overflow a n d resulting p a c k e t loss. Congestion-control m e c h a n i s m s c a n cause t h e s o u r c e s or switches to re­ d u c e t h e n u m b e r of p a c k e t s t h e y inject into t h e n e t w o r k w h e n c o n d i t i o n s of congestion are d e t e c t e d . If t h e sources u s e a w i n d o w m e c h a n i s m as in Go Back Ν or SRP, t h e n this m e c h a n i s m c a n b e a d a p t e d to serve t h e p u r p o s e s of c o n g e s t i o n control. Notice t h a t t h e s o u r c e s c a n d e t e c t c o n g e s t i o n or p a c k e t loss from t h e fact t h a t t h e i r packets t i m e out. T h u s as soon as several t i m e o u t s o c c u r in quick succession, t h e source s h o u l d r e d u c e t h e w i n d o w size. T h i s r e d u c t i o n auto­ matically r e d u c e s t h e n u m b e r of o u t s t a n d i n g packets, a n d h e n c e n e t w o r k con­ gestion. T h e r e is a conflict b e t w e e n k e e p i n g t h e w i n d o w size small to r e d u c e congestion a n d k e e p i n g it large e n o u g h to "fill t h e pipe" to m a i n t a i n high efficiency. I m a g i n e a t r a n s m i s s i o n b e t w e e n two hosts a n d a s s u m e t h a t t h e round-trip t i m e b e t w e e n t h e h o s t s is Τ s e c o n d s a n d t h e t r a n s m i s s i o n rate is R bits p e r second. T h e "delay-bandwidth" p r o d u c t of t h e c o n n e c t i o n is defined as R χ T. For instance, for a short wireless link, possible values m i g h t b e R = 30 Kbps a n d Τ = 10μ8, so t h a t Rx Τ = 0.3 bit. As a n o t h e r example, for a fast cross­ c o u n t r y b a c k b o n e link, t h e values m i g h t b e Τ = 30 m s and.R = 2.4 Gbps, so t h a t R χ Τ = 72 Mb. Suppose t h e d e l a y - b a n d w i d t h p r o d u c t is large a n d t h e w i n d o w size is large. T h e n m o s t of t h e o u t s t a n d i n g p a c k e t s are p r o p a g a t i n g t h r o u g h t h e links, r a t h e r t h a n waiting in t h e buffers. As a result, o n c e congestion is d e t e c t e d a n d t h e w i n d o w size is r e d u c e d , t h e m a n y p a c k e t s already in transit in t h e links are unaffected b y t h e r e d u c e d w i n d o w size. T h o s e p a c k e t s c o n t i n u e to arrive into t h e buffers, increasing congestion a n d p a c k e t loss. T h u s w i n d o w flow control is n o t effective w h e n individual c o n n e c t i o n s h a v e a large b a n d w i d t h - d e l a y product. In t h a t case, "open loop" congestion control in t h e form of so-called rate control is m o r e effective. We s t u d y t h e s e t e c h n i q u e s in detail in C h a p t e r s 8 a n d 9.

2.6

Basic Network Mechanisms

2.6.6

Resource Allocation Different applications r e q u i r e b e a r e r services of different qualities (delay, error rate, etc.). A n e t w o r k c a n g u a r a n t e e a n application a p a r t i c u l a r service quality only if it c a n dedicate r e s o u r c e s ( b a n d w i d t h , buffers) to t h a t application. A circuit-switched n e t w o r k d e d i c a t e s a fixed b a n d w i d t h to e a c h c o n n e c t i o n , so it c a n g u a r a n t e e m i n i m u m d e l a y to a n application w h o s e p e a k rate is less t h a n t h e fixed b a n d w i d t h . A d a t a g r a m n e t w o r k is c o n n e c t i o n l e s s , a n d t h e n e t w o r k switches do n o t h a v e t h e state information n e e d e d to distinguish p a c k e t s from different appli­ cations. As a result, t h e n e t w o r k c a n n o t dedicate r e s o u r c e s t h a t are specific to individual applications. S o m e c r u d e r e s o u r c e allocation is possible, however. For example, as will b e s e e n in C h a p t e r 4, t h e T C P / I P protocol p e r m i t s expe­ dited p a c k e t s to b e t r e a t e d w i t h a h i g h e r priority t h a n n o n e x p e d i t e d packets. As a n o t h e r example, switches c a n provide differentiated services to p a c k e t s of different t y p e s b y i m p l e m e n t i n g a priority or w e i g h t e d s c h e d u l i n g . In virtual circuit-switched n e t w o r k s , e a c h p a c k e t carries its virtual circuit identifier or VCI. T h i s allows switches to t r e a t p a c k e t s differently b a s e d o n t h e i r VCI. T h u s , at t h e t i m e its virtual circuit is set u p , a n application c a n negotiate w i t h t h e n e t w o r k for c e r t a i n service quality. T h e n e t w o r k switches c a n t h e n r e s e r v e b a n d w i d t h a n d buffers for t h a t virtual circuit so as to provide t h a t quality. G u a r a n t e e d service quality is t h u s a possibility. In C h a p t e r s 8 a n d 9 w e consider r e s o u r c e allocation m e c h a n i s m s t h a t provide this g u a r a n t e e . T&ble 2.2 s u m m a r i z e s t h e m a i n t e c h n i q u e s u s e d to i m p l e m e n t t h e b a s i c n e t w o r k operations. Function

Technique

Characteristics

Multiplexing

T i m e division (TDM)

Fixed bit rate channels, flexible, m i n i m u m overhead to identify channels

Frequency division (FDM)

Fixed bandwidth channels, very flexible, suited for analog signals

Statistical (SM)

Channel bandwidth o n demand, overhead for channel identifier

Circuit (CS)

Good for CBR traffic, no queuing delay, l o w utilization for bursty traffic

Switching

Reference

Ch. 5

(continued)

Network Services and Layered Architectures

Function

Technique

Characteristics

Reference

Packet (PS)

Good for connectionless message traffic, variable queuing delay, high utilization for bursty traffic, robust against link failures

Ch. 3

Virtual circuit (VC)

Good for connection-oriented VBR traffic, variable queuing delay, good utilization

Ch. 6, 8

Error control Detection

Flow control

Resource allocation

2.2

CRC codes can detect most errors

Retransmission

Alternating Bit, Go Back N, etc., used to control errors when reverse path is available

Error correction

Used when reverse path is not available, e.g., storage or high-delay satellite links

Window flow control

Combined with end-to-end error control protocol, effective in low delaybandwidth connections

Ch. 8, 9

Rate control

Effective in high delaybandwidth connections, used in ATM services

Ch. 8, 9

Routing, band­ Used in virtual circuit width, buffer networks to guarantee service allocation; admis­ quality sion control

Ch. 8, 9

Techniques and characteristics of basic network operations.

TABLE

27

LAYERED ARCHITECTURE A n architecture is a specific w a y of organizing t h e m a n y functions p e r f o r m e d b y a c o m p u t e r n e t w o r k w h e n it provides services s u c h as file transfer, e-mail, directory services, a n d t e r m i n a l e m u l a t i o n . In this section w e first i n t r o d u c e t h e layered a r c h i t e c t u r e of n e t w o r k func­ tions. We t h e n c o m m e n t o n t h e i m p l e m e n t a t i o n of layers.

2.7

Layered Architecture

27.1

Layers In m o s t n e t w o r k s , t h e functions are organized into layers. As s h o w n in Fig­ u r e 2.24, in a l a y e r e d d e c o m p o s i t i o n , services of l a y e r η are i m p l e m e n t e d b y protocol entities (processes) at l a y e r η u s i n g t h e services of l a y e r η — 1. Ex­ a m p l e s of s u c h a l a y e r e d d e c o m p o s i t i o n of c o m m u n i c a t i o n functions a b o u n d in e v e r y d a y situations. For instance, two executives c a n e x c h a n g e m e s s a g e s b y u s i n g t h e services of t h e i r secretaries. T h e secretaries t h e m s e l v e s exchange m e s s a g e s b y u s i n g facsimile m a c h i n e s . T h e m a c h i n e s t r a n s m i t facsimiles b y u s i n g t h e services of t h e t e l e p h o n e n e t w o r k . (See Figure 2.25.) We c o v e r e d a m o r e t e c h n i c a l e x a m p l e w h e n w e explained e r r o r control: b y u s i n g s o m e suitable protocol, a t r a n s m i t t e r a n d a receiver c a n i m p l e m e n t reliable p a c k e t t r a n s m i s s i o n s over a n u n r e l i a b l e t r a n s m i s s i o n link. By d e c o m p o s i n g n e t w o r k functions into layers, t h e n e t w o r k e n g i n e e r s partition a c o m p l e x design p r o b l e m into a n u m b e r of m o r e m a n a g e a b l e s u b p r o b l e m s — t h o s e of designing t h e different layers. T h i s d e c o m p o s i t i o n sim­ plifies t h e design a n d its verification. Moreover, t h e d e c o m p o s i t i o n p e r m i t s standardization, m a k e s possible t h e c o m p e t i t i v e i m p l e m e n t a t i o n s of different layers, a n d facilitates n e t w o r k i n t e r c o n n e c t i o n . For t h e l a y e r e d d e c o m p o s i t i o n to provide t h e s e benefits, t h e n e t w o r k e n g i n e e r s m u s t specify w i t h o u t a m ­ biguity t h e services p r o v i d e d b y different l a y e r s a n d t h e interfaces b e t w e e n

Node A 2.24 FIGURE

Node Β

In a layered architecture, protocol entities of layer η i m p l e m e n t services b y using the services i m p l e m e n t e d b y layer η — 1.

Network Services and Layered Architectures

».-·;·ί. ·-Γ:.-.ί.«ίΛΐ.ί. 7

Exchanging letters

IS

ecretary A Secretary

p|

Secretary Β

py
0, a n d it d e c r e m e n t s X b y 1 w h e n e v e r it h a s w r i t t e n into a location. Process P c a n r e a d w h e n e v e r X < N, a n d it i n c r e m e n t s X after it h a s r e a d a location. S e m a p h o r e s c a n also b e u s e d b y m u l t i p l e w r i t e r a n d r e a d e r processes. W h e n this possibility is i m p l e m e n t e d , special care m u s t b e t a k e n to avoid conflicting m a n i p u l a t i o n s of t h e s e m a p h o r e value; likewise, c a u t i o n is called for w h e n h a n d l i n g situations in w h i c h a process aborts before t h e s e m a p h o r e h a s b e e n reset. n

n

n +

n

n

n

Network Services and Layered Architectures

W h e n t h e processes u s e q u e u e s to c o m m u n i c a t e , p r o c e s s P +\ w r i t e s C into a q u e u e t h a t is r e a d b y p r o c e s s P . A q u e u e is organized as a first-in, first-out a r r a y of data. T h e q u e u e h a s s o m e r e s e r v e d capacity, a n d it c a n b e i m p l e m e n t e d b y t h e o p e r a t i n g s y s t e m as a l i n k e d list w i t h a p o i n t e r to t h e h e a d of t h e q u e u e a n d a n o t h e r to t h e tail of t h e q u e u e . Process P + i w r i t e s into t h e q u e u e (at t h e tail), a n d process P r e a d s from t h e q u e u e (at t h e h e a d ) . T h e operating s y s t e m c h e c k s t h a t t h e r e is data to b e r e a d w h e n P w a n t s to r e a d a n d t h a t t h e r e is space available w h e n P \ w a n t s to write. Typically, t h e operating s y s t e m c a n h a n d l e a large n u m b e r of q u e u e s b e t w e e n various p r o c e s s e s b y s h a r i n g a large m e m o r y a m o n g t h e s e q u e u e s . T h e capacities of t h e different q u e u e s can b e adjusted d y n a m i c a l l y b y creating a n e w l i n k e d list w h e n e v e r a n e w i n t e r p r o c e s s q u e u e is n e e d e d . T h e different q u e u e s c a n b e u s e d to pass m e s s a g e s t h a t s h o u l d b e h a n d l e d differently. For instance, o n e q u e u e m a y c o n t a i n high-priority m e s s a g e s a n d a n o t h e r low-priority m e s s a g e s . A n u m b e r of p r o c e s s e s m a y write into t h e s a m e q u e u e or r e a d from t h e s a m e q u e u e if, for instance, e a c h m e s s a g e in t h e q u e u e c o n t a i n s a n identification n u m b e r t h a t specifies t h e process for w h i c h it is i n t e n d e d . n

n

n

n

n

n+

Typically, P +\ does n o t t r a n s m i t t h e m e s s a g e Μ to P . Instead, it t r a n s m i t s a p o i n t e r to t h a t m e s s a g e t h a t indicates w h e r e Μ is in m e m o r y a n d its l e n g t h . T h e actual transfer of Μ m u s t o c c u r w h e n t h e m e s s a g e m u s t m o v e across dif­ ferent c o m p u t e r boards. In m a n y systems, t h e n e t w o r k interface b o a r d h a s its o w n m e m o r y t h a t stores t h e p a c k e t s r e a d y to b e t r a n s m i t t e d . In s u c h a n i m p l e ­ m e n t a t i o n , it m a y b e t h a t P is i m p l e m e n t e d b y t h e n e t w o r k interface b o a r d while P +\ is i m p l e m e n t e d b y t h e m a i n CPU. In t h a t case, P + i actually copies Μ to P across t h e c o m p u t e r b u s to w h i c h t h e interface b o a r d is attached. Obvi­ ously, t h e m e s s a g e m u s t also b e copied b y t h e b o t t o m layer, w h i c h i m p l e m e n t s t h e actual t r a n s m i s s i o n b e t w e e n c o m p u t e r s . n

n

n

n

n

n

Let u s n o w t u r n to t h e i m p l e m e n t a t i o n of t h e functions t h a t P p e r f o r m s . T h e entities P a n d P' i m p l e m e n t t h e protocol of l a y e r n. T h i s protocol specifies t h a t P m u s t c o m p u t e t h e h e a d e r H, start s o m e timer, a n d u p d a t e s o m e coun­ ters. W h e n a t i m e r expires, P typically initiates a n e w call to P _ i to r e t r a n s m i t t h e message. T h e e n t i t y P' m u s t r e a d t h e h e a d e r Η a n d p e r f o r m a set of oper­ ations s u c h as verifying t h a t t h e p a c k e t is correct, s e n d a n a c k n o w l e d g m e n t , a n d indicate to P' t h a t a p a c k e t h a s arrived. In s o m e protocols, t h e h e a d e r Η c o n t a i n s a n error-detection field w h o s e value d e p e n d s o n t h e m e s s a g e M. In t h a t case, to calculate H, t h e p r o c e s s P m u s t r e a d t h e m e s s a g e M, which, t o g e t h e r w i t h t h e calculation of H, r e q u i r e s a large n u m b e r of instructions. In o t h e r protocols, t h e h e a d e r Η c o n t a i n s a n error-detection field t h a t is i n d e p e n d e n t of Μ a n d t h a t p r o t e c t s o n l y t h e h e a d e r itself. T h e execution of s u c h a protocol is typically m u c h faster. n

n

n

n

n

n

n

n+l

n

2.7

Layered Architecture

T h i s discussion p o i n t s to t h e i m p l i c a t i o n s of b o t h t h e design a n d t h e i m p l e ­ m e n t a t i o n of protocols for t h e achievable r a t e s of execution of s u c h protocols. Fast protocols are d e s i g n e d to limit t h e n e e d for protocol entities to r e a d full m e s s a g e s . Protocol i m p l e m e n t a t i o n s are faster w h e n t h e y m i n i m i z e t h e n u m ­ b e r of actual m e s s a g e transfers b y passing p o i n t e r v a l u e s i n s t e a d of c o p y i n g t h e m e s s a g e s . Finally, protocol e x e c u t i o n s c a n b e s p e e d e d u p b y i m p l e m e n t i n g protocol entities o n d e d i c a t e d h a r d w a r e t h a t frees u p t h e m a i n CPU. Ideally, t h e execution of t h e protocols s h o u l d i m p o s e a m i n i m u m b u r d e n o n t h e m a i n CPU, a n d it s h o u l d b e fast e n o u g h to k e e p u p w i t h t h e c o m m u n i c a t i o n link a n d w i t h t h e source of t h e data to b e transferred. We m a k e this discussion m o r e c o n c r e t e w i t h t h e h e l p of Figure 2.27. T h e p a n e l o n t h e left s h o w s a basic h o s t c o m p u t e r a r c h i t e c t u r e . T h e CPU a n d t h e

Display

Cache CPU

VRAM

Disk RAM

Keyboard, mouse, etc. NIC Computer

2.27 FIGURE

Message transfers

The left panel gives a simple architecture of a host computer and its connection to the network. The right panel shows that four copies m a y be involved across the CPU b u s to r u n an application, reducing the host throughput.

Network Services and Layered Architectures

m a i n m e m o r y c o m m u n i c a t e over t h e CPU b u s ; t h e r e also is a n I / O b u s for c o m m u n i c a t i o n w i t h t h e n e t w o r k . A d e d i c a t e d N e t w o r k Interface Card (NIC) i m p l e m e n t s m a n y of t h e functions dealing w i t h t h e physical t r a n s m i s s i o n a n d r e c e p t i o n of packets. T h e p a n e l o n t h e right s h o w s t h a t four copies of a file across t h e CPU b u s are n e e d e d w h e n t h e file is t r a n s f e r r e d b y FTP (File Transfer Protocol), studied in C h a p t e r 4. T h e file, initially in u s e r space in m e m o r y , is first copied into s y s t e m space, w h e r e it is h a n d l e d b y FTP (1). T h e n FTP copies t h e file over to t h e next l a y e r protocol TCP (Transmission Control Protocol) (2). T h e TCP protocol e n t i t y fragments t h e file into IP ( I n t e r n e t Protocol) packets. T h e CPU n o w c o m p u t e s t h e CRC bits of e a c h packet, w h i c h r e q u i r e s a third copy (3). Finally, e a c h c o m p l e t e d IP p a c k e t is forwarded to t h e NIC (4). T h e NIC t r a n s m i t s t h e p a c k e t over t h e n e t w o r k . If t h e CPU b u s h a s a t h r o u g h p u t of 320 m e g a b y t e s p e r s e c o n d (MBps) (80-MHz clock a n d 4-byte-wide b u s ) , t h e four copies h a v e r e d u c e d this to 80 MBps. We h a v e i n t r o d u c e d t h e c o n c e p t of l a y e r e d a r c h i t e c t u r e s . We n o w describe a v e r y useful architecture, t h e O p e n Data N e t w o r k or ODN m o d e l .

2.8

OPEN DATA NETWORK MODEL T h e O p e n Data N e t w o r k or ODN m o d e l w a s r e c e n t l y p r o p o s e d b y a p a n e l of n e t w o r k e n g i n e e r s as a f r a m e w o r k w i t h i n w h i c h t h e t e l e p h o n e , computer, a n d CATV n e t w o r k s c a n b e located a n d c o m p a r e d . T h e p u r p o s e of t h e ODN m o d e l is n o t to d e v e l o p a s t a n d a r d like t h e OSI m o d e l studied in C h a p t e r 3, b u t to h e l p u n d e r s t a n d h o w it m a y b e possible to i n t e r c o n n e c t t h e s e t h r e e t y p e s of n e t w o r k s , despite t h e differences in t h e i r technologies, services, a n d m a r k e t s . T h e ODN m o d e l is displayed in Figure 2.28. It h a s four layers, called bit ways, bearer, middleware, a n d applications, as s h o w n o n t h e left side of t h e figure. E x a m p l e s of i m p l e m e n t a t i o n s of t h o s e l a y e r s in specific n e t w o r k s are listed o n t h e right. T h e service provided b y a bit way is t h e t r a n s p o r t of bit s t r e a m s over a link. T h e bit w a y m a y b e i m p l e m e n t e d b y a SONET link of t h e t e l e p h o n e n e t w o r k , b y a direct b r o a d c a s t satellite or DBS link, b y a CATV link from t h e h e a d station to a user, b y a cellular radio c h a n n e l , or b y s o m e o t h e r wireless c o n n e c t i o n . T h e bit w a y provided b y a specific link t e c h n o l o g y c a n b e c h a r a c t e r i z e d b y its speed, delay, a n d e r r o r rate. A bearer service is t h e end-to-end t r a n s p o r t of bit s t r e a m s in specific for­ m a t s . For example, in t h e ATM b e a r e r service, 53-byte cells are t r a n s p o r t e d

2.8

Open Data Network Model

end-to-end over virtual circuits. T h e format of t h e cells is fixed. In t h e n e t w o r k or IP l a y e r of t h e I n t e r n e t , t h e b e a r e r service t r a n s p o r t s variable-sized data­ g r a m s in a specified format from s o u r c e to d e s t i n a t i o n over a packet-switched n e t w o r k . As w e will see, t h e b e a r e r services l a y e r is t h e m o s t i m p o r t a n t l a y e r from t h e v i e w p o i n t of n e t w o r k i n t e r c o n n e c t i v i t y . T h e middleware services are g e n e r i c services t h a t are u s e d b y a large n u m b e r of applications. E x a m p l e s of s u c h m i d d l e w a r e services are file transfer, directories, a n d video servers. T h e s e services could b e p r o v i d e d b y individual u s e r c o m p u t e r s . This is i n d i c a t e d in Figure 2.29, w h i c h s h o w s m i d d l e w a r e services b e i n g p r o v i d e d b y t h e o p e r a t i n g s y s t e m . In a c a m p u s e n v i r o n m e n t , however, e c o n o m i e s of scale m o t i v a t e t h e installation of d e d i c a t e d s e r v e r s for directories a n d o t h e r databases, software "warehouses," a n d t h e like. For s o m e m i d d l e w a r e services, t h e e c o n o m i e s of scale are e v e n greater, a n d this m a y justify installing special h a r d w a r e a n d software in n e t w o r k n o d e s to provide t h o s e services. For example, cellular p h o n e s t a n d a r d s like GSM a n d IS54 c o m p r e s s a voice signal into a 16-Kbps bit s t r e a m . T h e t e l e p h o n e c o m p a n y switches c o n t a i n h a r d w a r e t h a t d e c o m p r e s s e s t h o s e bit s t r e a m s to r e c o v e r t h e original voice signal. (If this w e r e n o t d o n e , o n l y t e l e p h o n e sets e q u i p p e d w i t h d e c o m p r e s s i o n h a r d w a r e could receive t h e c o m p r e s s e d signal.) Similarly, CATV n e t w o r k s m a y t r a n s p o r t a c o m p r e s s e d video signal to t h e curbside w h e r e it is d e c o m p r e s s e d before distribution to individual u s e r s .

Network Services and Layered Architectures

Hostl

Host 2

User

Application "Connections" or "datagrams" Routers

Kernel

Middleware

Bearer services

Packets, messages Bearer

NIC

Bits Bit ways EM waves

UNI

2.29

UNI

Networks with the same bearer services can be connected via routers. The simpler the bearer services, the easier the network interconnection.

Finally, t h e application l a y e r provides t h e services t h a t u s e r s w a n t . Exam­ ples i n c l u d e e-mail, WWW (World Wide Web), video o n d e m a n d , a n d video­ conferencing. As indicated in Figure 2.29, t h e application l a y e r is usually i m p l e m e n t e d in t h e h o s t c o m p u t e r . T h i s often involves p r o p r i e t a r y software, s u c h as Eudora for e-mail a n d Intuit's Q u i c k e n for h o m e b a n k i n g . S o m e t i m e s specialized h a r d w a r e is n e e d e d to r u n t h e application. T h u s , "pay TV" r e q u i r e s a "set-top box" t h a t u n s c r a m b l e s or d e c o m p r e s s e s a TV signal. T h e set-top box h e l p s create a record for billing p u r p o s e s . Because applications often involve p r o p r i e t a r y a n d expensive "progam c o n t e n t " (e.g., a m o v i e ) as well as t r a n s p o r t services, provision m u s t b e m a d e for billing for this p r o g r a m c o n t e n t . M u c h profit will b e m a d e b y h a r d w a r e a n d software m a n u f a c t u r e r s w h o s e set-top boxes b e c o m e a de facto standard. If a m a n u f a c t u r e r ' s e q u i p m e n t is widely a d o p t e d b y users, p r o g r a m c o n t e n t providers will h a v e a n i n c e n t i v e to conform to t h a t e q u i p m e n t , a n d t h e e q u i p m e n t m a n u f a c t u r e r c a n t h e n extract a m o n o p o l y r e n t from t h e c o n t e n t providers for u s e of t h a t e q u i p m e n t . In 1995 t h e c o n t e n d e r s for this set-top "prize" i n c l u d e d t h e large software

2.9

Network Architectures

c o m p a n i e s , s u c h as Microsoft a n d Sybase, a n d t h e t e l e p h o n e c o m p a n i e s . O n t h e o t h e r side, c o m p a n i e s s u c h as Sun Microsystems, w h i c h h a v e n o t y e t d e v e l o p e d competitive products, are lobbying for g o v e r n m e n t r e g u l a t i o n t h a t will m a i n t a i n a n "open" application layer. We c a n n o w see h o w different l a y e r s c o o p e r a t e to p r o d u c e sophisticated u s e r services. A n application m a y m a k e u s e of m i d d l e w a r e services s u c h as file transfer. File transfer service in t u r n is i m p l e m e n t e d u s i n g a b e a r e r service s u c h as IP d a t a g r a m t r a n s p o r t . A n d d a t a g r a m t r a n s p o r t is i m p l e m e n t e d u s i n g a bit w a y p r o v i d e d b y a local a r e a n e t w o r k s u c h as E t h e r n e t . Finally, w e s t u d y n e t w o r k i n t e r c o n n e c t i v i t y . O b s e r v e t h a t t h e ODN m o d e l of Figure 2.28 h a s a n a r r o w waist at t h e b e a r e r service layer. T h i s is i n t e n d e d to suggest t w o things. First, a v e r y small set of b e a r e r services c a n b e p r o v i d e d b y a large variety of bit w a y i m p l e m e n t a t i o n s . Second, t h e small set of b e a r e r services is sufficiently versatile to s u p p o r t a large variety of applications. T h e small set of b e a r e r services greatly p r o m o t e s i n t e r n e t w o r k i n g , as c a n b e s e e n w i t h t h e h e l p of Figure 2.29. T h e figure s h o w s two hosts, Host 1 a n d H o s t 2, b e l o n g i n g to t w o s e p a r a t e n e t w o r k s . T h e s e t w o n e t w o r k s c a n b e i n t e r c o n n e c t e d b y a r o u t e r or switch, p r o v i d e d b o t h n e t w o r k s s u p p o r t t h e s a m e b e a r e r service. ( T h e r o u t e r is a t t a c h e d to t h e bit w a y s of b o t h n e t w o r k s . ) T h e application r u n n i n g o n H o s t 1 p r o d u c e s a bit s t r e a m t h a t is r e c o v e r e d b y t h e r o u t e r u s i n g t h e b e a r e r services of t h e first n e t w o r k . T h e r o u t e r t h e n forwards t h a t bit s t r e a m to Host 2 u s i n g t h e b e a r e r services of t h e s e c o n d n e t w o r k . We t h u s see t h a t t h e s i m p l e r t h e set of b e a r e r services, t h e easier it will b e to i n t e r c o n n e c t n e t w o r k s . T h e m o s t i m p o r t a n t e x a m p l e s of this are t h e I n t e r n e t Protocol (IP) a n d ATM, b o t h of w h i c h specify a single b e a r e r service. O n t h e o t h e r h a n d , t h e set of b e a r e r services s h o u l d b e versatile e n o u g h to s u p p o r t a w i d e r a n g e of applications. We will see t h a t in this r e s p e c t ATM is b e t t e r t h a n IP.

2.9

NETWORK ARCHITECTURES In this section w e discuss n e t w o r k a r c h i t e c t u r e s . O u r objective is to explain t h e interoperability of t h e different technologies. Figure 2.30 s h o w s a collection of t e l e p h o n e sets, c o m p u t e r s , a n d p e r s o n a l digital assistants (PDAs). (Television sets w o u l d fit i n this collection as well.) T h e s e devices are i n t e r c o n n e c t e d w i t h links t h a t are wireless, c o p p e r pairs, coaxial cables, or optical fibers. T h e s e l i n k s t r a n s p o r t d a t a a n d voice (possibly video) e i t h e r as bits in T D M frames, or in ATM cells, or in IP packets.

Network Services and Layered Architectures

PC PC

Ph

Ph

Ph

PC

PC

PC

Ph

LH

PC

Ph

D

PC

Ph

D

HiHi

Ξ-Θ-L* W Ph A C Cp D L PC pda

2.30 FIGURE

= ATM switch = Cable = Cellular phone =DSL = LAN = Computer = PDA

Ph Ph R S Τ W Wr

= Telephone = Router = SONET - TDM = Wireless base station = Wireless router

Networks combine m a n y different technologies. The power of the TCP/IP protocols is in enabling their interoperability on a large scale.

T h e local t e l e p h o n e lines u s e a T D M t e c h n o l o g y t h a t is n o t globally syn­ chronized. Long-distance t e l e p h o n e l i n e s u s e t h e s y n c h r o n i z e d t e c h n o l o g y over fibers (SONET). ATM cells c a n b e t r a n s p o r t e d over c o p p e r l i n e s e i t h e r directly if t h e lines are short or using DSL m o d e m s if t h e y are longer. ATM cells c a n also b e t r a n s p o r t e d over optical links. W h e n t h e fiber is short, a s i m p l e t r a n s m i s s i o n s c h e m e c a n b e used. W h e n t h e fiber is longer, o n e c a n u s e t h e SONET framing, clock recovery, a n d error detection, w i t h o u t h a v i n g to s y n c h r o n i z e t h e link. This u s e of SONET, s o m e t i m e s called SONET-Light, is b e c o m i n g p o p u l a r in 1999. If t h e t r a n s m i s s i o n is over a long distance, t h e ATM cells c a n b e s e n t over a SONET "circuit" (the p r o p e r t e r m is SONET path) to t h e n e x t ATM switch. T h i s c o m b i n a t i o n is t h e n called ATM over SONET. IP packets c a n b e t r a n s p o r t e d b e t w e e n PCs a n d r o u t e r s over a LAN, over a cable ( e q u i p p e d w i t h cable m o d e m s ) , over wireless links, or over optical fibers (using SONET-Light or SONET). ATM cells c a n also t r a n s p o r t t h e IP packets.

2.10

Network Bottlenecks

All t h e s e c o m b i n a t i o n s c a n s e e m excessively c o m p l i c a t e d . T h e i r coexis­ t e n c e is explained b y t h e n e e d to s u p p o r t different "legacy" s y s t e m s . For in­ stance, SONET w a s d e v e l o p e d to m a k e t h e long-distance t e l e p h o n e e q u i p m e n t simpler. Since this long-distance digital n e t w o r k is available, it m a k e s s e n s e to u s e it to t r a n s p o r t ATM cells a n d IP packets. O n t h e o t h e r h a n d , IP n e t w o r k s in t h e form of i n t r a n e t s a n d t h e I n t e r n e t are s o m e w h a t capable of t r a n s m i t t i n g voice. O n e t h e n e n d s u p w i t h voice over IP over ATM over SONET, w h i c h is i n d e e d a "baroque monstrosity," to q u o t e a r e s p e c t e d IP researcher. We discuss t h e plausible evolution of this t e c h n o l o g y in C h a p t e r 13.

2.10

NETWORK BOTTLENECKS Before w e e m b a r k o n a detailed s t u d y of n e t w o r k s , w e p a u s e to c o m m e n t o n t h e c u r r e n t major t e c h n o l o g y b o t t l e n e c k s in a c h i e v i n g a h i g h - p e r f o r m a n c e n e t w o r k . T h e rest of t h e b o o k elaborates o n t h e issues t h a t w e i n t r o d u c e in this brief section.

/s Bandwidth Plentiful and Free? T h e t r a n s m i s s i o n rate of links is rapidly increasing. Optical links c a n t r a n s m i t at 10 Gbps or m o r e over a b o u t 100 k m . Moreover, w i t h wave-division multiplexing, or WDM, a single fiber c a n c a r r y a large n u m b e r (16 to 64 c o m m e r c i a l l y , u p to 512 in t h e laboratory) of s u c h fast t r a n s m i s s i o n s . C o n s e q u e n t l y , a t r a n s m i s s i o n link w i t h 1 T b p s (equal to 1 0 b p s ) is feasible b e t w e e n two cities. If o n e large city h a s o n e h u n d r e d t h o u s a n d u s e r s t h a t are s i m u l t a n e o u s l y active, e a c h u s e r c a n in principle get a t r a n s m i s s i o n rate of 10 Mbps, w h i c h is m o r e t h a n a d e q u a t e , as w e k n o w from o u r e x p e r i e n c e w i t h local a r e a n e t w o r k s t h a t t r a n s m i t at s u c h a rate. This s i m p l e discussion a p p e a r s to justify a c o m m o n l y h e l d belief t h a t b a n d w i d t h is or will s o o n b e a n a b u n d a n t c o m m o d i t y . 1 2

A closer s t u d y of n e t w o r k o p e r a t i o n s reveals a m o r e c o m p l e x situation. Al­ t h o u g h t h e b a c k b o n e n e t w o r k is built w i t h optical fibers, m a n y access links still u s e c o p p e r or s h a r e d coaxial cable w h o s e t r a n s m i s s i o n r a t e s are c o m p a r a t i v e l y slow. Another, m o r e subtle source of difficulty is t h a t m a n y s o u r c e s of data traf­ fic are g r e e d y a n d t r y to fill u p t h e available b a n d w i d t h , as w e explain w h e n w e s t u d y TCP. C o n s e q u e n t l y , i n c r e a s i n g t h e b a n d w i d t h does n o t e l i m i n a t e t h e c o m p e t i t i o n for it n o r t h e delays a n d losses it entails. Moreover, in o r d e r to exploit t h e vast t r a n s m i s s i o n rate of optical links in t h e b a c k b o n e , switches,

Network Services and Layered Architectures

routing protocols, t r a n s p o r t protocols, a n d applications m u s t b e i m p r o v e d . T h e m a i n objective of this b o o k is to explain t h e s e i m p r o v e m e n t s .

Switch Modifications T h e switch m u s t m a k e a forwarding decision for e a c h p a c k e t t h a t arrives a n d m u s t forward t h e p a c k e t to t h e c o r r e s p o n d i n g o u t p u t port. Better look­ u p algorithms c a n s p e e d u p t h e forwarding decision. N e w switch a r c h i t e c t u r e s s p e e d u p t h e forwarding of packets. Different applications h a v e vastly different delay, t h r o u g h p u t , a n d loss r e q u i r e m e n t s . Switches are d e s i g n e d t h a t c a n h a n d l e packets according to t h e s e r e q u i r e m e n t s .

New Routing Protocols Tb simplify t h e task of t h e switches, n e w protocols are b e i n g designed. T h e s e protocols label t h e p a c k e t s according to t h e r e q u i r e m e n t s of t h e application a n d possibly b a s e d o n s o m e p r e c o m p u t e d r o u t i n g decisions. It is s i m p l e r for t h e switch to u s e t h e label t h a n t h e d e s t i n a t i o n a d d r e s s a n d a n application identification to d e t e r m i n e h o w it s h o u l d h a n d l e t h e packet. As t h e n u m b e r of subscribers to "push" p r o g r a m s increases, multicast r o u t i n g m a y b e c o m e a n i m p o r t a n t m e t h o d to i m p r o v e t h e efficiency of t h e routing.

Transport Protocols T h e e n d h o s t s (the source a n d t h e destination) control t h e p a c i n g of p a c k e t t r a n s m i s s i o n s . As t h e links get faster, t h e n u m b e r of p a c k e t s in transit in­ creases, w h i c h modifies t h e d y n a m i c s of t h e control m e c h a n i s m s . T h e control m e c h a n i s m s probably n e e d to b e modified to m a t c h t h e s e c h a n g i n g o p e r a t i n g conditions.

Applications Faster n e t w o r k s m a k e n e w applications possible. W h e r e a s a few y e a r s ago m o s t of t h e applications w e r e text-based, today's applications involve m u l t i m e d i a . As t h e n e t w o r k s p e e d c o n t i n u e s to increase, w e will see m o r e c o n v e r s a t i o n applications a n d Web sites w i t h r i c h e r c o n t e n t . More p e o p l e will u s e live radio a n d video p r o g r a m s . T e l e c o m m u t i n g a n d residential v i d e o c o n f e r e n c i n g m a y become more commonplace. It m a y b e t h a t for s o m e t i m e to c o m e , t h e s e n e w applications will m a i n t a i n t h e i r pace of increasing t h e traffic to t h e capacity of t h e n e t w o r k . If this predic­ tion h o l d s true, t h e n optimization of r e s o u r c e utilization will r e m a i n critical. Pricing of services m a y b e c o m e n e c e s s a r y to p r o t e c t essential applications.

2.13

Problems

2.11

93

SUMMARY N e t w o r k s provide c o m m u n i c a t i o n services n e e d e d to s u p p o r t u s e r applica­ tions. T h e s e services are p r o v i d e d from m o r e e l e m e n t a r y services in a l a y e r e d architecture, like t h e O p e n Data N e t w o r k m o d e l . S o m e of t h e l a y e r s are i m p l e ­ m e n t e d in t h e n e t w o r k switches, o t h e r s in h o s t c o m p u t e r s . T h e n e t w o r k itself provides b e a r e r services, t h a t is, t h e end-to-end t r a n s p o r t of bit s t r e a m s . T h e b e a r e r services are i m p l e m e n t e d b y n e t w o r k links a n d switches u s i n g m e c h ­ a n i s m s of multiplexing, switching, e r r o r control, flow control, a n d r e s o u r c e allocation. Applications g e n e r a t e c o n s t a n t or variable bit rate traffic or m e s s a g e ex­ c h a n g e s . T h e applications i m p o s e c e r t a i n r e q u i r e m e n t s o n t h e b e a r e r ser­ vices t h a t t r a n s p o r t this traffic. T h o s e r e q u i r e m e n t s are e x p r e s s e d in t e r m s of b a n d w i d t h , delay, a n d e r r o r rates. Circuit switching m e e t s t h e m o s t s t r i n g e n t r e q u i r e m e n t s in t e r m s of d e l a y a n d b a n d w i d t h b u t m a y l e a d to s u c h poor utili­ zation t h a t it b e c o m e s u n e c o n o m i c a l . D a t a g r a m switching is t h e m o s t efficient, b u t it m a y n o t b e able to provide g u a r a n t e e d d e l a y or b a n d w i d t h . Virtual circuit switching c a n c o m b i n e high utilization w i t h t h e ability to m e e t g u a r a n t e e s in delay a n d b a n d w i d t h . In C h a p t e r 3 w e s t u d y p a c k e t switching, in C h a p t e r 5 w e s t u d y circuit switching, a n d in C h a p t e r 6 w e s t u d y ATM n e t w o r k s , t h e m o s t i m p o r t a n t t y p e of virtual circuit switching n e t w o r k .

2.12

NOTES S h a n n o n ' s t h e o r e m s , w h i c h form t h e basis of i n f o r m a t i o n t h e o r y , are discussed in m a n y texts; see [CT91]. T h e n o t i o n of l a y e r e d architecture, m o d u l a r i t y , a n d h i e r a r c h y s h o w s u p in various p a r t s of c o m p u t e r s c i e n c e as well as in c o m m u n i c a t i o n n e t w o r k s ; see [T88, W98]. For details o n i m p l e m e n t a t i o n of protocols in UNIX, see [P93]. T h e O p e n Data N e t w o r k m o d e l a p p e a r s in [Kle94].

2.13

PROBLEMS 1. It is v e r y expensive to store a n d archive X r a y s for m e d i c a l diagnosis. T h e c u r r e n t s y s t e m u s e s large p h o t o g r a p h s . A n electronic i m a g e of c o m p a r a b l e

Network Services and Layered Architectures

quality w o u l d r e q u i r e a display of 1,000 χ 1,000 pixels, w i t h a 16- or 24bit grayscale, a n d w i t h a zoom-in capability. H o w expensive is a m o n i t o r capable of displaying so m u c h information? ( H o w m a n y pixels d o e s y o u r m o n i t o r display, a n d h o w m a n y grayscale levels does it p e r m i t ? ) If a radiologist is retrieving t h e X r a y from a n archive, t h e acceptable d e l a y is 10 s, say. W h a t s h o u l d b e t h e bit rate of t h e links c o n n e c t i n g t h e archive to t h e radiologist's office? If t h e X rays t h a t t h e radiologist is going to v i e w are k n o w n , say 10 m i n in advance, o n e could retrieve t h e X r a y s early a n d buffer t h e m locally. T h i s p e r m i t s a r e d u c t i o n in t h e bit rate at t h e cost of increasing local storage. W h a t is t h e bit r a t e / b u f f e r trade-off? Suppose t h e cost of i n c r e a s i n g t h e bit rate b y 1 Mbps is r t i m e s t h e cost of 1 Mbps of disk. For w h a t values of r is it w o r t h r e d u c i n g t h e link rate a n d increasing local disk storage? 2. T h e figure b e l o w illustrates t h e OSI seven-layer m o d e l for c o m m u n i c a t i o n b e t w e e n n o d e s A a n d C. T h e t w o i n t e r m e d i a t e "?" n o d e s are also s h o w n . Application Presentation Session Transport Network Data link Physical

T h e next figure shows t h r e e possibilities for t h e "?" n o d e s .

(a)

(b)

(c)

W h a t k i n d of device is s h o w n in (a), (b), a n d (c) of this

figure?

3. (a) Consider t h e n e t w o r k topology d e p i c t e d in t h e figure o n t h e n e x t page w h e r e t h e boxes labeled w i t h "?" are E t h e r n e t h u b s . (1) Is this a valid topology? W h y or w h y not?

2.13

Problems

95 (2) If this is a valid topology, c a n p a c k e t s e v e r circulate a r o u n d t h e loop? (3) If p a c k e t s e v e r do circulate a r o u n d t h e loop, w h a t k e e p s t h e m from circulating forever? (b) A s s u m e t h a t t h e boxes l a b e l e d w i t h " ? " are E t h e r n e t switches. A n s w e r t h e s a m e sub-questions from (a). (c) A s s u m e t h a t t h e boxes l a b e l e d w i t h "?" are I n t e r n e t Protocol routers. A n s w e r t h e s a m e s u b - q u e s t i o n s from (a).

a -)

"7"

Hub

Hub

(

Ό

4. As s e e n in Figure 2.1, w e m a y r e p r e s e n t t h e i n t e r c o n n e c t i o n of n e t w o r k e l e m e n t s as a n e t w o r k g r a p h w h o s e edges are t h e l i n k s a n d w h o s e vertices are t h e switches a n d u s e r n o d e s . Formally, w e r e p r e s e n t t h e n e t w o r k as a g r a p h G = (V, E) w h e r e V = {1, 2 , . . . , N] is t h e set of vertices a n d Ε C V χ V is t h e set of edges. T h e i n t e r p r e t a t i o n is t h a t (ι,;') e Ε if t h e r e is a o n e - w a y t r a n s m i s s i o n l i n k from switch i to switch;'. T h i s r e p r e s e n t a t i o n is useful to specify a n d verify m a n y n e t w o r k a l g o r i t h m s . Suppose t h e r e is a cost Cy > 0 associated w i t h e a c h l i n k (i,;) e Ε r e p r e ­ s e n t i n g d e l a y or dollar cost of t r a n s m i t t i n g o n e p a c k e t o v e r t h a t link. We a s s u m e t h a t t h e k n o w l e d g e of Cy is local, t h a t is, r o u t e r ι k n o w s o n l y t h e costs Cy of links t h a t originate at i. T h e p r o b l e m is to b u i l d a s h o r t e s t - p a t h r o u t i n g table at e a c h router. T h e table at i h a s t h e following form:

Destination router

Outgoing l i n k

Cost to destination

1

h

2

h

D

Ν

JN

AN

i2

T h e first r o w in t h e table is i n t e r p r e t e d like this. If a p a c k e t for d e s t i n a t i o n #1 arrives at r o u t e r i, it s h o u l d b e forwarded over l i n k (ι,;Ί). T h e m i n i m u m

Network Services and Layered Architectures

cost i n c u r r e d b y this packet from r o u t e r i o n w a r d is Dn. T h e o t h e r rows are i n t e r p r e t e d similarly. In o r d e r t h a t e a c h r o u t e r b e able to c o n s t r u c t its o w n table, it will n e e d to exchange s o m e information w i t h its neighbors. (a) C o n s t r u c t a distributed algorithm in w h i c h e a c h r o u t e r iterates over a two-phase cycle. In t h e first p h a s e it u p d a t e s its e s t i m a t e of t h e table; in t h e s e c o n d p h a s e it c o m m u n i c a t e s its table w i t h its m i n i m u m cost e s t i m a t e to its i m m e d i a t e u p s t r e a m neighbors. H o w is t h e table initialized? Show t h a t y o u r algorithm converges after a finite n u m b e r of iterations. (b) Obtain a n u p p e r b o u n d o n t h e n u m b e r of iterations n e e d e d for con­ vergence. (c) Is t h e r e a n y formal w a y in w h i c h y o u c a n say t h a t t h e i n f o r m a t i o n t h a t is e x c h a n g e d b e t w e e n t h e r o u t e r s is t h e minimum a m o u n t of information t h a t m u s t b e exchanged? 5. T h e figure b e l o w s h o w s two a p p r o a c h e s for i n t e r c o n n e c t i n g n o d e s : t h e fully c o n n e c t e d m o d e l a n d t h e shared-resource m o d e l . (a) If w e u s e t h e fully c o n n e c t e d m o d e l a n d w e w i s h to c o n n e c t η n o d e s , h o w m a n y links will w e r e q u i r e ? (b) W h a t if w e u s e t h e shared-resource m o d e l instead. U n f o r t u n a t e l y w e can't b u y a single h u b w i t h η ports; all w e c a n p u r c h a s e are six-port h u b s . Approximately, h o w m a n y h u b s a n d links will w e r e q u i r e to connect η nodes?

(a)

(b)

6. Consider t h e r i n g t o p o l o g y d e p i c t e d in t h e figure at t h e top of t h e n e x t page. This topology h a s t h e p r o p e r t y t h a t all n o d e s c a n still c o m m u n i c a t e e v e n w h e n t h e r e is a single link failure. Let's a s s u m e t h a t t h e links in t h e n e t w o r k fail i n d e p e n d e n t l y a n d w i t h probability ρ in (0, 1). (a) What is t h e probability t h a t all n o d e s c a n c o m m u n i c a t e ? A n s w e r this p a r t for t h e case of Ν = 6.

2.13

Problems

97 Hint: re-express this q u e s t i o n in t e r m s of t h e n u m b e r of links t h a t h a v e failed.

(b) A n s w e r t h e p r e v i o u s p a r t for a n y N. (c) If w e tell y o u t h a t two of t h e links h a v e failed, w h a t is t h e probability t h a t n o d e s A & Β c a n c o m m u n i c a t e ? A n s w e r this p a r t for t h e exact configuration s h o w n in t h e figure (i.e., Ν = 6 a n d Α δ ' Β l o c a t e d w h e r e indicated in t h e figure). (d) Let D b e t h e m i n i m u m distance p a t h from A to B. W h a t are t h e possible values for D? W h a t is t h e probability distribution of D? A n s w e r this p a r t for t h e exact configuration s h o w n in t h e figure. 7. Consider t h e n e t w o r k topology d e p i c t e d in t h e figure below. Let's a s s u m e t h a t t h e links in t h e n e t w o r k fail i n d e p e n d e n t l y a n d w i t h probability ρ in (0, 1)· (a) H o w m a n y links m a y fail before n e t w o r k c o n n e c t i v i t y is lost? (b) W h a t is t h e probability of losing n e t w o r k c o n n e c t i v i t y ? A n s w e r this p a r t w i t h Ν = 6. (c) A n s w e r t h e p r e v i o u s p a r t for a n y N.

8. In this p r o b l e m w e c o n s i d e r t h e topologies of t h e p r e v i o u s two figures. Unlike t h e p r e v i o u s p r o b l e m s , t h e r a n d o m n e s s h e r e is in t h e choice of n o d e s A & B. (All links are a s s u m e d to b e functioning.) Let Ν = 6. T h e n o d e s A & Β are i n d e p e n d e n t l y c h o s e n at r a n d o m a n d w i t h e q u a l probability, (i.e., as if w e rolled a fair die to select A a n d t h e n rolled again to select B.) (a) Consider t h e topology of t h e figure in P r o b l e m 6. Let D b e t h e distance b e t w e e n n o d e s A & B. W h a t is t h e e x p e c t e d v a l u e of D? (b) Repeat for t h e figure in P r o b l e m 7.

Network Services and Layered Architectures

(c) What part(s) of y o u r solution to t h e first p a r t w e r e y o u able to r e u s e in the second part? Why? 9. What is t h e u n c o m p r e s s e d bit rate for t h e NTSC TV signal? W h a t c o m p r e s ­ sion ratio is a c h i e v e d b y MPEG1? 10. W h a t is t h e p r o p a g a t i o n delay of a l i n k from a n e a r t h station to a geostation­ a r y satellite? W h a t w o u l d b e t h e end-to-end delay of a voice c o n v e r s a t i o n t h a t is r e l a y e d via s u c h a satellite? 11. A v e r y c o m m o n w a y to c o m p r e s s a n audio signal is t h e following: S a m p l e t h e a u d i o signal. D e n o t e t h e s a m p l e s b y xo, x\, . . . . T r a n s m i t XQ. T h e n t r a n s m i t x\ — xo, X2 — x\,.... T h e d y n a m i c r a n g e of t h e s a m p l e differences, xt — is m u c h s m a l l e r t h a n t h e s a m p l e s t h e m s e l v e s , so t h e s a m p l e differences c a n b e c o d e d into fewer bits to a c h i e v e t h e s a m e q u a n t i z a t i o n noise. Develop a m a t h e m a t i c a l m o d e l t h a t s h o w s this. 12. C o n s i d e r a video source t h a t p r o d u c e s a periodic VBR s t r e a m w i t h t h e following "on-off s t r u c t u r e . T h e source is "on" for 1 s w i t h a r a t e of 20 Mbps; it is t h e n "off for 2 s w i t h a bit rate of 1 Mbps. (a) What are t h e p e a k a n d average r a t e s of this source? S u p p o s e this s o u r c e is served at a c o n s t a n t b a n d w i d t h of c Mbps, w h e r e c is larger t h a n t h e average rate. Calculate t h e buffer size b = b(c) in MB n e e d e d to p r e v e n t a n y loss as a function of c. W h a t is t h e q u e u i n g d e l a y as a function of c? (b) If this traffic is p r o d u c e d b y a v i d e o c o n f e r e n c i n g application, w h i c h p e r m i t s a m a x i m u m delay of 200 m s , w h a t s h o u l d c b e ? (c) If this traffic is p r o d u c e d b y a video s e r v e r t h a t is d o w n l o a d i n g a one-hour-long video p r o g r a m , h o w m u c h disk storage do y o u n e e d ? Suppose y o u w a n t to play t h e p r o g r a m , b u t y o u r disk access r a t e is o n l y 10 Mbps. H o w m a n y parallel disks w o u l d y o u n e e d , a n d h o w w o u l d y o u store t h e video p r o g r a m o n t h e disk? (d) In t h e description above, t h e p e r i o d of t h e source is 3 s, a n d t h e r e is a d u t y cycle of 1/3. Consider a n o t h e r source w i t h t h e s a m e d u t y cycle b u t w i t h a s m a l l e r period. Would y o u say t h e s e c o n d s o u r c e is m o r e or less b u r s t y ? W h y ? 13. T h e analog p h o n e access line h a s a b a n d w i d t h of 4 k H z . T h e line c a n b e u s e d to t r a n s m i t digital voice at 64 Kbps or, u s i n g a m o d e m , to t r a n s m i t data at 9.6 or 11.4 Kbps. W h a t is t h e spectral efficiency in e a c h case? W h a t k i n d of m o d u l a t i o n s c h e m e w o u l d y o u u s e to i n c r e a s e spectral efficiency?

2.13

Problems

99

14. Give e x a m p l e s of applications t h a t c a n l e a d to m u l t i p l e x i n g gain r a n g e s displayed in Figure 2.7. 15. Explain w h y error-correction s c h e m e s are u s e d (instead of error d e t e c t i o n followed b y r e t r a n s m i s s i o n ) in data storage applications ( s u c h as audio CDs a n d m a g n e t i c disks) a n d in real-time applications (e.g., controlling a satellite). 16. A code w i t h m i n i m u m distance 5 c a n (a) (b) (c) (d) (e)

correct u p to 3 bit errors. d e t e c t u p to 4 bit errors. correct u p to 2 bit errors. d e t e c t u p to 3 bit errors. d e t e c t u p to 5 bit errors.

17. Packet switching is m o r e efficient t h a n circuit switching for b u r s t y traffic because (a) (b) (c) (d)

t h e switching delay is s m a l l e r for p a c k e t switches. p a c k e t switching u s e s faster links. b a n d w i d t h is n o t r e s e r v e d w h e n it is n o t n e e d e d . circuit switching u s e s slow m o d e m s .

18. T h e ABP protocol is u s e d w i t h p a c k e t s of size n. T h e t r a n s m i s s i o n link h a s a BER of p. ( A s s u m e t h e a c k n o w l e d g m e n t s are r e c e i v e d e r r o r free.) W h a t is t h e average n u m b e r of p a c k e t t r a n s m i s s i o n s p e r correctly r e c e i v e d packet? 19. Write a s i m u l a t i o n of ABP w i t h a loss probability of p. Plot t h e t h r o u g h p u t as a function οϊ ρ for a few v a l u e s of t h e ratio of t h e p r o p a g a t i o n t i m e over the transmission time. 20. Consider a 10,000-km r o u n d - t r i p r o u t e w i t h a t r a n s m i s s i o n Suppose a p r o p a g a t i o n t i m e of 5 / z s / k m . C o n s i d e r a p a c k e t H o w m a n y p a c k e t s are n e e d e d to fill u p t h e links along is t h e m i n i m u m w i n d o w size in t h e Go Back Ν protocol efficiency?

r a t e of 100 Mbps. size of 1,000 bits. the route? What to achieve 100%

21. Consider a t r a n s m i s s i o n of data from S to R below. T h e p a c k e t s h a v e a fixed size e q u a l to 1 KB. Each a c k n o w l e d g m e n t is in a p a c k e t w i t h 40 b y t e s (consisting of a n IP h e a d e r a n d a TCP h e a d e r ) . T h e p a c k e t s are t r a n s m i t ­ t e d reliably, a n d a c k n o w l e d g m e n t p a c k e t s are t r a n s m i t t e d correctly w i t h probability 0.95 each. Host R s e n d s b a c k a n ACK w i t h t h e next p a c k e t

Network Services and Layered Architectures

100

n u m b e r it expects. H o s t S h a s a t i m e o u t value e q u a l to a "round-trip t i m e " a n d i m p l e m e n t s Go Back Ν w i t h a w i n d o w size c o m p u t e d to "fill u p t h e pipe" exactly in t h e a b s e n c e of errors. T h e p r o p a g a t i o n t i m e is 1 m s in e a c h direction. Neglect processing t i m e s . 10 Mbps, 100% reliable 1-KB packets R

S 30 Kbps, PER = 5% 40-byteACKs

(a) W h a t is t h e w i n d o w size? (b) Calculate t h e average t h r o u g h p u t (in p a c k e t s p e r s e c o n d ) of t h e con­ n e c t i o n from S to R. 22. Consider t h e Go Back Ν protocol. Suppose t h a t t h e p a c k e t e r r o r probability is p. (Errors in different p a c k e t s are i n d e p e n d e n t . ) H o w w o u l d y o u calcu­ late t h e efficiency, a s s u m i n g t h a t Ν is c h o s e n so t h a t t h e p i p e is j u s t full? H o w will efficiency c h a n g e as Ν i n c r e a s e s ? 23. See t h e figure o n t h e next page. Node A is s e n d i n g a v e r y large (infinite?) a m o u n t of data to Node B. T h e link from A to Β h a s bit errors w i t h proba­ bility ρ in (0, 1), a n d t h e link from Β to A h a s bit errors w i t h probability q in (0, 1); t h e s e bit errors are m u t u a l l y i n d e p e n d e n t . A n e r r o r control pro­ tocol similar to Selective Repeat is b e i n g used. T h i s protocol is e q u i v a l e n t to Selective Repeat w i t h a n infinite w i n d o w size. Node A s e n d s data pack­ ets w i t h d data b y t e s a n d h h e a d e r b y t e s (d + h b y t e s total). Node Β s e n d s acknowledgment packets of a b y t e s . (a) W h a t is t h e probability t h a t a data p a c k e t is received correctly at n o d e B? (b) W h a t is t h e probability t h a t a data p a c k e t is received correctly b y n o d e Β a n d its s u b s e q u e n t a c k n o w l e d g m e n t is received correctly b y n o d e A? (c) W h a t is t h e e x p e c t e d n u m b e r of t i m e s a data p a c k e t will b e t r a n s m i t t e d b y n o d e A before it is successfully a c k n o w l e d g e d ? (d) Let's define efficiency to b e t h e fraction of t i m e n o d e A t r a n s m i t s useful data (i.e., n o t c o r r u p t e d o n t h e w a y to n o d e B, n o r a d u p l i c a t e c o p y previously delivered to n o d e B). H e a d e r s are n o t c o n s i d e r e d useful data. W h a t is t h e efficiency of this protocol over this link? (e) Does this analysis a p p l y to t h e s t a n d a r d Selective Repeat Protocol? If not, w h y ?

Β

A

Retransmission

24. Consider two t r a n s m i s s i o n links. O n e h a s a Bit Error Rate (BER) of 10~ ; t h e o t h e r h a s BER of 1 0 ~ . T h e p a c k e t s to b e t r a n s m i t t e d are 1,000 bits long. (We will ignore t h e a c k n o w l e d g m e n t s . ) A 32-bit CRC is a d d e d to t h e p a c k e t s to c h e c k for t r a n s m i s s i o n errors. T h e A R Q protocol is t h e A l t e r n a t i n g Bit Protocol (ABP) (i.e., t h e link b a n d w i d t h χ d e l a y p r o d u c t is small c o m p a r e d to t h e p a c k e t size). We will c o n s i d e r two different Forward Error Correction (FEC) codes. Both are all 31-bit BCH block codes; o n e u s e s 5 p a r i t y c h e c k bits to correct u p to o n e error, t h e o t h e r u s e s 10 p a r i t y c h e c k bits to correct u p to two errors. ( T h e first code l e a v e s 26 bits for u s e r data a n d CRC, t h e s e c o n d leaves 21 bits.) A n s w e r e a c h of t h e following q u e s t i o n s for t h e A R Q o n l y case, t h e FECI plus A R Q c a s e , a n d t h e FEC2 plus A R Q c a s e . 8

4

(a) H o w m a n y bits are t r a n s m i t t e d p e r p a c k e t ? (b) W h a t is t h e probability t h a t a p a c k e t is r e c e i v e d correctly if t h e first link is u s e d ? (c) W h a t is t h e probability t h a t a p a c k e t is r e c e i v e d correctly if t h e s e c o n d link is u s e d ? 25. Consider t h e r e t r a n s m i s s i o n protocol d e s c r i b e d in Figure 2.15. Suppose t h a t t h e a c k n o w l e d g m e n t following a correct t r a n s m i s s i o n arrives after a r a n d o m delay d. ( T h e r a n d o m n e s s is d u e to r a n d o m q u e u i n g d e l a y in t h e n e t w o r k . ) Let F b e t h e c u m u l a t i v e probability distribution of d, n a m e l y , F(t) = Prob{d < t}. T h u s t h e probability of receiving a n a c k n o w l e d g m e n t before t h e t i m e o u t is F ( T ) , a n d t h e probability of n o t receiving it before t h e t i m e o u t is 1 - F(T).

Network Services and Layered Architectures

102

(a) Show t h a t t h e e x p e c t e d n u m b e r of t i m e o u t s t h a t t h e s a m e p a c k e t is s e n t is 00

N(T) =

nF(T)[l

-

F(T)] ~ . n

l

n=l

Show t h a t N ( T ) d e c r e a s e s as Τ i n c r e a s e s . (b) Show t h a t t h e e x p e c t e d t i m e τ ( Τ ) before a n a c k n o w l e d g m e n t is re­ ceived before a t i m e o u t is b e t w e e n T[N(T) - 1] a n d TN(T). W h a t is t h e exact value of τ ( Τ ) ? (c) T h e t i m e o u t Τ is a design p a r a m e t e r . H o w w o u l d y o u c h o o s e it?

3

Packet-

CHAPTER

Switched Networks

TV e saw in C h a p t e r 2 t h a t m o r e sophisticated services d e m a n d e d b y u s e r applications are built from basic services in a l a y e r e d a r c h i t e c t u r e . We also discussed t h e ODN a r c h i t e c t u r e for c o m m u n i c a t i o n n e t w o r k s . In this c h a p t e r w e s t u d y t h e seven-layer O p e n S y s t e m s I n t e r c o n n e c t i o n (OSI) m o d e l for t h e logical l a y e r i n g of functions in data n e t w o r k s . T h e OSI m o d e l c a n b e r e g a r d e d as m o r e detailed specifications of t h e ODN m o d e l , a l t h o u g h t h e y w e r e d e v e l o p e d l o n g before t h e ODN m o d e l . We t h e n discuss t h e i m p o r t a n t i m p l e m e n t a t i o n s of t h e OSI m o d e l , begin­ n i n g w i t h t h e major local a n d m e t r o p o l i t a n a r e a n e t w o r k i m p l e m e n t a t i o n s of t h e data link layer, from E t h e r n e t a n d t o k e n r i n g to FDDI a n d DQDB (Dis­ t r i b u t e d Q u e u e Dual Bus), F r a m e Relay, a n d Switched Multimegabit Data Service (SMDS). O t h e r i m p o r t a n t i m p l e m e n t a t i o n s of t h e data link l a y e r are p r e s e n t e d in C h a p t e r s 5 a n d 6. We explain t h e I n t e r n e t a n d t h e T C P / I P net­ w o r k s in C h a p t e r 4. By t h e e n d of this chapter, y o u will b e able to e s t i m a t e t h e limitations of e a c h i m p l e m e n t a t i o n in t e r m s of speed, delay, a n d versatility a n d to j u d g e w h i c h i m p l e m e n t a t i o n b e s t m e e t s a n organization's n e e d s . You will also u n ­ d e r s t a n d t h e a d v a n c e s t h a t are likely to o c c u r in t h e n e a r future a n d w h a t t h e y offer. Section 3.1 describes t h e OSI m o d e l . Section 3.2 is d e v o t e d to E t h e r n e t a n d section 3.3 to t h e t o k e n ring n e t w o r k ; sections 3.4 t h r o u g h 3.7 describe t h e FDDI, DQDB, F r a m e Relay, a n d SMDS n e t w o r k s , respectively. T h e r e a d e r m a y skip particular sections w i t h o u t loss in c o n t i n u i t y .

Packet-Switched Networks

104

3.1

OSI AND IP MODELS In this section w e explain t h e O p e n S y s t e m s I n t e r c o n n e c t i o n (OSI) r e f e r e n c e m o d e l . T h e OSI is a seven-layer d e c o m p o s i t i o n of n e t w o r k functions p u b l i s h e d b y t h e ISO. We explain t h e m a i n functions p e r f o r m e d b y t h e s e layers. (See Figure 3.9 for a s u m m a r y . ) M a n y n e t w o r k s do n o t strictly follow t h e OSI m o d e l . In s o m e cases, t h e n e t w o r k s w e r e d e v e l o p e d before t h e OSI m o d e l w a s published. However, despite t h e s e differences, t h e OSI m o d e l h e l p s in u n d e r s t a n d i n g t h e design of packet-switched n e t w o r k a r c h i t e c t u r e s . T h e OSI m o d e l is u s e d to specify standards.

3.1.1

Layer 1: Physical Layer T h e b o t t o m m o s t layer is t h e physical layer. It i m p l e m e n t s a n u n r e l i a b l e bit link. A link consists of a transmitter, a receiver, a n d a m e d i u m over w h i c h signals are propagated. T h e s e signals are m o d u l a t e d e l e c t r o m a g n e t i c w a v e s t h a t p r o p a g a t e e i t h e r g u i d e d (by a c o p p e r cable, wire pair, or a n optical fiber) or u n g u i d e d in free space (as in radio). T h e t r a n s m i t t e r c o n v e r t s t h e bits into signals, a n d t h e physical l a y e r in t h e receiver c o n v e r t s t h e signals b a c k into bits. T h e receiver m u s t b e s y n c h r o n i z e d to b e able to r e c o v e r t h e successive bits. Tb assist t h e synchronization, t h e t r a n s m i t t e r inserts a specific bit p a t t e r n , called a preamble, at t h e b e g i n n i n g of t h e packet, indicated b y s y n c in Figure 3.1. T h e

LinkL A

Sync

Unreliable i bit link ;

jg J :

Β

Physical layer

l-E 3.1 FIGURE

The physical layer transmits bits b y converting t h e m into electrical or optical signals. In m a n y implementations, the physical layer uses synchronization bits (sync) to synchronize the receiver.

3.1

OSI and IP Models

105

bit link is u n r e l i a b l e b e c a u s e s y n c h r o n i z a t i o n e r r o r s a n d noise in t h e link c a n c o r r u p t a packet. Physical l a y e r s t a n d a r d s specify t h e m o d u l a t i o n s c h e m e (the relation b e ­ t w e e n t h e bits a n d t h e e l e c t r o m a g n e t i c signal), a n d t h e characteristics of t h e interface b e t w e e n t h e t r a n s m i t t e r a n d r e c e i v e r a n d t h e m e d i u m . A link's char­ acteristics i m p o s e a limit o n h o w fast it c a n t r a n s m i t data. Generally, wireless links are slower t h a n c o p p e r links, a n d c o p p e r links are slower t h a n optical links.

3.1.2

Layer 2: Data Link Layer Figure 3.2 illustrates t h e data link layer for a point-to-point link b e t w e e n two c o m p u t e r s . As w e explained in o u r discussion of e r r o r d e t e c t i o n (see sec­ t i o n 2.6.3), t h e t r a n s m i t t e r a n d r e c e i v e r c a n e x e c u t e a specific protocol to r e t r a n s m i t c o r r u p t e d packets. In t h e figure, t h e protocol entities are r e p r e ­ s e n t e d b y t h e boxes labeled 2. T h e s e entities are p r o g r a m s t h a t are u s u a l l y executed b y d e d i c a t e d electronic circuits (in t h e NIC or n e t w o r k interface card) b e c a u s e of t h e h i g h speed.

Reliable packet link

Β

Data link layer

+

Sequence no.

CRC

Physical layer

Sync

Unreliable bit link

3.2 FIGURE

The data link layer supervises the transmission of packets b y the physical layer. In a typical implementation, the data link layer adds a sequence n u m b e r and error-detection bits (CRC). As we explain in the text, networks with reliable links control the errors from source to destination instead of controlling t h e m on every link.

Packet-Switched Networks

106

T h e t r a n s m i t t e r a p p e n d s error-detection bits a n d m a y n u m b e r t h e packets. T h e figure shows t h e fields t h a t are a p p e n d e d to t h e p a c k e t t h a t t h e data link layer t r a n s m i t s ; first, t h e data link l a y e r in t h e t r a n s m i t t e r adds t h e errord e t e c t i o n bits (CRC) a n d m a y add a s e q u e n c e n u m b e r to t h e packet. T h e n t h e physical layer adds a s y n c h r o n i z a t i o n p r e a m b l e (sync). In t h e receiver, t h e physical layer strips t h e p r e a m b l e a n d gives t h e rest of t h e p a c k e t to t h e data link layer, w h i c h u s e s t h e error-detection bits to verify t h a t t h e p a c k e t is correct. If t h e data link l a y e r a r r a n g e s for r e t r a n s m i s s i o n s of e r r o n e o u s packets, it u s e s t h e packet s e q u e n c e n u m b e r s to d e t e r m i n e w h i c h p a c k e t should b e r e t r a n s m i t t e d . If t h e data link layer only d r o p s i n c o r r e c t packets, t h e n it d o e s n o t i n s e r t a s e q u e n c e n u m b e r . T h e data link l a y e r t h e n strips t h e error-detection bits a n d t h e s e q u e n c e n u m b e r (if p r e s e n t ) . T h e errors c a n b e controlled end-to-end i n s t e a d of hop-by-hop (at e a c h link). End-to-end control is preferable to link-level control w h e n t h e n e t w o r k u s e s links w i t h a small bit error rate. Indeed, in s u c h a situation, m o s t p a c k e t s r e a c h t h e i r d e s t i n a t i o n w i t h o u t errors, a n d it is wasteful to verify t h e m at e v e r y link. End-to-end e r r o r control is i m p l e m e n t e d at t h e t r a n s p o r t l a y e r (layer 4) with t h e s a m e m e c h a n i s m u s e d b y t h e data link layer. Link-level e r r o r control is preferable in links with a high bit error rate, as in wireless radio or satellite links. In t h e s e links error-correction bits m a y b e u s e d b e c a u s e r e t r a n s m i s s i o n of c o r r u p t e d p a c k e t s m a y i n c u r large delays. T h e fields t h a t t h e protocol entities add to t h e p a c k e t c o n t a i n control infor­ m a t i o n t h a t t h e protocols in t h e different layers u s e to m o n i t o r t h e t r a n s m i s ­ sions. T h e a p p e n d i n g of control fields to a p a c k e t is called encapsulation. The reverse process, stripping t h e control fields, is decapsulation. Observe t h a t en­ capsulation a n d d e c a p s u l a t i o n m a y often b e p e r f o r m e d w i t h o u t e x a m i n a t i o n of t h e packet, w h i c h simplifies t h e h a r d w a r e a n d r e d u c e s p a c k e t processing t i m e .

3.13

Sublayer 2a: Media Access Control Figure 3.3 shows c o m p u t e r s a t t a c h e d to a c o m m o n link. T h e s e c o m p u t e r s m u s t regulate t h e access to t h a t s h a r e d link. T h i s function, called access control, is p e r f o r m e d b y a sublayer called t h e media access control (MAC) sublayer. T h u s , t h e MAC sublayers in t h e c o m p u t e r s follow a set of rules—a p r o t o c o l to regulate access to t h e s h a r e d link. Because t h e link is shared, t h e MAC m u s t a p p e n d t h e physical a d d r e s s of t h e destination, t h e specific c o m p u t e r to w h i c h t h e packet is destined. T h e physical a d d r e s s identifies u n i q u e l y a

3.1

OSI and IP Models

107 Unreliable packet link

Media access control

2a

MAC addresses Physical layer

Sync

3.3 FIGURE

Shared link

2a

•

s

The figure shows three computers that share a c o m m o n link. Access to such link I regulated b y the media access control (MAC) sublayer. That sublayer adds the addresses of the source and destination to the packet before giving it to the physical layer for transmission.

a

c

o

m

m

o

n

s

device a t t a c h e d to a s h a r e d link. Such a physical a d d r e s s is n o t n e e d e d for a point-to-point link. T h e c o m m o n link m a y b e a c o p p e r cable to w h i c h t h e c o m p u t e r s are a t t a c h e d (as in E t h e r n e t a n d CATV), or a radio b r o a d c a s t c h a n n e l to w h i c h all t h e c o m p u t e r radio receivers are t u n e d . T h e MAC s u b l a y e r i m p l e m e n t s t h e u n r e l i a b l e t r a n s m i s s i o n of p a c k e t s b e t w e e n c o m p u t e r s a t t a c h e d to t h e c o m m o n link. MAC s t a n d a r d s specify t h e p a c k e t formats, t h e MAC a d d r e s s i n g s c h e m e , a n d t h e MAC protocol.

3.1.4

Sublayer 2b: Logical Link Control Figure 3.4 s h o w s t h e logical link control (LLC) sublayer. It u s e s t h e u n r e l i a b l e t r a n s m i s s i o n of p a c k e t s i m p l e m e n t e d b y t h e MAC s u b l a y e r to i m p l e m e n t e i t h e r o n l y e r r o r d e t e c t i o n or reliable p a c k e t t r a n s m i s s i o n b e t w e e n c o m p u t e r s a t t a c h e d to a s h a r e d link. T h e functions of t h e LLC are t h e s a m e as t h o s e t h e data link l a y e r executes for a point-to-point link. T h e MAC a n d LLC t o g e t h e r c o n s t i t u t e t h e data link l a y e r for m u l t i p l e access links: t h e y u s e t h e u n r e l i a b l e bit link of t h e physical l a y e r to i m p l e ­ m e n t a p a c k e t - t r a n s m i s s i o n service w i t h e r r o r d e t e c t i o n or a reliable packett r a n s m i s s i o n service b e t w e e n c o m p u t e r s a t t a c h e d to a c o m m o n link.

Packet-Switched Networks

108 A

I

Reliable packet link

β

2b

2b

Logical link sublayer

2a

2a

MAC sublayer

I, I CRC

No.

Unreliable packet link

3.4 FIGURE

T h e logical link control (LLC) adds error detection to the transmissions ^ ^ sublayer implements. (The LLC can also use a retransmission m e c h a n i s m to provide reliable packet transmissions, if desired.) a

t

e

M

A

C

Figure 3.5 shows a bndge ( c o m p u t e r C) b e t w e e n t w o E t h e r n e t n e t w o r k s . C o m p u t e r s A, B, a n d C a r e a t t a c h e d to o n e E t h e r n e t . T h e right p a r t of t h e figure shows t h e s e t h r e e c o m p u t e r s a t t a c h e d to t h e s a m e link. T h e MAC s u b l a y e r i n t h e t h r e e c o m p u t e r s i m p l e m e n t s u n r e l i a b l e p a c k e t t r a n s m i s s i o n s . T h e LLC detects t h e errors in t h e t r a n s m i s s i o n s . T h e situation is similar for c o m p u t e r s C, D, a n d Ε a t t a c h e d to t h e o t h e r E t h e r n e t . Consider a p a c k e t s e n t b y c o m p u t e r A a n d d e s t i n e d for c o m p u t e r E. Com­ p u t e r C m u s t store t h a t p a c k e t a n d r e t r a n s m i t it o n t h e s e c o n d E t h e r n e t . If t h e p a c k e t s e n t b y A is d e s t i n e d for c o m p u t e r B, C m u s t n o t r e t r a n s m i t it. T h e

3.5 FIGURE

Computer C in this figure is a bndge. It connects two local area networks b y copying packets from one to the other.

3.1

OSI and IP Models

109

decision of C is l i m i t e d to w h e t h e r to r e t r a n s m i t t h e p a c k e t or not. A b r i d g e is a c o m p u t e r e q u i p p e d w i t h t h e h a r d w a r e a n d software to p e r f o r m s u c h decisions a n d capable of r e t r a n s m i t t i n g p a c k e t s at a h i g h rate. T h e b r i d g e allows t h e c o m ­ p u t e r s o n t h e two E t h e r n e t s to function as if t h e y are o n t h e s a m e E t h e r n e t (see section 3.2.3).

3.1.5

Layer 3: Network Layer T h e data link l a y e r i m p l e m e n t s a p a c k e t link b e t w e e n c o m p u t e r s a t t a c h e d to a c o m m o n link. As w e e x p l a i n e d in o u r discussion of store-and-forward p a c k e t switching (see section 2.6.2), w h e n c o m p u t e r s are c o n n e c t e d b y a collection of point-to-point links, t h e y m u s t figure o u t w h e r e to s e n d t h e p a c k e t s t h a t t h e y receive: w h e t h e r to s e n d t h e m o u t over a n o t h e r l i n k and, if so, w h i c h one. (See Figure 3.6.) T h i s function—finding t h e p a t h t h e p a c k e t s m u s t follow—is called routing. Routing is o n e of t h e m a i n functions of t h e network layer T h e n e t w o r k l a y e r a p p e n d s u n i q u e n e t w o r k a d d r e s s e s of t h e source a n d d e s t i n a t i o n c o m p u t e r s . A n i m p o r t a n t a d d r e s s i n g s c h e m e in packetswitched n e t w o r k s is t h a t u s e d b y t h e I n t e r n e t . Circuit-switched n e t w o r k s , like t h e t e l e p h o n e n e t w o r k , u s e different a d d r e s s i n g s c h e m e s . T h u s , t h e n e t w o r k l a y e r u s e s t h e t r a n s m i s s i o n over point-to-point links p r o v i d e d b y t h e data link l a y e r to t r a n s m i t p a c k e t s b e t w e e n a n y t w o c o m p u t e r s a t t a c h e d in a n e t w o r k .

3.6 FIGURE

The network layer delivers packets b e t w e e n any two computers attached to network. That layer i m p l e m e n t s store-and-forward transmissions along successive links from the source to the destination.

t n e

s

a

m

e

Packet-Switched Networks

110

~2--

S 3-7__ FIGURE

D

r

Computer C in this figure is a router. It is designed to relay packets at a high r a t e

to

^

e

P

r o

P

e r

a n c

^

w

^

a

l°

w

delay.

Figure 3.7 shows a router a t t a c h e d to several links. W h e n t h e r o u t e r receives a packet, it m u s t decide o n t h e basis of t h e n e t w o r k a d d r e s s e s along w h i c h link it should r e t r a n s m i t t h e packet. T h i s r o u t i n g function is i m p l e m e n t e d b y t h e n e t w o r k layer. Observe t h a t t h e link b e t w e e n , say, C a n d D in Figure 3.7 m a y c a r r y packets b e t w e e n A a n d Ε a n d b e t w e e n Β a n d E. T h e s e p a c k e t s are statistically multiplexed b y t h e r o u t e r C (see section 2.6.1). T h e n e t w o r k a d d r e s s e s of t h e packets p e r m i t demultiplexing. N e t w o r k l a y e r s t a n d a r d s specify p a c k e t formats, addressing s c h e m e s , a n d r o u t i n g protocols.

3.1.6

Layer 4: Transport Layer T h e transport layer delivers messages b e t w e e n transport service access points (TSAPs or ports) in different c o m p u t e r s . Several p r o c e s s e s r u n n i n g o n a com­ p u t e r m a y b e exchanging m e s s a g e s w i t h p r o c e s s e s r u n n i n g o n o t h e r com­ p u t e r s . T h e TSAPs a p p e n d e d to t h e m e s s a g e s differentiate t h o s e i n f o r m a t i o n s t r e a m s . T h e t r a n s p o r t l a y e r m a y multiplex several low-rate t r a n s m i s s i o n s w i t h different TSAPs onto o n e virtual circuit or divide a high-rate c o n n e c t i o n into parallel virtual circuits. Some frequently u s e d applications s u c h as e-mail a n d file transfers are allocated fixed TSAPs (also called well-known ports), l b c o n n e c t to a process w i t h u n k n o w n TSAP, a r e m o t e process first c o n n e c t s to a process s e r v e r a t t a c h e d to a fixed TSAP. T h e s e r v e r t h e n indicates t h e TSAP of t h e d e s i r e d process. T h e t r a n s p o r t l a y e r delivery of m e s s a g e s is e i t h e r connection-oriented or con­ nectionless. A c o n n e c t i o n - o r i e n t e d t r a n s p o r t l a y e r delivers error-free m e s s a g e s in t h e correct order. Such a t r a n s p o r t l a y e r provides t h e following services:

3.1

OSI and IP Models

CONNECT, DATA, EXP_DATA, a n d DISCONNECT CONNECT sets u p a con­ n e c t i o n b e t w e e n TSAPs. DATA delivers a s e q u e n c e of m e s s a g e s in t h e correct o r d e r a n d w i t h o u t errors. EXP_DATA delivers u r g e n t m e s s a g e s b y m a k i n g t h e m j u m p a h e a d of t h e n o n u r g e n t m e s s a g e s in t h e two e n d n o d e s . DISCONNECT r e l e a s e s t h e c o n n e c t i o n . A c o n n e c t i o n l e s s t r a n s p o r t l a y e r delivers m e s s a g e s o n e b y one, possibly w i t h errors, a n d w i t h n o g u a r a n t e e o n t h e o r d e r of t h e m e s s a g e s . T h e service of a c o n n e c t i o n l e s s t r a n s p o r t l a y e r is UNIT_DATA, t h e c o n n e c t i o n l e s s delivery of a single m e s s a g e . T h e t r a n s p o r t l a y e r d e c o m p o s e s m e s s a g e s into p a c k e t s a n d c o m b i n e s pack­ ets into messages, possibly after r e s e q u e n c i n g t h e m . Such packetization m a y b e n e c e s s s a r y b e c a u s e t h e size of t h e m e s s a g e m a y b e larger t h a n t h e size of t h e p a c k e t s a c c e p t e d b y t h e n e t w o r k layer. R e s e q u e n c i n g m a y b e n e c e s s a r y b e c a u s e p a c k e t s m a y n o t b e r e c e i v e d in t h e o r d e r in w h i c h t h e y are t r a n s m i t ­ ted, l b p e r f o r m this packetization function, t h e t r a n s p o r t l a y e r n u m b e r s t h e p a c k e t s b e l o n g i n g to t h e s a m e m e s s a g e . T h e l a y e r also controls t h e flow of p a c k e t s to p r e v e n t t h e source from s e n d i n g p a c k e t s faster t h a n t h e d e s t i n a t i o n c a n h a n d l e t h e m . Moreover, t h e t r a n s p o r t l a y e r r e q u e s t s r e t r a n s m i s s i o n s of c o r r u p t e d p a c k e t s (see Figure 3.8). Tb b e able to c o m b i n e p a c k e t s into m e s s a g e s , t h e t r a n s p o r t l a y e r n u m b e r s packets. W h e n t h e source c o m p u t e r fails, it m a y lose track of t h e s e q u e n c e n u m b e r it h a d r e a c h e d . R e s u m i n g t h e n u m b e r i n g arbitrarily m i g h t l e a d to d e l a y e d p a c k e t s in t h e n e t w o r k h a v i n g t h e s a m e s e q u e n c e n u m b e r s as t h e p a c k e t s b e i n g t r a n s m i t t e d ; this w o u l d confuse t h e d e s t i n a t i o n . O n e solution to this p r o b l e m is to attach a "time to live," L, to e a c h p a c k e t a n d to d e c r e m e n t t h a t t i m e to live e v e r y t i m e t h e p a c k e t goes t h r o u g h a n e t w o r k n o d e . Packets w i t h a zero t i m e to live are discarded. T h u s , L is t h e m a x i m u m n u m b e r of h o p s for e a c h packet, a n d it c o r r e s p o n d s to a m a x i m u m lifetime of Τ s e c o n d s inside t h e n e t w o r k , b e c a u s e a p a c k e t will r e m a i n in e a c h n o d e for s o m e b o u n d e d t i m e . W h e n a source c o m p u t e r recovers from a crash, it c a n wait for Τ s e c o n d s to m a k e sure t h a t all t h e d e l a y e d p a c k e t s to t h e d e s t i n a t i o n w e r e discarded, t h u s avoiding a n y n u m b e r i n g ambiguity. T h e t r a n s p o r t l a y e r protocols in t h e two n o d e s agree o n a n initial s e q u e n c e n u m b e r b y u s i n g a t h r e e - w a y handshake. Figure 3.8 s u m m a r i z e s t h e functions of l a y e r s 4 to 7.

3.1.7

Layer 5: Session Layer T h e session layer s u p e r v i s e s t h e dialog b e t w e e n two c o m p u t e r s . It c a n set u p a c o n n e c t i o n prior to a n exchange of i n f o r m a t i o n b e t w e e n t h e m a c h i n e s . T h e session l a y e r c a n partition a transfer of a large n u m b e r of m e s s a g e s b y i n s e r t i n g

Packet-Switched Networks

112

7 6 5

- Communication services Application layer Local syntax, secure, efficient connections Presentation layer —Connections— Session layer

7

Commonly used applications

6

Encryption; compression; syntax conversion

5

Supervision of connection

4

Segmentation/reassembly; flow control; end-to-end error control

elivery of messag

4

3.8

Transport layer

3 2

3 2

1

1

This figure summarizes the functions and services of layers 4 to 7.

FIGURE

synchronization points. T h e s e s y n c h r o n i z a t i o n p o i n t s are specific p a c k e t s t h a t divide t h e s e q u e n c e of m e s s a g e s into groups. In case of c o m p u t e r malfunction, t h e t r a n s m i s s i o n c a n restart from t h e last s y n c h r o n i z a t i o n point.

3.1.8

Layer 6: Presentation Layer Application p r o g r a m s a n d c o m p u t e r devices u s e different c o n v e n t i o n s to r e p r e ­ s e n t information b y b i n a r y n u m b e r s . For instance, s o m e c o m p u t e r s r e p r e s e n t 16-bit w o r d s b y placing t h e m o s t significant b y t e before t h e least significant byte, w h e r e a s o t h e r c o m p u t e r s u s e t h e opposite c o n v e n t i o n . As a n o t h e r exam­ ple, different t e r m i n a l s u s e different control c h a r a c t e r s to specify b a c k s p a c e , line feed, a n d carriage r e t u r n . Also, different application p r o g r a m s m a y follow different r u l e s to e n c o d e data s t r u c t u r e s s u c h as m a t r i c e s a n d c o m p l e x n u m ­ b e r s . C o m p u t e r scientists refer to a set of r u l e s for r e p r e s e n t i n g i n f o r m a t i o n as a syntax. T h u s , different c o m p u t e r s u s e different syntaxes. T h e s y n t a x u s e d b y a c o m p u t e r is its local syntax. Suppose t h a t c o m p u t e r s u s i n g Ν different local syntaxes w a n t to c o m m u ­ nicate. O n e possible m e t h o d is to h a v e e v e r y c o m p u t e r p e r f o r m t h e Ν — I syntax conversions n e e d e d to c o m m u n i c a t e w i t h t h e o t h e r c o m p u t e r s . A n o t h e r m e t h o d is to adopt a c o m m o n transfer syntax a n d h a v e e v e r y c o m p u t e r p e r f o r m t h e conversion b e t w e e n its local syntax a n d t h e transfer syntax. T h e s e c o n d

3.1

OSI and IP Models

113

m e t h o d is obviously m o r e c o n v e n i e n t b e c a u s e it does n o t r e q u i r e a c o m p u t e r to b e a w a r e of all t h e possible o t h e r local syntaxes. C o m m u n i c a t i o n n e t w o r k s u s e t h e s e c o n d m e t h o d , a n d t h e p r e s e n t a t i o n l a y e r is r e s p o n s i b l e for t h e con­ version b e t w e e n local s y n t a x a n d transfer syntax. As s h o w n in Figure 3.8, o n e service p r o v i d e d b y t h e p r e s e n t a t i o n l a y e r is t h e e x c h a n g e of i n f o r m a t i o n b e ­ t w e e n c o m p u t e r s , e a c h u s i n g its o w n local syntax. A n additional task of t h e p r e s e n t a t i o n l a y e r is to e n c r y p t t r a n s m i s s i o n s t h a t m u s t b e secure. In abstract t e r m s , e n c r y p t i o n is a one-to-one t r a n s f o r m a t i o n of a m e s s a g e into a n e n c r y p t e d version. A n o t h e r i m p o r t a n t task of t h e p r e s e n t a t i o n l a y e r is data c o m p r e s s i o n . Such data c o m p r e s s i o n e l i m i n a t e s s o m e of t h e r e d u n d a n c y i n t h e i n f o r m a t i o n to b e t r a n s m i t t e d , t h e r e b y r e d u c i n g t h e n u m b e r of bits to b e transferred.

3.1.9

Layer 7: Application Layer T h e application layer provides f r e q u e n t l y n e e d e d c o m m u n i c a t i o n services s u c h as file transfer, t e r m i n a l e m u l a t i o n , r e m o t e login, directory service, a n d r e m o t e j o b execution. T h e s e services are u s e d b y t h e u s e r applications. For instance, a n e-mail p r o g r a m u s e s a file transfer service t h a t delivers a file to a list of a d d r e s s e s a n d informs t h e s e n d e r if s o m e a d d r e s s e s c a n n o t b e r e a c h e d . User applications, s u c h as e-mail or WWW, are r u n o n t o p of t h e application layer.

3.1.10

Summary T h e OSI m o d e l is a detailed a r c h i t e c t u r e t h a t specifies h o w c o m p l e x services s u c h as file transfer a n d r e m o t e login are to b e c o n s t r u c t e d o u t of t h e m o s t basic services of u n r e l i a b l e bit transfer in a seven-layer h i e r a r c h y . E a c h l a y e r adds functionality to t h e services of t h e l a y e r i m m e d i a t e l y b e l o w it. T h e OSI m o d e l is s u m m a r i z e d in Figure 3.9. We c a n u s e t h e OSI r e f e r e n c e m o d e l to a d d m o r e detail to t h e s i m p l e d i a g r a m of t h e n e t w o r k e l e m e n t s in Figure 2.1. In t h e figure, e l e m e n t s labeled "Switch" are c o n n e c t e d to o t h e r switches b y links. F r o m t h e OSI r e f e r e n c e m o d e l w e n o w k n o w t h a t a c o m p u t e r at a switch or at a u s e r c a n c o n n e c t to a n o t h e r c o m p u t e r at different layers. For example, if t h e switch e l e m e n t is a r o u t e r like c o m p u t e r C in Figure 3.7, it c o n n e c t s w i t h n e i g h b o r i n g c o m p u t e r s at t h e n e t w o r k layer. I n this case, t h e s w i t c h o n l y n e e d i m p l e m e n t l a y e r s 1, 2 a n d 3. If t h e switch e l e m e n t is a b r i d g e like c o m p u t e r C in Figure 3.5, it c o n n e c t s w i t h its n e i g h b o r s at t h e data link layer, so t h a t it o n l y n e e d i m p l e m e n t layers 1 a n d 2. T h e u s e r h o s t c o m p u t e r r u n s applications a n d so it m u s t i m p l e m e n t all s e v e n layers.

Packet-Switched Networks

114

End-to-end applications (e.g., e-mail, file transfer) Sessions in local syntax, secure, compressed Reliable end-to-end connections Reliable end-to-end message delivery End-to-end packet delivery

39 FIGURE

imam mm

Packet link

Packet link

Unreliable bit link

Unreliable bit link

Signal propagation

Signal propagation

The figure shows the services implemented b y the seven layers of the OSI reference model.

We can place t h e OSI m o d e l in t h e context of t h e O p e n Data N e t w o r k m o d e l of section 2.8 b y identifying l a y e r s 1 a n d 2 w i t h bit ways, l a y e r s 3 a n d 4 w i t h b e a r e r services, layers 4, 5, a n d 6 w i t h m i d d l e w a r e , a n d l a y e r 7 w i t h applications. Most n e t w o r k i m p l e m e n t a t i o n s do n o t follow t h e OSI m o d e l , b u t r a t h e r t h e four-layer IP m o d e l b a s e d o n t h e I n t e r n e t protocol l a y e r s s h o w n in Figure 4.2.

3.2

ETHERNET (IEEE 802.3) E t h e r n e t is t h e m o s t p o p u l a r local area n e t w o r k or LAN technology. More t h a n 100 million E t h e r n e t n o d e s h a v e b e e n deployed, a n d m o r e t h a n 50 million E t h e r n e t n e t w o r k interface cards are sold e a c h year. T h e installed b a s e a n d m a r k e t s h a r e of E t h e r n e t n e t w o r k s dwarfs t h o s e of o t h e r t y p e s of LANs. T h e large installed E t h e r n e t b a s e h a s also s p u r r e d a series of advances, s u c h as fast a n d gigabit E t h e r n e t a n d E t h e r n e t switches, w h i c h are b a c k w a r d - c o m p a t i b l e w i t h earlier versions of E t h e r n e t . E t h e r n e t is specified b y t h e IEEE 802.3 standards, p u b l i s h e d in 1985. A n E t h e r n e t LAN typically u s e s twisted pair wires or fiber optic cable. 10BASE-T

3.2

Ethernet (IEEE 802.3)

115

( E t h e r n e t over twisted pairs) t r a n s m i t s at 10 Mbps, a n d 100BASE-T or Fast E t h e r n e t t r a n s m i t s at 100 Mbps. Gigabit E t h e r n e t , w i t h a s p e e d of 1,000 Mbps (1 gigabit or 1 billion bps), u s u a l l y over optical fiber, is u s e d for e n t e r p r i s e - w i d e backbones. E t h e r n e t s are i n e x p e n s i v e a n d provide a relatively h i g h t h r o u g h p u t a n d low delays t h a t c a n s u p p o r t m a n y applications. Most i m p o r t a n t l y , E t h e r n e t provides inexpensive, relatively h i g h - s p e e d n e t w o r k access to individual users. ( T h e t o k e n ring n e t w o r k , in w h i c h u s e r s s h a r e s o m e of t h e t r a n s m i s s i o n costs, also provides i n e x p e n s i v e access. T h e CATV n e t w o r k , d e s c r i b e d in C h a p t e r 5, provides s h a r e d access over l o n g e r distances.)

3.2.1

Physical Layer T h e physical l a y e r of IEEE 802.3 n e t w o r k s specifies t h e electrical a n d m e c h a n ­ ical characteristics of t h e w i r i n g a n d of t h e e n c o d i n g of t h e bits. We describe t h e physical l a y e r s of t h e lOBASEx, lOOBASEx, a n d t h e lOOOBASEx n e t w o r k s , w i t h t r a n s m i s s i o n s p e e d s of 10, 100, a n d 1,000 Mbps, respectively.

WBASE-T T h e 10BASE-T wiring, s h o w n in Figure 3.10, m a y u s e t h e s a m e category 3 u n s h i e l d e d twisted pairs (UTPs or UTP-3s) t h a t t e l e p h o n e c o m p a n i e s u s e to wire buildings for t e l e p h o n e service. C o n s e q u e n t l y , if a n office b u i l d i n g or r e s i d e n c e h a s e n o u g h s p a r e UTPs, t h e s e c a n b e u s e d to install a 10BASE-T n e t w o r k . (This explains t h e p o p u l a r i t y of this form of E t h e r n e t . E c o n o m i e s of scale h a v e led to a s t e a d y r e d u c t i o n in t h e cost of E t h e r n e t chips a n d b o a r d s . I b d a y , t h e cost of w i r i n g is likely to b e m o r e t h a n t h e cost of t h e E t h e r n e t hardware.) In a 10BASE-T n e t w o r k , e a c h c o m p u t e r is e q u i p p e d w i t h a n e t w o r k inter­ face card t h a t is a t t a c h e d to a m e d i u m access u n i t (MAU) w i t h t w o twisted pairs. Two u n s h i e l d e d twisted pairs attach a n MAU to a 10BASE-T h u b . H u b s c a n b e a t t a c h e d t o g e t h e r w i t h two U T P s to b u i l d larger n e t w o r k s . A h u b is a r e p e a t e r : it t r a n s m i t s a p a c k e t r e c e i v e d o n o n e p o r t to all t h e o t h e r ports. T h e physical l a y e r e n c o d e s t h e bits as electrical signals u s i n g t h e Man­ c h e s t e r e n c o d i n g . In this encoding, t h e bits are r e p r e s e n t e d b y a two-value signal t h a t m a k e s a transition d u r i n g e v e r y bit t r a n s m i s s i o n . Specifically, a bit 0 is e n c o d e d as t h e low value for half of t h e bit t r a n s m i s s i o n interval followed b y t h e h i g h value for t h e s e c o n d half of t h e interval. A bit 1 is e n c o d e d as t h e high value for t h e first half of t h e interval followed b y t h e low value for t h e s e c o n d half of t h e interval. At 10 Mbps, t h e bit t r a n s m i s s i o n interval lasts 0.1 με.

Packet-Switched Networks

116

Two twisted pairs Medium access unit

Two UTPs (0.5 mm) < 100 m

Interface card MAU Data

Hub

Interface card Collision detected MAU

Hub Data Collision detection

Data Data

Data l_l MAU Interface card

3.10 FIGURE

MAU Collision detected Interface card

Physical layout of a 10BASE-T network. This network uses unshielded twisted pairs and can be wired with spare telephone pairs already in place in the building. The transmission rate is 10 Mbps. One can attach hubs together to build larger networks. T h e receiver u s e s t h e transitions to s y n c h r o n i z e to t h e r e c e i v e d signal. S u c h a code is said to b e self-synchronizing. T h e actual receiver h a r d w a r e consists of a phase-locked loop t h a t adjusts its p h a s e to m a t c h t h e t r a n s i t i o n s of t h e r e c e i v e d signal.

100BASE-T T h e 100BASE-T transceivers o p e r a t e at 100 Mbps over two pairs of UTP-3s, t h e s a m e as for 10BASE-T. ( T h u s 10BASE-T E t h e r n e t u s e r s c a n u p g r a d e to 100BASE-T

3.2

Ethernet (IEEE 802.3)

• Μ » » ^

_ wmm

•

w i t h o u t c h a n g i n g t h e i r cabling infrastructure.) T h e m a x i m u m distance b e ­ t w e e n two c o m p u t e r s is 100 m . Physical l a y e r s t a n d a r d s are also defined for o t h e r cabling a r r a n g e m e n t s , i n c l u d i n g t h e following: 100BASE-FX is for trans­ m i s s i o n over optical fiber, 100BASE-TX is for full-duplex t r a n s m i s s i o n over t w o pairs of U T P - 5 ( c a t e g o r y 5) cables or t w o pairs of s h i e l d e d t w i s t e d pairs (STP), 100BASE-T4 is for four UTP-3 cables. T h e m o d u l a t i o n s c h e m e s for all t h e s e cabling a r r a n g e m e n t s are different. Similar to 10BASE-T, t h e 100BASE-X family (except T4) s u p p o r t s s i m u l t a n e ­ o u s or full-duplex t r a n s m i s s i o n of 100 M b p s data s t r e a m s o n o n e l i n k s e g m e n t in e a c h direction. T h e s t a n d a r d s also specify a n auto-negotiation protocol t h a t p r o m o t e s flexible products, s u c h as dual-speed 10/100 Mbps n e t w o r k interface cards.

WOOBASEx T h e physical l a y e r s t a n d a r d for gigabit E t h e r n e t w a s a p p r o v e d in J u n e 1998. It also p e r m i t s several a r r a n g e m e n t s : 1000BASE-LX is for t r a n s m i s s i o n over single- or m u l t i m o d e fiber, 1000BASE-SX is for m u l t i m o d e fiber, a n d 1000BASECS is for s h i e l d e d c o p p e r cables. A future standard, 1000BASE-T, will c o n s i d e r t r a n s m i s s i o n over UTP.

Wireless Ethernet A n u m b e r of c o m m e r c i a l p r o d u c t s are available to set u p wireless E t h e r n e t n e t w o r k s . In s u c h a wireless E t h e r n e t n e t w o r k , all t h e stations s h a r e a radio c h a n n e l . T h e physical l a y e r s t a n d a r d s specify t h e f r e q u e n c y s p e c t r u m t h a t t h e radio c h a n n e l c a n use, a n d t h e m o d u l a t i o n s c h e m e . T h e r e are m a n y varieties of wireless LANs. Some LAN p r o d u c t s diffuse infrared light signals i n s t e a d of radio waves.

10BASE5 Prior to t h e i n t r o d u c t i o n of 10Base-T a n d 100Base-T, E t h e r n e t n e t w o r k s u s e d t h e 10BASE5 version. We briefly describe this version for historical i n t e r e s t a n d b e c a u s e s u c h n e t w o r k s are still a r o u n d . T h e w i r i n g of a 10BASE5 consists of s e g m e n t s . Each s e g m e n t is a l e n g t h of u p to 500 m of coaxial cable w i t h a d i a m e t e r of 10 m m a n d a characteristic i m p e d a n c e of 50 o h m s . S e g m e n t s are c o n n e c t e d b y r e p e a t e r s t h a t c a n b e u p to 1,000 m apart. No two c o m p u t e r s o n t h e n e t w o r k c a n b e m o r e t h a n 2,500 m apart. T h e t r a n s m i s s i o n rate is 10 Mbps; t h e physical l a y e r u s e s t h e M a n c h e s t e r e n c o d i n g .

117

Packet-Switched Networks

118

3.2.2

MAC T h e m e d i a access control sublayer of E t h e r n e t specifies t h e MAC a d d r e s s e s of n e t w o r k interfaces, t h e frame format, a n d t h e MAC protocol for s h a r i n g t h e cable. T h e MAC address of a n E t h e r n e t interface card is a 48-bit string set b y t h e m a n u f a c t u r e r t h a t is u n i q u e to t h a t card. T h e first 24 bits of t h e a d d r e s s are t h e "organizational u n i q u e identifier" t h a t t h e IEEE assigned to t h e m a n u f a c t u r e r s of t h e E t h e r n e t interface. Each m a n u f a c t u r e r t h e n m a k e s sure t h a t its interfaces h a v e different lower 24-bit identifiers. T h e frame format of E t h e r n e t p a c k e t s is s h o w n in Figure 3.11. T h e p r e a m ­ ble (PRE) s y n c h r o n i z e s t h e receiver. T h e start-of-frame d e l i m i t e r (SFD) in­ dicates t h e start of t h e frame. T h e d e s t i n a t i o n (DA) a n d source (SA) MAC a d d r e s s e s are 48-bit-long strings u n i q u e to e a c h interface card. T h e l e n g t h in­ dicator (LEN) e l i m i n a t e s t h e n e e d for a n end-of-frame d e l i m i t e r a n d p e r m i t s t h e u s e of a p a d d i n g field (PAD) to m a k e s u r e t h a t t h e frames h a v e at least 64 b y t e s . T h e cyclic r e d u n d a n c y c h e c k (CRC) e n a b l e s t h e receiver to d e t e c t m o s t t r a n s m i s s i o n errors, as w e explained in section 2.6.3. T h e logical link control (LLC) frame specifies t h e d e s t i n a t i o n (DSAP) a n d source (SSAP) service access points. A slightly different s t a n d a r d (called t h e DEC-Intel-Xerox or DIX Ethernet) r e p l a c e s t h e l e n g t h field b y a 2-byte t y p e field. T h e E t h e r n e t MAC protocol is CSMA/CD, Carrier Sense Multiple Access with Collision Detection. W h e n u s i n g CSMA/CD, a n o d e t h a t h a s a p a c k e t to t r a n s m i t waits until t h e c h a n n e l is silent before t r a n s m i t t i n g . Also, t h e n o d e aborts t h e t r a n s m i s s i o n as soon as it realizes t h a t a n o t h e r n o d e is t r a n s m i t t i n g . After aborting a t r a n s m i s s i o n , t h e n o d e waits for a r a n d o m t i m e a n d t h e n r e p e a t s t h e s e steps. (Note t h a t t h e CSMA/CD protocol is disabled in full-duplex operation. Such o p e r a t i o n is possible o n l y b e t w e e n devices e q u i p p e d w i t h buffers.)

64 to 1,518 bytes (512 to 12,144 bits) 7 PRE

1

6

SFD DA

6 SA

2

1

3.11

1 or 2

LEN DSAP SSAP CONT

·


3.14

^^J^

B l i e

2,500 m 608-bit packets

Two representative values of the efficiency of Ethernet. Typically, the efficiency is about 80%.

m e d i u m is n o t s h a r e d . However, t h e h u b , w h i c h r e p e a t s e a c h p a c k e t o n e v e r y port, c o n v e r t s it into a b r o a d c a s t m e d i u m . W h e n a h u b d e t e c t s a collision ( m o r e t h a n o n e port h a s a n i n c o m i n g signal), it s e n d s a j a m m i n g signal to all ports, to e m u l a t e collision d e t e c t i o n . W h e n several h u b s are i n t e r c o n n e c t e d to b u i l d a larger n e t w o r k , a p a c k e t is r e p e a t e d o n t h e ports of all t h o s e h u b s . T h e c o m p u t e r s a t t a c h e d to t h o s e h u b s form a single broadcast domain a n d s h a r e t h e total bit rate. Bridges a n d switches, b y contrast, selectively r e p e a t p a c k e t s r e c e i v e d at a port, t h e r e b y s e g m e n t i n g a b r o a d c a s t d o m a i n , so t h a t p a c k e t s o n different s e g m e n t s c a n b e s i m u l t a n e o u s l y t r a n s m i t t e d . T h e aggregate bit rate is t h e r e b y i n c r e a s e d . In a switched E t h e r n e t , a switch r e p l a c e s t h e h u b a n d collisions are n o t possible. However, for b a c k w a r d compatibility, t h e s a m e frame s t r u c t u r e is used.

3.23

LLC T h e IEEE 802.2 logical link control s t a n d a r d is u s e d for 802.3, 802.5, a n d o t h e r n e t w o r k s . T h e LLC s u b l a y e r p r o v i d e s c o n n e c t i o n - o r i e n t e d or c o n n e c t i o n l e s s ( a c k n o w l e d g e d or not) services. T h e LLC c a n also m u l t i p l e x different t r a n s m i s ­ sions t h a t are differentiated b y t h e service access p o i n t field (see Figure 3.11).

Packet-Switched Networks

122

Finally, t h e LLC i m p l e m e n t s t h e transparent routing of p a c k e t s b e t w e e n Ether­ n e t s attached t o g e t h e r with bridges. W h e n it provides a c o n n e c t i o n - o r i e n t e d or a n a c k n o w l e d g e d c o n n e c t i o n ­ less service, t h e LLC u s e s t h e CRC field to d e t e c t errors. Tb i m p l e m e n t a c o n n e c t i o n - o r i e n t e d service, t h e LLC u s e s t h e Go Back Ν protocol (see sec­ tion 2.6.3) to a r r a n g e for t h e t r a n s m i t t e r to r e t r a n s m i t p a c k e t s t h a t do n o t arrive error free.

3.2.4

LAN Interconnection LANs c a n b e i n t e r c o n n e c t e d b y h u b s , bridges, a n d switches. T h e s e are l a y e r 2 devices. Since h u b s are r e p e a t e r s , attaching two LANs to t h e s a m e h u b c o n v e r t s t h e m to a single LAN s e g m e n t so t h a t c o m p u t e r s o n b o t h LANs s h a r e t h e s a m e bit rate. Bridges a n d switches are u s e d to c o n n e c t LANs b u t i n s t e a d of r e p e a t i n g packets, t h e y forward t h e m selectively.

Bridges We u s e Figure 3.15 to explain h o w a p a c k e t w i t h a given MAC d e s t i n a t i o n ad­ dress finds its w a y in a n e t w o r k of E t h e r n e t s c o n n e c t e d b y bridges. T h e b r i d g e s

3.15 FIGURE

Ethernet attached b y bridges. Transparent routing enables a computer to transmit a packet to another computer on one of these Ethernets as if it were on the same Ethernet.

3.2

Ethernet (IEEE 802.3)

u s e a s i m p l e p r o c e d u r e called transparent routing. T r a n s p a r e n t r o u t i n g r e q u i r e s t h e b r i d g e s (such as n o d e C in t h e figure) to m a i n t a i n tables of MAC a d d r e s s e s of t h e n o d e s o n t h e i r E t h e r n e t s . Tb m a i n t a i n s u c h a table, a b r i d g e r e a d s t h e source a d d r e s s e s of t h e p a c k e t s b r o a d c a s t o n t h e E t h e r n e t s . Say t h a t a p a c k e t o n a n E t h e r n e t is local if it is d e s t i n e d for a n o t h e r n o d e of t h e s a m e E t h e r n e t . By m a i n t a i n i n g address tables, a b r i d g e l e a r n s w h i c h p a c k e t s are local. W h e n bridge C receives a p a c k e t t h a t it does n o t t h i n k is local, b r i d g e C r e t r a n s m i t s t h e p a c k e t o n t h e o t h e r E t h e r n e t . If t h e p a c k e t w a s in fact local, n o d a m a g e is d o n e since all t h e n o d e s o n t h e o t h e r E t h e r n e t will disregard t h e packet. T r a n s p a r e n t r o u t i n g is a v e r y c o n v e n i e n t p r o c e d u r e . However, it c a n m a k e packets loop forever b e t w e e n b r i d g e s u n l e s s s o m e care is t a k e n to avoid t h a t behavior. Consider Figure 3.16. A s s u m e t h a t t h e top-left n o d e in t h e figure s e n d s a p a c k e t d e s t i n e d to t h e r i g h t m o s t n o d e . T h e p a c k e t is s e e n b y two bridges, say A a n d B. Bridge A r e t r a n s m i t s t h e p a c k e t o n t h e s e c o n d E t h e r n e t b e c a u s e t h e p a c k e t is n o t local. W h e n bridge Β sees t h a t p a c k e t o n t h e s e c o n d E t h e r n e t , it r e t r a n s m i t s it o n t h e top E t h e r n e t b e c a u s e t h e p a c k e t is n o t local to t h e s e c o n d E t h e r n e t . T h u s bridge A gets t h e p a c k e t a s e c o n d time, a n d t h e bridges A a n d Β r e p e a t this s e q u e n c e of r e t r a n s m i s s i o n s indefinitely. Tb p r e v e n t s u c h a n infinite loop, t h e b r i d g e s u s e spanning tree routing. A s p a n n i n g t r e e in a g r a p h is a s u b g r a p h t h a t is a t r e e (i.e., loop-free) a n d t h a t s p a n s all t h e n o d e s (but n o t n e c e s s a r i l y all t h e bridges). If t h e b r i d g e s k n o w a t r e e t h a t s p a n s all

A

3 16 FIGURE

One method that bridges use to avoid making multiple copies with transparent routing is spanning tree routing. Only the bridges along the spanning tree shown by thick lines copy the packet.

123

Packet-Switched Networks

124

t h e n e t w o r k nodes, t h e n t h e y c a n avoid loops b y agreeing t h a t t h e p a c k e t s will b e r e t r a n s m i t t e d only b y t h e bridges o n t h e tree. A s p a n n i n g t r e e is s h o w n b y thicker lines in Figure 3.16. T h e bridges c o n s t r u c t a s p a n n i n g t r e e b y u s i n g a distributed a l g o r i t h m t h a t finds t h e shortest p a t h to destinations. T h e shortest p a t h to a n y d e s t i n a t i o n c a n n o t contain a n y loop. C o n s e q u e n t l y , t h e set of shortest p a t h s from all t h e bridges to a p a r t i c u l a r bridge m u s t b e a s p a n n i n g tree. We explain t h e s p a n n i n g t r e e a l g o r i t h m t h a t IEEE802 n e t w o r k s u s e o n t h e n e t w o r k of Figure 3.17. T h e k e y idea of t h e a l g o r i t h m is t h a t t h e root is elected at t h e s a m e t i m e t h a t t h e s p a n n i n g t r e e is c o n s t r u c t e d . T h e root is t h e bridge w i t h t h e smallest identification n u m b e r . I n this n e t w o r k , five E t h e r n e t LANs (a,..., e) are c o n n e c t e d w i t h six bridges ( E l , . . . , £ 6 ) . T h e b r i d g e s h a v e identification n u m b e r s (e.g., t h e i r IP addresses), a n d in o u r notation, t h e s e identification n u m b e r s are increasing from Bl to B6. Each bridge s e n d s m e s s a g e s of t h e following format: [ID.sender|ID.presumed_root|distance.presumed_root] w h e r e I D . s e n d e r is t h e identification n u m b e r of t h e bridge t h a t s e n d s t h e message, I D . p r e s u m e d _ r o o t is t h e identification n u m b e r of t h e b r i d g e t h a t t h e s e n d e r believes to b e t h e root of t h e s p a n n i n g tree, a n d d i s t a n c e . p r e s u m e d _ root is t h e c u r r e n t e s t i m a t e of t h e n u m b e r of h o p s b e t w e e n t h e s e n d e r a n d t h e p r e s u m e d root. Each bridge stores t h e b e s t m e s s a g e t h a t it h a s r e c e i v e d so far

a B1

B2

B3

B4

B6

B5

3.17 FIGURE

In this figure, a, . . . , e are Ethernet LANs and Bl, . . . , B6 are bridges.

3.2

Ethernet (IEEE 802.3)

125

from e a c h port. A m e s s a g e R2 receives t h e s e Ll__PDUs, it a s s e m b l e s t h e m into t h e L3_PDU. Subsequently,

Packet-Switched Networks

146

1J3__,,,,,,

FDDI networks can be interconnected using SMDS.

FIGURE

1. bridge b m u s t "convert" t h e L3_PDU b a c k into t h e original FDDI frame, place it o n FDDI 2, a n d t h e n r e m o v e it from t h a t ring w h e n it r e t u r n s ( n o t e t h a t t h e source address o n t h e frame is t h a t of u s e r 1 a n d n o t b y, 2

2

2. u s e r 2 copies this FDDI frame. Bridge b is o n FDDI 2 a n d o n t h e SMDS n e t w o r k . N e i t h e r u s e r 1 n o r u s e r 2 k n o w s t h a t t h e y are n o t o n t h e s a m e ring—the FDDI frame is forwarded t r a n s p a r e n t l y . Bridge b\ m u s t c a r r y o u t two functions: address c o n v e r s i o n a n d frame conversion. W h e n b\ r e a d s t h e FDDI frame a d d r e s s e d to u s e r 2, it m u s t recognize t h a t u s e r 2 is o n a "remote" r i n g t h a t c a n b e r e a c h e d via bridge b . T h i s a d d r e s s c o n v e r s i o n is carried o u t w i t h t h e h e l p of two address tables. T h e first h a s e n t r i e s of t h e form: ( r e m o t e u s e r MAC address, r e m o t e bridge SMDS address). T h e s e c o n d is a list of all MAC a d d r e s s e s o n its o w n ring. T h e bridge m u s t t h e n c o n v e r t t h e FDDI frame into a n L3_PDU. T h e m o s t c o m m o n w a y to do this is to e n c a p s u l a t e t h e e n t i r e FDDI frame as a payload in a n L3_PDU. T h i s is s h o w n in t h e top p a n e l of Figure 3.34. We will shortly see h o w t h e address table c a n b e c o n s t r u c t e d . After bridge b a s s e m b l e s t h e L3_PDU, it m u s t d e c a p s u l a t e it a n d r e c o v e r t h e FDDI frame. F r o m its a d d r e s s table, b recognizes t h a t t h e d e s t i n a t i o n address, u s e r 2, is o n its "own" ring. It t h e n places t h e frame o n t h e ring, a n d r e m o v e s it w h e n it r e t u r n s . 2

2

2

2

3.8

Summary

147

TCP

^

IP MAC

Bridge

MAC

^

IP

SIP

Γ

*

SIP

SMDS

FDDI

TCP

TCP IP

IP

IP MAC

MAC LAN 3.34

TCP

L3_PDU

FDDI

SIP

SIP

SIP = SMDS interface protocol

SMDS

I n t e r n e t w o r k i n g w i t h SMDS.

FIGURE

T h e two a d d r e s s tables, o n e w i t h r e m o t e MAC a n d SMDS b r i d g e addresses, t h e o t h e r w i t h MAC a d d r e s s e s o n its o w n ring, c a n b e built u p as follows. Each t i m e b\ receives a n L3_PDU, it n o t e s t h e SMDS a d d r e s s of t h e s e n d i n g b r i d g e and, from t h e e n c a p s u l a t e d FDDI frame, it obtains t h e MAC a d d r e s s of t h e source. In this w a y it b u i l d s t h e first table. T h e s e c o n d table is built s i m p l y from t h e source a d d r e s s o n e a c h FDDI frame o n its o w n ring. In case u s e r 1 u s e s t h e IP a d d r e s s of u s e r 2, t h e stations b\ a n d Z?2 will h a v e to b e routers. T h i s r e q u i r e s additional functions suggested in t h e l o w e r p a n e l of Figure 3.34.

3.8

SUMMARY T h e packet-switched n e t w o r k s s t u d i e d above s h o w a s t e a d y a d v a n c e in speed, connectivity, delay, a n d flexibility or ability to a c c o m m o d a t e m o r e k i n d s of traffic t y p e s . We s u m m a r i z e this d e v e l o p m e n t in I k b l e 3.1. All t h e n e t w o r k s in this table, except F r a m e Relay a n d SMDS, i m p l e m e n t layers 1 a n d 2 of t h e OSI m o d e l . In C h a p t e r 4 w e describe t h e I n t e r n e t Protocol or IP, w h i c h r o u g h l y c o r r e s p o n d to l a y e r s 3 a n d above of t h e OSI m o d e l . IP c a n b e i m p l e m e n t e d o n top of l a y e r 2 of t h e s e n e t w o r k s . F r a m e Relay a n d SMDS

Packet-Switched Networks

148

Name

3.1

Speed, connectivity

Delay

Application

Ethernet

10-100-1000 Mbps, local area

Random, increases with load

Transfer of messages between nearby computers

Tbken ring

4, 16 Mbps, local area

Random but bounded

TYansfer of messages between nearby computers, some real­ time traffic

FDDI

100 Mbps, LAN and campus

Random but bounded

LAN interconnections, real-time and CBR applications

DQDB

Unspecified

Random

Unspecified but similar to FDDI

Frame Relay

1.5 Mbps, wide area

Random, increases with load

Transfer of messages between distant computers

SMDS

1.5 to more than 45 Mbps

Random, traffic shaping

LAN interconnections, migration to ATM

Summary of advances in packet-switched networks.

TABLE

also i m p l e m e n t l a y e r 3 (the n e t w o r k layer). By m e a n s of a r o u t e r w i t h s o m e additional functionality, t h e s e n e t w o r k s c a n also i m p l e m e n t IP. Over its 100-year history, t h e services provided b y t h e t e l e p h o n e n e t w o r k s e e m h a r d l y to h a v e c h a n g e d . T h e quality of voice is m u c h i m p r o v e d , c o n n e c ­ tions c a n b e set u p t h r o u g h "direct dialing" to virtually a n y w h e r e in t h e world, service is very reliable, a n d costs h a v e steadily declined. N o n e t h e l e s s , u s e r s m u s t find t h e s e i m p r o v e m e n t s slow, cumulative, a n d u n r e m a r k a b l e . Tb a p p r e ­ ciate t h e progress of t h e t e l e p h o n e s y s t e m , o n e h a s to s t u d y t h e c h a n g e s in its infrastructure: t h e i n c r e a s e d t h r o u g h p u t a n d capabilities of its switches a n d its links. T h e 20-year history of packet-switched n e t w o r k s p r e s e n t s a s h a r p contrast. F r o m t h e v i e w p o i n t of u s e r applications, its progress h a s b e e n d r a m a t i c . Start­ ing from t h e h u m b l e b e g i n n i n g s of i n t e r c o n n e c t i n g a c o m p u t e r w i t h a p r i n t e r or a n o t h e r computer, t h e s e n e t w o r k s h a v e e x p a n d e d to e n a b l e u s e r s a r o u n d t h e world to c o m m u n i c a t e in t h e form of data, text, images, movies, a n d a n i m a t i o n . Several t e c h n i c a l a d v a n c e s c o o p e r a t e d in b r i n g i n g a b o u t this d r a m a t i c progress. T h e a c c e p t a n c e of t h e principles of l a y e r e d a r c h i t e c t u r e s for de-

3.10

Problems

149 veloping n e w services e n c o u r a g e d m o d u l a r i t y a n d r e u s e of existing services a n d p e r m i t t e d n e t w o r k e n g i n e e r s a n d application d e v e l o p e r s to w o r k i n d e p e n ­ dently. T h e choice of t r a n s p o r t i n g p a c k e t s as t h e b a s i c service t u r n e d o u t to b e ideally suited for i n t e r c o n n e c t i n g c o m p u t e r s a n d o t h e r devices, b e c a u s e it iso­ lated t h e rapid t e c h n i c a l a d v a n c e s in link s p e e d from t h e a d v a n c e s in software. Finally, t h e i n v e n t i o n of local a r e a n e t w o r k s s u c h as E t h e r n e t quickly r e d u c e d t h e cost of access at t h e s a m e t i m e as t h e i r s p e e d increased, a n d t h e capabil­ ities of w o r k s t a t i o n s a n d p e r s o n a l c o m p u t e r s k e p t u p w i t h t h e i m p r o v e m e n t s in access a n d speed.

3.9

NOTES T h e OSI m o d e l is described in detail in [T88, W98]. Standards for local area n e t w o r k s s u c h as E t h e r n e t , t o k e n ring, a n d FDDI are p u b l i s h e d b y t h e IEEE. Fast E t h e r n e t is described in [COUCR97, MW96], gigabit E t h e r n e t is described in [F98]. O n e wireless LAN p r o d u c t w i t h a CSMA MAC protocol is described in [CMM94]. E t h e r n e t switches a n d VLANs are discussed in [MW96] a n d in t e c h n i c a l d o c u m e n t a t i o n of switch m a n u f a c t u r e r s . F r a m e Relay, SMDS, a n d SONET (discussed in C h a p t e r 5) are d e s c r i b e d in [B95]. F r a m e Relay a n d ATM i n t e r w o r k i n g efforts are r e v i e w e d in [DE96].

3.10

PROBLEMS 1. T h e r e are m a n y cases in w h i c h a link is s h a r e d b y different devices. In s u c h cases t h e r e is s o m e explicit or implicit protocol t h a t r e g u l a t e s t h e access to t h e link. Discuss t h e following e x a m p l e s . (a) C o m p u t e r b a c k p l a n e b u s is u s e d b y m a n y devices (e.g., CPU, m e m o r y , I / O devices) to c o m m u n i c a t e . H o w is access to this c o m m o n b u s regulated? (b) In a cellular p h o n e s y s t e m , a fixed n u m b e r of c h a n n e l s are available for u s e b y all t h e m o b i l e p h o n e s w i t h i n t h e s a m e cell. ( T h e ratio of n u m b e r of p h o n e s to t h e n u m b e r of c h a n n e l s is t h e "pair gain/') H o w does a mobile p h o n e u s e r get access to a n idle c h a n n e l or l e a r n t h a t n o c h a n n e l is idle? (c) T h e r e are a fixed n u m b e r of seats in a t h e a t e r or b u s or airplane. M a n y u s e r s m a y w i s h to o c c u p y a seat. H o w is access regulated? Airlines

Packet-Switched Networks

150 1

3.35

ΓΡΤΓρΤ

Nil

Ν/2+1

Ν

By partitioning an Ethernet into work clusters, the network manager can accommodate more nodes.

sell m o r e tickets t h a n seats b e c a u s e p e o p l e cancel t h e i r flight. H o w is access r e g u l a t e d w h e n t h e r e is "overbooking"? (d) W h e n a driver w i s h e s to c h a n g e into a n adjacent l a n e in w h i c h t h e r e are o t h e r cars, t h e l a n e - c h a n g e m a n e u v e r n e e d s to b e c o o r d i n a t e d to p r e v e n t a n accident. W h a t "protocol" do drivers u s e to achieve coordi­ nation? 2. Figure 3.35 shows h o w n e t w o r k m a n a g e r s c a n partition a n E t h e r n e t net­ w o r k to a c c o m m o d a t e m o r e n o d e s . T h e figure a s s u m e s t h a t e a c h n o d e t r a n s m i t s .R bps, o n average, d u r i n g a r e p r e s e n t a t i v e p e r i o d of t i m e . T h e r e are Ν n o d e s to b e c o n n e c t e d . If t h e total t r a n s m i s s i o n rate Ν χ R is larger t h a n t h e rate t h a t c a n b e h a n d l e d b y o n e E t h e r n e t , t h e n t h e n e t w o r k m a n ­ ager c a n t r y to partition t h e n e t w o r k . For instance, if t h e efficiency of o n e E t h e r n e t c o n n e c t i n g t h e Ν n o d e s is 80% a n d if Ν χ R > 8 Mbps, t h e n o n e E t h e r n e t c a n n o t h a n d l e all t h e n o d e s . Let u s a s s u m e t h a t t h e Ν n o d e s c a n b e divided into two groups t h a t do n o t exchange m e s s a g e s frequently. For simplicity, say t h a t e a c h g r o u p s e n d s a fraction ρ of its m e s s a g e s to t h e o t h e r group. Let u s c o n n e c t t h e c o m p u t e r s in e a c h g r o u p w i t h a d e d i c a t e d E t h e r n e t , as s h o w n in t h e figure. T h e two E t h e r n e t s are c o n n e c t e d b y a bridge or b y a device called a switching huh t h a t p e r f o r m s t h e s a m e func­ tion b e t w e e n E t h e r n e t s e g m e n t s . Calculate t h e traffic o n e a c h E t h e r n e t . T h e a r r a n g e m e n t c a n h a n d l e t h e Ν n o d e s if this rate is less t h a n t h e effi­ ciency of e a c h E t h e r n e t t i m e s 10 Mbps. Of course, t h e bridge m u s t b e able to h a n d l e t h e traffic. 3. Discuss t h e flow of information at t h e correct layers in a c o n n e c t i o n from CI to C2.

3.10

Problems

151

ci

SI

Rl

R2

C2

Host

Ethernet Switch

Router

Router

Host

Application Transport Network Link/LAN Physical

4. (a) Consider t h e n e t w o r k topology d e p i c t e d in t h e figure b e l o w w h e r e t h e boxes labeled w i t h "?" are Ethernet hubs. (1) Redraw (roughly) t h e figure a n d indicate t h e e x t e n t of t h e E t h e r n e t "collision domains." (2) Does n o d e A's ARP table e v e r c o n t a i n e n t r i e s for n o d e Β a n d n o d e C n e t w o r k addresses? (3) W h e n n o d e A is s e n d i n g a p a c k e t to n o d e C, w h a t E t h e r n e t a d d r e s s does n o d e A u s e for t h e d e s t i n a t i o n ? (4) Suppose t h e r e is two-way c o m m u n i c a t i o n b e t w e e n n o d e s A a n d B. Does n o d e C get to see this traffic? (b) A s s u m e t h a t t h e boxes l a b e l e d w i t h "?" are Ethernet switches. A n s w e r t h e s a m e sub-questions from (a). (c) A s s u m e t h a t t h e boxes labeled w i t h "?" are I n t e r n e t Protocol routers. A n s w e r t h e s a m e sub-questions from (a).

® )

Hub

"?"

Hub

"?"

Hub

( \

Ό

T h e OSI t r a n s p o r t l a y e r s e g m e n t s a m e s s a g e into p a c k e t s a n d a p p e n d s a s e q u e n c e n u m b e r to e a c h p a c k e t so t h a t t h e r e c e i v e r c a n r e a s s e m b l e t h e original m e s s a g e . In practice, t h e s e q u e n c e n u m b e r is selected m o d u l o N, for s o m e i n t e g e r N, a n d c o n s e c u t i v e p a c k e t s are n u m b e r e d 0, 1, . . . , Ν — 1, 0, . . . . (For example, in SMDS t h e s e q u e n c e n u m b e r h a s 4 bits, so Ν = 16.) With this choice of s e q u e n c e n u m b e r s , t h e r e c e i v e r w o u l d b e

Packet-Switched Networks

152

u n a b l e to detect t h e loss of Ν consecutive p a c k e t s of t h e s a m e m e s s a g e . H o w large s h o u l d Ν b e to k e e p t h e probability of m i s s e d d e t e c t i o n small? H o w large is Ν for IP, ATM, SMDS? 6. In this p r o b l e m y o u design a protocol b e t w e e n two n o d e s t h a t recovers w h e n o n e of t h e n o d e s crashes. Node A sets u p a c o n n e c t i o n w i t h n o d e B. Node Β s e n d s packets to A u s i n g GBN. Node Β closes t h e c o n n e c t i o n w h e n it h a s s e n t all t h e packets. Make sensible a s s u m p t i o n s a b o u t t h e c h a n n e l s b e t w e e n t h e two n o d e s . 7. Modify Figure 3.21 to reflect t h e release after t r a n s m i s s i o n protocol u s e d b y t h e 16-Mbps t o k e n ring n e t w o r k . Show t h a t its efficiency is a p p r o x i m a t e l y (1 + a/N)~ , w h i c h a p p r o a c h e s 100% as Ν b e c o m e s v e r y large. l

8. T h e physical layer of a token bus network is a b r o a d c a s t coax cable like E t h e r n e t , b u t t h e MAC l a y e r is like t h e t o k e n ring. Each station o n t h e b u s h a s a n u m b e r 1 , 2 , . . . , say. After station i t r a n s m i t s its packets, it releases a token, indicating t h a t t h e t o k e n is i n t e n d e d for station i + 1. W h e n t h e last station, say N, finishes t r a n s m i s s i o n , it releases t h e token, indicating t h a t t h e next station is station 1. Analyze t h e efficiency of t h e t o k e n b u s network. 9. Suppose t h e r e are Ν stations o n a n E t h e r n e t . Suppose t h e probability is ρ t h a t a n y of t h e m h a s a packet r e a d y for t r a n s m i s s i o n a n d s u p p o s e t h a t different stations are i n d e p e n d e n t . W h a t is t h e probability p that m stations h a v e a p a c k e t r e a d y for t r a n s m i s s i o n , m = 0, 1, . . . , N? T h e average n u m b e r of p a c k e t s r e a d y for t r a n s m i s s i o n is λ :=pN. Suppose t h a t Ν oo, ρ -> 0 in s u c h a w a y t h a t Ν xp^ λ. For this a s y m p t o t i c case, s h o w t h a t t h e probability t h a t t h e r e are m p a c k e t s to t r a n s m i t is Poisson w i t h m e a n λ. m

10. Consider t h e wireless p a c k e t n e t w o r k s h o w n in t h e figure o n t h e next page. Four stations s h a r e a slotted ALOHA c h a n n e l b y t r a n s m i t t i n g w i t h proba­ bility 0.25 e a c h in e v e r y slot, i n d e p e n d e n t l y of o n e another. In addition to t h e risk of collision, p a c k e t s t h a t are t r a n s m i t t e d also face t r a n s m i s s i o n er­ rors. A s s u m e t h a t errors c o r r u p t p a c k e t s t h a t w e r e o t h e r w i s e successfully t r a n s m i t t e d w i t h probability 0.1. A c k n o w l e d g m e n t s are given priority a n d are c o r r u p t e d b y e r r o r s w i t h probability 0.1. Specifically, a s s u m e t h a t A t r a n s m i t s a packet to B. If t h e t r a n s m i s s i o n is successful (possibly w i t h errors), t h e n Β gets t h e c h a n n e l for o n e slot to t r a n s m i t o n e ACK: n o o t h e r station i n t e r r u p t s t h e t r a n s m i s s i o n of B's ACK. With this p r o c e d u r e , t h e t r a n s m i s s i o n of a p a c k e t b y A is c o m p l e t e w i t h s o m e probability after 2 slots. If t h a t t r a n s m i s s i o n is n o t c o m p l e t e , t h e n A k n o w s t h a t it m u s t try to t r a n s m i t t h a t p a c k e t later.

3.10

Problems

153

LU l£J L5J

LAJ I

l

I

I

I

I

I

I

I

I

I

I

I

I

I

I

I

» time

Calculate t h e average reliable t h r o u g h p u t of p a c k e t s over this c h a n n e l , in p a c k e t s p e r slot. 11. If y o u r c o m p u t e r is c o n n e c t e d to a n E t h e r n e t , find o u t w h a t t h e n e t w o r k utilization is, h o w m a n y collisions h a v e occurred, a n d o t h e r n e t w o r k statis­ tics t h a t are available. W h a t is t h e m a x i m u m s u s t a i n e d rate at w h i c h y o u r c o m p u t e r c a n g e n e r a t e n e t w o r k traffic? W h a t is limiting this rate? Is it t h e disk, I / O b u s , or s o m e o t h e r factor? 12. Recall t h e FDDI MAC protocol. (a) Show t h a t t h e m a x i m u m t i m e b e t w e e n successive t o k e n arrivals in t h e FDDI n e t w o r k is TTRT + TRANS. (b) Show t h a t this t i m e is 2 TTRT (neglecting p a c k e t t r a n s m i s s i o n t i m e ) w h e n s y n c h r o n o u s traffic is i n c l u d e d . (c) Show t h a t w h e n provision for s y n c h r o n o u s traffic is m a d e , t h e t o k e n arrivals are n o t periodic. W h a t is t h e m a x i m u m deviation in t h e t o k e n interarrival t i m e for s y n c h r o n o u s traffic? H o w m u c h buffering w o u l d b e n e e d e d to transfer c o n s t a n t bit rate s y n c h r o n o u s traffic? (d) Give a definition of fairness a n d argue t h a t FDDI p r o v i d e s fair access to all stations. (e) T h e DQDB MAC protocol a t t e m p t s to regulate access as if all t h e stations p l a c e d t h e i r p a c k e t s in a single q u e u e t h a t is s e r v e d o n a firstcome, first-served (FCFS) basis. Would y o u call FCFS access fair? W h y ? 13. We w a n t to analyze t h e efficiency of DQDB similar to t h e analysis of t h e t o k e n ring n e t w o r k in section 3.3.2. Suppose t h e r e are Ν e q u a l l y s p a c e d stations, indexed 1, 2, . . . , N, a r r a n g e d from left to right, a n d s u p p o s e PROP is t h e p r o p a g a t i o n t i m e b e t w e e n adjacent stations. Suppose Τ is t h e t r a n s m i s s i o n t i m e at e a c h station of a 53-byte frame. (For example, if t h e link s p e e d is 100 Mbps, t h e n Τ = 53 χ 8 χ 1 0 ~ s.) 8

(a) Suppose t h e h e a d station t r a n s m i t s F idle frames p e r second, b a c k to back. W h a t is F (in t e r m s of T)? (b) We define utilization as t h e fraction of frames t h a t c a r r y data. Suppose t h a t at all t i m e s at least o n e station h a s s o m e i n f o r m a t i o n to s e n d to its right. Show t h a t t h e utilization of b u s A in Figure 3.27 is 100%. (c) Consider two stations i a n d ;' w i t h i to t h e left of;. Suppose b o t h are t r a n s m i t t i n g to a station k o n t h e i r right. Suppose i a n d ; receive p a c k e t s

Packet-Switched Networks

154

to t r a n s m i t at t i m e s U a n d tj. Suppose t{ > tj, t h a t is, i's p a c k e t arrived later t h a n ; ' s packet. Show t h a t t h e DQDB protocol w o r k s in s u c h a w a y t h a t ; will t r a n s m i t its p a c k e t before i. N o w s u p p o s e t{ < tj. C o n s t r u c t a n example s u c h t h a t t h e protocol will t r a n s m i t i's p a c k e t after ; ' s . W h a t is t h e m a x i m u m a m o u n t of t i m e tj — U for w h i c h this could h a p p e n ? (Note: In a t r u e first-come, first-served s y s t e m , i s h o u l d t r a n s m i t before j w h e n e v e r ti < tj.) 14. Consider t h e expressions s h o w n in Figure 3.28. Show t h a t if t r a n s m i s s i o n from e a c h n o d e to t h e next c o r r u p t s a p a c k e t w i t h probability 1—2?, a n d if e a c h frame is r e t r a n s m i t t e d until it is successfully received, t h e n e a c h frame is t r a n s m i t t e d 1 /p t i m e s o n average before b e i n g successfully trans­ mitted. Calculate t h e p a c k e t e r r o r rate 1 — ρ for p a c k e t s of size 100 a n d 1,000 b y t e s a n d bit error rates of 1 0 " a n d 1 0 ~ . 4

8

15. Suppose in Figure 3.28 t h a t σ\ = 50 m s a n d 02 — 3 m s . Suppose k = 10, so t h e r e are 10 n o d e s . H o w small s h o u l d (1 —p), t h e p a c k e t e r r o r rate p e r link, b e before F r a m e Relay h a s less delay t h a n X.25? 16. Suppose a n FDDI n e t w o r k c o n n e c t s two E t h e r n e t s as s h o w n in Figure 3.36. Bridge b\ is a t t a c h e d to o n e E t h e r n e t a n d t h e FDDI, w h i l e 2?2 is a t t a c h e d to t h e o t h e r E t h e r n e t a n d t h e FDDI. T h e s e b r i d g e s are s u p p o s e d to pro­ vide t r a n s p a r e n t routing. Explain h o w c o m p u t e r A c a n s e n d p a c k e t s to Β using B's MAC address? Explain h o w t h e bridges c a n obtain t h e i r a d d r e s s conversion tables.

Bridge/router

A

Ο

3.36 Λ Μ Μ » ^

Bridges b\ and Z?2 must be designed so that computer A can transparently send packets to Β connected to a different Ethernet. Note that the MAC addresses on FDDI and Ethernet are different.

4 CHAPTER

The Internet and TCP/IP Networks

I his c h a p t e r explains t h e I n t e r n e t a n d n e t w o r k s t h a t u s e t h e s a m e protocols. We call t h e s e n e t w o r k s TCP/IP Networks b e c a u s e TCP a n d IP are t h e i r m o s t i m p o r t a n t protocols. For instance, c o m p a n i e s b u i l d isolated T C P / I P n e t w o r k s called intranets to c o n n e c t t h e i r c o m p u t e r s a n d servers. In this c h a p t e r w e first discuss t h e I n t e r n e t a n d its topology. We t h e n describe t h e IP a n d TCP protocols, a n d s o m e of t h e major applications, i n c l u d i n g t h e h t t p protocol u s e d in t h e World Wide Web. We also discuss p e r f o r m a n c e characteristics of t h e T C P / I P n e t w o r k s a n d p r o p o s e d u p g r a d e s of t h e IP a n d TCP protocols. Section 4.1 explains t h e I n t e r n e t . Section 4.2 clarifies t h e a d d r e s s i n g a n d r o u t i n g a n d reviews t h e l a y e r e d s t r u c t u r e of t h e protocols. Section 4.3 discusses t h e I n t e r n e t Protocol. Section 4.4 describes t h e TCP a n d U D P protocols a n d considers i m p o r t a n t applications. Section 4.5 looks at b o t h t h e success a n d t h e limitations of t h e I n t e r n e t . Section 4.6 discusses p e r f o r m a n c e issues of t h e Internet.

4.1

THE INTERNET T h e I n t e r n e t today i n t e r c o n n e c t s a large n u m b e r of c o m p u t e r s a n d n e t w o r k s t h r o u g h o u t t h e world. T h e r e w e r e 1 million s u c h c o m p u t e r s in early 1993, 5 million in 1995, 16 million in 1997, a n d over 50 million in 1999 organized in 2 million d o m a i n s .

156

The Internet and TCP/IP Networks

T h e I n t e r n e t h a s its origin in t h e ARPANET n e t w o r k s p o n s o r e d b y t h e U.S. D e p a r t m e n t of Defense starting in t h e 1960s. T h e ARPANET w a s a d a t a g r a m store-and-forward n e t w o r k t h a t t h e D e p a r t m e n t of Defense liked for its abil­ ity to r e r o u t e p a c k e t s a r o u n d failures. This feature m a k e s d a t a g r a m n e t w o r k s survivable. A n o t h e r i m p o r t a n t objective of ARPANET w a s to e n a b l e t h e inter­ c o n n e c t i o n of h e t e r o g e n e o u s n e t w o r k s . T h e t e c h n i c a l success of t h e I n t e r n e t is d u e to t h e large variety of applications (from.e-mail a n d telnet, to file transfer a n d WWW) t h a t IP c a n s u p p o r t o n t h e o n e h a n d and, o n t h e o t h e r h a n d , t h e m a n y different n e t w o r k s t h a t c a n i m p l e m e n t IP. (See Figure 4.2 in section 4.2.) We will discuss this k e y feature at t h e e n d of this section. A major factor t h a t c o n t r i b u t e d to t h e p o p u l a r i t y of t h e I n t e r n e t w a s t h e exploitation of n e t w o r k externalities. This w a s a c h i e v e d t h r o u g h early stan­ dardization a n d free distribution of its protocols a n d t h e i r software i m p l e ­ m e n t a t i o n s , w h i c h could r u n o n p e r s o n a l c o m p u t e r s as well as o n worksta­ tions a n d m a i n f r a m e c o m p u t e r s . N e t w o r k externalities w e r e c r e a t e d b y t h e National Science Foundation's subsidy of t h e c o n s t r u c t i o n a n d u s e of t h e In­ t e r n e t . Since 1995, e x p a n s i o n a n d d e v e l o p m e n t h a v e b e e n funded b y private enterprise. T h e spectacular growth of t h e I n t e r n e t is fueled largely b y t h e World Wide Web, w h i c h m a k e s m u l t i m e d i a information available at t h e click of a m o u s e . T h e vast increase in d e m a n d for I n t e r n e t traffic explains t h e rapid g r o w t h in n e t w o r k capacity. It is e s t i m a t e d t h a t in 1999 t h e v o l u m e of data traffic is c o m p a r a b l e to t h a t of voice. Moreover, b e c a u s e data traffic d o u b l e s e v e r y y e a r w h e r e a s voice traffic i n c r e a s e s o n l y b y a b o u t 10% a year, voice traffic will soon a m o u n t to only a small fraction of t h e total. T h e s e observations l e a d to t h e conclusion t h a t t h e future n e t w o r k s h o u l d b e o p t i m i z e d for data and, in t h e b a c k g r o u n d , should b e able to c a r r y voice traffic reliably a n d w i t h a small delay. T h e I n t e r n e t is a n e t w o r k of n e t w o r k s . It c o m p r i s e s b a c k b o n e n e t w o r k s of point-to-point links t h a t c o n n e c t regional gateways called network access points (NAPs). Routers a t t a c h e d to t h e NAPs are called points of presence (PoPs). Subscribers c o n n e c t to a PoP e i t h e r w i t h a dial-in m o d e m over t h e i r tele­ p h o n e line, or w i t h a digital subscriber l i n e (typically ADSL), a cable m o ­ d e m , or a leased digital line. Some b u s i n e s s e s c o n n e c t to a PoP w i t h a n op­ tical link. T h e c u s t o m e r ' s c o m p u t e r s are typically i n t e r c o n n e c t e d w i t h a local network. Over time, a n d i n c r e m e n t a l l y , link s p e e d s h a v e i n c r e a s e d from 56 Kbps to 1.5 Mbps to 45 Mbps. T h e r e c e n t explosive g r o w t h in d e m a n d is b e i n g m e t b y 155-Mbps, 622-Mbps, a n d h i g h e r - s p e e d (2.4 Gbps a n d 9.6 Gbps) SONET links,

l e a s e d from t e l e p h o n e c o m p a n i e s . Local a r e a n e t w o r k s p e e d s h a v e i n c r e a s e d as well to 100-Mbps E t h e r n e t LANs a n d FDDI rings, Gbps E t h e r n e t LANs, a n d s o o n to 10-Gbps E t h e r n e t LANs. T h e I n t e r n e t is u s e d for applications t h a t r e q u i r e (relatively) low t r a n s m i s s i o n r a t e s a n d t h a t c a n tolerate large delays. Recent a d v a n c e s are a i m e d at a c c o m m o d a t i n g applications t h a t n e e d h i g h e r n e t w o r k performance. Until April 1995, t h e b a c k b o n e of t h e I n t e r n e t w a s m a n a g e d u n d e r t h e auspices of t h e National Science F o u n d a t i o n . T h e topology of t h e b a c k b o n e w a s simple, w i t h four public NAPs a n d two private o n e s . W h e n c o m p e t i n g c o m m e r c i a l carriers took over t h e b a c k b o n e , t h e topology a n d r o u t e s b e c a m e m u c h m o r e complex. Tbday, t h e r e are 75 public NAPs a r o u n d t h e world, 12 of t h e m in t h e U n i t e d States. M a n y t e l e c o m m u n i c a t i o n c o m p a n i e s are b u i l d i n g U.S.-wide I n t e r n e t b a c k b o n e s . As of May 1999, t h e s e b a c k b o n e s i n c l u d e t h o s e of ANS, AT&T, BBN/GTE, CERFNET, DIGEX, EBONE, MCI, NETCOM, PSI, Qwest, Sprint, UUNET, a n d Verio. Figure 4.1 s h o w s t h e topology of t h e ANS b a c k b o n e . ANS (Advanced N e t w o r k s & Services, Inc.) w a s a subsidiary of A m e r i c a O n Line u n t i l 1998 w h e n it w a s a c q u i r e d b y WorldCom. You c a n find t h e m a p s of t h e o t h e r b a c k b o n e s at http://boardwatch.internet.com/isp. M u c h of t h e traffic is r o u t e d t h r o u g h private i n t e r c o n n e c t i o n s . T h e s e "private

4.1 FIGURE

ANS backbone. This is one of m a n y U.S.-wide backbones deployed by various telecommunication companies.

The Internet and TCP/IP Networks

158

peering" a r r a n g e m e n t s b e t w e e n ISPs c a n take place a n y w h e r e t h e t r a n s a c t i o n is m u t u a l l y c o n v e n i e n t , a n d t h e y a c c o u n t for a n e s t i m a t e d two-thirds of all I n t e r n e t traffic.

4.2

OVERVIEW OF INTERNET PROTOCOLS In this section w e describe t h e m a i n c o m p o n e n t s of t h e I n t e r n e t protocols. T h e I n t e r n e t protocol layers are s h o w n in Figure 4.2. T h e figure s h o w s t h e c o r r e s p o n d e n c e b e t w e e n t h e IP protocols a n d t h e OSI seven-layer m o d e l . T h e c o r r e s p o n d e n c e is inexact. M a n y n e t w o r k s i m p l e m e n t l a y e r s 1 a n d 2, i n c l u d i n g those studied earlier. Layer 3, t h e n e t w o r k layer, is i m p l e m e n t e d b y IP. Layer 4, t h e t r a n s p o r t layer, is i m p l e m e n t e d b y two protocols, U D P a n d TCP. T h e r e is n o direct c o u n t e r p a r t to t h e h i g h e r OSI layers. As Figure 4.2 shows, t h e I n t e r n e t protocols are internetworking protocols. T h a t is, t h e s e protocols are d e s i g n e d to glue t o g e t h e r m a n y different n e t w o r k s s u c h as E t h e r n e t LANs, FDDI, wireless n e t w o r k s , a n d point-to-point links. Each

Τ Η

e

Τ Τ

1 η

Ρ

e

FTP

S Μ

S Ν

Τ

Μ

Ρ

Ρ

RCP

R S Η

TCP IP,ICMP,ARP,RARP X.25, Ethernet, token ring, FDDI, Tl,...

FIGURE

0

8 i

Τ F

7

Τ

6

Ρ

η

t

4.2

r 1

5 UDP

4 3 1.2

Internet protocols are arranged in a layered hierarchy and compared with the OSI seven-layer model. On top of the basic IP or network layer and TCP/UDP or transport layer are generic applications such as Web page transfer (Hypertext Transfer Protocol or HTTP), file transfer (File Transfer Protocol or FTP), remote file copy (rep), remote terminal (Tfelnet), remote login (rlogin), network m a n a g e m e n t (Simple Network Management Protocol or SNMP), and e-mail (Simple Mail Transfer Protocol or SMTP). User applications such as WWW and Microsoft's Outlook are built on top of these generic applications.

4.2

Overview of Internet Protocols

of t h e s e n e t w o r k s u s e s its specific a d d r e s s i n g c o n v e n t i o n s , p a c k e t format, a n d protocols. T h e IP protocol sees t h e s e n e t w o r k s as i m p l e m e n t i n g virtual links b e t w e e n t h e i r n o d e s a n d does n o t c o n c e r n itself w i t h t h e characteristics of t h e s e links. T h e only a s s u m p t i o n s t h a t t h e IP protocol m a k e s a b o u t t h e s e virtual links is t h a t t h e y c a n t r a n s p o r t p a c k e t s of u p to s o m e m a x i m u m size. T h e virtual links m a y b e u n r e l i a b l e a n d h a v e a w i d e r a n g e of t r a n s m i s s i o n rates a n d delays. T h e IP protocol u s e s t h e s e virtual links t h a t t h e individual n e t w o r k s im­ p l e m e n t to deliver p a c k e t s end-to-end across a n u m b e r of s u c h links. T h i s end-to-end delivery n e c e s s i t a t e s t w o basic tasks: a d d r e s s i n g a n d routing. T h e IP protocol defines globally u n i q u e a d d r e s s e s for t h e h o s t c o m p u t e r s a n d routers. T h e s e a d d r e s s e s are i n d e p e n d e n t of t h o s e a l r e a d y defined b y t h e specific net­ w o r k s to w h i c h t h e h o s t s are attached. T h a t is, if a c o m p u t e r is o n a n E t h e r n e t LAN, t h e E t h e r n e t interface card of t h a t c o m p u t e r h a s a 48-bit MAC address. If this E t h e r n e t LAN is a t t a c h e d to a T C P / I P n e t w o r k , t h e n t h e E t h e r n e t interface of t h e c o m p u t e r also h a s at least o n e IP address. T h e IP a d d r e s s e s are organized in a hierarchical way. In this organization, t h e c o m p u t e r s are g r o u p e d into clus­ t e r s called IP subnets. F r o m t h e IP a d d r e s s of a computer, o n e c a n d e t e r m i n e its IP s u b n e t . T h e IP r o u t i n g u s e s this hierarchical s t r u c t u r e of t h e a d d r e s s e s to r e d u c e t h e size of t h e r o u t i n g tables t h a t t h e r o u t e r s m a i n t a i n . We explained t h a t t h e I n t e r n e t Protocol delivers p a c k e t s from e n d to e n d in t h e I n t e r n e t or, m o r e generally, in a T C P / I P n e t w o r k . Tb m a k e t h e s e p a c k e t deliveries m o r e directly usable b y applications, t h e protocols of l a y e r 4 per­ form a few additional tasks: multiplexing, e r r o r control a n d reordering, flow control, a n d congestion control. Multiplexing m e a n s a d d i n g a "port n u m b e r " to t h e p a c k e t s to distinguish p a c k e t s d e s t i n e d to different applications. A n appli­ cation "transmits" a n d "listens" to a specific port. Error control m e a n s a r r a n g i n g for p a c k e t s t h a t do n o t arrive correctly to b e r e t r a n s m i t t e d ; reordering m e a n s delivering t h e p a c k e t s in t h e i r o r d e r of t r a n s m i s s i o n . Flow control m e a n s stop­ ping t h e t r a n s m i s s i o n s w h e n t h e d e s t i n a t i o n r u n s o u t of buffer space. Finally, congestion control m e a n s slowing d o w n w h e n p a c k e t s are getting d r o p p e d in t h e n e t w o r k , w h i c h indicates t h a t t h e n e t w o r k is probably congested. We ex­ plain t h e s e additional tasks in t h e following pages. For now, w e n o t e t h a t UDP, t h e u s e r d a t a g r a m protocol, o n l y a d d s m u l t i p l e x i n g to t h e p a c k e t delivery a n d discards c o r r u p t e d packets. TCP, t h e t r a n s m i s s i o n control protocol, a d d s all t h e tasks t h a t w e m e n t i o n e d so t h a t it i m p l e m e n t s error-free, ordered, a n d wellp a c e d deliveries of packets. In t h e sections t h a t follow, w e e x a m i n e IP, UDP, a n d TCP. We also discuss HTTP, TFTP, a n d a few o t h e r application l a y e r protocols in s o m e detail.

159

The Internet and TCP/IP Networks

160

4.3

INTERNET PROTOCOL T h e n e t w o r k layer of I n t e r n e t , called t h e Internet Protocol (IP), is t h e m o s t i m p o r t a n t . T h e version in use, first i m p l e m e n t e d in 1984, is IPv4. (See RFC 791.) Version 5 w a s u s e d for s o m e e x p e r i m e n t s . Version 6, IPv6, is rarely i m p l e m e n t e d in 1999 a n d m a y b e i m p l e m e n t e d m o r e widely over t h e next few years. As w e explained in t h e p r e v i o u s section, IP organizes t h e a d d r e s s e s hierar­ chically a n d m a i n t a i n s t h e r o u t i n g tables of t h e r o u t e r s . In addition, IP r e p o r t s s o m e delivery p r o b l e m s . We start b y describing IPv4, t h e m u l t i c a s t a n d mobile version of IP, t h e n we explain t h e modifications of IPv6.

4.3.1

IPv4 In t e r m s of t h e O p e n Data N e t w o r k m o d e l of section 2.8, t h e IP b e a r e r service is t h e delivery of m e s s a g e s in t h e form of d a t a g r a m s of size u p to 2 b y t e s (64 Kbytes). T h i s delivery service carries n o g u a r a n t e e of service quality in t e r m s of error, delay, or b a n d w i d t h . Such service is called best-effort service, m e a n i n g t h e r e b y t h a t t h e n e t w o r k will a t t e m p t to do t h e b e s t it can. IP discards p a c k e t s w h o s e h e a d e r is c o r r u p t e d . Figure 4.3 illustrates t h e m a i n ideas of IP routing. T h e figure s h o w s two LANs t h a t are a t t a c h e d t o g e t h e r w i t h r o u t e r Rl a n d to t h e rest of t h e I n t e r n e t w i t h r o u t e r R2 at t h e PoP of t h e ISP. T h e link b e t w e e n Rl a n d R2 u s e s s o m e specific framing t h a t w e do n o t e x a m i n e h e r e . Each c o m p u t e r h a s a LAN a d d r e s s a n d a n IP address. T h e IP a d d r e s s e s of t h e c o m p u t e r s of LAN1 h a v e t h e form IPl.x a n d t h o s e of LAN2 h a v e t h e form IP2.Z/. T h e r o u t e r Rl m a i n t a i n s t h e routing table s h o w n in t h e figure. T h i s table specifies w h e r e Rl s h o u l d s e n d t h e p a c k e t s next. A s s u m e t h a t c o m p u t e r A w i t h IP a d d r e s s IP1.4 w a n t s to s e n d [ data 1 ] to c o m p u t e r Β w i t h IP a d d r e s s IP2.3. H e r e are t h e s t e p s t h a t are r e q u i r e d for this packet transfer: 1 6

1. Given t h e n a m e of c o m p u t e r B, c o m p u t e r A discovers its IP a d d r e s s IP2.3, b y using a directory service called DNS. 2. C o m p u t e r A places [ data 1 ] in a n IP p a c k e t w i t h source a d d r e s s IP1.4 a n d destination address IP2.3. This IP p a c k e t is [ IP1.4 | IP2.3 | data 1 ]. 3. C o m p u t e r A d e t e r m i n e s t h a t it m u s t s e n d [ IP1.4 | IP2.3| data 1 ] to R l . Tb m a k e this d e t e r m i n a t i o n , c o m p u t e r A n o t e s t h a t t h e IP a d d r e s s IP2.3 is n o t

4.3

Internet Protocol

1

'

— C o m p u t e r C: E R ^ — ^ IP5.3 Internet J

LinkL Porte

SlIiifillililBl

IPl IP2 Default

I

P

2

1

Portb

Porta Portb Porte

IP2.3

43 FIGURE

A TCP/IP network. The figure shows two LANs attached to the rest of the Internet.

o n LAN1 as it is n o t of t h e form I P l . * . C o m p u t e r A is configured w i t h t h e address IP1.1 of t h e "default gateway" Rl to w h i c h it m u s t s e n d p a c k e t s t h a t leave LAN1. 4. I b s e n d [ I P l . 4 | IP2.3 | data 1 ] to Rl over LAN1, c o m p u t e r A places it in a frame of t h e format r e q u i r e d b y LAN1. For instance, if LAN1 is a n E t h e r n e t LAN, t h e n t h a t format looks like [ m a c ( I P l . l ) | mac(IP1.4) | I P l . 4 | IP2.3| data 1 I CRC ], w h e r e m a c ( I P l .1) a n d m a c ( I P l .4) are t h e MAC a d d r e s s e s of t h e n e t w o r k interfaces of Rl a n d A o n LAN1, a n d CRC is t h e error-detection field. We designate t h a t frame b y [1] in t h e figure. 5. W h e n it gets t h e packet, Rl r e m o v e s it from its E t h e r n e t frame a n d recovers [ I P l .4 I IP2.3 I data 1 ]. Rl t h e n c o n s u l t s its r o u t i n g table a n d finds t h a t t h e s u b n e t w i t h a d d r e s s e s lF2.y is a t t a c h e d to p o r t b. 6. Tb s e n d [ I P l . 4 | IP2.3 | data 1 ] to c o m p u t e r Β over LAN2, Rl places it into a frame w i t h t h e format suitable for LAN2. We d e s i g n a t e t h a t frame b y [2] in t h e figure.

The Internet and TCP/IP Networks

162 A LANl

Rl

LAN2 Β

| [data 1]

1] Η

1,2

- M -

-[2]= [ IP1.4 I IP2.3 I data 1 ]

[data 2]

• 4.4_

=[IP1.4IIP5.3ldata2]

Symbolic representation of the transfers of packets at layers 1 through 3.

FIGURE

7. Eventually, c o m p u t e r Β gets t h e packet, r e m o v e s it from its LAN2 frame a n d from its IP e n v e l o p e a n d extracts [ data 1 ]. T h e top p a r t of Figure 4.4 s h o w s t h e p a c k e t transfers at t h e different layers. T h e packet [ data 1 ] arrives at t h e IP l a y e r in c o m p u t e r A. T h e IP l a y e r adds t h e IP a d d r e s s e s of t h e source a n d destination. T h e IP l a y e r also finds o u t t h a t t h e packet m u s t go to R l . T h e IP l a y e r t h e n gives this IP p a c k e t to t h e LANl layer t h a t delivers it to R l . T h i s transfer is in a format [1] suitable for L A N l . W h e n it gets t h e packet, Rl invokes its IP layer, w h i c h d e t e r m i n e s w h e r e to s e n d t h e packet next. T h e IP l a y e r gives t h e p a c k e t to t h e LAN2 layer, w h i c h delivers it to c o m p u t e r B. Note from this diagram t h a t IP views LANl a n d LAN2 as virtual links. Note also t h a t t h e LANs are n o t a w a r e of t h e IP addresses. Figure 4.3 also shows t h e transfer of [ data 2 ] from c o m p u t e r A to c o m p u t e r C w i t h IP address IP5.3 s o m e w h e r e o n t h e I n t e r n e t . T h e steps are similar, except for step 5. W h e n it gets t h e p a c k e t [ IP1.4 | IP5.3 | data 1 ], Rl c o n s u l t s its r o u t i n g table. Since n o r o w c o r r e s p o n d s to a d d r e s s e s of t h e form IP5.z, Rl s e n d s t h e p a c k e t o n t h e "default" port, p o r t c. Figure 4.4 s h o w s t h e transfers at t h e various layers. We designate b y [3] t h e p a c k e t [ IP1.4 | IP5.3 | data 1 ] in t h e format for LANl from A to R l , b y [4] t h a t p a c k e t in t h e format for l i n k L from

4.3

Internet Protocol

163

0

31 1

_l

VER

1

1

1

1

IHL

1

1

1

1

1

J

1

1

1

1

1

1

1

1

L... J

1

1

1

J

1

1

1

.

Total length

Service type Flag

Identification Time to live

1

Protocol

Fragment offset Header checksum

Source network address Destination network address

WSUiM^S^S^^^^^^^^ 4.5 IIHIHMIIIIIII.

• / +>PP

T h e IP header specifies t h e source a n d destination addresses a n d control fi id described in t h e text. (VER = version; IHL = h e a d e r length/32 bits.) e

s

Rl to R2, a n d b y [5] t h a t p a c k e t in t h e format for t h e link t h a t exits R2 toward t h e rest of t h e I n t e r n e t . Each IP p a c k e t h a s a n IP h e a d e r of at least 20 b y t e s . (See Figure 4.5.) T h e h e a d e r indicates t h e source a n d d e s t i n a t i o n n e t w o r k a d d r e s s e s of t h e m e s s a g e . T h e IP h e a d e r also specifies t h e t i m e to live of t h e packet. W h e n a r o u t e r gets a n IP message, it d e c r e m e n t s its t i m e to live a n d it discards t h e m e s s a g e if t h e t i m e to live r e a c h e s 0. T h i s p r o c e d u r e p r e v e n t s p a c k e t s from floating a r o u n d for a l o n g t i m e in t h e n e t w o r k i n case of r o u t i n g errors. T h e h e a d e r i n d i c a t e s t h e t y p e of service or TOS r e q u e s t e d b y t h e m e s s a g e : e i t h e r l o w delay, h i g h t h r o u g h p u t , or v e r y reliable. TOS is i g n o r e d i n c u r r e n t i m p l e m e n t a t i o n s , a l t h o u g h it could b e u s e d for Differentiated Services (RFC 2474) a n d also for Explicit Congestion Notification (RFC 2481). T h e IP h e a d e r also specifies t h e upper-level protocol (typically U D P or TCP) t h a t r e q u e s t e d t h e t r a n s m i s s i o n of t h e m e s s a g e so t h a t t h e r e c e i v e r k n o w s h o w to h a n d l e t h e m e s s a g e . Finally, t h e h e a d e r m a y r e q u e s t s o m e optional services s u c h as r e c o r d i n g t h e r o u t e followed b y t h e p a c k e t ( t h e list of n o d e s is t h e n w r i t t e n i n t h e IP h e a d e r as t h e p a c k e t progresses), following a r o u t e specified b y t h e source, or t i m e - s t a m p i n g of t h e m e s s a g e b y e a c h n o d e t h a t it goes t h r o u g h . We explain t h r e e i m p o r t a n t a s p e c t s of IP: addressing, f r a g m e n t a t i o n / r e a s s e m b l y , a n d routing.

Addressing T h e r e a r e four i m p o r t a n t a d d r e s s e s : MAC (layer 2) or hard­ w a r e addresses, IP or n e t w o r k (layer 3) addresses, d o m a i n - b a s e d a d d r e s s e s used, for example, in e-mail, a n d Universal Reference Locator or URL.

The Internet and TCP/IP Networks

164

Layer 2 a d d r e s s e s vary from o n e n e t w o r k t y p e to a n o t h e r : E t h e r n e t a n d Tbken Ring u s e 48-bit addresses. F r a m e Relay u s e s a 12-bit DLCI c o n n e c t i o n identifier, a n d SMDS u s e s t e l e p h o n e n u m b e r s . In o r d e r for t h e millions of c o m p u t e r s o n different n e t w o r k s o n t h e I n t e r n e t to c o m m u n i c a t e , t h e y m u s t u s e t h e u n i q u e I n t e r n e t or IP address assigned to t h e m . Strictly speaking, t h e IP address is assigned to a host's n e t w o r k interface. A computer, s u c h as a router, with m o r e t h a n o n e n e t w o r k interface, h a s a different IP a d d r e s s for e a c h interface. (A c o m p u t e r ' s l a y e r 2 interface to a n E t h e r n e t , say, will also h a v e a MAC address.) Tb facilitate routing of packets, t h e 4-byte IP address ( w r i t t e n as a set of four decimal n u m b e r s s e p a r a t e d b y a dot [.]) is divided into a two-part h i e r a r c h y : t h e first part is t h e n e t w o r k n u m b e r , t h e s e c o n d p a r t is t h e h o s t n u m b e r . A p a c k e t is routed, h o p b y h o p , from o n e n e t w o r k r o u t e r to a n adjacent r o u t e r u n t i l it r e a c h e s t h e destination n e t w o r k w h o s e r o u t e r forwards it to t h e d e s t i n a t i o n host. T h u s a c o m p u t e r w i t h IP a d d r e s s 123.32.239.151 h a s t h e 16-bit n e t w o r k n u m b e r 123.32 a n d t h e 16-bit h o s t n u m b e r 239.151. T h e original organization of IP a d d r e s s e s w a s class-based. A l t h o u g h t h e c u r r e n t usage of a d d r e s s e s is n o w classless (based o n s u b n e t s ) , w e explain t h e class-based addressing to motivate t h e c u r r e n t s t r u c t u r e a n d also b e c a u s e it is still occasionally used. In class-based addressing, t h e r e are five classes of ad­ dresses. Class A a d d r e s s e s h a v e 8-bit n e t w o r k n u m b e r s b e g i n n i n g w i t h 0 a n d 24-bit h o s t n u m b e r s , so o n e class A n e t w o r k m a y h a v e u p to 2 hosts. Class Β addresses h a v e 16-bit n e t w o r k n u m b e r s b e g i n n i n g w i t h t h e bit p a t t e r n 10 a n d 16-bit h o s t n u m b e r s , so o n e class Β n e t w o r k m a y h a v e 2 hosts. Class C a d d r e s s e s h a v e 24-bit n e t w o r k n u m b e r s b e g i n n i n g w i t h t h e p a t t e r n 110 a n d 8-bit h o s t n u m b e r s , so o n e class C n e t w o r k c a n h a v e 256 hosts. Class D ad­ dresses, b e g i n n i n g w i t h 1110, for multicast, are discussed o n p . 173. Class Ε addresses, b e g i n n i n g w i t h 1111, are r e s e r v e d for future use. T h e class Β n e t w o r k 123.32 c a n h a v e 2 = 65, 536 hosts. A r o u t e r in this n e t w o r k s h o u l d b e able to forward a p a c k e t a d d r e s s e d to a n y of t h o s e hosts. T h i s c a n lead to very large routing tables. (Routing tables are d e s c r i b e d below.) Subnetting simplifies t h e m a n a g e m e n t of a d d r e s s e s b y partitioning t h e ad­ dress space so as to define a single n e t w o r k t h a t i n c l u d e s m a n y different p h y s ­ ical s e g m e n t s . Subnetting g r o u p s all t h e IP a d d r e s s e s w i t h t h e s a m e l e a d i n g η-bits into o n e n e t w o r k . Each IP address is a c c o m p a n i e d b y a 32-bit subnet mask t h a t consists of o n e s followed b y 32-η zeros. T h e s u b n e t m a s k is w r i t t e n in t h e s a m e notation as IP addresses. For instance, 255.255.255.0 c o r r e s p o n d s to η = 24. I n t h e e x a m p l e of t h e c o m p u t e r w i t h IP address 123.32.239.151, o n e c a n define a s u b n e t t h a t consists of all t h e IP a d d r e s s e s t h a t start w i t h 123.32.239. This s u b n e t is a subset of t h e class Β n e t w o r k 123.32. With this s u b n e t defini2 4

1 6

1 6

43

Internet Protocol

IBBWWMWWMMIW

tion, h o s t A b e l o n g s to t h e s u b n e t 132.32.139 a n d it h a s n u m b e r 151. A r o u t e r w i t h i n n e t w o r k 123.32, w h e n it receives t h e p a c k e t for h o s t A, w o u l d h a v e t h e s i m p l e r task of d e t e r m i n i n g w h e t h e r A b e l o n g s to t h e r o u t e r ' s o w n s u b n e t . T h e r o u t e r does this b y taking t h e logical AND of A's IP a d d r e s s a n d t h e r o u t e r ' s s u b n e t m a s k 255.255.255.0, w h i c h yields A's s u b n e t n u m b e r 123.32.239. If this m a t c h e s t h e router's s u b n e t n u m b e r , t h e n A is a local address; o t h e r w i s e t h e r o u t e r looks u p its r o u t i n g table to d e t e r m i n e t h e a d d r e s s of t h e r o u t e r for A's s u b n e t . Typically, s u b n e t n u m b e r s c o r r e s p o n d to a LAN, in w h i c h case "local address" m e a n s a n a d d r e s s o n t h e r o u t e r ' s LAN. (For example, 123.32.239 is t h e s u b n e t n u m b e r of a n E t h e r n e t LAN.) A c h a n g e of a d d r e s s e s inside t h e s u b n e t d o e s n o t affect t h e routing tables of t h e r o u t e r s . Classless interdomain routing or CIDR (RFC 1519) e x t e n d s t h e idea of sub­ n e t s to variable-length s u b n e t m a s k s . CIDR u s e s t h e a d d r e s s s p a c e efficiently b y r e s e r v i n g only t h e n u m b e r of a d d r e s s e s t h a t are n e e d e d for a s u b n e t (approx­ imately). In CIDR, a g r o u p of h o s t s w h o s e a d d r e s s e s h a v e a c o m m o n prefix are g r o u p e d as a s u b n e t . T h i s c o m m o n prefix serves as a n e t w o r k n u m b e r for t h e g r o u p as a whole. For example, t h e 1024 a d d r e s s e s in t h e four class C n e t w o r k s 192.0.8.*, 192.0.9.*, 192.0.10.* a n d 192.0.11.*, h a v e t h e c o m m o n prefix 192.0.8 w i t h a s u b n e t m a s k of 255.255.252.0. A n o t h e r s u b n e t m i g h t c o r r e s p o n d to t h e prefix 192.0.2 a n d y e t a n o t h e r to t h e prefix 192.0.5. T h e r o u t i n g decision, to identify t h e correct subnet, is a longest-matching-prefix search. For instance, if t h e d e s t i n a t i o n a d d r e s s is 192.0.9.123, t h e r o u t e r m u s t find o u t t h a t it b e l o n g s to t h e s u b n e t w i t h prefix 192.0.8. T h e pool of IP a d d r e s s e s allocated to a n organization m a y b e insufficient to assign a n individual IP a d d r e s s to e v e r y host. Besides, o n l y a few h o s t s m a y c o n n e c t to t h e I n t e r n e t at a n y time, so t h a t t h e pool c a n b e shared, w i t h u n u s e d IP a d d r e s s e s assigned o n a t e m p o r a r y basis. (An e x a m p l e is a n I n t e r n e t Service Provider [ISP], w h i c h h a s m a n y subscribers, o n l y a few of w h o m are logged o n at a n y time.) T h e Dynamic Host Configuration Protocol or D H C P is u s e d for this p u r p o s e (RFC 2131). A host, n e e d i n g a n IP address, u s e s its MAC a d d r e s s to b r o a d c a s t a DHCP discover packet. D H C P s e r v e r s r e p l y w i t h a DHCP offer t h a t i n c l u d e s a n u n u s e d IP address, a n d t h e h o s t a c c e p t s o n e of t h o s e offers a n d b r o a d c a s t s its selection w i t h a DHCP request. T h e d e s i g n a t e d server c o m m i t s its offer w i t h a DHCP ack. T h e a d d r e s s h a s a t i m e to live a n d m u s t b e refreshed to r e m a i n valid. W h e n t h e h o s t is d o n e , it c a n s e n d a DHCP release; o t h e r w i s e t h e a d d r e s s is a u t o m a t i c a l l y r e l e a s e d w h e n its lifetime expires. A distributed directory service called t h e Domain Name System or DNS is provided b y t h e I n t e r n e t to t r a n s l a t e b e t w e e n IP a d d r e s s e s a n d n a m e s , a n d to control I n t e r n e t e-mail delivery. T h e n a m e s are hierarchically organized,

„

_

165

The Internet and TCP/IP Networks

166

a d m i n i s t e r e d in a decentralized m a n n e r , a n d t h e directory service is p r o v i d e d b y distributed DNS servers. D o m a i n n a m e s are organized in a tree. T h e top-level d o m a i n s are edu (ed­ ucational institutions), com ( c o m m e r c i a l ) , gov (U.S. federal g o v e r n m e n t ) , org (nonprofit organizations), net (mostly n e t w o r k providers), int ( s o m e i n t e r n a ­ tional organizations), mil (U.S. a r m e d forces), a n d t h e country-specific d o m a i n (e.g., fr, gr, jp etc.). T h e edu d o m a i n s e r v e r h a s p o i n t e r s to berkeley.edu n a m e servers, a n d o t h e r s like it. T h u s a d o m a i n is a s u b t r e e of this d o m a i n n a m e s y s t e m . T h e berkeley.edu d o m a i n i n c l u d e s s u b d o m a i n s like eecs.berkeley.edu, w h i c h includes h o s t n a m e s like diva.eecs.berkeley.edu. T h e registration a u t h o r ­ ity for berkeley.edu is allowed to assign s u b d o m a i n n a m e s like eecs.berkeley.edu a n d m a y delegate a u t h o r i t y to assign n a m e s u n d e r it, like diva.berkeley.eecs.edu. f

T h e a u t h o r s u s e a c o m p u t e r n a m e d diva.eecs.berkeley.edu. This n a m e consists of t h e institution n a m e (berkeley.edu) a n d t h e n a m e of t h e com­ p u t e r (diva.eecs). I b s e n d a m e s s a g e to o n e of t h e a u t h o r s , y o u s e n d it to [email protected] or to [email protected], respectively. You do n o t n e e d to k n o w w h e r e diva.eecs.berkeley.edu is located. T h e directory service t r a n s l a t e s t h e n a m e into t h e a d d r e s s 128.32.110.56, a n d t h e n e t w o r k l a y e r finds a p a t h t h a t leads to t h a t c o m p u t e r . T h e directory servers s u p e r v i s e disjoint zones of t h e n a m e - s t r u c t u r e tree. For reliability, m u l t i p l e servers are u s e d for e a c h zone. For t h e p u r p o s e of illus­ trating h o w DNS works, w e a s s u m e t h a t o n e zone r e p r e s e n t s t h e c h i l d r e n of t h e edu node, a n o t h e r r e p r e s e n t s t h e c h i l d r e n of t h e berkeley.edu node, a n d a n o t h e r r e p r e s e n t s t h e children of t h e eecs.berkeley.edu n o d e . Note t h a t t h e z o n e s do n o t h a v e to b e all t h e children of o n e n o d e b u t c a n b e defined m o r e generally. A s s u m e t h a t y o u w a n t to s e n d a p a c k e t to diva.eecs.berkeley.edu. Your local directory service s e n d s t h e r e q u e s t "find t h e IP a d d r e s s of diva.eecs.berkeley.edu" to t h e edu z o n e n a m e server. T h i s server k n o w s t h e a d d r e s s of t h e berkeley.edu zone server a n d forwards t h e r e q u e s t to t h a t server. T h i s server k n o w s t h a t eecs h a s its o w n server, a n d it forwards t h e r e q u e s t to t h a t server. Finally, t h e eecs server r e t u r n s t h e a d d r e s s of diva to t h e berkeley.edu server, w h i c h r e t u r n s it to t h e edu server, w h i c h finally s e n d s it to y o u r c o m p u t e r . T h e servers cache t h e i n t e r m e d i a t e a n s w e r s . For instance, t h e edu s e r v e r c a c h e s t h e address of t h e berkeley.edu server. If a r e q u e s t e d a d d r e s s is already c a c h e d in a server, t h e n it replies to t h e q u e r y w i t h o u t h a v i n g to forward t h e r e q u e s t . Cache e n t r i e s h a v e a t i m e to live to p r e v e n t obsolescence. T h e I n t e r n e t Corporation for Assigned N a m e s a n d N u m b e r s (ICANN, www.icann.org) w a s formed in 1998 to take over responsibility for IP a d d r e s s space allocation, protocol p a r a m e t e r a s s i g n m e n t , d o m a i n n a m e s y s t e m m a n ­ a g e m e n t , a n d root server s y s t e m m a n a g e m e n t .

43

Internet Protocol

W h e r e a s MAC a n d IP a d d r e s s e s refer to a h o s t interface, t h e Universal Reference Locator or URL specifies t h e location of r e s o u r c e s ( s u c h as files, directories, h t m l d o c u m e n t s , or database services) a n d a s c h e m e for retrieving it via t h e I n t e r n e t . A URL is w r i t t e n as

:

T h e c o m m o n s c h e m e s are ftp, h t t p , gopher, n e w s , mailto. T h e s c h e m e specific-part is of t h e form //:©:/

User, password, a n d url-path m a y b e omitted, a n d t h e p o r t n u m b e r m a y b e assigned b y default. For e x a m p l e ftp://ftp.cis.ohio-state.edu/pub/ is a URL of t h e a n o n y m o u s FTP archive of t h e C o m p u t e r a n d I n f o r m a t i o n Science D e p a r t m e n t of Ohio State University. Files l o c a t e d in t h e archive are a c c e s s e d b y t h e File Transfer Protocol (FTP). T h e default p o r t for FTP is 21. A n H T T P URL takes t h e form http://:/

w h e r e h o s t a n d p o r t are as above, p a t h is a n H T T P selector, a n d s e a r c h p a r t is a q u e r y . T h e default port for H T T P is 80.

Fragmentation/Reassembly T h e OSI m o d e l specifies t h a t l a y e r 4 (the t r a n s p o r t l a y e r ) p e r f o r m t h e f r a g m e n t a t i o n a n d r e a s s e m b l y of m e s s a g e s . How­ ever, in a T C P / I P n e t w o r k , IP f r a g m e n t s m e s s a g e s into IP p a c k e t s in t h e s o u r c e for t h e data link l a y e r a n d r e a s s e m b l e s IP p a c k e t s into m e s s a g e s at t h e desti­ n a t i o n . For instance, if t h e m e s s a g e s go t h r o u g h a n E t h e r n e t , t h e n t h e y m u s t b e divided into p a c k e t s of at m o s t 1.5 KB. A r o u t e r m a y h a v e to further frag­ m e n t IP p a c k e t s t h a t are larger t h a n t h e m a x i m u m transfer u n i t of t h e next link. I n a n y case, o n l y t h e d e s t i n a t i o n r e a s s e m b l e s t h e m e s s a g e s . T h e h e a d e r c o n t a i n s a n identification n u m b e r of t h e m e s s a g e to w h i c h t h e p a c k e t b e l o n g s a n d a n offset indication t h a t specifies t h e position of t h e data p o r t i o n of t h e IP p a c k e t inside t h e original m e s s a g e . T h e 3-bit flag indicates w h e t h e r t h e r o u t e r m a y fragment t h e p a c k e t or n o t a n d w h e t h e r t h e r e are additional fragments or t h e p a c k e t is t h e last fragment of t h e m e s s a g e . T h e f r a g m e n t a t i o n / r e a s s e m b l y function is s h o w n in Figure 4.6. Routing Tb r o u t e d a t a g r a m s , I n t e r n e t n o d e s called routers m a i n t a i n rout­ ing tables. T h e table specifies w h e r e t h e n o d e s h o u l d s e n d a d a t a g r a m next. If t h e d a t a g r a m is d e s t i n e d to a n e t w o r k to w h i c h t h e r o u t e r is n o t directly at­ tached, t h e n t h e r o u t i n g table specifies t h e next r o u t e r for t h a t n e t w o r k . If t h e

The Internet and TCP/IP Networks

168

Byte stream from T S A P m @ A to T S A P η @ Β

Fragment 1

Fragment 2

Sequence number: initial number = M,

M + N = M 1

{

Fragment 3

τ M + N = M

2

2

2

3

TCP m n Μ,

Fragment 1

m η M

2

Fragment 2

m η M

TCP header

Fragment 3 IP

ABO IP header

4.6 FIGURE

3

m η M

2

A Β Pj m η M

2

TCP header

TCP delivers a byte stream by first decomposing it into fragments (called segments) that are transported b y IP. TCP n u m b e r s the fragments, starting with some initial sequence n u m b e r and t h e n counting the bytes. IP divides the TCP fragments into packets. The IP header indicates the offset of the packet in the fragment.

d a t a g r a m is for a c o m p u t e r o n t h e s a m e n e t w o r k as t h e router, t h e n it consults a table to find t h e physical (MAC) address of t h e computer. To m a i n t a i n t h e table w i t h t h e a d d r e s s e s of c o m p u t e r s a t t a c h e d o n t h e s a m e n e t w o r k , t h e r o u t e r u s e s t h e Address Resolution Protocol (ARP). Using this protocol, a r o u t e r s e a r c h e s its table for a given c o m p u t e r n a m e w h o s e MAC address it is trying to locate. If t h e c o m p u t e r n a m e is n o t in t h e table, t h e r o u t e r broadcasts a m e s s a g e o n t h e n e t w o r k asking for t h e c o m p u t e r w i t h t h e given address to reply. T h e c o m p u t e r with t h a t a d d r e s s replies directly to t h e router, indicating its MAC address. T h e table e n t r i e s h a v e a t i m e to live, a n d t h e y are p u r g e d w h e n t h a t lifetime is exceeded. T h a t is h o w t h e r o u t e r u p d a t e s t h e table e n t r i e s to reflect c h a n g e s in t h e n e t w o r k . l b m a i n t a i n t h e routing table to t h e o t h e r n e t w o r k s , t h e r o u t e r s typically u s e o n e of two versions of t h e shortest-path algorithm: t h e Bellman-Ford al­ g o r i t h m a n d Dijkstra's algorithm. We explain t h e s e two versions below. TJo i m p l e m e n t that algorithm, t h e r o u t e r s exchange control i n f o r m a t i o n s u c h as t h e i r e s t i m a t e s of t h e l e n g t h of a shortest path. In addition, t h e r o u t e r s u s e a protocol called t h e Internet Control Message Protocol (ICMP) for exchanging

4.3 M i m

Internet Protocol

»^^

control information. ( T h e t e r m gateway is n o w r e p l a c e d b y router. A g a t e w a y is a n o d e t h a t passes data b e t w e e n devices w i t h similar functions b u t dissimilar i m p l e m e n t a t i o n s . In this sense, a r o u t e r is a l a y e r 3 gateway.) ICMP provides a n u m b e r of services t h a t c o m p u t e r s u s e for m o n i t o r i n g t h e n e t w o r k . O n e s u c h service is ping, w h i c h asks a n o d e to e c h o a packet, a n d traceroute, w h i c h ob­ tains t h e r o u t e to t h e n o d e a n d t h e d e l a y o n e a c h h o p over t h e route. A n o t h e r service entails a r o u t e r indicating to t h e source of a n IP p a c k e t t h a t t h e desti­ n a t i o n is u n r e a c h a b l e . Similarly, a r o u t e r indicates to t h e source w h e n o n e of its p a c k e t s exceeds its t i m e to live. A r o u t e r c a n also tell a source to redirect its p a c k e t s in case of difficulties, or to slow d o w n t r a n s m i s s i o n s .

Bellman-Ford Algorithm Figure 4.7 illustrates a distributed algorithm t h a t finds t h e shortest p a t h s from all t h e n o d e s in a n e t w o r k to a n y p a r t i c u l a r n o d e t h a t w e call t h e destination (RFC 1058). T h i s is t h e Bellman-Ford algorithm.

4.7 eiiiiiwiiii.

Using the Bellman-Ford algorithm, the nodes find the shortest path to destination. The nodes estimate the length of the shortest path to the destination. These estimates are shown inside the circles representing the nodes. When its estimate decreases, a node sends the n e w value to its neighbors. The figure shows, from left to right and top to bottom, the successive steps of the algorithm for the graph shown in the top-left diagram. Some intermediate steps are not shown.

a

169

The Internet and TCP/IP Networks

170

T h e algorithm a s s u m e s t h a t e a c h n o d e k n o w s t h e length of t h e l i n k s a t t a c h e d to itself. T h e s e l e n g t h s c a n b e a m e a s u r e of t h e t i m e t a k e n for a p a c k e t to b e t r a n s m i t t e d along t h e links, or t h e y c a n b e arbitrarily selected positive n u m ­ b e r s (e.g., 1 for e v e r y link). At e a c h step of t h e algorithm, e v e r y n o d e k e e p s track of its c u r r e n t e s t i m a t e of t h e l e n g t h of its s h o r t e s t p a t h to t h e d e s t i n a t i o n . W h e n e v e r a n o d e receives a m e s s a g e from a n o t h e r node, it u p d a t e s its esti­ m a t e . If t h e u p d a t e d e s t i m a t e is strictly s m a l l e r t h a n t h e p r e v i o u s estimate, t h e n t h e n o d e s e n d s a m e s s a g e c o n t a i n i n g t h e n e w e s t i m a t e to its neigh­ bors. A n o d e i e s t i m a t e s t h e l e n g t h of t h e shortest p a t h to t h e d e s t i n a t i o n b y computing L(i) = m i n { d ( i ; ) + L 0 ' ) }

j

l

l

w h e r e t h e m i n i m u m is over all t h e n e i g h b o r s ;' of n o d e i. In this expression, d(i,j) is t h e l e n g t h of t h e link from i t o ; a n d L(j) is t h e latest e s t i m a t e received f r o m ; of t h e l e n g t h of its shortest p a t h to t h e destination. Initially, all t h e e s t i m a t e s L(i) are set to infinity. At t h e first step, t h e destination r e d u c e s its e s t i m a t e to 0. T h e d e s t i n a t i o n t h e n s e n d s m e s s a g e s to its neighbors, w h o t h e n u p d a t e t h e i r estimates, a n d so on. Eventually, e v e r y n o d e i finds out t h e shortest distance L*(i) b e t w e e n itself a n d t h e destination. Moreover, t h e r e is a shortest p a t h from i to t h e d e s t i n a t i o n t h a t goes first to ; if a n d only if L*(i) = d(i,j) + L*(j). Shortest p a t h s n e e d n o t b e u n i q u e , a n d t h e n o d e s b r e a k ties arbitrarily. Figure 4.7 shows t h e c o n v e r g e n c e s t e p s of t h e algorithm. T h e shortest p a t h s are s h o w n b y t h i c k e r l i n e s in t h e last step, at t h e b o t t o m right of t h e figure. T h e shortest-path algorithm t h a t w e j u s t explained is u s e d b y b r i d g e s for s p a n n i n g t r e e routing. It is also u s e d b y t h e n e t w o r k l a y e r of s o m e n e t w o r k s to select good p a t h s along w h i c h to r o u t e packets. Typically, t h e s e n e t w o r k s define t h e l e n g t h of a link as a l i n e a r c o m b i n a t i o n of t h e average t r a n s m i s s i o n t i m e a n d of t h e r e c e n t backlog in t h e q u e u e of t h e t r a n s m i t t e r of t h e link. By r u n n i n g t h e shortest-path algorithm periodically, t h e n o d e s c a n a d a p t t h e i r r o u t i n g decisions to c h a n g i n g conditions. In this m a n n e r , t h e n e t w o r k r e a c t s to link a n d n o d e failures, a n d it also controls congestion. T h e a l g o r i t h m is called a distance vector algorithm b e c a u s e e a c h n o d e s e n d s to its n e i g h b o r s t h e e s t i m a t e of its distance to all n o d e s . T h e Bellman-Ford algorithm is distributed: e a c h n o d e h a s o n l y a partial k n o w l e d g e of t h e topology of t h e n e t w o r k . It c a n b e s h o w n t h a t t h e algo­ r i t h m m a y fail to converge if t h e delay b e t w e e n u p d a t e s is large a n d if d u r i n g t h a t t i m e a significant a m o u n t of traffic is r e r o u t e d . Tb avoid t h e s e difficul­ ties, t h e I n t e r n e t started u s i n g a centralized shortest-path a l g o r i t h m d u e to Dijkstra.

43

Internet Protocol

171

Dijkstra's Algorithm C o n s i d e r o n c e again t h e n e t w o r k s h o w n in Fig­ u r e 4.7. A s s u m e t h a t e a c h n o d e h a s a c o m p l e t e m a p of t h e n e t w o r k . Tb u p d a t e t h e m a p s , e a c h n o d e s e n d s to all t h e o t h e r n o d e s a m e s s a g e t h a t indicates t h e l e n g t h s of t h e links a t t a c h e d to it. A l g o r i t h m s t h a t rely o n global u p d a t e s of local link information are called link state algorithms. Each n o d e t h e n c a n i m p l e m e n t a shortest-path a l g o r i t h m to find t h e short­ est p a t h to e v e r y possible destination. T h e n o d e could i m p l e m e n t t h e BellmanFord a l g o r i t h m t h a t w e discussed earlier. It c a n also i m p l e m e n t t h e following algorithm d u e to Dijkstra. Conceptually, t h e algorithm, called open shortest path first, or OSPF, b y t h e I n t e r n e t c o m m u n i t y , c o n s t r u c t s a s p a n n i n g t r e e of s h o r t e s t p a t h s from a n y given root n o d e to t h e o t h e r n o d e s . Actual details of t h e a l g o r i t h m d e p e n d o n t h e i m p l e m e n t a t i o n . T h e steps of t h e a l g o r i t h m are s h o w n from left to right a n d from top to b o t t o m in Figure 4.8. Tb c o n s t r u c t a s p a n n i n g t r e e of shortest

No longer

4.8 mmmamx

A node uses Dijkstra's algorithm to construct a spanning tree of shortest paths f d e called the root. The network is shown in the top-left graph. We assume that each node has a complete m a p of the network. At each step, shown left to right and top to bottom, the algorithm picks the u n m a r k e d node with the smallest current label. The algorithm explores the neighbors of that node, after which it marks the node. The algorithm compares the label of each neighbor to the sum of the node's label plus the link's length. If that s u m is smaller, it b e c o m e s the new label and the link is added to the list of preferred links while the previously preferred link leading to that node is removed from the list of preferred links. T h e steps leading to the last graph are not shown. rom

a

n o

172

The Internet and TCP/IP Networks

p a t h s rooted at t h e "root" node, t h e algorithm first assigns a label 0 to t h e root n o d e a n d oo to e v e r y o t h e r n o d e . T h e label of a n o d e r e p r e s e n t s t h e l e n g t h of t h e shortest p a t h t h a t t h e algorithm h a s discovered so far from t h e root to t h a t node. At e a c h step, t h e algorithm e x a m i n e s t h e n o d e w i t h t h e smallest label a n d explores its n e i g h b o r s to see if it n e e d s to r e d u c e t h e i r label; in addition, t h e algorithm m a k e s a n o d e of t h e links t h a t are along s h o r t e s t p a t h s . T h e algorithm starts w i t h t h e root a n d e x a m i n e s its neighbors, t h a t is, t h o s e n o d e s t h a t c a n b e r e a c h e d in o n e s t e p from t h e root. T h e algorithm c o m p a r e s t h e label of e a c h of t h e s e neighbors w i t h t h e s u m of t h e root's label (0) a n d t h e l e n g t h of t h e link from t h e root to t h e n o d e . If t h e s u m is smaller, t h e a l g o r i t h m r e p l a c e s t h e label b y t h a t s u m a n d n o t e s t h a t t h e link from t h e root to t h e n o d e is o n t h e shortest p a t h discovered so far. T h e a l g o r i t h m m a r k s t h e root to r e m e m b e r t h a t it h a s explored its neighbors. T h e algorithm t h e n r e p e a t s t h e p r o c e d u r e w i t h t h e u n m a r k e d n o d e t h a t h a s t h e smallest label. If t h e label of a n o d e is r e d u c e d , t h e n t h e n e w link t h a t led to this r e d u c t i o n is n o t e d as b e i n g a p r e f e r r e d link while t h e link t h a t was previously o n a shortest p a t h is r e m o v e d from t h e list of "preferred" links.

Border Gateway Protocol In IP, t h e r o u t i n g u s e s a h i e r a r c h y . T h e net­ w o r k is d e c o m p o s e d into autonomous systems (ASs). A n AS is a g r o u p of net­ w o r k s or s u b n e t w o r k s u n d e r a single a d m i n i s t r a t i o n . Routers u s e d for informa­ tion exchange w i t h i n a u t o n o m o u s s y s t e m s are called interior routers, a n d t h e y u s e a variety of interior g a t e w a y protocols to a c c o m p l i s h this p u r p o s e . Routers t h a t m o v e information b e t w e e n a u t o n o m o u s s y s t e m s are called exterior routers, a n d t h e y u s e a n exterior g a t e w a y protocol for this p u r p o s e . Within e a c h AS, t h e r o u t i n g algorithm in e a c h area r o u t e r calculates t h e shortest p a t h s to all t h e o t h e r r o u t e r s in t h e areas to w h i c h it b e l o n g s u s i n g RIP, OSPF, or s o m e o t h e r shortest p a t h algorithm. T h e r o u t i n g b e t w e e n ASs u s e s t h e Border G a t e w a y Protocol (BGP) (RFC 1771,1772). ASs are a t t a c h e d t o g e t h e r b y o n e or m o r e routers, called border gate­ ways. Two ASs are c o n n e c t e d if t h e y s h a r e a link l a y e r n e t w o r k t h a t i n c l u d e s a b o r d e r g a t e w a y from e a c h AS. Each AS c o n t a i n s a router, called border gateway speaker, t h a t i m p l e m e n t s BGP. BGP s p e a k e r s u s e TCP (see section 4.4) to ex­ c h a n g e routing tables a n d t h e i r u p d a t e s . T h e s e r o u t i n g tables specify t h e p a t h s t h a t e a c h BGP s p e a k e r c u r r e n t l y u s e s to r e a c h o t h e r BGP s p e a k e r s . BGP speak­ ers exchange keep-alive m e s s a g e s to m a k e s u r e t h a t t h e p a t h s are still valid. Each BGP s p e a k e r c o m p a r e s t h e s e p a t h s to c o n s t r u c t preferable p a t h s b e t w e e n itself a n d t h e o t h e r BGP s p e a k e r s a n d s e n d s t h e s e p a t h s as u p d a t e s to t h e o t h e r BGP speakers. T h i s selection is left to e a c h a u t o n o m o u s s y s t e m . T h e selection m u s t p r e v e n t t h e creation of loops a n d take into a c c o u n t restrictions

4.3

Internet Protocol

173

t h a t ASs m a y h a v e a b o u t c a r r y i n g traffic from o t h e r ASs. T h u s BGP o p e r a t e s o n c o m p l e t e p a t h s i n s t e a d of s u m m a r i z i n g t h e m b y t h e i r distance to t h e d e s t i n a t i o n as t h e a l g o r i t h m s b y Dijkstra or Bellman-Ford do. BGP e n a b l e s a u t o n o m o u s s y s t e m s to refuse to c a r r y traffic originating from o t h e r given a u t o n o m o u s s y s t e m s a n d also to favor specific p a t h s for t h e i r o w n traffic. Each BGP s p e a k e r s h o u l d r e m e m b e r t h e set of feasible paths, a l t h o u g h it advertises o n l y t h e preferable o n e .

432

Multicast IP Multicast IP is a facility t h a t s e n d s a n IP p a c k e t to a g r o u p of h o s t s identified b y a g r o u p a d d r e s s (RFC 1112, 1584). Multicast r o u t i n g is i m p l e m e n t e d b y a subset of IP r o u t e r s called multicast routers. G r o u p a d d r e s s e s m a y b e e i t h e r p e r m a n e n t or t r a n s i e n t a n d are of class D (i.e., h a v e 1110 as t h e i r h i g h e r o r d e r 4 bits). E a c h h o s t c a n b e c o m e or stop b e i n g a m e m b e r of a g r o u p b y s e n d i n g j o i n a n d leave r e q u e s t m e s s a g e s specifying t h e g r o u p address of t h e g r o u p it w a n t s to j o i n or leave. A m u l t i c a s t r o u t e r l e a r n s w h i c h groups are active b y periodically ( e v e r y m i n u t e ) polling t h e m e m b e r s of its d o m a i n . T h i s polling procfess p r o c e e d s hierarchically u s i n g a protocol called t h e I n t e r n e t G r o u p M a n a g e m e n t Protocol (IGMP; RFC 1112). Specifically, t h e multicast r o u t e r s e n d s a q u e r y , "which g r o u p s do y o u b e l o n g to?" to all t h e h o s t s of its local d o m a i n (using a g r o u p a d d r e s s to w h i c h all t h e multicast-capable h o s t s b e l o n g ) . O n receiving t h e q u e r y , a h o s t starts a c o u n t - d o w n t i m e r w i t h a r a n d o m value b e t w e e n 0 a n d s o m e Τ ( T = 10 s is r e c o m m e n d e d ) . W h e n t h e t i m e r expires, t h e h o s t s e n d s a r e p o r t w i t h t h e g r o u p a d d r e s s as a d e s t i n a t i o n a d d r e s s for e a c h g r o u p it b e l o n g s to. T h e m u l t i c a s t r o u t e r listens to all t h o s e r e p o r t s d e s t i n e d to g r o u p addresses. If a h o s t h e a r s a r e p o r t from a n o t h e r host—signaling t h a t host's m e m b e r s h i p to a c o m m o n g r o u p — t h e n t h e h o s t does n o t s e n d its o w n r e p l y c o r r e s p o n d i n g to t h a t g r o u p . T h i s p r o c e d u r e limits t h e n u m b e r of r e p o r t s p e r q u e r y . W h e n it w a n t s to j o i n a n e w group, a h o s t s e n d s a r e p o r t for t h a t group, w i t h o u t waiting for a q u e r y . T h a t r e p o r t s h o u l d b e s e n t a few t i m e s at r a n d o m intervals to m a k e sure it gets to t h e multicast router. Note t h a t t h e m u l t i c a s t r o u t e r d o e s n o t m a i n t a i n a list of g r o u p m e m b e r s . It is r e q u i r e d t h a t h o s t s u s e t h e i r individual IP addresses, a n d n o t a g r o u p address, as t h e source a d d r e s s of t h e IP p a c k e t s t h e y send. For instance, t h e IP d a t a g r a m r e a s s e m b l y a l g o r i t h m a s s u m e s t h a t e a c h h o s t u s e s a different source address. T h e multicast r o u t e r s t h e n c o n s t r u c t a t r e e of m u l t i c a s t r o u t e r s from a h o s t to t h e d e s t i n a t i o n s t h a t b e l o n g to t h e m u l t i c a s t g r o u p . IP t u n n e l s (see next page) c o n n e c t t h e m u l t i c a s t r o u t e r s . T h i s t r e e is t h e t r e e of s h o r t e s t p a t h s

The Internet and TCP/IP Networks

174

from t h e source to all t h e possible d e s t i n a t i o n s p r u n e d b a c k to cover o n l y t h e d e s t i n a t i o n s of t h e m u l t i c a s t g r o u p . Tb calculate t h a t tree, t h e r o u t e r s r u n t h e Bellman-Ford algorithm, explained earlier, w h e r e t h e d e s t i n a t i o n is in fact t h e source of t h e multicast. In t h e p r u n e d tree, t h e p a c k e t s are d u p l i c a t e d at t h e last fork of t h e tree. Note t h a t t h e p r u n e d t r e e does n o t m i n i m i z e t h e s u m of t h e distances covered b y t h e replicated packets. T h e results of t h e t r e e calculations are c a c h e d in t h e r o u t e r s for s u b s e q u e n t packets. As hosts j o i n or leave a group, t h e r o u t e r s automatically u p d a t e t h e tree. If g r o u p m e m b e r s b e l o n g to t h e s a m e LAN, t h e m u l t i c a s t r o u t e r of t h e LAN c o n v e r t s t h e IP g r o u p address into a multicast LAN address. If g r o u p m e m b e r s b e l o n g to a c o m m o n n o n b r o a d c a s t m u l t i a c c e s s n e t w o r k ( s u c h as F r a m e Relay), t h e n t h a t n e t w o r k ' s m u l t i c a s t r o u t e r m u s t t r a n s m i t copies to t h e different g r o u p members. T h e IP Multicast Backbone, or MBone, is a n e t w o r k overlaid o n t h e I n t e r n e t , w h i c h is u s e d for interactive video a n d audio multicasts. MBone c o m p r i s e s multicast routers, c o n n e c t e d b y virtual links formed b y t u n n e l s . Suppose a multicast p a c k e t arrives at MBone r o u t e r A. A k n o w s t h a t it s h o u l d b e forwarded to MBone r o u t e r B, w h i c h is n o t directly c o n n e c t e d to A. T h e p a c k e t m u s t b e r o u t e d t h r o u g h a n e i g h b o r i n g n o d e C, w h i c h is n o t a m u l t i c a s t router. If A s e n d s t h e multicast packet directly to C, C w o u l d n o t k n o w h o w to forward it b e c a u s e t h e packet's d e s t i n a t i o n a d d r e s s is n o t B's IP a d d r e s s b u t a m u l t i c a s t g r o u p address t h a t C c a n n o t u n d e r s t a n d . Therefore, A e n c a p s u l a t e s t h e m u l t i c a s t p a c k e t in a n o t h e r IP d a t a g r a m w i t h d e s t i n a t i o n a d d r e s s Β a n d t h e n forwards it to C. Eventually t h e p a c k e t r e a c h e s Β w h e r e it is d e c a p s u l a t e d a n d t h e original multicast packet retrieved. T h i s e n c a p s u l a t i o n / d e c a p s u l a t i o n p r o c e s s c r e a t e s a virtual link b e t w e e n A a n d B, called a tunnel, m o r e precisely, a l a y e r 3 t u n n e l since e n c a p s u l a t i o n is into a l a y e r 3 or IP packet.

433

Reliable Multicast S o m e multicast applications r e q u i r e a reliable delivery. E x a m p l e s of s u c h ap­ plications i n c l u d e t h e distribution of software, of n e w s p a p e r s , a n d of o t h e r d o c u m e n t s . A n u m b e r of m e c h a n i s m s h a v e b e e n p r o p o s e d for reliable multi­ cast. We explain t h e m a i n ideas b e h i n d t h e s e m e c h a n i s m s . Consider o n e source t h a t multicasts a d o c u m e n t to a large n u m b e r of u s e r s . T h e first observation is t h a t a n a c k n o w l e d g m e n t - b a s e d s c h e m e is n o t practical: If all t h e d e s t i n a t i o n s w e r e to a c k n o w l e d g e t h e p a c k e t s t h e y received, t h e s e a c k n o w l e d g m e n t s w o u l d o v e r w h e l m t h e source. Also, m e r g i n g a c k n o w l e d g ­ m e n t s at t h e b r a n c h i n g n o d e s of t h e t r e e is n o t practical e i t h e r b e c a u s e t h o s e n o d e s w o u l d n e e d to c o u n t to m a k e s u r e t h e y got all t h e a c k n o w l e d g m e n t s

43

Internet Protocol

175

before a c k n o w l e d g i n g to t h e u p s t r e a m m u l t i c a s t n o d e . Also, in s u c h a m e r g ­ ing s c h e m e , if a u s e r s t o p p e d listening, t h e source w o u l d k e e p o n t r a n s m i t t i n g copies u n t i l t h e m e r g i n g n o d e l e a r n e d t h a t t h e u s e r h a d left t h e m u l t i c a s t g r o u p . If o n e p a c k e t gets d r o p p e d or c o r r u p t e d o n a link of t h e m u l t i c a s t tree, t h e n a large n u m b e r of u s e r s will n o t receive it. As a result, a n e g a t i v e acknowledg­ m e n t s c h e m e also risks o v e r w h e l m i n g t h e source. However, it is possible to m e r g e negative a c k n o w l e d g m e n t s . H e n c e , all p r o p o s e d s c h e m e s u s e negative a c k n o w l e d g m e n t s w i t h m e r g i n g . S o m e s c h e m e s u s e "designated receivers" t h a t act as caches. In s u c h a s c h e m e , e v e r y m u l t i c a s t n o d e is assigned a receiver. W h e n a negative a c k n o w l e d g m e n t r e a c h e s t h e node, it asks its d e s i g n a t e d re­ ceiver for a copy of t h e m i s s i n g p a c k e t i n s t e a d of p r o p a g a t i n g t h e negative acknowledgment upstream.

43.4

Mobile IP T h e objective of Mobile IP is to deliver p a c k e t s to mobile n o d e s automatically. A mobile n o d e is a h o s t or a r o u t e r t h a t c h a n g e s its p o i n t of a t t a c h m e n t from o n e n e t w o r k or s u b n e t w o r k to another. Using Mobile IP, a m o b i l e n o d e m a y c h a n g e location w i t h o u t c h a n g i n g its IP address. (See [P96].) A mobile n o d e is associated w i t h a fixed IP address, called its home IP address. A r o u t e r o n t h e m o b i l e n o d e ' s h o m e n e t w o r k delivers IP d a t a g r a m s to t h e mobile n o d e w h e n it is o n a foreign n e t w o r k , t h a t is, a w a y from its h o m e n e t w o r k . T h a t r o u t e r is called t h e home agent. Two p r o c e d u r e s c a n b e u s e d to deliver t h e d a t a g r a m s to t h e m o b i l e n o d e o n a foreign n e t w o r k . In t h e first p r o c e d u r e , t h e m o b i l e n o d e gets a t e m p o r a r y care-of address o n t h e foreign n e t w o r k from s o m e a s s i g n m e n t m e c h a n i s m a n d registers t h a t a d d r e s s w i t h its h o m e agent. W h e n t h e h o m e a g e n t gets a d a t a g r a m for t h e mobile node, it e n c a p s u l a t e s t h e d a t a g r a m in a n IP p a c k e t w i t h t h e care-of a d d r e s s as d e s t i n a t i o n address. In t h e s e c o n d p r o c e d u r e , t h e m o b i l e n o d e u s e s a foreign agent. T h e foreign a g e n t is a r o u t e r o n t h e n e t w o r k visited b y t h e mobile n o d e . W h e n it m o v e s to a n o t h e r n e t w o r k , t h e mobile n o d e registers w i t h a foreign a g e n t a n d obtains a care-of address from t h a t agent. T h e foreign a g e n t s e n d s its a d d r e s s to t h e m o ­ bile n o d e ' s h o m e agent. T h e h o m e a g e n t e n c a p s u l a t e s t h e d a t a g r a m s d e s t i n e d to t h e mobile n o d e in IP p a c k e t s t h a t it s e n d s to t h e foreign agent. T h e foreign a g e n t d e c a p s u l a t e s t h e p a c k e t s a n d delivers t h e m to t h e m o b i l e n o d e w i t h t h e care-of address. Note t h a t t h e first s c h e m e d o e s n o t r e q u i r e a g e n t s b u t is n o t a u t o m a t i c . T h e e n c a p s u l a t i o n / d e c a p s u l a t i o n p r o c e s s is a n o t h e r e x a m p l e of t u n n e l i n g . In e i t h e r p r o c e d u r e , w h e n r e t u r n i n g h o m e , t h e m o b i l e n o d e deregisters w i t h

The Internet and TCP/IP Networks

176

its h o m e agent. Also, t h e mobile n o d e u s e s its h o m e IP a d d r e s s as t h e source address of its IP packets. Agents ( h o m e or foreign) advertise t h e i r availability b y periodically send­ ing agent advertisement messages. Moreover, mobile n o d e s m a y solicit s u c h a m e s s a g e b y s e n d i n g a n agent solicitation message. As it receives t h e s e m e s s a g e s , t h e mobile n o d e c a n d e t e r m i n e w h e t h e r it is o n its h o m e n e t w o r k or o n a for­ eign n e t w o r k . O n its h o m e n e t w o r k , t h e mobile n o d e o p e r a t e s w i t h o u t t h e mobility services of Mobile IP, t h a t is, b y u s i n g t h e s t a n d a r d IP. T h e advantage of t h e first p r o c e d u r e is t h a t it does n o t r e q u i r e foreign agents. Its disadvantage is t h a t n e t w o r k s m u s t m a i n t a i n a pool of a d d r e s s e s available for visiting mobile n o d e s . T h e mobility m a n a g e m e n t m e s s a g e s are a u t h e n t i c a t e d to p r e v e n t redirec­ tion attacks, s u c h as a third p a r t y stealing m e s s a g e s b y giving its a d d r e s s as t h e n e w mobile n o d e address.

4.3.5

IPv6 T h e c u r r e n t version of IP is version 4. Its m a i n limitations are a l i m i t e d n u m b e r of addresses, a c o m p l e x header, a difficult m e c h a n i s m for i n t r o d u c i n g exten­ sions a n d options, a l i m i t e d n u m b e r of different services, a n d p o o r security a n d privacy. IPv6 is b e i n g d e s i g n e d to o v e r c o m e t h e s e limitations. T h e t r a n s i t i o n from IPv4 to IPv6 is e x p e c t e d to take m a n y years. In t h e m e a n t i m e , t h e two versions c a n coexist b y t u n n e l i n g : IPv4 forwards IPv6 p a c k e t s w i t h o u t m o d i ­ fication. (For details, see RFC 2460, "IPv6 specification.") A n IPv6 version of ICMP is d e s i g n e d (RFC 2463). T h e IPv6 h e a d e r consists of a 40-byte h e a d e r followed b y u p to six optional e x t e n s i o n h e a d e r s . T h e 40-byte h e a d e r is s h o w n in Figure 4.9. T h e version field c o n t a i n s t h e b i n a r y r e p r e s e n t a t i o n of t h e n u m b e r 6, indicating t h a t t h e p a c k e t is a n IPv6 packet. T h e r o u t e r u s e s t h e traffic class field to d e t e r m i n e t h e i m p o r t a n c e a n d u r g e n c y of t h e packet, s o m e w h a t like t h e TOS field c a n b e u s e d in Ipv4. T h e flow label m a y b e u s e d to describe t h e characteristics of t h e traffic to w h i c h t h e p a c k e t belongs. T h e payload l e n g t h indicates t h e n u m b e r of b y t e s in t h e p a c k e t t h a t follows t h e h e a d e r ; zero h e r e indicates t h a t t h e actual p a c k e t l e n g t h is specified in a hop-by-hop e x t e n s i o n h e a d e r (see next page). T h e next h e a d e r indicates w h i c h of t h e six e x t e n s i o n h e a d e r s follows this header, if any, a n d w h e t h e r t h e p a c k e t is for U D P or for TCP, otherwise. T h e e x t e n s i o n h e a d e r h a s a similar next h e a d e r field. T h e h o p limit is d e c r e m e n t e d b y o n e b y e a c h r o u t e r to p r e v e n t p a c k e t s from looping forever in t h e n e t w o r k ; t h e p a c k e t is discarded w h e n t h e value r e a c h e s zero, so t h a t t h e m a x i m u m n u m b e r of h o p s is 254. T h e source address a n d d e s t i n a t i o n a d d r e s s h a v e 16 b y t e s (RFC 1884), e n o u g h for

43

Internet Protocol

1

177 4

Version

32 T.Class Payload length

Flow label Next header

Hop limit

Source address

Destination address

4.9 FIGURE

The IPv6 header consists of the 40-byte header shown here, followed b y up to six extension headers.

5 χ 1 0 a d d r e s s e s p e r h u m a n being, w h i c h s h o u l d suffice for a while; t h e r e is n o explicit s u p p o r t for mobile hosts. Note t h a t t h e r e is n o e r r o r - d e t e c t i o n field in t h e h e a d e r : IPv6 relies o n higher-layer protocols for e r r o r control. In IPv6, t h e r o u t e r s do n o t fragment packets. W h e n a p a c k e t is too l o n g for a r o u t e r to h a n d l e , t h e r o u t e r s e n d s a n ICMP m e s s a g e to t h e source asking it to fragment t h e p a c k e t a n d r e s e n d it. T h e e x t e n s i o n h e a d e r s are t h e following: 2 8

•

Hop-by-hop Options: I n f o r m a t i o n t h a t e v e r y r o u t e r m u s t e x a m i n e (usage to b e d e t e r m i n e d later).

•

Routing: Specifies a set of r o u t e r s t h a t t h e d a t a g r a m m u s t visit (a list of u p to 24 IPv6 addresses).

•

Fragmentation: This h e a d e r specifies w h i c h fragment of a larger p a c k e t this IPv6 p a c k e t contains. By u s i n g this e x t e n s i o n header, t h e d e s t i n a t i o n c a n r e a s s e m b l e t h e p a c k e t as in IPv4.

• Authentication: T h i s h e a d e r c o n t a i n s a c h e c k s u m t h a t e n a b l e s t h e destina­ tion to a u t h e n t i c a t e t h e sender. •

Encapsulating Security Payload: T h i s h e a d e r c o n t a i n s t h e e n c r y p t i o n k e y n u m b e r a n d t h e e n c r y p t e d payload. T h e d e s t i n a t i o n c a n r e c o v e r t h e orig­ inal payload from t h e s e two i t e m s u s i n g a d e c r y p t i o n a l g o r i t h m t h a t is left to t h e choice of t h e user.

•

Destination Options: T h e possible u s e of this header, i n t e n d e d o n l y for t h e destination, h a s n o t b e e n defined so far.

The Internet and TCP/IP Networks

178

This h e a d e r s t r u c t u r e is v e r y flexible in t h a t it leaves t h e door o p e n for extensions. It is also r a t h e r efficient in t h a t t h e m i n i m a l h e a d e r is s i m p l e r t h a n in IPv4 b y t h e r e m o v a l of t h e c h e c k s u m a n d of fragmentation. With time, t h e I n t e r n e t d e s i g n e r s will d e v e l o p m e c h a n i s m s t h a t exploit this format to i m p l e m e n t a large r a n g e of services.

4.4

TCP AND UDP T h e I n t e r n e t u s e s two different t r a n s p o r t layers: U D P a n d TCP. U D P a n d TCP data u n i t s from t h e t r a n s p o r t layer are forwarded to t h e IP layer, w h i c h e n c a p s u l a t e s t h e m in IP packets. UDP, t h e User Data Protocol, is a c o n n e c t i o n l e s s service t h a t provides for multiplexing a n d for error detection, w i t h o u t g u a r a n t e e d delivery or duplicate p r e v e n t i o n . UDP is useful for s i m p l e q u e r y - r e s p o n s e applications b e c a u s e n o t i m e is lost for TCP c o n n e c t i o n e s t a b l i s h m e n t . U D P is also u s e d for applications w h e r e reliability is n o t critical or w h e r e r e t r a n s m i s s i o n delays are u n a c c e p t ­ able. For instance, t e l e p h o n y over IP u s e s UDP, as do audio a n d video s t r e a m i n g applications. T h e UDP h e a d e r consists of 16-bit source a n d d e s t i n a t i o n port n u m b e r s , a 16-bit l e n g t h field t h a t gives t h e total l e n g t h (data a n d h e a d e r ) in bytes, a n d a 16-bit c h e c k s u m over t h e h e a d e r a n d data (RFC 768). TCP, t h e Transmission Control Protocol, is a c o n n e c t i o n - o r i e n t e d , duplex, reliable, b y t e - s t r e a m service w i t h flow control. Byte-stream service m e a n s t h a t w i t h i n a TCP c o n n e c t i o n , t h e source s e n d s a s e q u e n c e of b y t e s for delivery to t h e destination. TCP buffers a g r o u p of t h e s e b y t e s into a segment a n d gives it to IP. T h e size of t h e buffer is MSS, or m a x i m u m s e g m e n t size. T h e default value of MSS is 536. A s e g m e n t h a s n o s e m a n t i c significance. Reliable service m e a n s g u a r a n t e e d delivery, in order, a n d w i t h o u t duplicates. TCP u s e s t h e Go Back Ν protocol (GBN), discussed in section 2.6.3, to e n s u r e reliable service. A version of TCP, SACK TCP, u s e s t h e Selective Repeat Protocol (SRP). Figure 4.10 shows t h e TCP h e a d e r (RFC 761). T h e 16-bit source a n d destina­ tion port n u m b e r s p e r m i t multiplexing of several c o n n e c t i o n s b e t w e e n hosts. Well-known ports, s u c h as 21 for FTP a n d 80 for HTTP, are fixed. Every b y t e of data s e n t over a TCP c o n n e c t i o n h a s a 32-bit s e q u e n c e n u m ­ b e r given b y t h e s u m of t h e s e g m e n t s e q u e n c e n u m b e r a n d its position in t h e s e g m e n t . T h e s e q u e n c e n u m b e r of t h e first s e g m e n t of a c o n n e c t i o n is a g r e e d o n b y a three-way h a n d s h a k e : (1) t h e source a n n o u n c e s t h e s e q u e n c e n u m ­ ber; (2) t h e d e s t i n a t i o n a c k n o w l e d g e s t h a t n u m b e r ; a n d (3) t h e source s e n d s t h e first s e g m e n t w i t h t h a t n u m b e r . T h i s p r o c e d u r e p r e v e n t s t h e m i s u n d e r -

4.4

TCP and UDP

179 _l

I

I

I

I

I

I

I

I

I

I

I

I

I

I

I

1

1

1 L

—I

I

I

31

I L_

Destination port

Source port Sequence number Acknowledgment number Reserved

Offset

Checksum

4.10 ORORR^T!:

Flags

Window Urgent pointer

TCP header. This header indicates the source and destination port numbers, ^ d ] contains the sequence and acknowledgment n u m b e r s needed b y the SRP that TCP implements. The window size specified b y the header is negotiated b y the hosts. 6

n e a

e r

a

s o

standing t h a t w o u l d b e c a u s e d b y a d e l a y e d packet. C o n n e c t i o n s are r e l e a s e d b y o n e two-way h a n d s h a k e for e a c h direction. T h e 32-bit a c k n o w l e d g m e n t n u m b e r is c u m u l a t i v e so t h a t a n acknowledg­ m e n t of s e q u e n c e n u m b e r η indicates t h a t all b y t e s u p to b u t n o t i n c l u d i n g η h a v e b e e n received, a n d η is t h e next e x p e c t e d b y t e n u m b e r . ( T h u s , if se­ q u e n c e n u m b e r η is a c k n o w l e d g e d t h r e e t i m e s , say, t h e s e n d e r m a y infer t h a t t h e p a c k e t c o n t a i n i n g t h a t b y t e is lost.) T h e 4-bit offset gives t h e n u m b e r of 32-bit w o r d s in t h e TCP header, includ­ ing options, a n d indicates w h e r e t h e data b e g i n s . Six flags m a y b e set: URG, ACK, PSH, RST, SYN, FIN. If URG is set, t h e 16-bit u r g e n t p o i n t e r indicates t h e position of t h e first b y t e of n o n u r g e n t data in t h e s e g m e n t . ( P r e s u m a b l y , all t h e p r e c e d i n g b y t e s are u r g e n t . ) ACK is set o n c e t h e c o n n e c t i o n is established, indicating validity of t h e a c k n o w l e d g m e n t n u m b e r . PSH is set to indicate t h a t t h e s e n d buffer w a s e m p t i e d b y t h e p a c k e t c o n t a i n i n g PSH = 1. W h e n set, RST i m m e d i a t e l y t e r m i n a t e s t h e c o n n e c t i o n . SYN is set to establish a c o n n e c t i o n , indicating t h a t t h e s e g m e n t carries t h e initial s e q u e n c e n u m b e r . FIN is set to r e q u e s t n o r m a l t e r m i n a t i o n of t h e c o n n e c t i o n : t h e twow a y h a n d s h a k e r e q u i r e s a FIN in e a c h direction. T h e c h e c k s u m over t h e data a n d h e a d e r provides error detection. T h u s , t h e steps of a TCP c o n n e c t i o n from A to Ε are as follows: 1. A s e n d s a SYN to Β to indicate t h a t it w a n t s to o p e n a c o n n e c t i o n . T h e c o n n e c t i o n is identified b y t h e s o u r c e a n d d e s t i n a t i o n IP a d d r e s s e s a n d TCP p o r t n u m b e r s , as well as b y a n initial s e q u e n c e n u m b e r t h a t A d e t e r m i n e s

The Internet and TCP/IP Networks

180

from its clock. T h i s s e q u e n c e n u m b e r p r e v e n t s confusion t h a t m i g h t b e c a u s e d b y d e l a y e d SYN packets. 2. Β s e n d s a SYN.ack b a c k to A to a c k n o w l e d g e t h e start of t h e c o n n e c t i o n a n d to indicate t h e initial s e q u e n c e n u m b e r it will u s e w h e n s e n d i n g to A. 3. A s e n d s t h e first data packet, w h i c h also a c k n o w l e d g e s t h e r e c e p t i o n of t h e SYN.ack. T h i s packet c o m p l e t e s t h e t h r e e - w a y h a n d s h a k e . 4. A k e e p s s e n d i n g p a c k e t s a n d Β a c k n o w l e d g e s e v e r y correct p a c k e t it re­ ceives w i t h a n ACK w h o s e s e q u e n c e n u m b e r is t h a t of t h e next b y t e t h a t it expects to receive. A a n d Β u s e t h e GBN protocol for this exchange. (Note t h a t in m a n y TCP i m p l e m e n t a t i o n s , t h e r e c e i v e r delays t h e t r a n s m i s s i o n of a c k n o w l e d g m e n t s . ) A adjusts t h e w i n d o w size as w e explain in section 4.6. 5. W h e n o n e of t h e hosts, say B, w a n t s to close t h e c o n n e c t i o n , it s e n d s a FIN to t h e o t h e r h o s t A. A t h e n s e n d s b a c k a FIN.ack to c o m p l e t e a two-way h a n d s h a k e close of t h e c o n n e c t i o n from Β to A. Finally, A s e n d s a FIN to Β a n d Β replies w i t h a FIN.ack to close t h e o t h e r side of t h e c o n n e c t i o n . T h e options field is u s e d at c o n n e c t i o n e s t a b l i s h m e n t to negotiate a variety of options, including MSS. T h e 16-bit window, W , is u s e d for flow control. It is set b y t h e receiver at t h e m a x i m u m n u m b e r of u n a c k n o w l e d g e d b y t e s it is willing to accept from t h e sender. It is usually set at t h e size of t h e receiver's buffer t h a t h o l d s out-ofo r d e r packets. If t h e buffer is full, t h e receiver m a y r e s e t t h e w i n d o w to zero. T h e last b y t e t h e s e n d e r can s e n d is t h e a c k n o w l e d g m e n t s e q u e n c e n u m b e r plus t h e w i n d o w size. At a n y time, t h e source m u s t u s e SRP w i t h w i n d o w size equal to t h e s m a l l e r of W a n d t h e w i n d o w size W calculated b y a congestion avoidance algorithm, studied i n section 8.3.4. A n u m b e r of s u c h a l g o r i t h m s h a v e b e e n proposed, a n d w e discuss t h e i r m a i n o p e r a t i o n s i n section 4.6.1. Roughly, W is i n c r e a s e d w h e n congestion is n o t indicated a n d d e c r e a s e d w h e n t h e source detects congestion. max

max

4.4.1

Applications O n top of t h e TCP l a y e r are g e n e r i c applications, a s s h o w n i n Figure 4.2. We describe s o m e of t h e s e . A n application u s e s TCP o r UDP via s y s t e m calls t h a t a r e p a r t of t h e operating s y s t e m . For instance, in UNIX a n d in Windows, t h e s y s t e m calls are socket-based. A socket c a n b e t h o u g h t of as a q u e u e a t t a c h e d to t h e TCP or U D P protocols. Tb give a s e n s e of h o w t h e s e s y s t e m calls are used, i m a g i n e t h a t y o u w a n t to s e n d m e s s a g e s from a c o m p u t e r w i t h IP a d d r e s s I P l to a c o m p u t e r w i t h

4.4

TCP and UDP

181

IP address IP2. In o n e case, w e s e n d a m e s s a g e stored in buffer Bl u s i n g UDP w i t h source port n u m b e r PI a n d d e s t i n a t i o n p o r t n u m b e r P2. T h e c o m p u t e r IP2 w r i t e s t h e m e s s a g e into a buffer B2. T h e pseudo-code at IP2 is as follows: c r e a t e s o c k e t S2 % c r e a t e t h e s o c k e t

object

b i n d S2 t o IP2.P2 a s UDP % a t t a c h t h e s o c k e t t o UDP a t IP2 and P2 r e c e i v e S2 i n t o B2 °/ w r i t e what a r r i v e s a t t h e s o c k e t i n t o B2 0

c l o s e S2 % d e l e t e t h e s o c k e t a f t e r t h e t r a n s a c t i o n i s

completed

H e r e is t h e pseudo-code at I P l : c r e a t e s o c k e t SI

% create the socket

object

b i n d SI t o I P l . P I as UDP % a t t a c h t h e s o c k e t t o UDP a t I P l and PI send Bl t o IP2.P2 through SI % w r i t e t h e message from Bl t o t h e s o c k e t c l o s e SI % d e l e t e t h e s o c k e t

Now consider t h e case of a TCP c o n n e c t i o n from I P l w i t h p o r t n u m b e r PI to IP2 w i t h port n u m b e r P2. T h e pseudo-code at IP2 is as follows: c r e a t e s o c k e t S2 % c r e a t e t h e s o c k e t

object

b i n d S2 t o IP2.P2 a s TCP % a t t a c h t h e s o c k e t t o TCP a t IP2 and P2 l i s t e n t o S2 % g e t ready t o r e c e i v e p a c k e t s accept % accept a connection read S2 i n t o B2 % w r i t e i n t o B2 what you r e c e i v e on S2 c l o s e S2 % t e r m i n a t e t h e

connection

H e r e is t h e pseudo-code at I P l : c r e a t e s o c k e t SI

% create the socket

object

bind SI t o I P l . P I a s TCP % a t t a c h t h e s o c k e t t o TCP a t I P l and PI c o n n e c t SI t o IP2.P2 % open t h e

connection

send Bl t o SI % w r i t e t h e m e s s a g e s from Bl t o t h e s o c k e t c l o s e SI % d e l e t e t h e s o c k e t

T h e actual code d e p e n d s o n t h e o p e r a t i n g s y s t e m . M a n y textbooks provide examples.

4.4.2

FTP FTP, t h e File Transfer Protocol, e n a b l e s u s e r s to transfer files b e t w e e n c o m p u t ­ ers. As Figure 4.11 shows, FTP o p e n s two c o n n e c t i o n s b e t w e e n t h e c o m p u t e r s :

The Internet and TCP/IP Networks

182 Interactive: send, get, transfer, cd Modes: stream, block, compressed

4.11

frrTf™

The File Transfer Protocol (FTP) sets u p two connections: one for the commands, the other for the data exchange.

o n e c o n n e c t i o n for t h e c o m m a n d s a n d replies a n d t h e o t h e r for t h e data trans­ fers. FTP is interactive. Its c o m m a n d s are send, get, transfer, and cd ( c h a n g e directory). FTP transfers files in t h r e e m o d e s : stream, block, a n d compressed. In t h e s t r e a m m o d e , FTP h a n d l e s information as a string of b y t e s w i t h o u t sepa­ rating b o u n d a r i e s . In t h e block m o d e , FTP d e c o m p o s e s t h e i n f o r m a t i o n into blocks of data. In t h e c o m p r e s s m o d e , FTP u s e s t h e Lempel-Ziv a l g o r i t h m to c o m p r e s s data.

4.4.3

SMTP, rlogin, Τ FTP, and HTTP T h e first two of t h e s e applications r u n o n top of FTP, t h e t h i r d is b a s e d o n UDP, a n d t h e last o n TCP. T h e Simple Mail Transfer Protocol (SMTP) is u s e d for e-mail. SMTP accepts m e s s a g e s w i t h a list of destinations. W h e n it does n o t s u c c e e d in delivering a message, SMTP retries a n u m b e r of t i m e s , a n d it notifies t h e s e n d e r if it c a n n o t deliver t h e message. T h e remote login service, rlogin, e n a b l e s a u s e r to access a r e m o t e m a c h i n e . Rlogin establishes t h e c o n n e c t i o n to t h e r e m o t e m a c h i n e , e x c h a n g e s t h e re­ q u i r e d authorization information, a n d e v e n t u a l l y logs in t h e user. W h e n rlogin is used, t h e e c h o i n g is p e r f o r m e d b y t h e r e m o t e m a c h i n e . T h a t is, t h e c h a r a c t e r s

4.5

Internet Success and Limitation

« « • • ^ ^

183

t y p e d b y t h e u s e r are s e n t b a c k b y t h e r e m o t e m a c h i n e before b e i n g displayed. Special control c o m m a n d s s u c h as stop a n d resume are s e n t as u r g e n t data. T h e Trivial File Transfer Protocol (TFTP) transfers data as blocks of 512 b y t e s . T F T P s e n d s o n e block of 512 b y t e s a n d waits for a n a c k n o w l e d g m e n t . T F T P retries after a t i m e o u t until it s u c c e e d s a n d t h e n p r o c e e d s to t h e n e x t block. T F T P n u m b e r s t h e blocks s e q u e n t i a l l y from 1. T h i s r o b u s t protocol o p e r a t e s e v e n w h e n t h e t r a n s p o r t l a y e r is of l o w quality. However, it is n o t efficient a n d is useful o n l y w h e n a lightweight protocol is n e e d e d , for i n s t a n c e as a s i m p l e b o o t ROM application for loading k e r n e l s over t h e n e t w o r k . T h e Hypertext Transfer Protocol ( H T T P ) is t h e basis for access over t h e World Wide Web to r e s o u r c e s r e f e r e n c e d b y t h e i r URL. T h e s u c c e s s of t h e Web is vir­ tually d u e to t h e flexibility of HTTP. T h e H T T P protocol is a r e q u e s t / r e s p o n s e protocol o n top of TCP. A client, often a browser, o p e n s a TCP c o n n e c t i o n w i t h a server a n d s e n d s a r e q u e s t in t h e form of a m e t h o d (e.g., GET, POST, HEAD, DELETE), t h e URL of t h e r e s o u r c e (the object to w h i c h t h e m e t h o d is to b e applied), a n d t h e protocol version (e.g., H T T P / 1 . 1 ) , possibly followed b y m o d ­ ifiers. T h e s e r v e r r e s p o n d s w i t h a status line, i n c l u d i n g t h e m e s s a g e ' s protocol version a n d a success or error code (e.g., OK, P a y m e n t r e q u i r e d ) followed b y a m e s s a g e c o n t a i n i n g s e r v e r information, a n d possibly, a b o d y giving informa­ tion a b o u t t h e data (e.g., h o w it is coded) a n d t h e data itself (e.g., a n HTML d o c u m e n t ) . T h e b r o w s e r m u s t b e able to i n t e r p r e t t h e data. I n t h e first version of HTTP, t h e r e s p o n s e w a s a m e s s a g e in h y p e r t e x t m a r k - u p l a n g u a g e (HTML). In t h e c u r r e n t version, t h e form of t h e H T T P r e q u e s t a n d r e s p o n s e is extensi­ ble so t h a t n e w m e t h o d s a n d r e s o u r c e t y p e s (images, video) c a n b e i n t r o d u c e d . T h e complexity of b r o w s e r p r o g r a m s is d u e to t h e large n u m b e r of m e t h o d s a n d r e s o u r c e t y p e s a n d display formats t h a t m u s t b e h a n d l e d .

4.5

INTERNET SUCCESS AND LIMITATION We studied t h e I n t e r n e t suite of protocols in t e r m s of t h e variety of services t h a t it s u p p o r t s . We also s a w t h a t t h o s e protocols a r e organized in a m u l t i l a y e r e d a r c h i t e c t u r e r e s e m b l i n g t h e seven-layer OSI m o d e l . Figure 4.2 s u m m a r i z e s b o t h of t h e s e aspects. It lists t h e different protocols in a m a n n e r t h a t s h o w s t h e i r l a y e r e d d e p e n d e n c e a n d h o w t h o s e l a y e r s are r e l a t e d to t h e OSI m o d e l . In order to explain t h e t e c h n i c a l basis for t h e a s t o u n d i n g success of t h e I n t e r n e t a n d to foresee its limitations, it will b e m o r e r e v e a l i n g to view t h e

The Internet and TCP/IP Networks

184

4.12 mmncmxi:

The Internet implementation of the Open Data Network model shows a simple bearer service, supported by m a n y networks and supporting a variety of applications. The bearer service is provided b y simple routers, facilitating network interconnection. Applications are provided by hosts, facilitating experimentation and incremental evolution.

I n t e r n e t protocols as a n i m p l e m e n t a t i o n of t h e O p e n Data N e t w o r k m o d e l of section 2.8. Such a view is p r e s e n t e d in Figure 4.12. T h e n a r r o w "waist" of t h e figure r e p r e s e n t s t h e single a n d s i m p l e IP b e a r e r service: t h e end-to-end t r a n s p o r t of IP d a t a g r a m s . T h i s service c a n b e p r o v i d e d b y r o u t e r s capable of t h e basic tasks of storing p a c k e t s a n d r o u t i n g t h e m to a p p r o p r i a t e outgoing links after consulting a routing table. Most significantly, t h e r e are n o p e r f o r m a n c e r e q u i r e m e n t s , so t h a t slow a n d old r o u t e r s c a n coexist w i t h new, h i g h - p e r f o r m a n c e routers. This "backward compatibility" is enor­ m o u s l y valuable a n d explains w h y n e t w o r k s w i t h widely differing p e r f o r m a n c e c a n b e i n t e r c o n n e c t e d (see Figure 1.13). D o m a i n n a m e s a n d IP a d d r e s s e s are assigned o n a decentralized basis, so n e t w o r k g r o w t h is o p p o r t u n i s t i c r a t h e r t h a n p l a n n e d . T h e benefits of c o n n e c t i n g to t h e I n t e r n e t grow w i t h its size at t h e s a m e t i m e as e q u i p m e n t costs (LANs, links, c o m p u t e r s ) decline b e c a u s e of scale e c o n o m i e s , resulting in a doubling of t h e size of t h e I n t e r n e t e a c h year. T h e brilliance of t h e IP design lies in its simplicity: b e c a u s e IP d a t a g r a m s are self-contained, r o u t e r s do n o t n e e d or k e e p a n y state i n f o r m a t i o n a b o u t those datagrams. As a result, t h e n e t w o r k b e c o m e s v e r y robust. If a r o u t e r fails, d a t a g r a m s in t h e r o u t e r m a y b e lost, b u t n e w d a t a g r a m s w o u l d a u t o m a t i c a l l y b e r o u t e d properly, w i t h n o special p r o c e d u r e s . (By contrast, if t h e b e a r e r service w e r e c o n n e c t i o n - o r i e n t e d like ATM, r o u t e r s w o u l d h a v e to m a i n t a i n connection-state information, a n d if a r o u t e r w e r e to fail, t h a t state i n f o r m a t i o n

4.5

Internet Success and Limitation

w o u l d b e lost, m a k i n g it v e r y difficult to restore t h e c o n n e c t i o n . ) Simple r u l e s in r o u t e r s c a n h e l p r o u t e traffic a r o u n d c o n g e s t e d p a r t s of t h e n e t w o r k , giving t h e n e t w o r k t h e capability to a d a p t to c h a n g e s in traffic. T h e t h i r d feature of t h e figure—its w i d e t o p — r e p r e s e n t s t h e rich variety of applications t h a t t h e IP b e a r e r service, t o g e t h e r w i t h U D P a n d TCP, c a n s u p p o r t . T h e last d e c a d e of successful, sophisticated applications like t h e World Wide Web h a s s h o w n t h a t U D P / I P d a t a g r a m service a n d T C P / I P reliable, b y t e - s t r e a m c a n serve as building blocks for c o m p l e x services, p r o v i d e d t h a t t h e e n d h o s t s are sophisticated. T h e m o s t r e m a r k a b l e aspect of this feature is t h a t t h e application software resides e n t i r e l y in t h e e n d h o s t s a n d n o t in t h e r o u t e r s . T h i s m e a n s t h a t t h e s a m e basic service, i m p l e m e n t e d b y s i m p l e routers, c a n s u p p o r t t h e s e sophisticated applications. T h u s t h e n e t w o r k h a r d w a r e a n d software h a v e a m u c h longer t e c h n i c a l a n d e c o n o m i c life t h a n does t h e e n d host. I n d e e d , in t h e mid-1990s, significant n u m b e r s of I n t e r n e t r o u t e r s are 15 to 20 y e a r s old. T h i s also implies t h a t p a r t s of t h e I n t e r n e t c a n e x p e r i m e n t w i t h n e w applications o n a d v a n c e d h o s t s u s i n g t h e real n e t w o r k , while o t h e r p a r t s of t h e I n t e r n e t c o n t i n u e u n d i s t u r b e d to r u n old applications o n primitive hosts. T h i s ability to e x p e r i m e n t w i t h n e w applications h a s greatly h e l p e d t h e proliferation of n e w applications. T h u s t h e t e c h n i c a l basis for t h e I n t e r n e t ' s success is its r e l i a n c e o n s i m p l e r o u t e r s to transfer individual d a t a g r a m s a n d o n a d v a n c e d e n d h o s t s to r u n sophisticated applications. T h e s i m p l e infrastructure is c o m p a t i b l e w i t h a w i d e r a n g e of applications. T h e d e v e l o p e r s of successful applications c a n distribute t h e m for fame or profit, w i t h o u t r e q u i r i n g a n y c h a n g e from t h e infrastructure. T h e contrast w i t h t h e t e l e p h o n e n e t w o r k could n o t b e m o r e striking. T h e r e t h e n e t w o r k "intelligence" is located in its expensive switches, w h i l e t h e e n d h o s t s (the t e l e p h o n e sets) are primitive, w i t h little functionality. T h e i n t r o d u c t i o n of n e w services r e q u i r e s c h a n g e s in t h e i n f r a s t r u c t u r e — c h a n g e s t h a t are slow a n d expensive. H e n c e e x p e r i m e n t s are costly a n d infrequent, a n d n e w services are i n t r o d u c e d after m u c h deliberation a n d p l a n n i n g . T h e limitations of t h e I n t e r n e t c a n b e foreseen from t h e figure. T h e IP b e a r e r service c a n n o t provide a n y g u a r a n t e e s in t e r m s of d e l a y or b a n d w i d t h or loss. Routers treat all p a c k e t s in t h e s a m e way. (This "equal service for all" is, p e r h a p s ironically, called best-effort service.) T h i s is a n i n n a t e feature: t h e a b s e n c e of state i n f o r m a t i o n m e a n s t h a t p a c k e t s c a n n o t b e differentiated b y t h e i r application or c o n n e c t i o n , a n d so r o u t e r s w o u l d b e u n a b l e to provide additional r e s o u r c e s to m o r e d e m a n d i n g applications. T h e t e c h n i c a l challenge is to e x p a n d t h e services offered b y U D P / T C P a n d IP to provide g u a r a n t e e s in a w a y t h a t p r e s e r v e s t h e I n t e r n e t ' s a c c o m m o d a t i o n

185

The Internet and TCP/IP Networks

186

of b a c k w a r d compatibility a n d i n c r e m e n t a l change. We n o w discuss s o m e r e c e n t proposals to m e e t this challenge. T h e proposals c a n b e r o u g h l y classified in two t y p e s : those t h a t a s s u m e IP b e a r e r service b u t e x p a n d t h e services offered b y UDP a n d TCP to b e t t e r suit applications; a n d t h o s e t h a t alter t h e stateless n a t u r e of IP routers. Most proposals are of t h e first type. UDP service offers u n r e l i a b l e delivery of individual p a c k e t s w i t h little overhead, a n d TCP service offers reliable, o r d e r e d flow of b y t e s t r e a m s at t h e cost of considerable o v e r h e a d . Application r e q u i r e m e n t s m a y n o t readily m a p into e i t h e r of t h e s e two services. For example, video s t r e a m i n g applications m a y b e b e t t e r m a t c h e d to a service t h a t offers unreliable, o r d e r e d flow of b y t e s t r e a m s . Some H T T P applications, o n t h e o t h e r h a n d , m a y b e b e t t e r m a t c h e d to a reliable, u n o r d e r e d , delivery of m e s s a g e s .

4.6

PERFORMANCE OF TCP/IP NETWORKS We s t u d y t h e p e r f o r m a n c e of T C P / I P n e t w o r k s in section 8.3.4. In this section, w e highlight t h e m a i n p e r f o r m a n c e characteristics of t h e s e n e t w o r k s b e c a u s e t h e y justify a n u m b e r of modifications t h a t are b e i n g p r o p o s e d for t h e T C P / I P protocols. We first consider t h e w i n d o w a d j u s t m e n t m e c h a n i s m of TCP. T h i s m e c h a ­ n i s m d e t e r m i n e s t h e traffic t h a t TCP sources p r o d u c e . In 1999, a b o u t 8 5 % of t h e traffic o n t h e I n t e r n e t is TCP. O u r discussion shows t h a t TCP is b i a s e d in favor of c o n n e c t i o n s w i t h a small round-trip t i m e . I b correct this bias, variations of TCP h a v e b e e n p r o p o s e d a n d so h a v e modifications of t h e r o u t e r s . Next, w e exam­ ine t h e n o t i o n of differentiated service a n d discuss possible i m p l e m e n t a t i o n s e i t h e r in TCP or in IP.

4.6.1

Window Adjustment in TCP T h e t r a n s m i s s i o n rate of a TCP source d e p e n d s o n t h e w i n d o w size. Recall t h a t t h e w i n d o w size is t h e m a x i m u m n u m b e r of b y t e s t h a t t h e source c a n s e n d in a set of packets for w h i c h it h a s n o t r e c e i v e d a c k n o w l e d g m e n t s . As w e explained, t h e d e s t i n a t i o n c o m p u t e r s e n d s b a c k a n a c k n o w l e d g m e n t for every correct p a c k e t t h a t it receives. A s s u m e t h a t t h e r o u n d - t r i p t i m e of t h e c o n n e c t i o n is Τ seconds. T h a t is, Τ is t h e t i m e from t h e t r a n s m i s s i o n of a p a c k e t until t h e r e c e p t i o n of its a c k n o w l e d g m e n t b y t h e source. If t h e w i n d o w size is

4.6

Performance of TCP/IP Networks

187

W bytes, t h e n t h e source s e n d s W b y t e s in Τ seconds, a n d its t r a n s m i s s i o n rate is W/T b y t e s / s . It is t e m p t i n g for t h e source to c h o o s e a v e r y large v a l u e for W to i n c r e a s e its t h r o u g h p u t . However, doing so m a k e s it likely t h a t a r o u t e r will n o t b e able to t r a n s m i t as fast as t h e source. Ideally, t h e source s h o u l d adjust its w i n d o w size so t h a t it gets its fair s h a r e of t h e t r a n s m i s s i o n r a t e of t h e r o u t e r s . T h e precise m e a n i n g of "fair share" c a n b e debated, b u t a s i m p l e i n t e r p r e t a t i o n is t h a t if Ν c o n n e c t i o n s s h a r e a r o u t e r t h a t is t h e i r o n l y b o t t l e n e c k a n d t r a n s m i t s w i t h rate C, t h e n e a c h c o n n e c t i o n s h o u l d b e able to t r a n s m i t w i t h a rate close to C/N. It w o u l d b e v e r y unfair if s o m e c o n n e c t i o n s could t r a n s m i t at a m u c h larger rate t h a n t h e others. Also, if t h e w i n d o w sizes a r e too small, t h e n t h e c o n n e c t i o n s a r e n o t taking a d v a n t a g e of t h e link capacities. S u m m i n g u p , w e see t h a t t h e objectives of t h e w i n d o w a d j u s t m e n t m e c h a n i s m a r e to s h a r e t h e links fairly a n d efficiently a m o n g t h e c o n n e c t i o n s . T h e TCP w i n d o w a d j u s t m e n t m e c h a n i s m starts b y i n c r e a s i n g t h e w i n d o w size e x p o n e n t i a l l y fast to "discover" t h e available t r a n s m i s s i o n rate. W h e n t h e source fails to get a n a c k n o w l e d g m e n t , it s u s p e c t s t h a t a r o u t e r h a s d r o p p e d a p a c k e t a n d it r e d u c e s its w i n d o w size b y 50%. F r o m t h a t p o i n t on, t h e source i n c r e a s e s its w i n d o w size b y o n e u n i t e v e r y r o u n d - t r i p t i m e . T h e a l g o r i t h m t h e n c o n t i n u e s as before. Figure 4.13 s h o w s t h e resulting evolution of t h e w i n d o w size of a c o n n e c t i o n t h a t goes t h r o u g h a router, a s s u m i n g t h a t it is t h e only active c o n n e c t i o n . F r o m this s i m p l e graph, w e c a n infer t h e following conclusions. First, a c o n n e c t i o n w i t h a small r o u n d - t r i p t i m e is m u c h m o r e aggressive t h a n o n e w i t h a large o n e . C o n s e q u e n t l y , if c o n n e c t i o n s w i t h different r o u n d - t r i p t i m e s s h a r e a router, t h e o n e s w i t h t h e s m a l l e r r o u n d - t r i p t i m e s a r e likely to grab m o s t of t h e link capacity. We elaborate o n this o b s e r v a t i o n i n section 8.3.4.

Window size

Slope: 1 /round-trip time

D

..^JL1JL,.,_ FIGURE

Window size of a single TCP connection going through a single router.

The Internet and TCP/IP Networks

188

Second, n o t e t h a t TCP loses a p p r o x i m a t e l y o n e p a c k e t e v e r y D = TW/2 s e c o n d s w h e r e Τ is t h e round-trip t i m e (see figure). In D seconds, t h e c o n n e c t i o n s e n d s a p p r o x i m a t e l y W/2 + (W/2 + 1) + · · · + 2W/2 « 3W /8 units. A s s u m e t h a t o n e u n i t is Κ average packets. T h u s , if w e designate b y R = 3KW /8D = 3KW/4T t h e t h r o u g h p u t of t h e c o n n e c t i o n (in p a c k e t s / s ) a n d b y L = 1 / ( 3 W /8) = 8/3W its loss rate, t h e n w e find t h a t 2

2

2

R-

λ

~

'

2

2 5 Κ

T\fZ'

T h a t identity shows t h a t t h e t h r o u g h p u t is inversely p r o p o r t i o n a l to t h e s q u a r e root of t h e loss rate. This observation is s o m e t i m e s u s e d to d e t e r m i n e w h e t h e r a p r o p o s e d protocol is "TCP-friendly." If t h e t h r o u g h p u t of t h e protocol is larger t h a n t h a t of TCP for t h e s a m e loss rate, t h e n o n e decides t h a t t h e protocol is n o t TCP-friendly, w h i c h is to say t h a t it is grabbing a n unfairly large fraction of t h e r o u t e r b a n d w i d t h . Of course s u c h a s u m m a r y j u d g m e n t neglects t h e effect of interactions b e t w e e n protocols. More i m p o r t a n t l y , it decides t h a t a protocol t h a t is "better" t h a n TCP is n o t desirable. Observe also t h a t if t h e link from t h e d e s t i n a t i o n to t h e source is slow, t h e n it m a y limit t h e rate at w h i c h t h e d e s t i n a t i o n c a n a c k n o w l e d g e packets, w h i c h in t u r n limits t h e rate at w h i c h t h e source c a n s e n d packets. W h e n a noisy link (e.g., wireless) drops p a c k e t s b e c a u s e of t r a n s m i s s i o n er­ rors, TCP slows d o w n b e c a u s e it a s s u m e s t h a t t h e losses are d u e to congestion. As a result, t h e p e r f o r m a n c e of TCP is v e r y poor w h e n links are noisy.

4.6.2

Suggested Improvements for TCP Tb i m p r o v e t h e b e h a v i o r s h o w n in Figure 4.13, I n t e r n e t r e s e a r c h e r s suggested t h e following i m p r o v e m e n t s for TCP: 1. Increase t h e w i n d o w size at a rate t h a t is i n d e p e n d e n t of t h e round-trip time. 2. Base t h e w i n d o w a d j u s t m e n t m e c h a n i s m o n delays, n o t o n losses. 3. H a v e t h e r o u t e r s m a r k t h e p a c k e t s to indicate t h a t t h e y are getting con­ gested i n s t e a d of waiting until t h e y h a v e to drop packets. 4. Base t h e decision to d r o p p a c k e t s o n t h e n u m b e r of p a c k e t s of t h e s a m e c o n n e c t i o n in t h e router, n o t s i m p l y o n t h e total n u m b e r of packets. 5. H a v e t h e r o u t e r s provide t h e sources w i t h a m o r e explicit indication of t h e available t r a n s m i s s i o n rate.

4.6

Performance of TCP/IP Networks

6. Merge a c k n o w l e d g m e n t s o n a slow r e v e r s e link. 7. I n t r o d u c e a link error control m e c h a n i s m for n o i s y links. 8. Cheat a n d modify y o u r TCP w i n d o w a d j u s t m e n t m e c h a n i s m .

T h e first modification w o u l d e l i m i n a t e t h e b i a s in favor of c o n n e c t i o n s w i t h s h o r t e r round-trip t i m e s . However, e x p e r i m e n t s s h o w t h a t it is n o t e a s y to choose a rate of increase of t h e w i n d o w size t h a t w o r k s well over a w i d e r a n g e of r o u n d - t r i p t i m e s . If t h a t r a t e of i n c r e a s e is small c o m p a r e d w i t h t h e c u r r e n t rate, t h e n c o n n e c t i o n s are slow. If t h e rate is excessive, t h e n t h e protocol loses too m a n y p a c k e t s w h e n u s e d b y c o n n e c t i o n s w i t h a large r o u n d - t r i p t i m e . T h e s e c o n d proposal consists in c o m p a r i n g t h e r o u n d - t r i p t i m e w i t h t h e m i n i m u m r o u n d - t r i p t i m e o b s e r v e d over m a n y successive packets. T h e idea is t h a t t h e increase in r o u n d - t r i p t i m e m e a s u r e s t h e a m o u n t of q u e u i n g in t h e routers. TCP-Vegas is a protocol t h a t u s e s this m e c h a n i s m . T h e practical difficulty is t h a t t h e e s t i m a t e d m i n i m u m r o u n d - t r i p t i m e i n c l u d e s t h e q u e u i n g t i m e at t h e start of t h e c o n n e c t i o n . If all t h e s o u r c e s w e r e to u s e this m e c h a n i s m , t h e y m i g h t k e e p o n i n c r e a s i n g t h e congestion. T h e third m e c h a n i s m (called explicit congestion notification) r e q u i r e s t h a t t h e r o u t e r s b e able to m a r k p a c k e t s a n d t h a t t h e r e c e i v e r s e n d b a c k t h e m a r k s in t h e a c k n o w l e d g m e n t s . This m e c h a n i s m is desirable b e c a u s e it signals a source to slow d o w n w i t h o u t p a y i n g t h e p e n a l t y of forcing r e t r a n s m i s s i o n s . T h e actual a d j u s t m e n t of t h e w i n d o w size b a s e d o n t h e m a r k s m u s t b e d o n e carefully to avoid t h e bias in favor of c o n n e c t i o n s w i t h a s h o r t r o u n d - t r i p t i m e . T h e fourth proposal could limit t h e bias of TCP. Unfortunately, this m e c h a ­ n i s m is m o r e c o m p l e x to i m p l e m e n t . It r e q u i r e s t h a t t h e r o u t e r b e able to k e e p track of c o n n e c t i o n s a n d to c o u n t p a c k e t s of t h e different c o n n e c t i o n s . T h e fifth proposal n e c e s s i t a t e s a sophisticated r o u t e r t h a t c a n provide useful indications to t h e sources. O n e proposal is to m a r k a p a c k e t if it is likely to c a u s e a future loss of a packet. T h a t is, as a p a c k e t arrives, t h e r o u t e r p r e d i c t s w h e t h e r this p a c k e t will force it to d r o p o n e m o r e p a c k e t t h a t will arrive later. Such decisions m u s t b e b a s e d o n t h e c u r r e n t load of t h e router. T h e m e r g i n g of a c k n o w l e d g m e n t s o n a slow r e v e r s e link, for i n s t a n c e b y h a v i n g o n e ACK s e n t for e v e r y Ν packets, c a n e l i m i n a t e t h e effect of t h e a s y m m e t r y of t h e rates. T h e link e r r o r control protocol c o n v e r t s a n u n r e l i a b l e link into a reliable one, w h i c h i m p r o v e s t h e p e r f o r m a n c e of TCP. U n f o r t u n a t e l y t h e last proposal is i m p l e m e n t e d in a n u m b e r of c o m m e r c i a l p r o d u c t s t h a t p r o u d l y advertise b e t t e r n e t w o r k p e r f o r m a n c e . It is e a s y to c h e a t

189

The Internet and TCP/IP Networks

190

in t h e TCP i m p l e m e n t a t i o n b y increasing t h e initial w i n d o w size, t h e slope of t h e w i n d o w i n c r e a s e rate, or t h e factor b y w h i c h t h e w i n d o w size is d e c r e a s e d .

4.6.3

Suggested Improvements for IP O n e major a d v a n t a g e of IP is its simplicity, w h i c h m a k e s t h e n e t w o r k scalable a n d t h e e q u i p m e n t easily upgradable. T h e vision is t h a t if o n e b u i l d s a v e r y fast network, p e r f o r m a n c e p r o b l e m s go away. For instance, if all t h e links w o r k at 10 Gbps a n d h a v e a load of less t h a n 90%, t h e n t h e average d e l a y p e r n o d e should b e less t h a n 10 p a c k e t t r a n s m i s s i o n t i m e s , or less t h a n 80 μδ for 1-KB packets. If a p a t h across t h e I n t e r n e t goes t h r o u g h at m o s t 20 n o d e s , t h e average end-to-end delay s h o u l d b e less t h a n 1.6 m s p l u s t h e u n a v o i d a b l e p r o p a g a t i o n delay. This level of p e r f o r m a n c e is m o r e t h a n a d e q u a t e for c o n v e r s a t i o n a n d interactive applications. T h e r e are a few difficulties w i t h this vision. T h e first o n e is t h a t m a n y links will n o t b e o p e r a t i n g at 10 Gbps for a n u m b e r of y e a r s . T h a t is t h e case in particular for t h e access links t h a t c o n n e c t t h e u s e r ' s LAN or c o m p u t e r to t h e ISP router. T h e rate of s u c h links is typically b e l o w 1 Mbps. A n o t h e r p r o b l e m is t h a t t h e sources of TCP traffic are g r e e d y a n d t r y to grab all t h e available b a n d w i d t h . T h e s e s o u r c e s create b o t t l e n e c k s b y s e n d i n g b u r s t s of packets until losses tell t h e m to slow d o w n . A t h i r d p r o b l e m is t h a t t h e d e m a n d for b a n d w i d t h grows w i t h t h e supply. As t h e n e t w o r k p e r f o r m a n c e i m p r o v e s , t h e u s e r s derive m o r e benefits from u s i n g t h e n e t w o r k a n d t h e y t e n d to u s e it m o r e . Also, applications t h a t m a k e u s e of t h e i m p r o v e d p e r f o r m a n c e a p p e a r quickly a n d i n c r e a s e t h e n e t w o r k load. In v i e w of t h e s e difficulties, a n u m b e r of m e t h o d s h a v e b e e n d e s i g n e d to g u a r a n t e e t h e level of p e r f o r m a n c e t h a t s o m e applications r e q u i r e . T h e s e m e t h o d s involve s e p a r a t i n g different classes of traffic a n d h a n d l i n g t h e m dif­ ferently in t h e routers.

4.6.4

Queuing Algorithms A p a c k e t s c h e d u l e r sorts p a c k e t s t h a t m u s t b e forwarded t h r o u g h a link inter­ face into s e p a r a t e q u e u e s according to t h e i r QpS class. T h e a l g o r i t h m s a t t e m p t to provide t h e r e q u e s t e d QpS b y a p p r o p r i a t e l y s c h e d u l i n g t h e t r a n s m i s s i o n of packets. In first-come first-served or FCFS q u e u i n g , p a c k e t s are t r a n s m i t t e d in t h e o r d e r t h e y are received. T h i s widely u s e d a l g o r i t h m m a k e s n o distinction

Performance of TCP/IP Networks

4.6 i,«

B M

a m o n g p a c k e t s w i t h different QpS r e q u i r e m e n t s . Sources t h a t deliver m o r e or larger p a c k e t s will receive g r e a t e r b a n d w i d t h a n d m a y inflict large delays o n sources t h a t deliver fewer packets. Inpriority queuing, p a c k e t s are s o r t e d into different priority q u e u e s . H i g h e r priority p a c k e t s are t r a n s m i t t e d a h e a d of lower priority p a c k e t s . W h e n t h e r e is a large v o l u m e of h i g h e r priority traffic, l o w e r priority traffic m a y get v e r y little b a n d w i d t h a n d suffer large delays. In class-based queuing, p a c k e t s are s o r t e d into q u e u e s , o n e q u e u e p e r class. Packets w i t h i n e a c h q u e u e are t r a n s m i t t e d FCFS. Different q u e u e s are s e r v e d in round-robin fashion. T h e n u m b e r of p a c k e t s t r a n s m i t t e d from e a c h class, or t h e b a n d w i d t h d e v o t e d to e a c h class in e a c h r o u n d , is d e t e r m i n e d b y t h e attributes of t h a t class. T h u s t h e a l g o r i t h m g u a r a n t e e s t h a t e a c h class will receive a c e r t a i n fraction of t h e link b a n d w i d t h . Such g u a r a n t e e is n o t p r o v i d e d b y t h e two p r e v i o u s algorithms. However, w i t h i n a class, "misbehaving" s o u r c e s t h a t deliver m o r e or larger p a c k e t s will inflict delays o n s o u r c e s t h a t deliver fewer or s m a l l e r packets. In this s e n s e , t h e s c h e d u l i n g a l g o r i t h m m a y b e unfair. Fair queuing is a t e r m applied to a set of a l g o r i t h m s t h a t a t t e m p t to allocate b a n d w i d t h fairly a m o n g all flows w i t h i n a c e r t a i n class. A flow m a y b e defined, for example, as data flowing b e t w e e n a source-destination pair, a n d t h e class as a collection of s u c h flows. A fair b a n d w i d t h allocation m i g h t t h e n b e defined as o n e in w h i c h low-volume flows w i t h i n t h e class get t h e i r e n t i r e r e q u e s t e d allocation, w h e r e a s h i g h - v o l u m e flows s h a r e t h e r e m a i n i n g b a n d w i d t h equally. S o m e q u e u i n g a l g o r i t h m s are a n a l y z e d in C h a p t e r 8. T h e QpS e x p e r i e n c e d b y a flow d e p e n d s o n t h e s c h e d u l i n g a l g o r i t h m in all t h e s w i t c h e s in its route. Since t h e IP l a y e r forwards individual datagrams, p a c k e t s w i t h i n a flow c a n n o t b e assigned a fixed r o u t e a n d so it is difficult to g u a r a n t e e QpS w i t h i n a n IP n e t w o r k . O n e advantage of circuit- a n d virtual circuit-switched n e t w o r k s is t h a t t h e y assign r o u t e s to flows a n d c a n b e t t e r provide QpS g u a r a n t e e s . IP switching m a y also provide t h o s e g u a r a n t e e s . Q u e u i n g algorithms t h a t allocate b a n d w i d t h a m o n g different q u e u e s c a n b e a u g m e n t e d b y active queue management, in w h i c h p a c k e t s m a y b e selectively discarded. Such a m e c h a n i s m m a y d e l e t e a l o w e r priority p a c k e t from a q u e u e w h e n a h i g h e r priority arrives a n d t h e q u e u e is full.

4.6.5

^. _ 191

^^

Label Switching With i n c r e a s i n g link speeds, limits o n r o u t e r t h r o u g h p u t b e g i n to c o n s t r a i n n e t w o r k growth. Routers are slower t h a n l a y e r 2 ( E t h e r n e t , ATM) switches, b e c a u s e t h e look-up of large r o u t i n g tables is t i m e - c o n s u m i n g . O n e r e c e n t a d v a n c e s e e k s to b r i n g to r o u t e r s t h e s p e e d of l a y e r 2 switches for r o u t i n e

192

The Internet and TCP/IP Networks

p a c k e t forwarding, while m a i n t a i n i n g t h e capability t h a t r o u t e r s provide in t e r m s of protocols, packet processing, a n d security. T h i s involves a t e c h n i q u e called label switching. T h e k e y idea to label switching is t h a t t h e i n c o m i n g p a c k e t carries a label t h a t is directly t r a n s l a t e d b y t h e switch h a r d w a r e into t h e n u m b e r of t h e switch's outgoing port or interface. I m p l e m e n t i n g this idea in r o u t e r s involves two c o m p o n e n t s . First, p a c k e t s w i t h i n a particular flow m u s t b e b o u n d to a label t h a t is quickly t r a n s l a t e d (in h a r d w a r e or software) into t h e r o u t e r ' s outgoing layer 2 interface. Second, a n algorithm m u s t distribute labels a m o n g t h e r o u t e r s so t h a t a c o n s i s t e n t r o u t e is assigned to t h e flow. (It w o u l d b e b a d if, for instance, t h e r o u t e c r e a t e d loops.) By b i n d i n g a flow to a label, a n d a label to a route, all packets in t h e flow follow t h e s a m e route. T h u s , this t e c h n i q u e achieves s o m e of t h e benefits of virtual-circuit switching. In particular, it provides t h e o p p o r t u n i t y to g u a r a n t e e QpS to a flow. We give a highly abbreviated description of two i m p l e m e n t a t i o n s . O n e is called IP switching (RFC 1953); t h e o t h e r is tag switching (RFC 2105) w i t h a vari­ ation called multiprotocol label switching (MPLS, RFC 2547). Either t e c h n o l o g y c a n b e d e p l o y e d w i t h i n a s u b n e t w o r k of a p p r o p r i a t e l y e q u i p p e d r o u t e r s . IP switching r e q u i r e s a n u n d e r l y i n g ATM link layer, so t h e label a t t a c h e d to a particular flow c o r r e s p o n d s to a n ATM virtual c i r c u i t / p a t h identifier (VCI/VPI) (see C h a p t e r 6). T h e flow classification is b a s e d o n a T C P / U D P port pair (all p a c k e t s w i t h t h e s a m e port pair c o m p r i s e t h e s a m e flow) or, w i t h a greater level of aggregation, o n a source-destination h o s t pair. A label distri­ b u t i o n protocol is i n v o k e d e a c h t i m e a n e w flow is identified. Because t h e r e is significant o v e r h e a d a t t a c h e d to t h e protocol, it is c o u n t e r p r o d u c t i v e to identify short-lived flows. Routers e m p l o y a h e u r i s t i c to d e t e r m i n e if a p o t e n t i a l flow will b e long-lived. T h e r o u t e r c o u n t s t h e n u m b e r of p a c k e t s b e l o n g i n g to a po­ tential flow t h a t arrive w i t h i n a short t i m e . If t h a t n u m b e r exceeds a threshold, t h e flow is identified a n d t h e r o u t e r invokes t h e label distribution protocol. T h e protocol distributes t h e flow-label b i n d i n g from r o u t e r to n e i g h b o r i n g router. If t h e n u m b e r is b e l o w t h e threshold, t h e r o u t e r forwards t h e IP p a c k e t s in t h e n o r m a l m a n n e r . T h e b i n d i n g b e t w e e n a flow a n d its label is valid for a short t i m e (60 seconds, say), so t h a t t h e VCI/VPI c a n b e assigned to a n o t h e r flow. In tag switching, a short fixed-length tag is a t t a c h e d to p a c k e t s in a flow in a w a y t h a t d e p e n d s o n t h e u n d e r l y i n g link l a y e r p a c k e t format. T h e link l a y e r n e e d n o t b e ATM. Routers at t h e e n t r a n c e to t h e tag-switched s u b n e t w o r k insert t h e tag, a n d r o u t e r s at t h e exit r e m o v e t h e tag. Flow classification is m o r e flexible t h a n in IP switching. (For example, a m u l t i c a s t flow c a n b e assigned a tag, b a s e d o n its class D address.) Like t h e VCI/VPI labels u s e d in IP switching, tags are local to links, so w h e n a p a c k e t leaves a router, it m a y

4.6

Performance of TCP/IP Networks

acquire a n e w tag. T h e tag to o u t p u t interface b i n d i n g is p l a c e d in a table. I k b l e access is faster t h a n w i t h r o u t i n g tables b e c a u s e t h e r e are fewer, fixed-length entries, w h i c h c o n t r i b u t e s to t h e g r e a t e r t h r o u g h p u t of tag-switched routers. Unlike in IP-switching, flow classification a n d tag distribution are carried o u t b y r o u t e r protocols similar to o t h e r i n t e r i o r g a t e w a y protocols. Also, u n l i k e in IP-switching, t h e tag a s s i g n m e n t s m a y b e p e r m a n e n t .

4.6.6

Suggested Improvements for Other Protocols RSVP RSVP (Resource Reservation Protocol) is a r e s o u r c e r e s e r v a t i o n s e t u p protocol d e s i g n e d for multicast, m u l t i m e d i a data s t r e a m s or flows (RFC 2205). A flow is specified b y attributes s u c h as source-destination pair, m e a n data rate, latency, a n d quality of service (QpS) (RFC 1363). For example, a n MPEG video s t r e a m t h a t averages 3 to 7 Mbps a n d p e a k s to 12 Mbps at 30 frames p e r second, r e q u i r e s certain QoS for a d e q u a t e r e c e p t i o n . A n application signals to t h e n e t w o r k its QoS r e q u i r e m e n t , a n d t h e protocol r e s e r v e s r e s o u r c e s in t h e n e t w o r k switches. A source s e n d s a m e s s a g e (called path message) t h a t gives t h e m u l t i c a s t d e s t i n a t i o n address, t h e flow specification, a n d a t e m p l a t e for identifying w h i c h p a c k e t s b e l o n g to t h e flow. Receivers j o i n a m u l t i c a s t t r e e b y s e n d i n g m e s s a g e s (called reservation messages) w i t h a description of t h e QoS t h e y expect. T h e r o u t e r s forward u p s t r e a m a least u p p e r b o u n d of t h e QoS r e q u i r e m e n t s m a d e d o w n s t r e a m . T h u s , if a r o u t e r gets a r e q u e s t from a d o w n s t r e a m n o d e for a 200-ms delay version of a s t r e a m a n d a n o t h e r from a different n o d e for a 100-ms d e l a y version, t h e r o u t e r forwards t h e 100-ms d e l a y r e q u e s t u p s t r e a m . ( T h e precise m e a n i n g is a d m i t t e d l y fuzzy.) O n c e t h e r o u t e r gets t h e 100-ms d e l a y version from u p s t r e a m , it s e n d s it to t h e d o w n s t r e a m n o d e s . A similar idea is u s e d for t h r o u g h p u t r e q u e s t s . If a r o u t e r does n o t find e n o u g h s p a r e r e s o u r c e s to c a r r y t h e s t r e a m w i t h t h e r e q u e s t e d QoS, it s e n d s a reject m e s s a g e to t h e receiver. P r e s u m a b l y , a switch a l g o r i t h m u s e s t h e s e r e q u e s t s for QoS of different s t r e a m s to d e t e r m i n e t h e r e s o u r c e s t h a t s h o u l d b e r e s e r v e d a n d h o w to s c h e d u l e t h e t r a n s m i s s i o n s of packets. Tb a d a p t to changes, RSVP specifies t h a t s o u r c e s periodically s e n d m e s ­ sages t h a t describe t h e i r s t r e a m s a n d receivers periodically s e n d r e s e r v a t i o n r e q u e s t s . T h e d o w n s t r e a m m e s s a g e s s e n t b y t h e source are r o u t e d b y multi­ cast IP. T h e r e s e r v a t i o n m e s s a g e s are s e n t u p s t r e a m along t h e r e v e r s e d tree.

193

194

The Internet and TCP/IP Networks

All t h e s e m e s s a g e s c a r r y a t i m e o u t value t h a t t h e n o d e s u s e to set t i m e r s a n d to delete t h e c o r r e s p o n d i n g information w h e n t h e t i m e r s expire. RSVP m e c h a n i s m s n e e d e d to i m p l e m e n t QpS r e q u e s t s i n c l u d e (1) a packet classifier t h a t d e t e r m i n e s t h e QoS class of e a c h p a c k e t in a c c o r d a n c e w i t h t h e reservations t h a t are m a d e , (2) admission control t h a t d e t e r m i n e s w h e t h e r t h e n o d e h a s sufficient r e s o u r c e s to provide a r e q u e s t e d QoS, a n d (3) a packet scheduler link-layer mechanism t h a t achieves t h e r e q u e s t e d QoS u s i n g o n e of several q u e u i n g algorithms. Real-time protocol or RTP provides unreliable, end-to-end delivery services for unicast or multicast data w i t h real-time characteristics, s u c h as interactive audio a n d video (RFC 1889). T h o s e services i n c l u d e p a y l o a d t y p e identifi­ cation, s e q u e n c e n u m b e r i n g , t i m e stamping, a n d delivery m o n i t o r i n g . Ap­ plications m a y r u n RTP o n top of UDP to m a k e u s e of its m u l t i p l e x i n g a n d error-indication services. T h e s e q u e n c e n u m b e r s i n c l u d e d in RTP allow t h e receiver to r e c o n s t r u c t t h e s e n d e r ' s p a c k e t s e q u e n c e , b u t s e q u e n c e n u m b e r s m i g h t also b e u s e d to d e t e r m i n e t h e p r o p e r location of a packet, for e x a m p l e in video decoding, w i t h o u t necessarily decoding p a c k e t s in s e q u e n c e . Applica­ tions that r u n o n top of RTP i n c l u d e VIC, a real-time, m u l t i m e d i a application for video c o n f e r e n c i n g over t h e I n t e r n e t , a n d VAT, a n audio c o n f e r e n c i n g application. T h e w i d e s p r e a d u s e of t h e H T T P / 1 . 0 protocol over TCP h a s c r e a t e d prob­ l e m s . Typical Web pages today c o n t a i n a n HTML d o c u m e n t a n d m a n y small e m b e d d e d images. Since H T T P / 1 . 0 o p e n s a n d closes a n e w TCP c o n n e c t i o n for e a c h operation, this practice m e a n s a high fraction of p a c k e t s are s i m p l y TCP control packets u s e d to o p e n a n d close a c o n n e c t i o n . T h e creation of m a n y par­ allel, short-lived TCP c o n n e c t i o n s , a n d t h e i r i n t e r a c t i o n w i t h TCP congestion control algorithms, leads to poor p e r f o r m a n c e a n d congestion. We s u m m a r i z e several proposals to a m e l i o r a t e t h e s e p r o b l e m s . Transactional TCP or T / T C P (RFC 1644) is d e s i g n e d for c l i e n t / s e r v e r in­ teraction in w h i c h a r e q u e s t is followed b y a single r e s p o n s e . T h e client a n d server cache a c o n n e c t i o n c o u n t n u m b e r (CC). A n e w SYN TCP client r e q u e s t packet i n c l u d e s t h e n e w c o n n e c t i o n c o u n t a n d data ( r e q u e s t ) . T h e n e w CC is larger t h a n t h e c a c h e d CC, a n d so t h e server k n o w s t h a t this is a n e w r e q u e s t a n d not a n old, d e l a y e d packet. T h e server's reply p a c k e t i n c l u d e s acknowl­ e d g m e n t of t h e n e w CC a n d data ( r e s p o n s e ) . In this way, T / T C P b y p a s s e s t h e l a t e n c y of t h e t h r e e - w a y h a n d s h a k e . H T T P / 1 . 1 (RFC 2068) o v e r c o m e s s o m e of t h e p r o b l e m s of H T T P / 1 . 0 b y o p e n i n g a p e r s i s t e n t TCP c o n n e c t i o n a n d p i p e l i n i n g r e q u e s t s w i t h o u t waiting for r e s p o n s e s . T h e r e s p o n s e s are still serialized, however, w h i c h d o e s n o t a d e q u a t e l y s u p p o r t s i m u l t a n e o u s r e n d e r i n g of i n l i n e d objects. A n I n t e r n e t -

4.6

Performance of TCP/IP Networks

Draft proposal, WebMUX, allows several protocols to m u l t i p l e x so t h a t m u l t i p l e objects from a Web s e r v e r c a n b e s i m u l t a n e o u s l y fetched over a single TCP C o n n e c t i o n . In this way, m e t a d a t a to objects c a n b e s e n t to clients w i t h o u t o t h e r m e t a d a t a waiting for t h e rest of t h e first object r e q u e s t e d . H T T P / 1 . 1 a n d WebMUX are session m a n a g e m e n t protocols t h a t leave u n c h a n g e d t h e u n d e r l y i n g t r a n s p o r t layer, TCP. By contrast, t h e idea b e h i n d application-level framing or ALF is t h a t applica­ tions s h o u l d b e involved in t h e data t r a n s m i s s i o n process, b e c a u s e applications k n o w t h e r e q u i r e m e n t s of t h e i n f o r m a t i o n b e i n g t r a n s m i t t e d as well as w h a t to do w h e n information is lost, m i s o r d e r e d , or delayed. T h u s i n f o r m a t i o n s h o u l d b e packetized into application data u n i t s or ADUs, w h i c h s h o u l d b e t r a n s m i t ­ t e d as i n d e p e n d e n t e l e m e n t s (instead of as a c o n t i n u o u s b y t e s t r e a m as TCP a s s u m e s ) . T h e application s h o u l d t h e n d e t e r m i n e w h e t h e r ADUs s h o u l d b e r e t r a n s m i t t e d , discarded, or ordered. All t h e s e proposals modify end-to-end t r a n s p o r t protocols or a t t e m p t to b y p a s s t h e b y t e s t r e a m - o r i e n t e d s e m a n t i c s of TCP to b e t t e r m a t c h t h e re­ q u i r e m e n t s of applications. However, t h e y leave u n c h a n g e d t h e IP service of point-to-point, unreliable p a c k e t transfer. T h i s m e a n s t h a t applications c a n n o t b e given a n y QpS g u a r a n t e e involving end-to-end p e r f o r m a n c e s u c h as b o u n d s o n delay, delay jitter, or b a n d w i d t h . T h e Integrated Services or IS m o d e l is a p r o p o s e d e x t e n s i o n to t h e I n t e r n e t a r c h i t e c t u r e a n d protocols to provide i n t e g r a t e d services t h a t s u p p o r t real-time as well as t h e c u r r e n t non-real-time service of IP (RFC 1633). T h e service m o d e l considers t h e n e e d s of applications a n d t h o s e of n e t w o r k operators. T h e m o d e l provides for two categories of QoS for real-time applications. Guaranteed service is characterized b y a perfectly reliable or worst-case u p p e r b o u n d o n delay. Controlled-load service (RFC 2711) supplies a QoS c o m p a r a b l e to a lightly l o a d e d best-effort service. T h e c o n c e r n s of n e t w o r k o p e r a t o r s are a d d r e s s e d b y controlled link-sharing, w h i c h g u a r a n t e e s a specified b a n d w i d t h to certain classes of flows. T h e r e f e r e n c e i m p l e m e n t a t i o n f r a m e w o r k for t h e IS m o d e l i n c l u d e s m e c h ­ a n i s m s for r e s o u r c e reservation, q u e u e m a n a g e m e n t , a n d a d m i s s i o n control. I b i m p l e m e n t t h e IS m o d e l will r e q u i r e flows to b e b o u n d to routes, a n d t h e state of t h e flow to b e m a i n t a i n e d at t h e different r o u t e r s along t h e route. T h e RSVP protocol a p p e a r s to provide t h e s e m e c h a n i s m s . T h e i n t r o d u c t i o n of ATM n e t w o r k s in t h e I n t e r n e t b a c k b o n e provides a n i m p e t u s for i m p l e m e n t a t i o n of t h e IS m o d e l a n d t h e RSVP protocol (RFC 2382). Some of t h e m e c h a n i s m s m e n t i o n e d above m a y b e i m p l e m e n t a b l e b y policy-based routing. I n s t e a d of r o u t i n g solely b y d e s t i n a t i o n address, r o u t e r s i m p l e m e n t policies to allow or d e n y p a t h s b a s e d o n i d e n t i t y of t h e e n d s y s t e m ,

195

The Internet and TCP/IP Networks

196

protocol, a n d size of packets. T h e r o u t e r tags t h e p a c k e t s b y setting t h e IP p r e c e d e n c e or TOS values of t h e IP header. Load balancing is a g e n e r a l m e t h o d to i m p r o v e t h e s p e e d of Web servers. T h e idea is to u s e a n u m b e r of servers t h a t c o n t a i n t h e files. T h e URL r e q u e s t s arrive at a c o m p u t e r t h a t redirects t h e m to o n e of t h e servers. T h i s d i s p a t c h e r k e e p s track of t h e load of t h e servers a n d s e n d s t h e r e q u e s t s to t h e least l o a d e d one. This m e t h o d c a n also b e u s e d w h e n t h e s e r v e r s do n o t all c o n t a i n all t h e files b u t h a v e i n s t e a d a l i m i t e d a m o u n t of replication. T h e s e parallel s e r v e r s can e v e n b e m a d e adaptive so t h a t t h e y c a n decide w h e n it is useful to replicate information to increase t h e t h r o u g h p u t of t h e s y s t e m . Translators, c o m p u t e r s t h a t modify t h e format of files, c a n b e useful in specialized applications. For instance, consider t h e u s e r of a wireless p a l m t o p device w h o r e q u e s t s a large image from a server. Using a translator, t h e r e q u e s t goes to a translator t h a t gets t h e i m a g e a n d c o m p r e s s e s it before s e n d i n g it to t h e wireless device. M a n y s u c h t r a n s l a t i o n s are possible. For instance, e-mail c a n b e t r a n s l a t e d into a n audio clip to b e accessible from a t e l e p h o n e ; voice mail c a n b e t r a n s l a t e d into text; video clips c a n b e r e d u c e d to a few still images; a n d so on. Telephony servers c a n b e built to provide t h e u s u a l t e l e p h o n e services for voice over IP. T h e s e services i n c l u d e c o n f e r e n c e calls, call forwarding, call back, as well as 800 services t h a t r e r o u t e t h e calls d e p e n d i n g o n t h e source a n d o n t h e t i m e of day, a n d m a n y others. Caches are s i m p l e devices t h a t c a n i m p r o v e t h e p e r f o r m a n c e of a n e t w o r k . W h e n a c a c h e is used, t h e Web r e q u e s t s are i n t e r c e p t e d b y a c a c h e t h a t stores files previously r e q u e s t e d . If t h e c a c h e c o n t a i n s t h e n e w request, t h e n it c a n reply directly. If not, t h e c a c h e forwards t h e r e q u e s t to t h e URL a n d copies t h e reply. W h e n t h e cache gets full, it c a n u s e a n y s t a n d a r d r e p l a c e m e n t a l g o r i t h m to m a k e r o o m for a n e w r e q u e s t . For instance, t h e c a c h e c a n delete t h e least r e c e n t l y u s e d files. Typically, c a c h e e n t r i e s h a v e a t i m e to live after w h i c h t h e y are deleted. O n e c a n i m a g i n e a n e t w o r k of c a c h e s t h a t exchange lists of links of w h i c h t h e y h a v e copies. Such a n e t w o r k could t h e n b e o p t i m i z e d to m i n i m i z e t h e r e s p o n s e t i m e a n d t h e traffic o n t h e Web.

4.7

SUMMARY T h e I n t e r n e t is u n d e r g o i n g a n explosive g r o w t h in t h e n u m b e r of n o d e s , users, traffic, a n d applications. This g r o w t h is m a d e possible b y t h e simplicity of t h e d a t a g r a m service of IP a n d b y t h e hierachical n a m i n g a n d a d d r e s s i n g s c h e m e s .

However, t h e choice of t r a n s p o r t i n g d a t a g r a m s as t h e b a s i c service, w h i c h m a y b e t h e I n t e r n e t ' s m o s t i m p o r t a n t feature, m a y also t u r n o u t in t h e long r u n to limit its g r o w t h to applications t h a t do n o t r e q u i r e p e r f o r m a n c e g u a r a n t e e s in t e r m s of d e l a y or b a n d w i d t h . ( T h e p o t e n t i a l for s u c h applica­ tions, however, is e n o r m o u s as t h e g r o w t h of WWW suggests. T h e J u n e 1998 C o m m e r c e N e t / N i e l s e n s u r v e y found t h a t t h e n u m b e r of I n t e r n e t u s e r s in t h e U n i t e d States a n d C a n a d a 16 y e a r s of age or older r e a c h e d 79 million, u p from 58 million eight m o n t h s earlier.) It is e a s y to design n e w I n t e r n e t protocols a n d r o u t e r s to m e e t t h e n e e d s of t h e s e applications. T h e t e c h n i c a l challenge is to do t h a t in s u c h a w a y t h a t t h e n e w protocols a n d r o u t e r s c a n coexist w i t h t h e I n t e r n e t ' s vast legacy. If t h a t c h a l l e n g e c a n b e successfully m e t , t h e I n t e r n e t could claim to provide u n i v e r s a l n e t w o r k i n g . At t h e s a m e time, a d v a n c e s in circuit-switched n e t w o r k s are e n o r m o u s l y e x t e n d i n g t h e i r service capability. We s t u d y t h o s e n e t w o r k s in t h e n e x t chapter.

4.8

NOTES A n i n t e r p r e t i v e a c c o u n t of I n t e r n e t h i s t o r y b y s o m e of its founders is given in [LC + 97]. Details a b o u t t h e history a n d g r o w t h statistics are available from t h e I n t e r n e t Society Web site at http://www.isoc.org/internet/historyVindex.html. Most T C P / I P d o c u m e n t s a n d I n t e r n e t s t a n d a r d s are p u b l i s h e d in a series called Request for C o m m e n t or RFC. T h o s e d o c u m e n t s are available on-line at t h e URL http://www.rfc-editor.org/. T h e d o c u m e n t s are cited as RFC #### w h e r e #### is t h e RFC n u m b e r . URL is specified in RFC 1738, a n d a generalization of it called Universal Reference Identifier (URI) is specified in RFC 2396. T h e Routing I n f o r m a t i o n Protocol or RIP, b a s e d o n t h e Bellman-Ford Algo­ r i t h m is described in RFC 1058. Its e x t e n s i o n to m u l t i c a s t is d e s c r i b e d in RFC 1075. Information a b o u t MBone c a n b e o b t a i n e d at http://www.mhone.com. A s i m u l a t i o n s t u d y of IP switching is r e p o r t e d in [LN97]. T C P / I P protocols are discussed in [T88, W98, C95]. T h e ideas b e h i n d TCP w e r e first p u b l i s h e d in [CK74]. For details a b o u t t h e software i m p l e m e n t a t i o n s of T C P / I P in UNIX, see [C95]. T h e RSVP protocol is also d e s c r i b e d in [ZDESZ93]. VIC is available at www-nrg.ee.lhl.gov/vie, a n d VAT is available at www-nrg.ee.lbl.gov/vat. ALF w a s i n t r o d u c e d in [C1T90], a n d a n i m p l e m e n t a t i o n is d e s c r i b e d in [CKD98]. A proposal to u s e t h e ALF idea for t h e Web is discussed in [GCMW99].

The Internet and TCP/IP Networks

198

A c o m p a r i s o n of IP-based a n d ATM-based service m o d e l s t h a t provide s o m e real-time g u a r a n t e e s is carried out in [CWSA95]. XTP is discussed in [X92], a n d VMTP is described in [CW89]. T h e H T T P / 1 . 1 protocol is specified in RFC 2068.

4.9

PROBLEMS 1. Show t h a t t h e shortest-path-first a l g o r i t h m of Figure 4.8 p r o d u c e s a span­ n i n g t r e e w i t h t h e shortest length, w h e r e t h e l e n g t h of a t r e e is defined as t h e s u m of t h e l e n g t h s of all its links. 2. Define t h e t h r e e - w a y h a n d s h a k e of t h e TCP initialization protocol i n m o r e detail, u s i n g t i m e o u t s . Explain h o w it avoids m i s u n d e r s t a n d i n g s c a u s e d b y a d e l a y e d packet. 3. Propose a TCP w i n d o w flow-control s c h e m e t h a t r e d u c e s t h e w i n d o w size w h e n congestion is h i g h a n d i n c r e a s e s it w h e n c o n g e s t i o n is low. Also p r o p o s e a m e t h o d to m e a s u r e congestion. Propose a s i m p l e m a t h e m a t i c a l m o d e l t h a t c a n b e u s e d to analyze t h e p e r f o r m a n c e of y o u r s c h e m e . 4. A collection of n e t w o r k s a n d s u b n e t w o r k s is s h o w n in t h e figure below. All t h e links are E t h e r n e t links. T h e lower case a d d r e s s e s are MAC addresses, a n d t h e u p p e r case a d d r e s s e s are IP a d d r e s s e s in t h e n e t . s u b n e t notation. For instance, r21 is t h e E t h e r n e t MAC a d d r e s s of t h a t interface of R2.

4.9

Problems

199 (a) Fill in t h e following tables: ARP table ofr22:

Routing table of R2: Address Nl.l C D N2

Next Hop

IP

Port

Rl C D R3

r22 r21 r23 r24

Rl

MAC rll

(b) Specify t h e a d d r e s s e s in t h e E t h e r n e t a n d IP h e a d e r s of a p a c k e t from A to Η as it travels from R2 to R3: Ethernet Addresses: (To, From) r31,r24

H, A

\ TP Addresses: (To, From)

5. R a n d o m Early D e t e c t i o n (RED) (a) (b) (c) (d)

r e d u c e s t h e c h a n c e s of successive losses of a c o n n e c t i o n . modifies t h e r e t r a n s m i s s i o n m e c h a n i s m of TCP. modifies t h e w i n d o w a d j u s t m e n t a l g o r i t h m of TCP. r e d u c e s t h e b i a s of TCP in favor of c o n n e c t i o n s w i t h a s h o r t round-trip time.

6. C o n s i d e r t h e n e t w o r k below. Each switch i m p l e m e n t s a r o u n d - r o b i n ser­ vice of its t h r e e q u e u e s . T h a t is, t h e switch serves o n e cell from q u e u e 1, t h e n o n e cell from q u e u e 2, t h e n o n e cell from q u e u e 3, a n d r e p e a t s this cycle forever. It m a y h a p p e n t h a t a q u e u e overflows if it is n o t s e r v e d fast e n o u g h . T h i s m e c h a n i s m p r o t e c t s s t r e a m s aginst m i s b e h a v i n g c o n n e c ­ tions. T h e switch skips e m p t y q u e u e s . A packets/s •

Ql

0.5x10 packets/s •

Q2

0 . 3 x l 0 packets/s •

Q3

1 0 packets/s 6

6

6

Calculate t h e m a x i m u m value of A in t h e figure so t h a t n o cell from t h a t s t r e a m is lost b e c a u s e of buffer overflow.

The Internet and TCP/IP Networks

200

7. A s s u m e t h a t t h e I n t e r n e t is m a d e u p of o n l y class Β n e t w o r k s (recall t h a t t h e r e are u p to 2 class Β n e t w o r k s ) . We further simplify b y ignoring s u b n e t t i n g a n d a s s u m i n g t h a t t h e r e is o n e r o u t e r p e r class Β n e t w o r k . For simplicity, w e a s s u m e t h a t all t h e local a d d r e s s e s are E t h e r n e t addresses. 1 4

(a) What is t h e size (in b y t e s ) of e a c h e n t r y of a r o u t e r table a n d of a n ARP table? A n e n t r y of a r o u t i n g table is (Net, IP a d d r e s s of n e x t router, p o r t n u m b e r ) . A s s u m e t h a t t h e port n u m b e r takes o n e b y t e . A n e n t r y of a n ARP table is (Host part of local address, E t h e r n e t address). [Note: Round off to a n i n t e g e r n u m b e r of bytes.] (b) Estimate t h e m a x i m u m size (in kilobytes) of t h e r o u t i n g table of e a c h r o u t e r a n d of t h e ARP tables of e a c h router. 8. I m a g i n e a perfect n a m i n g s y s t e m w h e r e t h e root h a s Ν d o m a i n s , e a c h domain has Ν subdomains, each subdomain has Ν sub-subdomains, and e a c h s u b - s u b d o m a i n h a s Ν n a m e s . Say t h a t t h e r e are 100 million n a m e s . Estimate t h e n u m b e r of e n t r i e s of e a c h n a m e server table. 9. T h e I n t e r n e t Protocol provides (a) (b) (c) (d)

reliable p a c k e t delivery b e t w e e n I n t e r n e t hosts. unreliable p a c k e t delivery b e t w e e n I n t e r n e t hosts. reliable p a c k e t delivery b e t w e e n h o s t s o n t h e s a m e LAN or Link. unreliable p a c k e t delivery b e t w e e n h o s t s o n t h e s a m e LAN or Link.

10. Consider t h e n e t w o r k s h o w n in Figure 4.14. Tb simplify, w e u s e t h e classb a s e d addressing Net.Host, w i t h n o s u b n e t t i n g a n d n o CIDR. (a) Fill in t h e routing tables a n d t h e ARP tables. A r o u t i n g e n t r y is (Net, IP address of next router, port n u m b e r ) . A n e n t r y for a p o r t ARP table is (IP address, local address). (b) Fill in t h e h e a d e r s of t h e p a c k e t s 1, 2, 3, 4, a n d 5 for a t r a n s m i s s i o n f r o m N 1 . 3 to N2.3. 11. T h e effect of congestion o n TCP is to (select o n e or m o r e a n s w e r s ) (a) increase t h e loss rate s e e n b y applications. (b) limit t h e n u m b e r of c o n n e c t i o n s t h a t c a n b e set u p at o n e t i m e . (c) r e d u c e t h e t h r o u g h p u t of c o n n e c t i o n s . 12. T h e TCP state d i a g r a m is s h o w n in Figure 4.15 (page 202). T h e states are n u m b e r e d o n t h e diagram. C o n s i d e r two hosts: A a n d B. We indicate b y (ΑΙ, B3) t h e situation w h e n TCP is in state 1 in h o s t A a n d in state 3 in h o s t B. More generally, w e u s e t h e notation (A;, B ). A s s u m e t h e following scenario. Host A o p e n s a TCP c o n n e c t i o n to h o s t B. This start of c o n n e c t i o n is n o t r e c e i v e d b y h o s t B. Host A tries again. ;

4.9

Problems

201

Routing table

ARP table of Port 1

Headers of packets (use convention [destination I source I ...]): 1 2 3 4 5

4.14

Network addressing and routing.

FIGURE

T h i s time, h o s t Β gets t h e m e s s a g e a n d a c k n o w l e d g e s . H o s t A t h e n s e n d s p a c k e t s to B, w h o a c k n o w l e d g e s t h e m . H o s t A t h e n closes t h e c o n n e c t i o n from A to B, a n d Β a c k n o w l e d g e s . Host Β t h e n closes t h e c o n n e c t i o n , a n d A a c k n o w l e d g e s t h e close b u t its a c k n o w l e d g m e n t does n o t r e a c h B. Host Β t h e n r e p e a t s its close m e s s a g e , w h i c h this t i m e is correctly a c k n o w l e d g e d . Indicate t h e list of pairs of states of t h e TCP protocols in t h e t w o hosts. For instance, y o u r a n s w e r m i g h t look like (Al, B l ) , (A2, B l ) , (A2, B4), (A3, B4), (A3, B5). 13. Consider a r o u t e r w i t h a single FIFO q u e u e . A r a n d o m n u m b e r X of TCP c o n n e c t i o n s go t h r o u g h t h e router. E a c h c o n n e c t i o n s e n d s b u r s t s of p a c k e t s according to Go Back Ν protocol. Model t h e q u e u e as a discrete t i m e q u e u e w i t h b a t c h arrivals. A s s u m e t h a t X r e m a i n s c o n s t a n t d u r i n g a typical b u s y cycle of t h e r o u t e r q u e u e . T h e b a t c h A(n) is t h e s u m of t h e b a t c h e s t h a t t h e

The Internet and TCP/IP Networks

202 Notation: Input/Output

Anything/Reset [1]

Begin •

Closed Passive open (

SYN/SYN+ACK

) Close \

Active Open/SYN

Listen

Close/ Timeout/ Reset

[2] Send/SYN [3]

SYN received

[4] SYN sent

SYN/SYN+ACK SYN+ACK/ACK

Close/ FIN

Established [5]

[8]

FIN Wait-1

Closing

Close/FIN [9] [11]

[10] 4.15

FIN Wait-2

FIN/ACK

Timed Wait

Close Wait

FIN/ACK

ACK

ACK/

[6]

[7]

ACK

Last ACK

Timeout (2 lifetimes)

TCP State Diagram.

X c o n n e c t i o n s p r o d u c e . At t h e t i m e scale of a b u s y period, given X, m o d e l t h e A(n) as t h e s u m of X i n d e p e n d e n t , identically distributed (iid) r a n d o m variables B(l), . . . , B(X) a n d t h e successive A(n) are iid given X. (a) Analyze t h e average delay t h r o u g h t h e q u e u e a n d t h e average q u e u e size in t e r m s of t h e distribution of B(l) a n d X. (b) C o m m e n t o n t h e m o d e l a n d o n t h e results of t h e analysis. 14. T h e proposals for achieving a satisfactory end-to-end quality w i t h scalable protocols for t h e I n t e r n e t a p p e a r to b e b a s e d o n t h e intuition that, as t h e n u m b e r of c o n n e c t i o n s increases, t h e relative variability of t h e QpS gets smaller. Using this intuition, I n t e r n e t r e s e a r c h e r s p r o p o s e to control QoS b y using call a d m i s s i o n control a n d r e s o u r c e allocation in t h e access n e t w o r k (e.g., controlled load, i n t e g r a t e d services, RSVP) a n d to u s e a s i m p l e CoS in t h e b a c k b o n e (e.g., DiffServ). Propose a m o d e l a n d analyze it to confirm or contradict this intuition. 15. Indicate t h e successive steps of t h e OSPF a l g o r i t h m (Dijkstra) to find t h e shortest p a t h s from t h e source S to all t h e o t h e r n o d e s . All t h e links h a v e l e n g t h 1. For e a c h step, label t h e nodes, m a r k t h e n o d e s t h a t s h o u l d b e m a r k e d , a n d select t h e a p p r o p r i a t e links. Break ties b y exploring n o d e s

4.9

Problems

203 w i t h s a m e smallest label in t h e o r d e r from left to right a n d from top to bottom. inf

inf

16. A source s e n d s a p a c k e t to Ν subscribers. For i = 1 , . . . , N, s u b s c r i b e r i gets t h e p a c k e t w i t h probability 1 — p(i), i n d e p e n d e n t l y of o n e another. (a) H o w m a n y t i m e s m u s t t h e source s e n d t h e p a c k e t u n t i l e v e r y sub­ scriber gets it? (b) N o w a s s u m e t h a t t h e m u l t i c a s t occurs along a tree. T h e losses are i n d e p e n d e n t o n e a c h b r a n c h of t h e tree. We c a n define t h e probability 1 - p(i) t h a t subscriber i gets t h e packet. However, t h e losses are n o l o n g e r i n d e p e n d e n t . Can y o u d e t e r m i n e w h e t h e r t h e n u m b e r of t i m e s t h a t t h e source m u s t s e n d t h e p a c k e t is larger t h a n in t h e p r e v i o u s case or smaller? 17. W h a t is t h e effect of t r a n s m i s s i o n e r r o r s o n t h e t h r o u g h p u t of TCP? Propose a m o d e l a n d s o m e analysis.

CircuitSwitched Networks

packet-switched n e t w o r k s t h a t w e s t u d i e d in C h a p t e r s 3 a n d 4 are well suited for m e s s a g e traffic. Virtual circuit-switched n e t w o r k s , d e s i g n e d for variable bit rate traffic, are d i s c u s s e d in C h a p t e r 6. C o n s t a n t bit r a t e traffic is b e t t e r t r a n s p o r t e d b y circuit-switched n e t w o r k s , w h i c h w e s t u d y in this chapter. T h e m o s t i m p o r t a n t circuit-switched n e t w o r k s are t h e t e l e p h o n e , cable TV or CATV, a n d s o m e cellular t e l e p h o n e a n d satellite n e t w o r k s . We discuss o n l y t h e t e l e p h o n e a n d CATV n e t w o r k s . T h i s c h a p t e r p r e s e n t s t h e u n d e r l y i n g c o n c e p t s a n d t h e t e c h n o l o g i e s de­ p l o y e d in t h e 1980s a n d 1990s t h a t c h a n g e d t h e t e l e p h o n e s y s t e m from a voice-only n e t w o r k to o n e t h a t s u p p o r t s all t y p e s of traffic. You will b e c o m e familiar w i t h t h e m o s t i m p o r t a n t t e c h n o l o g i e s : SONET, DWDM, optical local loop, digital subscriber line solutions i n c l u d i n g ISDN a n d xDSL, a n d INA. You will also l e a r n h o w CATV s y s t e m s are c h a n g i n g from a o n e - w a y video distri­ b u t i o n s y s t e m into a n e t w o r k t h a t also offers c o m m u n i c a t i o n in t h e u p s t r e a m direction. T h e t e l e p h o n e a n d CATV n e t w o r k s , w h i c h u n t i l 1990 s e r v e d sepa­ rate m a r k e t s , h a v e since b e g u n to c o m p e t e a n d collaborate. It is still too e a r l y to assess t h e full i m p a c t of t h e 1996 T e l e c o m m u n i c a t i o n s Bill. A notable e x a m p l e of t h a t i m p a c t is AT&T's $37 billion acquisition of t h e cable TV c o m p a n y TCI a n d t h e $58 billion acquisition of M e d i a O n e , w i t h t h e objective of d e v e l o p i n g a two-way s y s t e m t h a t c a n h a n d l e voice, data, a n d video services. T h i s c h a p t e r explains t h e technological basis u n d e r l y i n g this c h a n g e . Tbday's t e l e p h o n e n e t w o r k consists of a high-speed digital b a c k b o n e net­ work. Most subscribers gain access to this n e t w o r k u s i n g dedicated, a n a l o g

Circuit-Switched Networks

206

twisted wire-pair called t h e local or subscriber loop. T h e b a n d w i d t h o n t h e s e loops h a s b e e n restricted to 3 k H z to m a k e it suitable for voice. M o d e m s t h a t convert bit s t r e a m s into voicelike signals c a n u s e this b a n d w i d t h to t r a n s m i t data at rates u p to 28.8 Kbps. T h e capacity of a c h a n n e l is given b y t h e formula bandwidth χ l o g ( l + signal/noise). For a 3 k H z c h a n n e l w i t h a s i g n a l / n o i s e ra­ tio of 30 dB or 1,000, this capacity w o r k s o u t to 30 Kbps. ( T h e r e c e n t 56 Kbps m o d e m s are discussed in 5.5.2.) More t h a n 90% of u s e r s gain I n t e r n e t access t h r o u g h voice m o d e m s . 2

We i n t r o d u c e in section 5.1 s o m e s i m p l e c o n c e p t s t h a t are u s e d to describe circuit-switched n e t w o r k s . More elaborate m a t h e m a t i c a l m o d e l s are d e v e l o p e d in C h a p t e r s 8 a n d 9. Five i n n o v a t i o n s h a v e dramatically e x p a n d e d t h e capabilities of t e l e p h o n e n e t w o r k s : SONET ( S y n c h r o n o u s Optical Network), DWDM ( D e n s e WaveDivision Multiplexer), optical local loop, various digital subscriber loop (DSL) solutions including ISDN (Integrated Services Digital N e t w o r k ) a n d ADSL ( A s y m m e t r i c DSL), a n d INA (Intelligent N e t w o r k A r c h i t e c t u r e ) . SONET provides h i g h e r s p e e d s t h a n t h e c u r r e n t t r a n s m i s s i o n s y s t e m . SONET also simplifies multiplexing e q u i p m e n t . SONET is discussed in sec­ tion 5.2. In t e r m s of t h e O p e n Data N e t w o r k m o d e l , SONET i m p l e m e n t s a "bit way," as indicated in Figure 2.28. SONET provides end-to-end fixed-rate c h a n n e l s at s p e e d s t h a t are m u l t i p l e s of 55.84 Mbps. T h o s e c h a n n e l s c a n b e u s e d in a flexible m a n n e r to s u p p o r t a variety of b e a r e r services. SONET is "backward-compatible" b e c a u s e it c a n c a r r y c u r r e n t t e l e p h o n e bit s t r e a m s . It is also "forward-compatible," since it c a n b e u s e d to t r a n s p o r t ATM cells. T h e t r a n s p o r t of ATM cells over SONET is t h e basis of t h e Broadband I n t e g r a t e d Ser­ vices Digital N e t w o r k or BISDN, w h i c h s o m e regard as t h e "universal" n e t w o r k . T h e r e also are proposals to t r a n s p o r t IP d a t a g r a m s directly over SONET. Five y e a r s ago, wave-division multiplexing w a s d e m o n s t r a t e d in laborato­ ries. Tbday c o m m e r c i a l DWDM p e r m i t s light of 40 different w a v e l e n g t h s , e a c h m o d u l a t i n g u p to 10 Gbps of information, a n d carried o n t h e s a m e fiber. In this way, t h e capacity of existing fiber is i n c r e a s e d 40 t i m e s at a fraction of t h e cost of adding n e w fiber. In May 1999, Nortel a n n o u n c e d a n optical amplification s y s t e m t h a t c o m b i n e s 160 different wavelengths, e a c h c a r r y i n g 10 Gbps of data, for a total of 1.6 terabits p e r second, exceeding all t h e traffic o n t h e I n t e r n e t . DWDM is described i n section 5.3. T h e third i n n o v a t i o n in t h e t e l e p h o n e n e t w o r k is t h e partial or full replace­ m e n t of t h e existing c o p p e r subscriber loop b y optical fiber. T h e r e p l a c e m e n t will e l i m i n a t e t h e c u r r e n t s p e e d b o t t l e n e c k intrinsic to copper, a n d u s e r s t h e n will b e able to s e n d a n d receive all forms of i n f o r m a t i o n (video, image, text,

Circuit-Switched Networks

voice) r e q u i r i n g h i g h speed. O n l y w h e n high-speed access is w i d e l y d e p l o y e d will BISDN or t h e "information s u p e r h i g h w a y " b e c o m e a reality t h a t m a n y c a n share. In section 5.4 w e s t u d y t h r e e t e c h n o l o g i e s for optical local loop: t h e tra­ ditional digital loop carrier; passive optical n e t w o r k s (PON) t h a t c o m b i n e a broadcast downstream channel with a shared upstream channel; and hybrid solutions. It is possible to r e m o v e t h e 3-kHz b a n d w i d t h restriction o n m o s t a n a l o g subscriber loops a n d m a k e u s e of t h e 1-MHz b a n d w i d t h available o n u n l o a d e d twisted pairs. T h e fourth i n n o v a t i o n consists of various digital subscriber line or DSL solutions t h a t u s e this b a n d w i d t h . ISDN, a n d ADSL a n d its v a r i a n t s (called xDSL), give u s e r s a c o m b i n a t i o n of data a n d voice c h a n n e l s m u l t i p l e x e d a n d m a d e available at a single interface at a n aggregate bit r a t e of several Mbps, over t h e existing c o p p e r subscriber loop. DSL t e c h n o l o g i e s t h e r e b y e x t e n d t h e e c o n o m i c life of t h e existing c o p p e r plant. We describe DSL t e c h n o l o g i e s i n section 5.5. SONET, DWDM, optical local loop, a n d DSL t e c h n o l o g i e s e n h a n c e t h e t e l e p h o n e n e t w o r k ' s i n f o r m a t i o n transfer services. INA is a n i n n o v a t i o n of a different sort. Tb a p p r e c i a t e INA, w e m u s t t h i n k of t h e t e l e p h o n e n e t w o r k as a collection of ( h a r d w a r e a n d software) resources, e a c h of w h i c h c a n execute a n u m b e r of activities. For example, circuits c a n b e c o m b i n e d to form a p a t h a n d u s e d to transfer a c o n s t a n t bit rate voice s t r e a m ; a r e c o r d i n g device c a n b e activated to record a m e s s a g e or to play b a c k a previously r e c o r d e d m e s s a g e ; a switch c a n s e n d a "ring" signal to a t e l e p h o n e ; t h e access-line interface at t h e switch c a n r e c o r d t h e digits dialed b y a subscriber; a n d so on. T h e s e activities c a n b e c o m b i n e d into s e q u e n c e s to p r o d u c e valuable n e w services s u c h as t h e 800 n u m b e r service, call forwarding, voice mail, a n d caller identification. INA facilitates t h e specification a n d i m p l e m e n t a t i o n of t h e s e activity s e q u e n c e s . INA is described in section 5.6. T h e s e c o n d m o s t i m p o r t a n t circuit-switched n e t w o r k is CATV. Until t h e mid-1990s CATV w a s a o n e - w a y a n a l o g distribution n e t w o r k w i t h a large b a n d ­ w i d t h going " d o w n s t r e a m " to users, little or n o b a n d w i d t h going "upstream" from users, a n d w i t h n o switching capability. R e c e n t technological i n n o v a t i o n s in n e t w o r k i n g a n d signal p r o c e s s i n g m a y r e d u c e or e l i m i n a t e t h e s e h a n d i c a p s . If t h a t h a p p e n s , CATV will b e c o m e a formidable c o m p e t i t o r to t h e t e l e p h o n e c o m p a n i e s b e c a u s e , u n l i k e t h e t e l e p h o n e n e t w o r k , CATV a l r e a d y h a s a high­ s p e e d "local loop"; b e c a u s e it h a s a c u s t o m e r b a s e t h a t is c o m p a r a b l e to t h a t of t h e p h o n e c o m p a n i e s ; a n d b e c a u s e it c a n u s e its large s u b s c r i b e r fees to finance t h e n e c e s s a r y i n v e s t m e n t . CATV is d i s c u s s e d in section 5.7. A s u m m a r y of this c h a p t e r is given in section 5.8.

^Qy

208

Circuit-Switched Networks mmmmmmmmmmmm—mmmmmm-

5.1

WWWWBBWWWa^WMBfcWWMWWBBW^WWMWiWIJMMWMWiai

W

M

W

M

B

PERFORMANCE OF CIRCUIT-SWITCHED NETWORKS Circuit-switched n e t w o r k s consist of switches i n t e r c o n n e c t e d b y links. T h e capacity or b a n d w i d t h of e a c h link is divided into fixed-rate circuits (e.g., 64Kbps circuits for t h e t e l e p h o n e n e t w o r k ) . Alternatively, w e c a n view e a c h link as c o m p r i s i n g several multiplexed circuits. T e l e p h o n e n e t w o r k s e m p l o y time-division multiplexing, while CATV, s o m e cellular t e l e p h o n e , a n d satellite n e t w o r k s today e m p l o y frequency- a n d time-division multiplexing. Users c o n n e c t to a n e t w o r k switch b y d e d i c a t e d or s h a r e d access links. A u s e r wishing to exchange i n f o r m a t i o n m a k e s a r e q u e s t to t h e n e t w o r k for a con­ nection or call to a n o t h e r u s e r at a specified address. (Different n e t w o r k s h a v e different addressing c o n v e n t i o n s . For example, t h e N o r t h A m e r i c a n t e l e p h o n e n e t w o r k h a s a 10-digit address.) O n receiving t h e call request, t h e n e t w o r k switches exchange information according to specified a l g o r i t h m s in o r d e r to m a k e two related decisions: t h e call admission decision d e t e r m i n e s w h e t h e r to a d m i t or reject t h e request; a n d if t h e decision is to admit, t h e routing al­ gorithm selects a r o u t e or s e q u e n c e of idle circuits to c o n n e c t t h e two u s e r s . O n c e t h e c o n n e c t i o n is established, t h e u s e r s exchange information. ( T h e t i m e t a k e n b y t h e switches to establish a c o n n e c t i o n is t h e c o n n e c t i o n setup time. D a t a g r a m n e t w o r k s h a v e n o s e t u p time.) W h e n t h e exchange is over, t h e u s e r r e q u e s t s c o n n e c t i o n t e r m i n a t i o n , a n d t h e switches r e t u r n t h e circuits to t h e idle state. A c o n n e c t i o n r e q u e s t t h a t is rejected b y t h e n e t w o r k is said to b e blocked. Requests m a y b e blocked b e c a u s e t h e switches are u n a b l e to find a r o u t e w i t h idle circuits or b e c a u s e idle circuits are k e p t in r e s e r v e for future c o n n e c t i o n r e q u e s t s . T h e m o s t i m p o r t a n t p e r f o r m a n c e m e a s u r e of t h e n e t w o r k is t h e blocking probability, defined as t h e c h a n c e t h a t a call r e q u e s t is rejected d u r i n g t h e b u s i e s t traffic hour. D u r i n g t h e information exchange period, u s e r data is t r a n s f e r r e d in frames at a fixed rate. T h e frames c a r r y a c o n n e c t i o n identifier, a n d t h e switches m a i n t a i n a state table relating t h e c o n n e c t i o n identifier to t h e n e x t l i n k in its route. Therefore, frames in t h e s a m e c o n n e c t i o n travel over t h e s a m e route. F r a m e s in different c o n n e c t i o n s t h a t go t h r o u g h t h e s a m e l i n k are multiplexed. At t h e switch, t h e different c o n n e c t i o n s o n e a c h i n c o m i n g l i n k are demultiplexed, a n d different c o m b i n a t i o n s are r e m u l t i p l e x e d to form t h e s t r e a m for e a c h outgoing link.

5.1

Performance of Circuit-Switched Networks

209

Because t h e n e t w o r k allocates a fixed b a n d w i d t h to e a c h c o n n e c t i o n , t h e frames face n o q u e u i n g delay. H e n c e t h e total end-to-end d e l a y is t h e propa­ gation t i m e p l u s a small processing d e l a y in e a c h switch along t h e route. This total d e l a y is b o t h small a n d constant. T h a t is w h y circuit-switched n e t w o r k s are well suited for real-time, c o n s t a n t bit rate traffic. I n t e r m s of t h e O p e n Data N e t w o r k m o d e l of section 2.8, t h e b e a r e r service offered b y t h e n e t w o r k is t h e real-time transfer of information w i t h specified p e a k rate. T h i s b e a r e r service m a y also b e u s e d for variable bit rate traffic w h o s e p e a k rate is s m a l l e r t h a n t h e c o n n e c t i o n b a n d w i d t h . T h i s m a y l e a d to a low utilization, defined as t h e ratio of t h e average rate of i n f o r m a t i o n transfer divided b y t h e b a n d w i d t h of t h e c o n n e c t i o n . If t h e utilization is significantly b e l o w 1, t h e n link capacity is v e r y poorly utilized. For example, in Tfelnet applications, t h e u s e r ' s t y p i n g s p e e d limits t h e i n f o r m a t i o n transfer to 40 b p s . If this application is r u n over a 64-Kbps t e l e p h o n e line, t h e utilization is 40/64,000, w h i c h is less t h a n 0 . 1 % . T h e b e a r e r service m a y also b e u s e d to transfer m e s s a g e s . In this case, t h e large s e t u p t i m e m a y m a k e this u s e of t h e service n o n c o m p e t i t i v e relative to m e s s a g e transfer t h r o u g h packet-switched networks with no setup time. Several q u e s t i o n s of n e t w o r k p l a n n i n g , r o u t e a s s i g n m e n t , a n d b l o c k i n g probability c a n b e f o r m u l a t e d u s i n g t h e m a t h e m a t i c a l m o d e l of t h e n e t w o r k d e s c r i b e d next. See Figure 5.1. (We s t u d y stochastic m o d e l s in C h a p t e r 9.) We r e p r e s e n t a circuit-switched n e t w o r k as a g r a p h w h o s e n o d e s i = 1, . . . , η d e n o t e t h e switches, a n d t h e r e is a n edge b e t w e e n n o d e s ι a n d ; if t h e r e is a t r a n s m i s s i o n link c o n n e c t i n g switches i a n d ;. We s u p p o s e t h a t t h e n e t w o r k provides c o n n e c t i o n s w i t h a fixed b a n d w i d t h . (For example, t h e tele­ p h o n e n e t w o r k provides 64-Kbps c o n n e c t i o n s . ) T h e capacity of t h e l i n k j o i n i n g i a n d ; is d e n o t e d b y Qj, t h e n u m b e r of s i m u l t a n e o u s (two-way) c o n n e c t i o n s or calls t h a t this link c a n a c c o m m o d a t e . (For example, a T-l l i n k c a n a c c o m m o d a t e 24 voice calls; see Tkble 1.1.) A r o u t e r is a s e q u e n c e of c o n n e c t e d links starting at a n origin switch a n d e n d i n g at a d e s t i n a t i o n switch. We w r i t e (i,;) e r if link (i,j) is p a r t of t h e r o u t e r. Let R d e n o t e t h e set of all routes. At a n y given t i m e t t h e r e are a n u m b e r of s i m u l t a n e o u s active c o n n e c t i o n s t h r o u g h different routes. We d e n o t e t h e s e c o n n e c t i o n s b y t h e vector n(r) = ( n ( r ) , r e R], w h e r e n (f) is t h e n u m b e r of c o n n e c t i o n s t h r o u g h r o u t e r at t i m e r. T h e n u m b e r of calls t h r o u g h l i n k (i,j) is t h e n u m b e r of calls t h r o u g h all r o u t e s t h a t pass t h r o u g h (i,;): r

{r\(i,j)er}

r

Circuit-Switched Networks

210

Switch /

5.1 ^Tr

Link has capacity Q and carries riy(t) calls at time t. There are n (t) calls along route r, which is designed to carry C calls. ;

T O

r

r

T h e n u m b e r of c o n n e c t i o n s t h r o u g h a link is limited b y its capacity a n d so n(t) m u s t satisfy t h e constraint: w

y(0 =

Σ

M 0 < C y , for all (i,;).

(5.1)

{r\(ij)zr}

Relation (5.1) d e t e r m i n e s t h e m a x i m u m possible set of c o n n e c t i o n s t h a t c a n b e a c c o m m o d a t e d b y t h e n e t w o r k . T h e vector n(t) c h a n g e s r a n d o m l y over t i m e . W h e n a call along r o u t e r t e r m i n a t e s , n (t) d e c r e a s e s b y 1; if a n e w call is p l a c e d along r o u t e r, t h e n n (t) i n c r e a s e s b y 1. Suppose t h a t t h e n e t w o r k is d e s i g n e d to carry C s i m u l t a n e o u s calls along r o u t e r, r eR. T h e n t h e link capacities Qj m u s t b e so large t h a t r

r

r

Σ

r< H>

c

c

forallreR.

(5.2)

[Wj)er]

T h e s y s t e m of inequalities (5.2) establishes a relation a m o n g t h e capacity invested in t h e t r a n s m i s s i o n system, t h e choice of routes, a n d t h e m a x i m u m n u m b e r of s i m u l t a n e o u s calls t h a t t h e n e t w o r k c a n a c c o m m o d a t e . T h e r a n d o m a m o u n t of t i m e t h a t a c o n n e c t i o n e n d u r e s is t h e call holding time. T h e r a n d o m t i m e b e t w e e n t h e arrival of two c o n s e c u t i v e c o n n e c t i o n r e q u e s t s is t h e interarrival t i m e . It is c u s t o m a r y to d e n o t e t h e average h o l d i n g

5.2

SONET

time by μ a n d t h e average interarrival t i m e b y λ " , so λ is t h e average r a t e of call r e q u e s t s (calls p e r s e c o n d ) . - 1

1

Fix a r o u t e r, a n d let k , μ b e t h e call p a r a m e t e r v a l u e s for t h a t route. Suppose p is t h e b l o c k i n g probability for t h e s e calls. T h e n t h e r a t e of earned calls is (1 -p )K calls p e r s e c o n d . E a c h of t h e s e calls lasts μ~ s e c o n d s o n average. So t h e average n u m b e r of active c o n n e c t i o n s along r o u t e r is r

Υ

r

ι

r

(1 -p )—

X

r

r

= : (1

~Pr)pr

t

Mr

w h e r e p is t h e n u m b e r of call r e q u e s t s d u r i n g o n e call h o l d i n g t i m e . p is a n i m p o r t a n t m e a s u r e of traffic i n t e n s i t y . Its u n i t of m e a s u r e m e n t is n a m e d after Erlang for h i s p i o n e e r i n g w o r k in traffic e n g i n e e r i n g . r

r

Evidently, w e m u s t h a v e (1 — p )p < C , all r eR, w h i c h p l a c e s a l o w e r b o u n d o n t h e b l o c k i n g probabilities {p }. T h i s l o w e r b o u n d , however, is q u i t e optimistic, as it is b a s e d o n average v a l u e s a n d i g n o r e s t h e statistical fluctua­ tions in t h e r e q u e s t arrivals a n d call h o l d i n g t i m e s . I n C h a p t e r 9 w e c o m p u t e m o r e a c c u r a t e e s t i m a t e s of b l o c k i n g probabilities. T h e m o d e l d e s c r i b e d above, t o g e t h e r w i t h r e l a t e d p r o b l e m s p r e s e n t e d in section 5.10, c a n b e u s e d as e x a m p l e s of h o w to f o r m u l a t e a n d resolve q u e s t i o n s of p l a n n i n g a n d o p e r a t i o n s of circuit-switched n e t w o r k s . T h e p a r a m e t e r s of costs a n d capacity t h a t e n t e r i n t h o s e q u e s t i o n s d e p e n d o n t e c h n o l o g y . We n o w describe t h e m a i n technological i n n o v a t i o n s i n circuit-switched n e t w o r k s , b e g i n n i n g w i t h SONET. r

r

r

r

52

SONET We v i e w SONET ( S y n c h r o n o u s Optical N e t w o r k ) as a bit w a y i m p l e m e n t a t i o n , providing end-to-end t r a n s p o r t of bit s t r e a m s . As its n a m e suggests, t h e devel­ o p m e n t of SONET w a s s p u r r e d b y a d v a n c e s in t h e a c c u r a c y of clocks a n d i n optical t r a n s m i s s i o n . SONET is a n ANSI s t a n d a r d . (SDH is a n ITU s t a n d a r d . ) It e n c o d e s bit s t r e a m s into optical signals t h a t are p r o p a g a t e d over optical fiber. SONET'S h i g h s p e e d a n d its frame s t r u c t u r e p e r m i t it to s u p p o r t a v e r y flexible set of b e a r e r services. T h e s t a n d a r d specifies t h e frame s t r u c t u r e as well as t h e characteristics of t h e optical signal. We will discuss o n l y t h e frame s t r u c t u r e . T h e m o s t i m p o r t a n t feature of t h e s t a n d a r d is t h a t all clocks in t h e net­ w o r k are locked to a c o m m o n m a s t e r clock, so t h a t t h e s i m p l e time-division multiplexing (TDM) s c h e m e of Figure 2.5 c a n b e u s e d . Multiplexing in SONET

Circuit-Switched Networks

212 Common clock

Synchronized byte streams

Multiplexing by byte interleaving

5.2 FIGURE

SONET sources are synchronized to a c o m m o n master clock. Different streams are multiplexed b y byte interleaving.

is d o n e b y b y t e interleaving, as d e p i c t e d in Figure 5.2. As a result, as w e see in t h e figure, if e a c h of Ν i n p u t s t r e a m s h a s t h e s a m e rate R, t h e m u l t i p l e x e d s t r e a m h a s rate NR. Because s o u r c e s are s y n c h r o n i z e d , t h e buffers at e a c h in­ p u t line will b e v e r y small as t h e y h a v e o n l y to a c c o m m o d a t e t h e effect of jitter. (Jitter is t h e inevitable, small variation in t h e successive clock ticks r e c o v e r e d from a periodic signal c o r r u p t e d b y s o m e noise.) In N o r t h A m e r i c a a n d J a p a n , t h e basic SONET signal is STS-1 ( S y n c h r o n o u s Transport Signal-1). It h a s a bit rate of 51.84 Mbps. H i g h e r r a t e signals are m u l t i p l e s of this rate; see Figure 5.3. In Europe, t h e b a s i c r a t e is STS-3 or 155.52 Mbps, a n d t h e STS h i e r a r c h y is called t h e SDH or S y n c h r o n o u s Digital H i e r a r c h y , starting at 155.52 Mbps. T h e simplicity of t h e STS h i e r a r c h y m a k e s a contrast w i t h t h e c u r r e n t dig­ ital signal (DS) h i e r a r c h y s h o w n in I&ble 1.2, w h i c h , b e c a u s e of t h e w a y it a c c o m m o d a t e s a s y n c h r o n o u s s t r e a m s , m a k e s m u l t i p l e x i n g m u c h m o r e com­ plex. As Figure 5.3 suggests, moreover, SONET is b a c k w a r d - c o m p a t i b l e in t h e s e n s e t h a t it c a n t r a n s p o r t c u r r e n t t e l e p h o n e signals: t h e 1.544-Mbps DS-1 signal in N o r t h A m e r i c a a n d J a p a n a n d t h e 2.0-Mbps E-l signal i n E u r o p e . SONET'S frame s t r u c t u r e is also forward-compatible in t h a t it c a n s u p p o r t t h e t r a n s p o r t of ATM cells. T h e frame also provides c h a n n e l s for organization, a d m i n i s t r a t i o n , a n d m a n a g e m e n t in a u n i f o r m m a n n e r . SONET'S frame s t r u c t u r e a n d multiplexing m e t h o d s also simplify e q u i p ­ m e n t . T h e b y t e - i n t e r l e a v e d time-division m u l t i p l e x i n g m a k e s d e m u l t i p l e x i n g

5.2

SONET

213

2.488 Gbps

622 Mbps

155 Mbps

51 Mbps

51 Mbps

STS-48

STS-12

STS-3

STS-1

139 Mbps

6.3 Mbps

1.5 Mbps Japan

5.3 FIGURE

2.0 Mbps

DS-1 North America

E-l

Europe

The STS-n signal has a rate equal to η χ 51.84 Mbps. In Europe the hierarchy starts at 155.52 Mbps. All the standards b e c o m e compatible at speeds of 155 Mbps.

easy b e c a u s e , as suggested in Figure 5.4, individual s t r e a m s c a n b e d e t e r m i n e d s i m p l y b y position in t h e frame. A SONET ADM ( a d d / d r o p m u l t i p l e x e r ) is also less difficult to design a n d build. T h e function of a n ADM is to d r o p o n e of t h e i n c o m i n g m u l t i p l e x e d

Byte-interleaved streams, Rate = 3R Rate = R Rate = 3R

SONET streams are demultiplexed byte by byte b y counting. FIGURE

Circuit-Switched Networks

ADM

5.5 FIGURE

The a d d / d r o p multiplexer (ADM) replaces one of the incoming streams with another.

s t r e a m s a n d replace it w i t h a n o t h e r stream; see Figure 5.5. (In t h e c u r r e n t digital signal h i e r a r c h y of Table 1.2, a n ADM r e q u i r e s a full demultiplexermultiplexer pair.) T h u s , for several functions, SONET e q u i p m e n t is c h e a p e r than current equipment. T e l e p h o n e n e t w o r k s n e e d to h a v e e n o u g h r e d u n d a n c y a n d a u t o m a t i c pro­ c e d u r e s t h a t u s e this r e d u n d a n c y to restore quickly t h e integrity of t h e n e t w o r k in t h e e v e n t of failures. Two e x a m p l e s will illustrate t h e m a g n i t u d e of t h e losses t h a t c a n o c c u r w h e n restoration p r o c e d u r e s fail. A p o w e r outage in a n u n ­ m a n n e d central office in Hinsdale, Illinois, o n M o t h e r ' s Day, 1988, affected 500,000 c u s t o m e r s w h o m a d e t h r e e a n d a half million calls p e r day. Full ser­ vice w a s n o t restored for o n e m o n t h . I n N o v e m b e r 1988, a c o n s t r u c t i o n c r e w accidentally s e v e r e d a major fiber-optic cable in N e w J e r s e y , d i s r u p t i n g m u c h of t h e long-distance traffic along t h e East Coast, b l o c k i n g m o r e t h a n t h r e e mil­ lion call a t t e m p t s . T h e r e are m a n y a p p r o a c h e s to r e d u n d a n c y , d e p e n d i n g o n t h e n e t w o r k . We shall briefly discuss t h e i n c r e a s e d reliability possible w h e n SONET s y s t e m s are d e p l o y e d as dual rings. It is b e l i e v e d t h a t dual rings m a y restore service w i t h i n 50 m s of a n e t w o r k failure. T h e basic idea is illustrated in t h e left p a n e l of Figure 5.6. O n e fiber carries t h e w o r k i n g ring in w h i c h t h e signal m o v e s counterclockwise. In t h e s t a n d b y or protection fiber, t h e signal m o v e s in a clockwise direction. Normally, t h e r e is n o signal in t h e p r o t e c t i o n ring. If a cable b e t w e e n a n y t w o c e n t r a l offices is cut or if t h e r e is a n o d e failure, t h e failure is d e t e c t e d b y t h e a u t o m a t i c p r o t e c t i o n

5.2

SONET

215 DCS

Working Protection

5.6 FIGURE

The panel on the left shows the configuration of a SONET ring. The panel ^ right shows how digital cross-connect systems (DCSs) can be used to create logical links and reconfigure the network topology.

o

n

e

s y s t e m , w h i c h p e r f o r m s a loop-back function u s i n g t h e s t a n d b y fiber. (This is v e r y similar to t h e FDDI r e c o v e r y p r o c e d u r e . ) In t h e p a n e l o n t h e right in t h e figure, a n ADM is r e p l a c e d b y a digital cross-connect s y s t e m (DCS), w h i c h p e r m i t s a n i n c o m i n g l i n k to b e directly c o n n e c t e d to a n outgoing link. T h e DCS h a s c r e a t e d a logical direct link from ADM1 to ADM2, a n d a n o t h e r logical link from ADM2 to A D M 1 , so t h a t t h e configuration of t h e right p a n e l is e q u i v a l e n t to a r i n g a m o n g t h e t h r e e ADMs. In this way, b y setting DCS i n t e r c o n n e c t i o n s appropriately, o n e c a n effectively c h a n g e t h e topology of t h e n e t w o r k . Such c h a n g e s are c a r r i e d o u t in o r d e r to m e e t shifts in d e m a n d over a 24-hour p e r i o d or s o m e o t h e r period.

52.1

SONET Frame Structure We n o w describe t h e m a i n features of t h e SONET frame s t r u c t u r e . Figure 5.7 depicts typical t r a n s m i s s i o n e q u i p m e n t c o n n e c t i n g a n e t w o r k e l e m e n t (on t h e left) w h e r e a SONET frame is a s s e m b l e d to a n o t h e r e l e m e n t (on t h e right) w h e r e t h a t frame is d i s a s s e m b l e d . (Two t e r m s in t h e figure m a y b e obscure. A regenerator (REG) is a device t h a t boosts t h e p o w e r of t h e optical signal. T h e device h a s t h r e e c o m p o n e n t s : t h e first c o m p o n e n t c o n v e r t s t h e optical signal into a n electrical signal, t h e s e c o n d amplifies t h e electrical signal, a n d t h e third c o n v e r t s t h e amplified electrical signal b a c k into a m o r e powerful optical signal. T h e s e c o n d possibly obscure

Circuit-Switched Networks

216 BIM SONET MPX

OC-1

REG OC-3

ADM OC-3

REG OC-3

)

BIM OC-3

OC-1

( )

SONET MPX

Sections Lines Path BIM = Byte-interleaved multiplexer

57

REG = Regenerator

The main SONET network elements.

FIGURE

t e r m is OC-n or Optical Carrier-η. It is t h e n a m e of t h e facility t h a t t r a n s m i t s t h e STS-n signal.) Figure 5.7 also gives certain definitions u s e d in t h e SONET frame over­ h e a d structure. Section is t h e portion of t h e t r a n s m i s s i o n facility b e t w e e n a t e r m i n a l n e t w o r k e l e m e n t a n d a r e g e n e r a t o r or b e t w e e n two r e g e n e r a t o r s . A terminal point is t h e p o i n t after signal r e g e n e r a t i o n at w h i c h p e r f o r m a n c e m a y b e m o n i t o r e d . Line is t h e t r a n s m i s s i o n m e d i u m (optical fiber) a n d associated e q u i p m e n t r e q u i r e d to t r a n s p o r t information b e t w e e n t w o c o n s e c u t i v e net­ w o r k e l e m e n t s , o n e of w h i c h originates t h e line signal a n d t h e o t h e r of w h i c h t e r m i n a t e s it. (Line c o r r e s p o n d s to w h a t w e call a link.) A l i n e h a s a c e r t a i n bit rate. Path at a given rate is a logical c o n n e c t i o n b e t w e e n a p o i n t at w h i c h a s t a n d a r d frame for t h e signal is a s s e m b l e d a n d a p o i n t at w h i c h t h e s t a n d a r d frame is disassembled. (Path c o r r e s p o n d s to a n end-to-end route.) T h e l a y e r e d o v e r h e a d s t r u c t u r e associated w i t h t h e s e definitions is given in Figure 5.8. T h e four l a y e r s of t h e s t r u c t u r e are t h e path, line, section, a n d p h o t o n i c layers. T h e i r m a i n functions are s u m m a r i z e d in Tkble 5.1. Consider a n e x a m p l e w h e r e two p a t h l a y e r p r o c e s s e s are exchanging DS-3 frames. ( T h e 45-Mbps DS-3 signal b e l o n g s to t h e c u r r e n t t e l e p h o n e signal hier­ archy.) T h e DS-3 frames, plus Ρ Ο Η ( p a t h o v e r h e a d ) , are m a p p e d into a n STS-1 SPE ( s y n c h r o n o u s payload e n v e l o p e ) , w h i c h is t h e n given to t h e line layer. T h e line l a y e r m a y multiplex several different p a y l o a d s (in addition to t h e STS-1 SPE containing t h e DS-3 signal) from t h e p a t h l a y e r (frame a n d f r e q u e n c y aligning e a c h o n e ) a n d a d d LOH (line o v e r h e a d ) . In addition to multiplexing, t h e line o v e r h e a d provides o t h e r functions, s u c h as p r o t e c t i o n switching. Fi­ nally, t h e SOH (section o v e r h e a d ) provides framing a n d s c r a m b l i n g prior to t r a n s m i s s i o n b y t h e p h o t o n i c layer. T h e p h o t o n i c l a y e r c o n v e r t s t h e electrical bit s t r e a m from t h e section l a y e r into a n optical signal.

5.2

SONET

217 Services (DS-/i, Video,...)

Service + ΡΟΗ

Path

SPE + LOH

Line

STS-n + SOH Section Light Photonic Regenerator 5.8

Multiplexer

The four layers of the SONET overhead.

FIGURE

Going b o t t o m u p , e a c h l a y e r b u i l d s o n t h e service p r o v i d e d b y t h e l a y e r below. T h e p h o t o n i c l a y e r p r o v i d e s optical t r a n s m i s s i o n at s o m e rate; t h e section l a y e r provides framing a n d s c r a m b l i n g for t h e bits b e i n g t r a n s m i t t e d ; t h e l i n e l a y e r provides line m a i n t e n a n c e a n d p r o t e c t i o n a n d m u l t i p l e x i n g of STS-1 signals; t h e p a t h l a y e r provides a m a p p i n g from services (e.g., DS-3 frames, ATM cells, digital voice s t r e a m s ) into STS-1 SPEs. I n t e r m s of t h e O p e n Data N e t w o r k m o d e l , t h e SONET p a t h l a y e r i m p l e m e n t s t h e b e a r e r service consisting of end-to-end t r a n s p o r t of bit s t r e a m s in t h e format of t h e SPE. Certain n e t w o r k e q u i p m e n t m a y n o t participate in s o m e of t h e u p p e r - l a y e r functions. For example, t h e m u l t i p l e x e r in Figure 5.8 o p e r a t e s at t h e l i n e layer,

5.1 TABLE

Layer

Function

Path

Services; end-to-end error detection

Line

Multiplexing, with frame and frequency alignment; protection switching; data links

Section

Framing, scrambling, data links

Photonic

Electrical to optical and optical to electrical signal conversion

The layers of SONET and some of their functions.

Circuit-Switched Networks

218 90 bytes

Path overhead (ΡΟΗ)

9 rows

Frame η + 1 _ J Synchronous payload envelope (SPE) _SjS__

T h e SONET frame.

FIGURE

a n d t h e r e g e n e r a t o r o p e r a t e s at t h e section layer. A p h o t o n i c amplifier (which directly amplifies t h e optical signal w i t h o u t c o n v e r t i n g it into a n electrical signal as a r e g e n e r a t o r does) o p e r a t e s at t h e p h o t o n i c layer. Figure 5.9 p r e s e n t s overall features of t h e SONET frame. T h e 810-byte STS-1 frame is organized into n i n e rows of 90 b y t e s ( t r a n s m i t t e d left to right, top to b o t t o m ) . A frame is 125 ps in duration, c o r r e s p o n d i n g to o n e 8-kHz voice s a m p l e . (This gives t h e STS-1 rate of 810 b y t e s / f r a m e χ 8,000 frames/s or 51.840 Mbps.) T h e first t h r e e c o l u m n s (27 b y t e s ) are for SOH a n d LOH. T h u s a n SPE is 9 rows χ 87 c o l u m n s , of w h i c h t h e first c o l u m n (9 b y t e s ) is d e v o t e d to Ρ Ο Η . T h e m o s t u n u s u a l feature is t h a t a n SPE does n o t n e e d to b e aligned to a single STS-1 frame: it m a y "float" a n d o c c u p y parts of t w o c o n s e c u t i v e frames as s h o w n in t h e figure. Two b y t e s in LOH are allocated to a p o i n t e r t h a t indicates t h e offset in b y t e s b e t w e e n t h e p o i n t e r a n d t h e first b y t e of t h e SPE. Figure 5.10 explains t h e function of s o m e individual o v e r h e a d b y t e s . T h e s t a n d a r d defines t h e function of e a c h b y t e a n d assigns it to a l a y e r a n d position in t h e frame. Each of m o s t of t h e unidentified b y t e s provides a 64-Kbps data link for u s e b y t h e t r a n s m i s s i o n e q u i p m e n t for s e n d i n g m e s s a g e s , c o m m a n d s , a n d a l a r m s for t r a n s m i s s i o n functions. E x a m p l e s of t h e s e functions are testing, protection switching, a n d fault isolation. T h e first t h r e e rows are assigned to SOH. Two b y t e s (called Α Ι , A2 in t h e standard) are for framing e a c h STS-1. A n o t h e r b y t e (CI) is t h e STS-1 c h a n n e l ID. It is a u n i q u e n u m b e r assigned to e a c h c h a n n e l prior to b y t e i n t e r l e a v i n g

5.2

SONET

219

^10 FIGURE

into a n STS-n signal a n d stays w i t h t h a t c h a n n e l until it is d e i n t e r l e a v e d (de­ multiplexed). It is u s e d to d e t e r m i n e t h e position of t h e o t h e r STS-1 signals. Byte Bl is a bit-interleaved p a r i t y code 8 (BIP-8) u s i n g e v e n parity. T h e sec­ tion BIP-8 is calculated over all bits of t h e p r e v i o u s STS-n after scrambling. It is defined only for t h e first STS-1 t r i b u t a r y of a n STS-n signal. T h e r e m a i n i n g six rows are for LOH. Byte B2 is a n o t h e r p a r i t y code u s e d for line-error m o n i t o r i n g . Bytes H I , H 2 are allocated to a p o i n t e r indicating t h e offset in b y t e s b e t w e e n t h e p o i n t e r a n d t h e first b y t e of t h e STS SPE. Byte H 3 is u s e d for f r e q u e n c y justification, t h a t is, to c o m p e n s a t e for clock deviations. Even t h o u g h all sources are s y n c h r o n i z e d to t h e s a m e m a s t e r clock, deviations do occur. W h e n t h e rate of t h e source (payload) is h i g h e r t h a n t h e local STS-1 rate, H 3 is u s e d to add a n extra b y t e to t h e SPE; w h e n t h e source is slower, a b y t e is d e l e t e d from t h e SPE. We explain h o w this w o r k s in detail u s i n g Figure 5.11. In a n y frame ( H I , H2) p o i n t s to t h e start of t h e SPE in t h a t frame. Suppose t h a t t h e payload s p e e d is h i g h e r t h a n t h e frame speed. T h e n , o n occasion, o n e extra b y t e is a d d e d to t h e payload. T h i s is d o n e b y d e c r e m e n t i n g t h e p o i n t e r (in frame η + 1) b y 1, so t h a t SPE η + 2 starts sooner. T h e extra b y t e is placed in H 3 .

Circuit-Switched Networks

220

Η1Η2Ή3

Points to start tofSPEn

Frame η

Points to start of SPE η + 1

Data o f SPE η H1H2H3

Frame η + 2

SPE η + 1

Points to start of SPE/i + 2

H1H2 H3

Points to start| ofSPEn+ 1

Empty

H1H2H3

Points to startl of SPE n + 2

SPE n + 2 Pointer in frame η + 1 decremented by 1

5.11 FIGURE

Points to start of SPE η

SPE η H1H2

Frame η + 1

Η1Η2Ή3

Pointer in frame η + 1 incremented by 1

Frequency justification: w h e n the payload rate is higher than the frame rate, ^ * * to provide an extra byte as on the left; w h e n the payload rate is lower, a byte is left e m p t y or "stuffed" as on the right. s u s e c

Suppose i n s t e a d t h a t t h e payload bit rate is slower t h a n t h e frame rate. T h e n , o n occasion 1 b y t e is n o t m a d e available to t h e payload. T h i s is d o n e b y i n c r e m e n t i n g t h e p o i n t e r (in frame η + 1) b y 1, so t h a t SPE n + 2 starts later. T h e b y t e next to H 3 is n o w left e m p t y or "stuffed" a n d n o t allowed to b e p a r t of t h e payload e n v e l o p e . T h e Ρ Ο Η is assigned a n d r e m a i n s w i t h t h e p a y l o a d u n t i l t h e p a y l o a d is de­ multiplexed. T h e first b y t e of ΡΟΗ, J l , is u s e d to r e p e a t a ( u s e r - p r o g r a m m a b l e ) 64-byte, fixed-length string so t h a t a path-receiving t e r m i n a l c a n verify its con­ t i n u e d c o n n e c t i o n to t h e i n t e n d e d transmitter. T h e t h i r d byte, G l , is u s e d to c o n v e y b a c k to a n originating STS p a t h t e r m i n a l t h e p a t h status a n d perfor­ m a n c e . This feature p e r m i t s t h e status a n d p e r f o r m a n c e of t h e c o m p l e t e duplex p a t h to b e m o n i t o r e d at e i t h e r e n d or at a n y p o i n t along t h a t p a t h . Figure 5.12 illustrates h o w a n STS-3 frame w o u l d b e u s e d to c a r r y (approxi­ m a t e l y ) 44 ATM cells of 53 b y t e s each. A n i n t e r e s t i n g feature is t h a t n o framing bits are provided to delimit t h e b o u n d a r y of t h e ATM cells w i t h i n t h e payload. We will see in C h a p t e r 6 h o w t h e CRC bits of t h e ATM cell h e a d e r are u s e d to find t h e cell b o u n d a r y . In s u m m a r y , SONET is a t r a n s m i s s i o n s t a n d a r d t h a t a s s u m e s a syn­ c h r o n o u s m o d e (all signals s y n c h r o n i z e d to t h e s a m e clock). Its flexible frame

5.2

SONET

221 ATM cell

% τ τ τ τ 44 ATM cells

10 (OH) 5.12 FIGURE

260 bytes

An STS-3 frame accommodates 44 ATM cells. No framing bits are provided to delimit the cell boundary.

s t r u c t u r e a c c o m m o d a t e s b o t h a s y n c h r o n o u s traffic s u c h as ATM a n d syn­ c h r o n o u s transfer m o d e (STM) traffic s u c h as voice. T h e s e t w o k i n d s of traffic c a n b e s e p a r a t e d a n d r e c o m b i n e d at a n a d d / d r o p m u l t i p l e x e r or a switch, as suggested in Figure 5.13. W h e t h e r a p a r t i c u l a r service will follow a n STM or ATM m o d e will d e p e n d o n t h e traffic characteristics.

A'

C

Β

I I I

A

C SONET switch

Β

I I A'

Packet processor

I

Copy video broadcast DS-1 processor

5.13 FIGURE

At a SONET switch, streams of different traffic types are demultiplexed, appropriately processed, and t h e n remultiplexed.

Circuit-Switched Networks

222 Optical Networking and Future of SONET

T h e s y n c h r o n o u s clock in t h e n o d e s of a SONET n e t w o r k facilitates b y t e interleaving, r e d u c i n g t h e cost of time-division multiplexing e q u i p m e n t . T h a t feature e n c o u r a g e d t h e w i d e s p r e a d a d o p t i o n of SONET in t e l e p h o n e n e t w o r k s . Motivated b y t h e n e e d s of t e l e p h o n y , SONET also provides signaling c h a n n e l s for fault detection, a u t o m a t i c protection, a n d m a n a g e m e n t . SONET r e d u n d a n t ring a r c h i t e c t u r e s provide reliability. High-speed packet-switching r o u t e r s a n d switches c o m b i n e p a c k e t s from m u l t i p l e links b y statistical m u l t i p l e x i n g r a t h e r t h a n b y time-division multi­ plexing, a n d so t h e signals o n t h o s e links n e e d n o t b e s y n c h r o n i z e d . (Nor is t h e r e a n e e d for t h e still-expensive multiplexers a n d circuit switches.) At t h e physical layer, high-speed packet-switched n e t w o r k s m a y u s e optical fiber ter­ m i n a t e d b y SONET line interface cards i n c l u d i n g t h e features t h a t p r o m o t e reliability, b u t m a y d i s p e n s e w i t h t h e central clock a n d t h e features for fre­ q u e n c y justification n e e d e d for time-division multiplexing. T h e s e "lightweight" SONET links are u s e d in Gigabit E t h e r n e t s a n d ATM. Efforts are also u n d e r w a y to place IP p a c k e t s directly into SONET frames (bypassing t h e ATM l a y e r ) . In this approach, IP d a t a g r a m s are e n c a p s u l a t e d into Point-to-Point Protocol (PPP) packets, w h i c h are t h e n framed u s i n g HDLC to d e l i n e a t e t h e PPP p a c k e t w i t h i n a SONET payload e n v e l o p e . As t h e QpS g u a r a n t e e s of ATM b e c o m e well established, ATM will c a r r y b o t h voice a n d data. At t h a t point, t h e flexibility a n d efficiency of statistical multiplexing, c o m b i n e d w i t h t h e lower cost of ATM switches c o m p a r e d w i t h T D M circuit switches a n d multiplexers, will b r i n g a n e n d to t h e g r o w t h in t h e u s e of circuit switching for t h e b a c k b o n e t e l e p h o n e a n d data n e t w o r k s . T h u s t h e future b a c k b o n e n e t w o r k s (for b o t h data a n d voice) are likely to d e p l o y ATM or IP switches a n d r o u t e r s over "lightweight" SONET links (not SONET n e t w o r k s ) . SONET l i n k s a n d interfaces will also b e u s e d as t h e physical l a y e r in high-speed LANs. T h u s ironically, a l t h o u g h it w a s t h e s y n c h r o n o u s n a t u r e of SONET net­ w o r k s t h a t e n c o u r a g e d large-scale SONET d e p l o y m e n t a n d r e d u c e d t h e cost of SONET line interfaces, it is t h e latter t h a t will survive in n e t w o r k s t h a t are a s y n c h r o n o u s . ATM a n d IP n e t w o r k s over SONET links will a c c o u n t for a n in­ creasingly large fraction of future g r o w t h for o n e further additional r e a s o n : it is m u c h easier to i n t e r c o n n e c t packet-switched n e t w o r k s t h a n to i n t e r c o n n e c t T D M circuit-switched n e t w o r k s . By i n t e r c o n n e c t i n g w i t h existing n e t w o r k s , a small ATM or IP n e t w o r k c o m p a n y c a n effectively c o m p e t e w i t h a m u c h larger TDM-based t e l e p h o n e c o m p a n y .

5.3

Dense Wave-Division Multiplexing (DWDM)

F u t u r e high-speed n e t w o r k s will evolve from T D M circuit-switched infras­ t r u c t u r e into p a c k e t t e c h n o l o g y t h a t c o m b i n e s h i g h - p e r f o r m a n c e cell-switching a n d routing, w i t h statistical multiplexing, initially over SONET links. O v e r t i m e SONET functions will b e m o v e d into t h e data e q u i p m e n t .

5.3

DENSE WAVE-DIVISION MULTIPLEXING (DWDM) A n optical link consists of a l a s e r t h a t e m i t s a light of a c e r t a i n w a v e l e n g t h , m o d u l a t e d b y a data s t r e a m . T h e m o d u l a t e d lightwave is s e n t d o w n a n optical fiber. At t h e o t h e r e n d of t h e fiber, t h e r e c e i v e d lightwave is d e m o d u l a t e d , a n d t h e data s t r e a m is recovered. T h e optical fiber h a s a b a n d w i d t h of a b o u t 25 χ 1 0 Hz, as w e explain in C h a p t e r 11. However, e l e c t r o n i c c o m p o n e n t s today are l i m i t e d to m o d u l a t i o n s p e e d s of 2.5 Gbps, so m o s t of t h e fiber b a n d w i d t h is u n u s e d . 1 2

In wave-division multiplexing, lasers of different w a v e l e n g t h s are m o d u ­ l a t e d b y s e p a r a t e data s t r e a m s , a n d t h e m o d u l a t e d lightwaves are passively c o m b i n e d ( a d d e d u p ) a n d s e n t d o w n t h e s a m e fiber. At t h e o t h e r e n d t h e dif­ ferent w a v e l e n g t h s are s e p a r a t e d out, individually d e m o d u l a t e d , a n d all t h e data s t r e a m s are recovered. T h u s wave-division m u l t i p l e x i n g c r e a t e s several different c h a n n e l s over t h e s a m e fiber, e a c h w i t h a bit rate of u p to 2.5 Gbps (SONET OC-48), e x p e c t e d to i n c r e a s e to 10 Gbps (OC-192). C o m m e r c i a l d e n s e wave-division multiplexers (DWDMs) c o m b i n i n g u p to 16 w a v e l e n g t h s w e r e i n t r o d u c e d in 1996; 40-channel availability w a s an­ n o u n c e d in 1998; a 160-channel c o m m e r c i a l s y s t e m , e a c h w i t h a 10 Gbps data rate, w a s a n n o u n c e d b y Nortel in 1999. Since DWDM e q u i p m e n t c a n b e u s e d w i t h existing fibers, t h e capacity of t h e l i n k s in t h e t e l e p h o n e a n d I n t e r n e t back­ b o n e s h a s dramatically a n d s u d d e n l y i n c r e a s e d from 2.5 Gbps to 100 Gbps. Most DWDM e q u i p m e n t is d e s i g n e d to w o r k w i t h SONET, b u t p r o d u c t s h a v e b e e n a n n o u n c e d w i t h a n "open configuration" t h a t c a n also t r a n s p o r t ATM cells or IP packets, w i t h o u t a n i n t e r v e n i n g s y n c h r o n i z a t i o n l a y e r like SONET. As w e will see, however, full utilization of this capacity will r e q u i r e innova­ tions in switching t e c h n o l o g y t h a t c a n i n c o r p o r a t e DWDM links into a n optical network. T h e success of DWDMs is d u e to t h e d e v e l o p m e n t of t h e distributed feed­ b a c k laser s t r u c t u r e t h a t provides a stable, m o n o c h r o m a t i c light source, a n d filters t h a t s e p a r a t e lightwaves s p a c e d 100 GHz apart. T h e d e v e l o p m e n t of

223

224

Circuit-Switched Networks

w i d e b a n d optical amplifiers t h a t boost t h e p o w e r of all multiplexed lightwaves s i m u l t a n e o u s l y (without optical/electrical c o n v e r s i o n ) h a s i n c r e a s e d t h e s p a n of optical links, further r e d u c i n g cost a n d i n c r e a s i n g reliability. W h e n DWDM links form a n e t w o r k , as in Figure 2.1, a switching ele­ m e n t is n e e d e d to transfer t h e bit s t r e a m s m o d u l a t i n g t h e light signal from a n i n c o m i n g link to a n outgoing link. I b d a y , t h e s e switches are digital. T h e in­ c o m i n g multiplexed signal is d e m u l t i p l e x e d into c o m p o n e n t lightwaves, e a c h of w h i c h is d e m o d u l a t e d to recover t h e individual bit s t r e a m s . T h e bit s t r e a m s are electronically switched, individual lasers are m o d u l a t e d , a n d t h e lightwaves are r e m u l t i p l e x e d into t h e fibers at t h e o u t p u t ports. Several efforts are u n ­ d e r w a y to b y p a s s this costly digital switching, b y p e r f o r m i n g t h e o p e r a t i o n s p u r e l y in t h e optical d o m a i n . Without t h e creation of s u c h optical switching e l e m e n t s , t h e u s e of DWDM links will b e l i m i t e d to long-haul point-to-point links. T h e first c o m m e r c i a l optical a d d / d r o p multiplexers h a v e b e e n a n n o u n c e d . T h e s e e l e m e n t s c a n add or d r o p a small n u m b e r of individual w a v e l e n g t h s at a single site w i t h o u t d e m u l t i p l e x i n g t h e entire wave-division m u l t i p l e x e d signal. T h e s e a d d / d r o p multiplexers c a n b e u s e d at t h e n e t w o r k t e r m i n a t i o n n o d e s a n d will p r o m o t e t h e u s e of DWDM in m e t r o p o l i t a n areas. Wavelength-selective optical cross c o n n e c t s (WSXC or OXC) h a v e already b e e n d e m o n s t r a t e d . T h e s e are switches w i t h a certain n u m b e r of i n p u t a n d out­ p u t ports. Each i n p u t port accepts a signal multiplexing say 8 w a v e l e n g t h s . T h e different w a v e l e n g t h s are d e m u l t i p l e x e d at t h e i n p u t port, s w i t c h e d t h r o u g h t h e fabric to arrive at different o u t p u t ports, w h e r e t h e y are r e m u l t i p l e x e d . T h u s a signal arriving at o n e i n p u t p o r t o n o n e w a v e l e n g t h is t r a n s m i t t e d essentially u n c h a n g e d from a n o u t p u t port, after p o w e r loss c o m p e n s a t i o n . In this w a y a virtual link c a n b e c r e a t e d c o m p r i s i n g a single w a v e l e n g t h r o u t e d t h r o u g h sev­ eral WSXCs. This is called wavelength routing. T h e limitation of this a p p r o a c h is t h a t two signals m o d u l a t i n g t h e s a m e w a v e l e n g t h arriving o n different ports c a n n o t b e t r a n s m i t t e d o n t h e s a m e o u t p u t port. T h e limitation of w a v e l e n g t h r o u t i n g is t h a t t h e s a m e w a v e l e n g t h m u s t b e u s e d in t h e virtual link. This limitation could b e o v e r c o m e b y t h e d e v e l o p m e n t of wavelength converters t h a t c o n v e r t a n i n p u t w a v e l e n g t h into a different w a v e l e n g t h while b e i n g t r a n s p a r e n t to t h e m o d u l a t i n g data. C h a n g i n g t h e configuration of cross c o n n e c t s is still a slow process. It is u s e d o n l y to track large, predictable c h a n g e s in t h e aggregate p a t t e r n of traffic. So at this t i m e this t e c h n o l o g y is n o t suitable to d y n a m i c a l l y reconfigure virtual links at t h e t i m e scale of individual c o n n e c t i o n s . F u r t h e r in t h e future are optical switches t h a t c a n switch individual ATM cells carried b y a lightwave at o n e i n p u t p o r t to a lightwave at a n o t h e r port.

5.4

Fiber to the Home - 1 »

Such devices m u s t p e r f o r m cell h e a d e r processing a n d cell buffering in t h e optical d o m a i n , in addition to fast switching. WDM t e c h n o l o g y is further d i s c u s s e d in section 11.2.

5.4

FIBER TO THE HOME In this section w e c o n s i d e r a p p r o a c h e s d e s i g n e d to a d d r e s s t h e "last mile" p r o b l e m : A s s u m i n g t h a t a h i g h - s p e e d b a c k b o n e n e t w o r k is available, h o w c a n residential or b u s i n e s s c u s t o m e r s b e c o n n e c t e d to it in a cost-effective m a n n e r ? T h e a p p r o a c h e s c o n s i d e r e d h e r e rely o n r e p l a c e m e n t of t h e c o p p e r subscriber loops b y optical s y s t e m s . T h e s e s y s t e m s are k n o w n as fiber in the loop (FITL) or fiber to the home ( F T T H ) . CATV s y s t e m s are s o m e t i m e s also g r o u p e d u n d e r this h e a d i n g . T h e r e is a n e c o n o m i c justification for u s i n g fibers in t h e subscriber loop w h e n fibers are less expensive t h a n c o p p e r or w h e n n e w services r e q u i r e a larger b a n d w i d t h t h a n t h a t p r o v i d e d b y copper. T h e cost of installing a fiber loop is c o m p a r a b l e to t h a t of a c o p p e r loop. As a result, local o p e r a t i n g c o m p a n i e s often d e p l o y optical loops for n e w installations. Some are r e p l a c i n g old c o p p e r loops w i t h fibers w h e n m a i n t e n a n c e is n e e d e d or anticipated. T h e u s e of fibers d o e s entail additional cost of electric to optical conversion. Most observers believe t h a t t h e installation of fiber will b e justified b y t h e a n t i c i p a t e d n e w services t h a t r e q u i r e high-speed access. E x a m p l e s of t h e s e services are ISDN, TV distribution, h o m e shopping, a n d h i g h - s p e e d I n t e r n e t ac­ cess. Working at h o m e ( t e l e c o m m u t i n g ) a n d distance l e a r n i n g m a y c o n t r i b u t e greatly to t h e d e m a n d for h i g h - b a n d w i d t h services. However, i n c r e a s e d access s p e e d s over t h e existing c o p p e r loops a n d CATV will r e t a r d t h e d e p l o y m e n t of optical loops.

5.4.1

The Optical Loop Carrier System Most t e l e p h o n e s are c o n n e c t e d to t h e c e n t r a l office b y a t w i s t e d wire-pair. In t h e 1970s, T-l carrier l i n e s w e r e i n t r o d u c e d into loop cables, t e r m i n a t e d b y digital loop carrier s y s t e m s , o n e in t h e c e n t r a l office a n d t h e o t h e r in a r e m o t e t e r m i n a l (RT). Subscriber signals, carried over t w i s t e d pairs, w e r e time-division m u l t i p l e x e d at t h e RT into t h e T-l carrier to a n d from t h e central office. In t h e early 1990s, optical fiber t r a n s m i s s i o n b e g a n to b e installed b e t w e e n t h e c e n t r a l office a n d t h e RI& in place of t h e T-l lines.

225

Circuit-Switched Networks

226

Feeder

Central office

-

Distribution

DDM

1000

SLC Series 5 RT

SLC Series 5 DT

RT = Remote terminal, DT = Distant terminal, DDM 1000 = Digital multiplexer

5.14

The AT&T subscriber loop system.

FIGURE

T h e time-division multiplexer at t h e RT c a n b e software-configured to provision m u l t i p l e s of 64 Kbps service to individual subscribers, as n e e d e d . N e w carrier s y s t e m s offer m o r e flexible subscriber access i n c l u d i n g F r a m e Relay a n d ATM. In addition to this flexibility, t h e loop carrier s y s t e m s e l i m i n a t e t h e cost of t h e line interface cards n e e d e d p e r subscriber wire-pair t e r m i n a t i n g in t h e central office. We n o w describe o n e carrier s y s t e m in m o r e detail. Figure 5.14 gives a block d i a g r a m of t h e AT&T Subscriber Loop Carrier (SLC) Series 5 s y s t e m . In this system, t h e signals to a n d from different subscribers are m u l t i p l e x e d elec­ tronically (TDM) at a r e m o t e t e r m i n a l (RT). Fibers i n s t e a d of wire-pairs c o n n e c t t h e r e m o t e t e r m i n a l to t h e c u s t o m e r p r e m i s e s . T h u s t h e optical subscriber sig­ nal is c o n v e r t e d to a n electric signal at t h e RT, t h e different electric signals are multiplexed, a n d t h e multiplexed signal is c o n v e r t e d to a n optical signal a n d s e n t to t h e central office. T h e characteristics of t h e optical subscriber loops are indicated in Figure 5.15. ( T h e s e characteristics m a y n o t s e e m m e a n i n g f u l . In C h a p t e r 11 w e s t u d y w h a t t h e y m e a n . ) T h e s y s t e m allocates a DS-1 (1.544-Mbps) c h a n n e l to e a c h c u s t o m e r loca­ tion. As a result, additional l i n e s a n d ISDN services are easily p r o v i d e d to t h e customers.

5.4.2

Passive Optical Networks (PONs) T h e loop carrier s y s t e m s deliver a fixed bit rate, dedicated, duplex c h a n n e l to e a c h subscriber u s i n g TDM. T h e passive optical n e t w o r k s or PONs d e s c r i b e d in

5.4

Fiber to the Home

227 Directional coupler

Splices

-x—X-

Single-mode fiber, 3-dB directional couplers InGaAs laser diode, 1.3 mm, P = -20 dBm T

InGaAs PIN diodes with - 4 6 dBm sensitivity at 1.5 Mbps 5.15

Details of the AT&T subscriber loop system.

FIGURE

this section consist of a b r o a d c a s t d o w n s t r e a m c h a n n e l a n d a s h a r e d u p s t r e a m c h a n n e l . T h e d o w n s t r e a m c h a n n e l multiplexes several data s t r e a m s u s i n g TDM, a n d subscriber t e r m i n a l s select t h e s t r e a m i n t e n d e d for t h e m . In t h e u p s t r e a m c h a n n e l , data s t r e a m s from individual subscribers are s i m p l y a d d e d t o g e t h e r u s i n g passive optical couplers. In o r d e r to p r e v e n t o v e r l a p p i n g of different data s t r e a m s , e a c h subscriber s t r e a m is allocated a fixed t i m e slot. A MAC protocol g o v e r n s this time-slot allocation. Cable m o d e m s y s t e m s for CATV u s e a similar a p p r o a c h . T h e i m p o r t a n t difference is t h a t in PONs, t h e b r o a d c a s t a n d m u l t i p l e x i n g functions are carried out passively in t h e optical d o m a i n . We describe in detail t h e British Tfelecom T P O N (Ifelephony o n PON) system, a n d t h e n briefly discuss m o r e r e c e n t proposals. Figure 5.16 depicts t h e T P O N s y s t e m . T h i s s y s t e m u s e s passive (optical) m u l t i p l e x i n g to r e d u c e t h e cost p e r customer. T h e T P O N topology is a "dou­ b l e star" t h a t a t t a c h e s 128 (16 χ 8) c u s t o m e r s o n e a c h fiber from t h e c e n t r a l office. T h e two splitters in t h e figure are optical (contrast w i t h Figure 5.14, w h e r e multiplexing a n d d e m u l t i p l e x i n g are d o n e electronically, r e q u i r i n g electrial/optical conversion). Figure 5.17 s h o w s t h e frame s t r u c t u r e of T P O N for t h e d o w n s t r e a m signal from t h e c e n t r a l office to t h e c u s t o m e r s . T h e frames are r e p e a t e d 100 t i m e s p e r second, a n d e a c h frame c o n t a i n s 80 slots. T h u s , slots are r e p e a t e d 8,000 t i m e s p e r second, so t h a t e a c h b y t e (8 bits) r e p e a t e d in all t h e slots c o n s t i t u t e s a 64Kbps c h a n n e l , w h i c h m a y b e u s e d for voice or data. E a c h 2,496-bit slot t h u s provides 2,352 8-Kbps c h a n n e l s for 128 u s e r s p l u s o v e r h e a d c h a n n e l s . T h e s e c h a n n e l s c a n b e p a c k a g e d as voice c h a n n e l s , ISDN services, data c h a n n e l s , or

Circuit-Switched Networks

228

Distribution system 5 km

1:16

/

500 m

Splitter

Central office

1

0

0

m

/

Splitter

\ 5.16

The British Telecom TPON system.

FIGURE

204,800 bits, 10 ms, 20 Mbps

^

Synchronization Maintenance

BF 1

BF 80

BF2

Ranging 5,120 bits

Voice, data traffic 2,352

•Spare

Overhead 128

16 1 1

2,496 bits, 122 jus Capacity is 2,352 χ 8-Kbps channels divided into packages of 1 8 x 8 = 144-Kbps ISDN channels 8 x 8 = 64-Kbps voice channels

5.17 FIGURE

The TPON frame structure.

5.4

Fiber to the Home

5.18 FIGURE

t h e i r c o m b i n a t i o n s . T h e s y s t e m u s e s t h e additional bits in e a c h slot a n d e a c h frame for s y n c h r o n i z a t i o n a n d m a i n t e n a n c e . S o m e of t h e s e bits are also u s e d for ranging m e a s u r e m e n t s n e e d e d for multiplexing of t h e signals c o m i n g from t h e c u s t o m e r s . As s h o w n in Figure 5.18, t h e signals from t h e c u s t o m e r s are multiplexed in t i m e . T h e 128 subscriber u n i t s t i m e t h e i r periodic t r a n s m i s s i o n s so t h a t t h e i r signals are i n t e r l e a v e d w i t h o u t o v e r l a p p i n g w h e n t h e y r e a c h t h e c e n t e r of t h e star (the c e n t r a l office). (Signal overlap is a n a l o g o u s to a collision o n t h e E t h e r n e t . ) Tb r e d u c e t h e c h a n c e of overlap, a "guard b a n d " m u s t b e provided a r o u n d e a c h individual t r a n s m i s s i o n , w h i c h w a s t e s p o t e n t i a l bandwidth. T P O N services offer d e d i c a t e d b a n d w i d t h to subscribers u s i n g TDM, sim­ ilarly to loop carrier s y s t e m s a n d DSL. T h e e c o n o m i c viability of T P O N h a s e n c o u r a g e d m o r e a m b i t i o u s E u r o p e a n proposals. ΑΡΟΝ, or ATM over PON, s y s t e m s a t t e m p t to c o m b i n e t h e a d v a n t a g e s of ATM-based statistical multiplexing a n d t h e b r o a d c a s t n a t u r e of PON, to provide m u c h h i g h e r bit rates. (A r e c e n t proposal is for 600 Mbps d o w n s t r e a m a n d 300 Mbps u p s t r e a m . ) T h e physical double-star l a y o u t of ΑΡΟΝ is similar to T P O N . However, t h e u s e of optical amplifiers p e r m i t s a g r e a t e r splitting ratio a n d l o n g e r distances. T h e frame s t r u c t u r e is also similar to T P O N . Fixed-size frames are divided into slots allocated to ATM cells p l u s o v e r h e a d . Subscriber u n i t s periodically c o n d u c t r a n g i n g m e a s u r e m e n t s a n d a two-bit g u a r d b a n d is p r o v i d e d to p r e v e n t overlap of signals from different subscribers.

229

Circuit-Switched Networks

230

O n e n e w issue m u s t b e a d d r e s s e d in o r d e r to a c c o m m o d a t e statistical m u l ­ tiplexing. Since subscribers are n o t given fixed slots in w h i c h to t r a n s m i t , a MAC protocol is n e e d e d t h a t allows subscribers to r e q u e s t a n d b e allocated slots in w h i c h to t r a n s m i t . Reservation bits in u p s t r e a m slots are set b y sub­ scribers to indicate t h e i r r e q u e s t s . D o w n s t r e a m slots c a r r y grants to individual subscribers. Grants a n d u p s t r e a m r e q u e s t s are s y n c h r o n i z e d in s u c h a w a y t h a t in t h e u p s t r e a m direction a n e a r - c o n t i n u o u s s t r e a m of cells is m u l t i p l e x e d o n t h e fiber.

5.43

Passive Photonic Loop (PPL) T h e passive p h o t o n i c loop (PPL) u s e s wave-division multiplexing or WDM in­ stead of TDM. PPL allocates o n e pair of w a v e l e n g t h s of light to e a c h subscriber. T h u s η subscribers n e e d In different w a v e l e n g t h s . A subscriber t r a n s m i t s u s i n g o n e w a v e l e n g t h a n d receives t h e o t h e r w a v e l e n g t h . A block d i a g r a m of t h e PPL s y s t e m is s h o w n in Figure 5.19. In t h e figure, subscriber 1 is allocated wave­ l e n g t h s λι, λ + ι , subscriber 2 is allocated w a v e l e n g t h s λ2, λ +2, a n d so on. T h e subscriber c a n m o d u l a t e light of o n e w a v e l e n g t h to s e n d i n f o r m a t i o n to others; η

η

To switch

~~1

I

/ \

• •

From switch

5.19 TTT??™

The PPL architecture allocates one pair of wavelengths of light to each subscriber.

5.4

Fiber to the Home

231

a n d o t h e r s c a n m o d u l a t e light of t h e s e c o n d w a v e l e n g t h to s e n d i n f o r m a t i o n to t h e subscriber. T h e r e is a m o d u l a t i o n b a n d w i d t h a r o u n d e a c h w a v e l e n g t h , w h i c h defines t h e m a x i m u m t r a n s m i s s i o n s p e e d . I n this sense, WDM is similar to frequency-division m u l t i p l e x i n g (FDM), d e s c r i b e d in section 2.6.1. T h e b a n d ­ w i d t h available to e a c h subscriber is h u g e , o n t h e o r d e r of h u n d r e d s of GHz, a l t h o u g h c u r r e n t t e c h n o l o g y limits utilization to a few Gbps. ( T h e t e c h n o l o g y is discussed in C h a p t e r 11.) T h e n u m b e r of subscribers t h a t PPL c a n a c c o m m o d a t e is l i m i t e d b y t h e n u m b e r of w a v e l e n g t h s t h a t c a n reliably b e differentiated u s i n g t u n a b l e lasers a n d optical filters. C u r r e n t t e c h n o l o g y places a n u p p e r b o u n d of 50. PPL n e t w o r k s today exist o n l y as l a b o r a t o r y e x p e r i m e n t s . M a n y a d v a n c e s are n e e d e d before s u c h n e t w o r k s will b e c o m m e r c i a l l y viable.

5.4.4

Hybrid Scheme H y b r i d s c h e m e s t h a t c o m b i n e time-division a n d frequency- or wave-division multiplexing are also possible. O n e s u c h s y s t e m is s h o w n in Figure 5.20. In this system, a n u m b e r of bit s t r e a m s are t r a n s p o r t e d over ATM, u s i n g time-division a n d statistical multiplexing. T h e resulting ATM cell s t r e a m s are multiplexed w i t h s o m e TV signals e i t h e r b y s u b c a r r i e r m u l t i p l e x i n g (described in C h a p t e r 11) or b y WDM. Such a s y s t e m c a n b e u s e d to s u p e r p o s e CATV a n d ATM n e t w o r k s o n t h e s a m e fiber s y s t e m . Since t h e s y s t e m s c a n m a k e

ATM MUX

Subscriber

5.20 FIGURE

A hybrid scheme.

WDM/SCM

Circuit-Switched Networks

232

u s e of parts of existing distribution s y s t e m s for CATV a n d t e l e p h o n e , t h e s e s y s t e m s m a y prove to b e v e r y cost-effective to d e p l o y in t h e next stage of service integration.

5.5

DIGITAL SUBSCRIBER LINE (DSL) If t h e 3-kHz restriction placed o n t h e twisted pair c o n n e c t i n g a subscriber's t e l e p h o n e a n d t h e central office is r e m o v e d , it is possible to u s e t h e 1-MHz b a n d w i d t h t h a t is available for t r a n s m i t t i n g data. Digital subscnber line or DSL is t h e g e n e r i c t e r m for a set of technologies t h a t u s e this b a n d w i d t h for digital t r a n s m i s s i o n . T h e t e r m w a s first u s e d for basic rate i n t e g r a t e d services digital n e t w o r k or ISDN, w h i c h provides 128 Kbps access. In t h e early 1990s, t e l e p h o n e c o m p a n i e s initiated efforts to u s e t h e 1-MHz b a n d w i d t h for t r a n s m i s s i o n of video over twisted pairs. T h e service n e e d e d g r e a t e r b a n d w i d t h in t h e d o w n s t r e a m direction. T h e efforts c u l m i n a t e d in t h e d e v e l o p m e n t of t h e asymmetric digital subscnber line or ADSL. T h e video distribution application t u r n e d o u t n o t to b e c o m m e r c i a l l y viable. However, a growing d e m a n d for ADSL p r o d u c t s b e g a n in 1997. I b d a y , h i g h - s p e e d I n t e r n e t access is s e e n as t h e m a i n application of ADSL, especially for t h e so-called SOHO (single office, h o m e office) m a r k e t . Analog v o i c e - m o d e m s offer 28.8 Kbps. ISDN i n c r e a s e s this to 128 Kbps, c o m p a r e d w i t h ADSL s p e e d s of b e t w e e n 1.5 Mbps a n d 8.0 Mbps. Unlike ISDN, ADSL s t a n d a r d s are n o t widely accepted, a n d t h e r e are several c o m p e t i n g s c h e m e s , collectively k n o w n as xDSL.

5.5.1

ISDN T e l e p h o n e c o m p a n i e s are i m p l e m e n t i n g t h e I n t e g r a t e d Services Digital Net­ w o r k (ISDN) according to s t a n d a r d s defined in successive r e c o m m e n d a t i o n s c u l m i n a t i n g in [1120]. T h e objective of ISDN is to offer n e w digital t r a n s m i s ­ sion services to subscribers. T h e t e l e p h o n e n e t w o r k u s e s digital t r a n s m i s s i o n for voice a n d a packet-switched X.25 n e t w o r k for t h e t r a n s p o r t of signaling in­ formation. ISDN m a k e s t h e s e i n t e r n a l t r a n s p o r t facilities available to u s e r s as n e w services. ISDN offers a variety of b e a r e r services built o n top of t h e first t h r e e OSI layers, higher-level services (called teleservices), a n d s u p p l e m e n t a r y services. T h e teleservices are application-layer services in t e r m s of t h e O p e n Data N e t w o r k m o d e l . T h e s u p p l e m e n t a r y services are c o n c e r n e d w i t h call control

5.5

Digital Subscriber Line (DSL)

233

Packet-switched service

TE

ISDN switch

Circuit-switched service Point-to-point service

Telephone network

5.21 FIGURE

ISDN switch

TE

CCS service

ISDN architecture. Circuit, packet, unswitched point-to-point, and call-control services are accessed through a c o m m o n interface.

functions r a t h e r t h a n c o m m u n i c a t i o n p e r se, a n d do n o t fit directly in t h e O p e n Data N e t w o r k m o d e l . T h e m a i n b e a r e r services are t h e t r a n s p o r t of audio a n d digitized voice (at 64 Kbps), circuit-switched digital c h a n n e l s at r a t e s t h a t are m u l t i p l e s of 64 Kbps, packet-switched virtual circuits, a n d c o n n e c t i o n l e s s service (datagrams). T h e teleservices i n c l u d e telex, facsimile, videotex, a n d teletex t r a n s m i s s i o n s w i t h specific coding a n d end-to-end protocols. A m e s s a g e - h a n d l i n g service a n d a directory service are also b e i n g defined for ISDN. T h e s u p p l e m e n t a r y services i n c l u d e t e l e p h o n e services s u c h as caller identification, call forwarding, call waiting, a n d c o n f e r e n c e calling. Figure 5.21 gives a s c h e m a t i c of t h e ISDN architecture. Circuit-switched, packet-switched, d e d i c a t e d point-to-point, a n d call-control ( c o m m o n c h a n n e l signaling or CCS) services are b r o u g h t t o g e t h e r at a n ISDN switch a n d accessed b y t h e u s e r t h r o u g h a c o m m o n t e r m i n a l e q u i p m e n t (TE). T h e u s e r interfaces to ISDN are defined as c o m b i n a t i o n s of t h r e e t y p e s of c h a n n e l s : B, D, a n d H. (See Figure 5.22.) T h e Β c h a n n e l is a 64-Kbps c h a n n e l t h a t t r a n s p o r t s a circuit-switched c o n n e c t i o n , a n X.25 service (packetswitched, virtual circuit), or a p e r m a n e n t digital point-to-point c o n n e c t i o n . T h e D c h a n n e l is a 16-Kbps or a 64-Kbps c h a n n e l u s e d for signaling i n f o r m a t i o n (call control) a n d for low bit rate packet-switched services. A n Η c h a n n e l is a 384-Kbps, 1,536-Kbps, or 1,920-Kbps c h a n n e l u s e d like a Β c h a n n e l b u t for higher-rate services. T h e ISDN s t a n d a r d s specify t h e basic access a n d t h e

Circuit-Switched Networks

234 Channel types: B: 64 Kbps CS, X.25 (PS, VC), or permanent signaling and low rate PS D: 16 Kbps or 64 Kbps H: 384 Kbps, 1,536 Kbps, or 1,920 Kbps as Β Basic access: 2B + D(16) Primary access: 30B + D (64) in Europe 23B + D (64) United States, Japan, Canada

5.22

Basic interface: Pseudo-ternary (1 = 0 V, 0 = ± 0.75 V alt.) Frame level: 192 Kbps, sync, DC balancing (144 Kbps) Data link: B/PS: LAPB (GBN, ACK + NACK) B/CS and Permanent: user's choice D: LAPD: acknowledged = GBN, VC unacknowledged: datagram with discard Network: routing, mpx, congestion control, call control

ISDN services and standards.

FIGURE

p r i m a r y access for users. T h e basic access is 2B + D; it consists of two fullduplex Β c h a n n e l s a n d a full-duplex 16-Kbps D c h a n n e l . T h e p r i m a r y access is 30B + D (64 Kbps) in Europe, a n d it is 23B + D (64 Kbps) in t h e U n i t e d States, J a p a n , a n d Canada. T h e ISDN s t a n d a r d s specify a n e t w o r k - u s e r interface t h a t c a n b e accessed directly b y ISDN t e r m i n a l e q u i p m e n t s u c h as digital t e l e p h o n e s and, via ter­ m i n a l a d a p t e r s or ISDN routers, b y o t h e r devices s u c h as c o m p u t e r s . We n o w s u m m a r i z e s o m e of t h e ISDN s t a n d a r d s for i m p l e m e n t i n g t h e lower t h r e e OSI layers. For basic access, t h e physical l a y e r of ISDN specifies a n 8-pin c o n n e c t o r to attach to t h e n e t w o r k , a p s e u d o - t e r n a r y e n c o d i n g (1 is r e p r e s e n t e d b y 0 volt a n d 0 b y alternatively + 0 . 7 5 volt a n d —0.75 volt), a frame format t h a t i n c l u d e s s y n c h r o n i z a t i o n a n d DC-level b a l a n c i n g bits, a n d a line rate of 192 Kbps c o r r e s p o n d i n g to t h e 144 Kbps of u s e r data rate (2 χ 64 + 16 Kbps) p l u s t h e o v e r h e a d bits. In addition, t h e physical l a y e r specifies a contention-resolution protocol for access to t h e D c h a n n e l b y u p to eight t e r m i n a l s a t t a c h e d to a c o m m o n (multidrop) line. (See Figure 5.22.) T h e data link l a y e r of ISDN is LAPD for t h e D c h a n n e l a n d LAPB for packetswitched c o n n e c t i o n s o n t h e Β c h a n n e l . For circuit-switched or p e r m a n e n t c o n n e c t i o n s o n t h e Β c h a n n e l , t h e u s e r s c a n choose t h e data link protocol a n d c a n u s e t h e I.465/V.120 protocol defined b y t h e CCITT for s u c h c o n n e c t i o n s . LAPD provides u n a c k n o w l e d g e d a n d a c k n o w l e d g e d information-transfer services. T h e frame s t r u c t u r e is essentially t h a t of X.25: bit-oriented frames t h a t start a n d e n d w i t h a n 8-bit flag t h a t is avoided inside t h e frame b y bitstuffing; a 16-bit CRC is u s e d for e r r o r detection; 16-bit a d d r e s s e s are u s e d to

5.5

Digital Subscriber Line (DSL)

distinguish u s e r s c o n n e c t e d to t h e s a m e interface a n d different c o n n e c t i o n s w i t h a given u s e r (i.e., different service access points). T h e u n a c k n o w l e d g e d service is i m p l e m e n t e d as a d a t a g r a m ; e r r o n e o u s frames are discarded. T h e a c k n o w l e d g e d service is i m p l e m e n t e d as a virtual circuit w i t h Go Back Ν link e r r o r control. T h e receiver c a n t u r n t h e s e n d e r off a n d o n b y s e n d i n g it "receiver n o t ready" a n d "receiver ready" frames. T h e I.465/V.120 data l i n k protocol is a modified version of LAPD t h a t provides a s y n c h r o n o u s data transfer, HDLC s y n c h r o n o u s data transfer, a n d b i t - t r a n s p a r e n t a s y n c h r o n o u s transfer. Tb u s e this transfer protocol, t h e u s e r s first set u p a circuit o n t h e Β c h a n n e l b y u s i n g t h e D c h a n n e l . W h e n t h e transfer is c o m p l e t e , t h e u s e r s release t h e circuit also b y u s i n g t h e D c h a n n e l . T h e n e t w o r k l a y e r of ISDN specifies t h e routing, multiplexing, a n d con­ gestion control, in addition to t h e call-control m e s s a g e s . In s u m m a r y , ISDN is a n a t t e m p t to diversify t h e b e a r e r services offered b y t h e t e l e p h o n e n e t w o r k . T h e different services are p r o v i d e d b y different n e t w o r k s ( r a t h e r t h a n in o n e single n e t w o r k ) accessed t h r o u g h a c o m m o n ISDN switch. Diversity is l i m i t e d b e c a u s e t h e s e services are built o n top of t h e traditional 64-Kbps c h a n n e l s , w h i c h c o n s t r a i n s t h e bit s t r e a m r a t e s t h a t can be supported.

5.5.2

ADSL A s y m m e t r i c Digital Subscriber Line (ADSL) is a m o d e m t e c h n o l o g y t h a t u s e s existing twisted pairs to create t h r e e c h a n n e l s : a h i g h - s p e e d d o w n s t r e a m c h a n ­ nel, a m e d i u m - s p e e d u p s t r e a m or d u p l e x c h a n n e l , d e p e n d i n g o n t h e i m p l e ­ m e n t a t i o n of t h e ADSL architecture, a n d a POTS (Plain Old T e l e p h o n e Service) or a n ISDN c h a n n e l . T h e high-speed c h a n n e l r a n g e s from 1.5 to 6.1 Mbps, w h i l e duplex r a t e s r a n g e from 16 to 640 Kbps. ADSL service p r o v i d e r s offer different c o m b i n a t i o n s of u p s t r e a m / d o w n s t r e a m bit rates, a n d m a n y do n o t offer POTS. T h e c h a n n e l s m a y b e further subdivided u s i n g T D M into 64 Kbps or ISDN services. We first describe t h e physical l a y e r a n d t h e n t h e additional r e q u i r e m e n t s for t h e n e t w o r k layer. T h e r e is a n ADSL m o d e m at e a c h e n d of t h e wire-pair, o n e at t h e subscriber e n d a n d t h e o t h e r at t h e central office. T h e c e n t r a l office m o d e m is called ATU-C (ADSL Terminal Unit-Central Office); t h e subscriber m o d e m is called ATU-R (Remote). T h e 1-MHz b a n d w i d t h of t h e wire-pair is divided into t h r e e regions. F r e q u e n c i e s b e l o w 4 k H z are split off into a s e p a r a t e c h a n n e l for voice (if POTS is provided). F r e q u e n c i e s above 4 k H z are divided into a n u p s t r e a m or duplex c h a n n e l (10-100 k H z ) a n d a h i g h e r f r e q u e n c y d o w n s t r e a m c h a n n e l

235

Circuit-Switched Networks

236 voice

voice

wire-pair

ATU-C

splitter

voice

0

_J>23__

splitter

upstream

10

ATU-R

downstream

100

1,000 kHz

ADSL m o d e m s divide the bandwidth of the wire-pair into three regions.

FIGURE

(100-1,000 k H z ) . T h u s ADSL is a p a s s b a n d t e c h n i q u e , b y p a s s i n g t h e 0 to 4 k H z voice b a n d . By contrast, ISDN u s e s b a s e b a n d m o d u l a t i o n , o v e r l a p p i n g w i t h t h e voice b a n d . (See Figure 5.23.) A m o d u l a t i o n s c h e m e c o n v e r t s t h e serial bit s t r e a m into a n electrical signal suitable for t r a n s m i s s i o n . T h e ANSI ADSL s t a n d a r d specifies Discrete Multitone (DMT) m o d u l a t i o n in w h i c h t h e 1-MHz b a n d w i d t h is p a r t i t i o n e d into 256 4-kHz s u b c h a n n e l s . A s e p a r a t e carrier in e a c h b a n d is q u a d r a t u r e a m p l i t u d e - m o d u l a t e d . A competitive, less c o m p l e x t e c h n i q u e called carrierless amplitude-modulation-phase modulation or CAP h a s r e c e n t l y e m e r g e d . In this t e c h n i q u e t h e r e is a single d o w n s t r e a m a n d a single u p s t r e a m c h a n n e l . C h a n n e l p e r f o r m a n c e over a twisted pair c a n vary greatly d e p e n d i n g u p o n t h e l e n g t h a n d condition of t h e wire. A pair is u n s u i t a b l e for ADSL if it i n c l u d e s a loading coil, or if it t e r m i n a t e s in a 4-kHz band-limiting digital l o o p carrier (DLC). P e r h a p s one-third of all pairs are u n s u i t a b l e for this r e a s o n . T h e per­ f o r m a n c e of t h e r e m a i n i n g pairs is h i g h l y variable d e p e n d i n g again u p o n t h e i r l e n g t h a n d t h e extent of far- a n d n e a r - e n d cross-talk. O v e r c o m i n g t h e s e impair­ m e n t s r e q u i r e s m o r e sophisticated signal processing t e c h n i q u e s or a c c e p t i n g lower bit rates. Tb t r a n s p o r t data p a c k e t s over ADSL r e q u i r e s link l a y e r protocols. T h e r e are two ANSI standards. T h e first u s e s Point-to-Point Protocol (PPP) variablel e n g t h data u n i t s w i t h i n a n HDLC framed s t r u c t u r e (RFC 1662). T h e s e c o n d

5.5

Digital Subscriber Line (DSL)

237

Router or Switch

DSLAM ADSL

ADSL

ATU-C Η

ATU-C Η Central office

5.24

ATM link

ATM network

0

ISP

Corporate gateway

ADSL subscribers are connected via the DSLAM to a backbone network.

FIGURE

follows t h e ATM F o r u m ' s s t a n d a r d for ATM F r a m e UNI, also w i t h i n a n HDLC framed s t r u c t u r e . A n ADSL service p r o v i d e r locates a Digital Subscriber Loop Access Multi­ plexer (DSLAM) in t h e c e n t r a l office. (If POTS is also provided, t h e voice signal split off at t h e ATU-C is s e n t to t h e t e l e p h o n e switch.) T h e DSLAM c o n c e n t r a t e s a n u m b e r of ADSL subscriber l i n e s to a single ATM l i n e c o n n e c t e d via a r o u t e r or l a y e r 2 switch to t h e provider's ATM b a c k b o n e n e t w o r k t h r o u g h w h i c h a subscriber is c o n n e c t e d to a c o r p o r a t e g a t e w a y ( t e l e c o m m u t i n g applications) or a n I n t e r n e t service p r o v i d e r (Figure 5.24). As w i t h cable TV, ADSL service n e e d s a n e x p e n s i v e m o d e m . A t e c h n i c i a n is n e e d e d to test if t h e existing wire-pair is suitable for ADSL. A n e w proposal called DSL-lite suggests a splitter-less ADSL. T h e objective is to define a service t h a t m a k e s installation easy, w i t h o u t r e q u i r i n g a t e c h n i c i a n . DSL-lite will n o t offer t h e full ADSL speeds, b u t will o p e r a t e at s p e e d s of u p to 1.5 Mbps downstream. T h e r e are several o t h e r v a r i a n t s of DSL. A n earlier technology, h i g h data rate DSL (HDSL), r e q u i r e s t w o t w i s t e d pairs, capable of 1.5 M b p s in e a c h di­ rection. (HDSL u s e s b a s e b a n d signaling.) Rate-adaptive DSL (RADSL) m o d e m s test t h e l i n e at start-up a n d a d a p t t h e i r o p e r a t i n g s p e e d to t h e fastest t h e l i n e c a n h a n d l e . Very high data rate DSL (VDSL) m o d e m s o p e r a t e at data r a t e s u p to 50 Mbps over short twisted pair c o n n e c t i o n s . A l t h o u g h t h e basic DSL t e c h n o l o g y w a s d e v e l o p e d at Bellcore 15 y e a r s ago, it w a s ignored b y t h e t e l e p h o n e i n d u s t r y u n t i l t h e e m e r g e n c e of two competitive forces. Competitive local exchange carriers or CLECs c a m e into existence w i t h t h e passage of t h e 1996 C o m m u n i c a t i o n s Act a n d b e g a n to offer DSL service. More i m p o r t a n t w a s t h e t h r e a t p o s e d b y cable TV c o m p a n i e s t h a t b e g a n to offer I n t e r n e t access. In reaction, t h e i n c u m b e n t t e l e p h o n e c o m p a n i e s

Circuit-Switched Networks

238

are n o w selling ADSL, p r i c e d aggressively to u n d e r c u t b o t h cable TV a n d CLEC offerings. T h e ADSL m a r k e t s e g m e n t a t i o n t h a t t e l e p h o n e c o m p a n i e s are a t t e m p t i n g is discussed in C h a p t e r 10.

V34 and V.90 Modems Most subscribers gain I n t e r n e t access from t h e i r h o m e s via analog voice m o d e m s t h a t t r a n s m i t t h e i r data over t h e analog local loop. T h e b a n d w i d t h of this loop is l i m i t e d to 3 k H z . At t h e subscriber end, t h e m o d e m c o n v e r t s b i n a r y data into a n analog signal. At t h e local c e n t r a l office, t h e analog signal is s a m p l e d a n d e n c o d e d into a 64 Kbps digital signal. T h i s analog-to-digital con­ version (ADC) i n t r o d u c e s q u a n t i z a t i o n noise a n d limits t h e b i n a r y data rate to about 30 Kbps. T h e 64 Kbps digital signal g e n e r a t e d at t h e local c e n t r a l office is s e n t t h r o u g h t h e t e l e p h o n e n e t w o r k , again c o n v e r t e d into a n analog signal, t r a n s ­ m i t t e d over a n o t h e r local loop, r e c e i v e d b y t h e s e r v e r m o d e m , w h e r e t h e original b i n a r y data s t r e a m is recovered. T h e a r r a n g e m e n t is s h o w n in Fig­ u r e 5.25. T h e ITU s t a n d a r d for this m o d e m is called V.34. T h e s y m m e t r i c a r r a n g e m e n t of t h e figure i m p l i e s t h a t d o w n s t r e a m traffic (from server to subscriber) is also l i m i t e d b y t h e ADC c o n v e r s i o n at t h e server's central office, so t h a t d o w n s t r e a m traffic is also l i m i t e d to 30 Kbps. ISP servers today, however, b y p a s s this ADC conversion, a n d t h e i r m o d e m s g e n e r a t e digital signals at 64 Kbps t h a t r e a c h t h e subscriber's c e n t r a l office. T h e digitalto-analog (DAC) c o n v e r s i o n t h a t occurs t h e r e is lossless, a n d so subscribers c a n receive 64 Kbps in t h e d o w n s t r e a m direction (although u p s t r e a m traffic is still l i m i t e d to 30 Kbps). In actual fact n o n l i n e a r i t i e s a n d noise i n t r o d u c e d in t h e DAC at t h e subscriber's local office limits t h e d o w n s t r e a m s p e e d to 56 Kbps.

Local loop ADC Analog Data

DAC r modem

5.25 ^J^^^

Local central office

Local central office

Server modem

The arrangement for the V.34 Kbps modem. The ADC conversion limits speed to 30 Kbps. In the V.90 m o d e m the ADC conversion at the server end is avoided, permitting a downstream speed of 56 Kbps.

5.6

Intelligent Networks

T h e n e w s t a n d a r d is k n o w n as V.90. A m o n g t h e i n t e r e s t i n g features of this s t a n d a r d are protocols t h a t d e t e r m i n e w h e t h e r t h e subscriber m o d e m a n d t h e local loop are able to s u p p o r t t h e Kbps rate. If t h e y c a n n o t , t h e s e r v e r m o d e m r e v e r t s to a V.34 m o d e m .

5.6

INTELLIGENT NETWORKS In o u r historical s u m m a r y in section 1.1.1 w e n o t e d t h a t m o d e r n switches in t h e t e l e p h o n e n e t w o r k s are p r o g r a m m a b l e c o m p u t e r s , w h i c h m a k e s t h e m v e r y flexible. By s e n d i n g i n s t r u c t i o n s to a switch, o n e c a n modify its configuration. T h i s contrasts w i t h t h e earlier e l e c t r o m e c h a n i c a l switches, w h o s e functions w e r e built into t h e i r h a r d w a r e . In m o d e r n switches, t h e control is s e p a r a t e d from t h e h a r d w a r e t h a t executes t h e e l e m e n t a r y switching o p e r a t i o n s . T h i s s e p a r a t i o n of control a n d b a s i c o p e r a t i o n s is also p r e s e n t in t h e o t h e r n e t w o r k e l e m e n t s . As a result, t h e s e e l e m e n t s , too, are p r o g r a m m a b l e . T h e s e p a r a t i o n e n a b l e s t e l e p h o n e c o m p a n i e s to d e v e l o p t h e i r o w n services a n d i m p l e m e n t t h e m o n switches a n d o t h e r n e t w o r k e l e m e n t s o b t a i n e d from different e q u i p m e n t v e n d o r s . Intelligent Network or IN is t h e n a m e given to this n e t w o r k of p r o g r a m m a b l e e l e m e n t s , organized to facilitate i n t r o d u c t i o n of n e w services. I n this section w e explain a n a r c h i t e c t u r e m o d e l for intelligent n e t w o r k s p r o p o s e d b y t e l e c o m ­ m u n i c a t i o n e n g i n e e r s . Since 1985, t h e IN c o n c e p t h a s evolved t h r o u g h several p r o p o s e d a r c h i t e c t u r e s . A n u m b e r of t e l e p h o n e c o m p a n i e s h a v e i m p l e m e n t e d t h e i r o w n version of IN. S o m e wireless n e t w o r k o p e r a t o r s are also i m p l e m e n t ­ ing IN for mobile subscribers. Finally, t h e capability to i m p l e m e n t n e w services t h a t IN offers to a t e l e p h o n e c o m p a n y c a n in p a r t b e d e l e g a t e d to c u s t o m e r s w h o c a n u s e t h e capability to design t h e i r o w n services. In section 5.6.1 w e e x a m i n e services t h a t t h e t e l e p h o n e c o m p a n i e s provide. We explain h o w t h e provision of a service c a n b e d e c o m p o s e d into a s e q u e n c e of b a s i c steps. T h a t is, service provision c a n b e v i e w e d as a script, or p r o g r a m , w h o s e e l e m e n t a r y steps are o p e r a t i o n s p e r f o r m e d b y t h e n e t w o r k e l e m e n t s . In section 5.6.2 w e e x a m i n e a n a r c h i t e c t u r e m o d e l for intelligent n e t w o r k s . I n section 5.6.3 w e list t h e basic actions t h e n e t w o r k p e r f o r m s w h e n it i m p l e m e n t s a service.

5.6.1

Service Examples Figure 5.26 illustrates t h e plain old t e l e p h o n e service or POTS. POTS is t h e basic t e l e p h o n e call service. W h e n t h e n e t w o r k i m p l e m e n t s a t e l e p h o n e call, its e l e m e n t s execute t h e s e q u e n c e of s t e p s listed in t h e figure. I n t h e old

239

Circuit-Switched Networks

240

Steps in POTS:

5.26 GURE

1. Determine route from dialed digits 2. Create and join legs 1, 2, 3 3. Verify called party is available 4. Conversation between A and Β 5. Detect termination by participant set "on-hook" 6. Free legs 1, 2, 3

The figure notes the basic steps necessary to provide POTS or plain old telephone service.

t e l e p h o n e n e t w o r k , w i t h e l e c t r o m e c h a n i c a l switches, t h e call functions of t h e n e t w o r k e l e m e n t s are w i r e d into t h o s e e l e m e n t s . For example, w h e n a u s e r dials a t e l e p h o n e n u m b e r , t h e successive digits g e n e r a t e a string of i m p u l s e s t h a t activate a s e q u e n c e of rotary switches in t h e central offices of t h e t e l e p h o n e c o m p a n y . W h e n a n electrical c u r r e n t is s e n t to a n on-hook p h o n e , this c u r r e n t activates t h e bell. T h i s a r r a n g e m e n t of t h e h a r d w a r e is inflexible. For instance, a c u s t o m e r c a n n o t instruct t h e switches to block calls dialed from specific n u m b e r s or to forward a call to a n o t h e r n u m b e r at certain h o u r s of t h e day. Figures 5.27 a n d 5.28 illustrate t h e o p e r a t i o n s p e r f o r m e d b y t h e m o d e r n n e t w o r k w h e n it i m p l e m e n t s t h e call forwarding service. A c u s t o m e r u s i n g call forwarding c a n i n s t r u c t t h e t e l e p h o n e n e t w o r k to forward a n i n c o m i n g call to a n o t h e r t e l e p h o n e set w h e n t h e n o r m a l p h o n e is n o t picked u p after it rings (say) t h r e e t i m e s . Figure 5.27 s h o w s two t e l e p h o n e sets A a n d Β c o n n e c t e d via t w o switches 1 a n d 2. T h e c u s t o m e r at set A calls set B. T h e c u s t o m e r at Β h a s i n s t r u c t e d t h e t e l e p h o n e n e t w o r k to forward a call to s o m e set B' after t h r e e rings. T h i s i n s t r u c t i o n is stored in a database of switch 2 to w h i c h set Β is attached. Moreover, t h e p h o n e n u m b e r of set Β is stored in a trigger table. W h e n a call for a t e l e p h o n e set r e a c h e s switch 2, t h e switch c h e c k s its trigger table. If t h a t set is n o t in t h e trigger table, t h e n t h e call is h a n d l e d in t h e s t a n d a r d way. If t h e set is in t h e trigger table, t h e n t h e switch places a r e q u e s t to its database. T h e database r e s p o n d s to t h e r e q u e s t b y providing t h e i n s t r u c t i o n s t h a t t h e switch m u s t follow to h a n d l e t h e call. T h u s , w h e n t h e call for set Β r e a c h e s switch 2, t h e switch is triggered, a n d it r e q u e s t s t h e i n s t r u c t i o n s from its database. Note t h a t t h e t e l e p h o n e n e t w o r k c a n i m p l e m e n t s u c h services b e c a u s e t h e controls are s e p a r a t e d from t h e actual o p e r a t i o n s of t h e switches.

5.6

Intelligent Networks

241 Switch 1

Legl

Switch 2

Leg 2

Trigger

B

1. A dials address Β 2. Switch 1 creates and joins legs 1, 2 3. Address Β sets off trigger at switch 2 Switch 2

Database Request Β

Response Β Β

5.27 FIGURE

5.6.2

Instructions

Call forwarding: w h e n a call for Β reaches switch 2, the switch searches a database for instructions.

Intelligent Network Architecture T h e Intelligent Network Architecture (INA) is a m o d e l for organizing t h e p r o g r a m ­ mable network elements and the communication between those elements. The objectives of t h a t m o d e l are in p a r t t h e s a m e as t h o s e of t h e OSI r e f e r e n c e m o d e l

Leg 3

4. Database response Β: if no "off hook" in three rings, forward call to B' 5. Switch 2 waits, eventually frees leg 3, creates and joins leg 3'

5.28 FIGURE

Call forwarding, continued: if the set Β is not picked u p after three rings, the call is forwarded to B'.

242

Circuit-Switched Networks

for c o m p u t e r n e t w o r k s : d e c o m p o s i t i o n a n d standardization. INA is m o r e com­ plex in t h a t it involves n o t only c o m m u n i c a t i o n protocols a n d data e l e m e n t s as OSI does, b u t o t h e r n e t w o r k resources, i n c l u d i n g h a r d w a r e e l e m e n t s . O n t h e o t h e r h a n d , t h e OSI m o d e l is quite abstract, w h e r e a s t h e INA a r c h i t e c t u r e is specifically d e s i g n e d for t e l e p h o n e n e t w o r k s . As we discussed above, t h e m a i n feature of a n intelligent n e t w o r k is t h a t t h e control functions a n d t h e network resources are s e p a r a t e d . T h e n e t w o r k h a s t r a n s ­ mission a n d o t h e r resources, including subscriber lines, t r u n k lines, switch ports, databases, a n d voice recorders. T h e control functions are call-control functions a n d resource-control functions. E x a m p l e s of t h e s e functions i n c l u d e c o n n e c t i n g t h r e e n e t w o r k u s e r s in a c o n f e r e n c e call, playing a recording, a n d collecting digits dialed b y a user. Tb i m p l e m e n t a control function, t h e n e t w o r k executes a s e q u e n c e of a t o m i c o p e r a t i o n s called functional components. INA classifies t h e n e t w o r k e l e m e n t s into t h r e e t y p e s : service switching points (SSPs), service control points (SCPs), a n d intelligent peripherals (IPs). (See Figure 5.29.) T h e s e e l e m e n t s exchange information over t h e common channel signaling (CCS) packet-switched n e t w o r k , u s i n g t h e signaling system 7 (SS7) protocol. INA calls t h e CCS n e t w o r k switches signal transfer points (STPs).

5.6

Intelligent Networks

T h e figure s h o w s t h r e e t e l e p h o n e sets a n d o n e IP a t t a c h e d to t h e n e t w o r k of SSPs. Regular l i n e s d e n o t e t h e links u s e d b y t h e voice signals. T h e b o l d l i n e s r e p r e s e n t links of t h e CCS n e t w o r k . T h e SCPs c o n t a i n databases a n d i n s t r u c t i o n s for special applications. T h e n e t w o r k e l e m e n t s are capable of p e r f o r m i n g different functions. T h e SSP c a n d e t e c t t h e n e e d for IN service b a s e d o n a n originating line trigger s u c h as off-hook, triggers a p p l y i n g to all calls s u c h as for 800 or 911 calls, a n d t e r m i n a t i n g line triggers s u c h as for call forwarding u s e r s . W h e n s u c h a trigger is activated, t h e SSP s e n d s a r e q u e s t to t h e SCP for i n s t r u c t i o n s . T h e SSP c a n identify, monitor, a n d allocate t r a n s m i s s i o n r e s o u r c e s (legs) c o n n e c t e d to it. A n IP c a n identify, monitor, a n d allocate n o n t r a n s m i s s i o n r e s o u r c e s c o n n e c t e d to it. T h e SCP d e t e c t s IN service r e q u e s t s forwarded b y t h e SSP. It t h e n i n t e r p r e t s t h e r e q u e s t according to a service logic program (SLP). A service logic interpreter (SLI) executes t h e SLP. A given SLI c a n execute m u l t i p l e SLPs c o n c u r r e n t l y . T h e execution of t h e SLP involves i n v o k i n g functional c o m p o n e n t s a n d m o n i t o r i n g r e s o u r c e s . Finally, at c o m p l e t i o n of t h e execution, t h e SLI notifies t h e SSP.

5.63

Functional Components We briefly describe t h e functional c o m p o n e n t s , t h e a t o m i c actions o n n e t w o r k resources, in five t y p e s of n e t w o r k o p e r a t i o n s .

Control of Processing T h e functional c o m p o n e n t s for control of p r o c e s s i n g are of t w o t y p e s : to provide i n s t r u c t i o n s w h e n t h e SSP asks t h e SCP to take control of t h e call processing a n d to effect t h e release of control w h e n t h e SCP r e t u r n s t h e control to t h e SSP after servicing t h e r e q u e s t .

Connection Request A c o n n e c t i o n r e q u e s t involves t h e following functional c o m p o n e n t s : creating a leg b e t w e e n a n SSP a n d a n o t h e r n e t w o r k e l e m e n t , j o i n i n g a leg to a n ongoing call, splitting a leg from a n ongoing call, a n d freeing a leg to release t h e resource.

User Interaction Request Two t y p e s of functional c o m p o n e n t s are i n v o k e d in u s e r i n t e r a c t i o n s : s e n d i n g information s u c h as a p r e r e c o r d e d a n n o u n c e m e n t to a call p a r t i c i p a n t a n d receiving specific information s u c h as dialed digits from a call participant.

243

Circuit-Switched Networks

244 Network Resource Status Request

N e t w o r k r e s o u r c e status r e q u e s t s are u s e d b y t h e SLP in processing s o m e call control. Monitoring is a functional c o m p o n e n t t h a t i n s t r u c t s t h e SSP for notification of a particular e v e n t o n a specified leg, s u c h as on-hook, flashhook, a n d off-hook.

Network Information Revision Request N e t w o r k information revision r e q u e s t s e n a b l e t h e SLP to c h a n g e t h e data stored in t h e SSP tables. In s u m m a r y , INA is a c u l m i n a t i o n of a long d e v e l o p m e n t in w h i c h net­ w o r k e l e m e n t functions or o p e r a t i o n s are s e p a r a t e d from t h e control of t h o s e functions. T h i s s e p a r a t i o n p e r m i t s t h e c r e a t i o n of n e w services as p r o g r a m m a ­ ble s e q u e n c e s of functional c o m p o n e n t s . Sophisticated c u s t o m e r s c a n p r o g r a m t h e s e s e q u e n c e s b y t h e m s e l v e s . A v e r y i m p o r t a n t e x a m p l e is 800 n u m b e r ser­ vices. C o m p a n i e s , s u c h as credit card a n d direct o r d e r c o m p a n i e s t h a t provide direct c u s t o m e r services over t h e 800 n u m b e r p h o n e , c a n r o u t e c u s t o m e r calls to different p a r t s of t h e c o u n t r y or to different operators d e p e n d i n g o n t h e t i m e of day, t h e subscriber's location, a n d o t h e r i n f o r m a t i o n p r o v i d e d b y t h e subscriber via t h e t e l e p h o n e k e y p a d .

5.7

CATV CATV, originally Community Antenna TV, n o w refers to a n y cable or h y b r i d optical fiber a n d cable (HFC) s y s t e m u s e d to deliver video signals to a c o m m u ­ nity. T h e s t a n d a r d CATV distribution s y s t e m is illustrated in Figure 5.30. T h e h e a d station distributes analog video p r o g r a m s . Several p r o g r a m s are frequencydivision multiplexed at t h e h e a d e n d over a coaxial cable n e t w o r k . Each pro­ g r a m occupies a 6-MHz c h a n n e l , w i t h t h e s p e c t r u m b e t w e e n 50 a n d 550 M H z a c c o m m o d a t i n g u p to 80 c h a n n e l s . Every subscriber receives t h e s a m e pro­ grams. T h r e e i n n o v a t i o n s h a v e t r a n s f o r m e d CATV from a video distribution sys­ t e m to o n e t h a t c a n provide interactive, i n t e g r a t e d services. T h e first innova­ tion, discussed in section 5.7.1, u p g r a d e s CATV into a two-way c o m m u n i c a t i o n s y s t e m . T h e s e c o n d i n t r o d u c e s link l a y e r functions t h a t provide u s e r s access to a s h a r e d data link. It is described in 5.7.2. T h e third consists of digital c o m p r e s ­ sion s c h e m e s t h a t e n a b l e video to b e t r a n s m i t t e d at relatively l o w s p e e d s . In section 5.7.3 w e discuss t h r e e services t h a t CATV operators m a y offer: I n t e r n e t

- Β 1

Amplifier

Video signals

Coax Head end

5.30 FIGURE

In the standard CATV network, the head station distributes the video programs i l cable network. Amplifiers are inserted into the distribution cable in order to boost signal power. All subscribers receive the same programs.

o

v

e

r

a

c o a x

a

access, video o n d e m a n d , a n d t e l e p h o n y . In section 5.7.4 w e elaborate o n t h e MPEG s t a n d a r d s u s e d for video c o m p r e s s i o n .

57.1

Layout T h e physical l a y o u t of a n u p g r a d e d n e t w o r k is s k e t c h e d in Figure 5.31. T h e n e t w o r k b e t w e e n t h e h e a d stations a n d t h e u s e r s consists of optical fiber t e r m i n a t i n g at a fiber n o d e to w h i c h is a t t a c h e d a local coaxial n e t w o r k t h a t c o n n e c t s to 500 h o m e s . T h e h e a d stations of t h e service p r o v i d e r s h a v e access to video servers, w e b servers, a n d t h e I n t e r n e t , via a b a c k b o n e n e t w o r k t h a t is n o t s h o w n . T h e total b a n d w i d t h is i n c r e a s e d to 750 MHz, as s h o w n in Figure 5.32. T h e m o r e i m p o r t a n t difference, s e e n in Figure 5.31, is t h a t signals c a n travel in b o t h directions. T h e fiber n o d e c o n v e r t s t h e d o w n s t r e a m optical signal

5.31 GURΕ

Reverse channel

6 (US) or 8 (Europe) MHz analog channels (VSB)

5

50

42

6-MHz digital channels QPSK upstream: 10.8 Mbps QAM-64 downstream: 34 Mbps 16-VSB downstream: 42 Mbps

550

550

750

The network connects the head end stations of the service providers to the l i p c t . The downstream signal is carried over an optical fiber to the fiber node, where it is converted into an electrical signal and distributed over coaxial cable. User upstream signals use the same physical network. u

s

e

r

e c

u

m

n

Circuit-Switched Networks

246

V

Video signals Control signals

Optical —· Electrical Electrical—Optical

Coax Fiber

Amplifier

532

C

Head end

The network connects the head stations of the service providers to the user equipment. Signals are sent downstream from head stations to users, and upstream to head stations.

originating at t h e h e a d e n d into a n electrical signal, a n d t h e u p s t r e a m electrical signal originating at t h e u s e r s into a n optical signal. Figure 5.33 gives a n alternative distribution t e c h n o l o g y in w h i c h all or p a r t of t h e distribution s y s t e m is wireless. In t h e wireless cable s y s t e m , sub­ scribers directly access t h e signal b r o a d c a s t from t h e h e a d e n d station. In t h e

Κ

WLC Head end

2-3 GHz, < 40 km, 33 6-MHz analog channels, or 28-30 GHz, < 25 km, 100W TWT, 49 20-MHz FM analog HFWL

28 or 40 GHz 1W, solid state

Κ Head end

*

Fiber

Downstream digital; 2-3 Gbps

Upstream 300-600 Mbps 700

5.33 FIGURE

500 homes

Fiber node

800

1,000 MHz

The wireless cable (WLC) system replaces the current distribution system. The hybrid/fiber wireless (HFWL) system replaces the local coaxial distribution.

5.7

CATV

247 h y b r i d / f i b e r wireless s y s t e m , a digital video signal is s e n t to t h e c u r b over opti­ cal fiber (as in Figure 5.31), a n d t h e local coaxial distribution s y s t e m is r e p l a c e d b y a local wireless s y s t e m . T h e wireless p o r t i o n of t h e s e n e t w o r k s w o u l d ex­ t e n d over short distances. T h e s e s y s t e m s m a y b e less e x p e n s i v e t h a n cable w h e n t h e r e is a h i g h geographical c o n c e n t r a t i o n of u s e r s .

57.2

CATV Layered Network Figure 5.34 s k e t c h e s a plausible t h r e e - l a y e r d e c o m p o s i t i o n of t h e functions of t h e CATV n e t w o r k . T h e d e c o m p o s i t i o n is different for d o w n s t r e a m a n d u p s t r e a m traffic. We b e g i n w i t h t h e physical layer. Figure 5.32 s h o w s p r o p o s e d u s e s of t h e available f r e q u e n c y s p e c t r u m of 5 to 750 MHz. T h e d o w n s t r e a m a n d u p s t r e a m signals o c c u p y different f r e q u e n c y b a n d s . T h i s s e p a r a t i o n of f r e q u e n c y b a n d s p r e v e n t s t h e amplifiers from amplifying t h e i r o w n o u t p u t , w h i c h w o u l d l e a d to saturation of t h e amplifiers a n d to oscillation. T h e d o w n s t r e a m signals o c c u p y t h e s p e c t r u m from 50 to 750 MHz. Con­ v e n t i o n a l analog b r o a d c a s t s t h a t c a n b e r e c e i v e d b y existing television sets o c c u p y 6-MHz c h a n n e l s b e t w e e n 50 a n d 550 MHz. T h e s p e c t r u m b e t w e e n 550 a n d 750 M H z m a y c a r r y digital MPEG-2 p r o g r a m s , data s t r e a m s , a n d d o w n ­ s t r e a m t e l e p h o n y . Using QAM-64 or 16-VSB m o d u l a t o r s , e a c h 6-MHz analog c h a n n e l is c o n v e r t e d into a digital l i n k w i t h a bit rate of 27 to 38 Mbps. Such a link c a n b e u s e d to c a r r y 6 to 10 MPEG-2 p r o g r a m s at r a t e s of 3.5 Mbps, or to t r a n s p o r t digital data to users. T h e MPEG p r o g r a m s m a y b e d e c o d e d b y settop boxes. Transmission of u s e r data r e q u i r e s cable m o d e m s . T h e d o w n s t r e a m

Downstream signal Ν

Upstream signals

Circuitswitched Multiple access

LLC/MAC Ρ

QAMorVSB

QPSK

HFC

5.34 GURΕ

Plausible three-layer decomposition of network functions. The downstream signals are carried b y a circuit-switched network. The upstream signals are sent using a multiple access scheme.

248

Circuit-Switched Networks

signals are all broadcast. As s h o w n in Figure 5.34, t h e d o w n s t r e a m signals are circuit-switched. U p s t r e a m signals from u s e r s o c c u p y t h e 5 to 42 MHz s p e c t r u m . T h i s s p e c t r u m is usually divided into 2-MHz c h a n n e l s . Because t h e n e t w o r k h a s a t r e e a n d b r a n c h structure, t h e t r a n s m i s s i o n p a t h from u s e r s to t h e h e a d e n d is shared. T h e effect is t h a t t h e signal r e c e i v e d at t h e h e a d e n d is t h e s u m of t h e u s e r signals, so a MAC protocol is n e e d e d to provide collisionfree access, as is described in t h e next p a r a g r a p h . Moreover, t h e t r e e a n d p a t h s t r u c t u r e c a u s e s t h e addition of noise from all t h e users. Both n a r r o w - b a n d noise from e l e c t r o m a g n e t i c radiation a n d nonlinearities, a n d i m p u l s e noise from appliances, are p r e s e n t . T h e bits are m o d u l a t e d u s i n g q u a d r a t u r e p h a s e shift k e y i n g (QPSK) a n d forward error-correction is used, yielding a s h a r e d link w i t h a bit rate of a b o u t 3 Mbps for e a c h 2-MHz c h a n n e l . T h e a d v a n t a g e of QPSK is t h a t b y choosing t h e carrier frequency of t h e sine w a v e s judiciously, t h e n e t w o r k e n g i n e e r c a n m a k e sure t h a t t h e m o d u l a t e d bit s t r e a m does n o t interfere w i t h t h e video signals. A n o t h e r advantage is t h a t this m o d u l a t i o n s c h e m e r e q u i r e s o n l y a small r a n g e of frequencies to t r a n s m i t t h e bits. O n l y a small r a n g e of frequencies is g e n e r a t e d b e c a u s e t h e m o d u l a t e d signal c h a n g e s only o n c e e v e r y η bits. For transfer of data b e t w e e n u s e r s a n d t h e h e a d end, cable o p e r a t o r s (called Multiple Service Operators or MSOs) offer a s h a r e d 3-Mbps (over a 2-MHz c h a n n e l ) u p s t r e a m link a n d a b r o a d c a s t 38-Mbps (over a 6-MHz c h a n n e l ) d o w n s t r e a m link. T h e r e are several p r o p r i e t a r y MAC protocols for access to t h e s h a r e d link. A n e m e r g i n g standard, called Data-Over-Cable Service Interface Specifications (DOCSIS) is b e i n g d e v e l o p e d b y t h e M u l t i m e d i a Cable N e t w o r k Systems (MCNS) c o n s o r t i u m . T h e goal is to t r a n s p a r e n t l y t r a n s m i t IP traffic b e t w e e n t h e u s e r cable modem a n d t h e cable modem termination system or CMTS at t h e h e a d e n d . We briefly describe DOCSIS. U p s t r e a m frames a n d d o w n s t r e a m frames are s y n c h r o n i z e d (as in t h e PON systems), a n d cable m o d e m s u s e r a n g i n g m e a s u r e m e n t s so t h a t signals from different u s e r s do n o t overlap. T h e frames are divided into minislots, s o m e of w h i c h are k e p t aside for u s e r r e s e r v a t i o n r e q u e s t s . T h e r e q u e s t i n c l u d e s t h e m o d e m ID a n d t h e a m o u n t of b a n d w i d t h r e q u e s t e d . W h e n a u s e r m a k e s a request, t h e CMTS m a y g r a n t t h e u s e r a c e r t a i n n u m b e r of minislots in t h e n e x t frame. T h e grants are carried in a d o w n s t r e a m frame. However, m o r e t h a n o n e cable m o d e m m a y r e q u e s t at t h e s a m e time, resulting in a collision. T h e colliding m o d e m s l e a r n a b o u t this, b e c a u s e t h e y do n o t receive a n y grants. T h e y m u s t b a c k off for a r a n d o m a m o u n t of t i m e before m a k i n g a n o t h e r r e q u e s t . T h e s t a n d a r d specifies h o w u s e r E t h e r n e t p a c k e t s or ATM cells are to b e framed.

5.7

CATV

249 Cable modem

CMTS

IP Data link

PHY

Forward

DP Cable MAC

Cable MAC

PHY

PHY

To backbone network

5.35 FIGURE

Cable network transmission

Bridge

IEEE 802.3 lOBaseT To user computer

The CMTS can forward user layer 2 or layer 3 packets to the backbone network.

T h e cable m o d e m is c o n n e c t e d via t h e CMTS to a b a c k b o n e n e t w o r k . T h e CMTS m a y serve as a l a y e r 2 b r i d g e or provide n e t w o r k l a y e r connectivity, as s h o w n in Figure 5.35.

573

Services over CATV T h e large b a n d w i d t h in t h e d o w n s t r e a m b r o a d c a s t direction c a n s u p p o r t n o t o n l y 80 traditional, 6-MHz TV c h a n n e l s , b u t also several h u n d r e d MPEG-2 c h a n n e l s . T h i s possibility is realized as a service called video on demand or video dial-tone. Subscribers b r o w s e t h r o u g h a large collection of video p r o g r a m s a n d r e q u e s t a program, u s i n g t h e i r set-top boxes. T h e h e a d e n d t r a n s p o r t s t h e r e q u e s t e d MPEG digital s t r e a m over a n available c h a n n e l . T h e set-top box d e m o d u l a t e s a n d d e c o m p r e s s e s t h e r e c e i v e d bit s t r e a m a n d g e n e r a t e s t h e NTSC or HDTV signal for display o n t h e TV set. T h e initial e n t h u s i a s m for this service w a s greatly r e d u c e d following a n unsuccessful field trial b y T i m e Warner. T h e lack of success m a y h a v e b e e n d u e to t h e h i g h cost of t h e set-top boxes, or t h e n e a r - s u b s t i t u t e s ( s u c h as video rentals) t h a t are available. Video o n d e m a n d survives in a n a t t e n u a t e d form in hotels a n d as pay-per-view. A later d e v e l o p m e n t h a s b e e n t h e u s e of CATV for I n t e r n e t access. Inter­ n e t service providers, t o g e t h e r w i t h cable operators, offer subscribers I n t e r n e t access over a s h a r e d 3-Mbps u p s t r e a m link a n d a 38-Mbps d o w n s t r e a m broad­ cast link. Typically, 10 subscribers s h a r e t h e r e s o u r c e s at a n y t i m e . Providers e n h a n c e p e r f o r m a n c e b y Web c a c h i n g a n d offer directory a n d o t h e r services. Subscribers m u s t p u r c h a s e a cable m o d e m a n d u s u a l l y p a y a m o n t h l y flat rate. T h e growth of subscribers to this service h a s b e e n i m p r e s s i v e . However, b e ­ cause t h e links are shared, a few h e a v y u s e r s c a n c a u s e c o n g e s t i o n a n d d e g r a d e

Circuit-Switched Networks

250

service for all. This s e e m s to b e a c o m m o n o c c u r r e n c e , a n d service providers are i m p o s i n g restrictions o n usage. AT&T's acquisition of TCI is fueling s p e c u l a t i o n t h a t subscribers will soon b e offered t e l e p h o n e service over CATV u s i n g t h e next g e n e r a t i o n ca­ ble m o d e m s . T h e attraction for AT&T m i g h t b e its ability to offer lucrative long-distance t e l e p h o n e service w i t h o u t p a y i n g access charges to t h e local telephone company. Because t h e u p s t r e a m c h a n n e l is so noisy, a t e c h n i c i a n is n e e d e d to install a cable m o d e m . This m a k e s d e p l o y m e n t of cable m o d e m s expensive. Wide d e p l o y m e n t of cable m o d e m s will r e q u i r e t e c h n i c a l a d v a n c e s t h a t m a k e t h e m as easy to install as t e l e p h o n e m o d e m s .

5.7.4

MPEG T h e Moving Pictures Expert G r o u p h a s defined a n u m b e r of s t a n d a r d s for video c o m p r e s s i o n . T h e MPEG c o m p r e s s i o n algorithms u s e t h e r e d u n d a n c y w i t h i n e a c h frame a n d t h e similarity of successive frames. We give a brief historical a c c o u n t of t h e algorithms, starting w i t h JPEG. JPEG (Joint Photographic Experts Group) is t h e ITU protocol for coding c o n t i n u o u s - t o n e still images. JPEG is u s e d to s h o w i m a g e s t h a t do n o t move, as in h o m e s h o p p i n g TV ads. JPEG e m p l o y s a spatial digital cosine t r a n s f o r m or DCT c o m p r e s s i o n s c h e m e . H.261 is t h e ITU s t a n d a r d for v i d e o c o n f e r e n c i n g applications at ISDN trans­ mission rates. T h e coding s c h e m e u s e s DCT, DPCM (Differential Pulse Code Modulation), a n d m o t i o n c o m p e n s a t i o n . T h e m o t i o n c o m p e n s a t i o n a l g o r i t h m a s s u m e s relatively slow a n d restricted m o v e m e n t , so q u i c k e r m o v e m e n t ap­ p e a r s j e r k y . T h e first frame is a n I (intra-frame) frame; e a c h s u b s e q u e n t frame u s e s inter-frame prediction (P frames), u s i n g only t h e n e a r e s t frame for pre­ diction. T h e q u a n t i z a t i o n step-size is adjusted to achieve a target bit rate. T h e Motion Picture Experts G r o u p (MPEG) w a s founded u n d e r ISO. MPEG-1 defines a coding s c h e m e u s i n g DTC a n d m o t i o n c o m p e n s a t i o n similar to H.261. It is targeted for u s e b y progressively s c a n n e d m e d i a , s u c h as CDROM. MPEG-1 provides frame-based r a n d o m access of video, fast forward/fast reverse searches, reverse playback of video, a n d ability to edit. In addition to t h e I a n d Ρ frames, MPEG-1 i n t r o d u c e d B-frames, w h i c h u s e m o t i o n picture prediction b a s e d o n t h e two n e a r e s t already coded frames. Progressive s c a n n i n g m e a n s t h e first line is displayed first, t h e n t h e sec­ ond, a n d so on. MPEG-2 is a s u p e r s e t of MPEG-1 w i t h special c o n s i d e r a t i o n of interlaced s c a n n i n g u s e d in b r o a d c a s t TV, a n d scalable video e x t e n s i o n s for

5.7

CATV

251

5.36 oweemiiii

The motion-compensation algorithm compares successive frames. A frame is divided into blocks, and the corresponding blocks in the next frame are moved until they match those in the previous frame best. The algorithm transmits the block displacement vectors and the residual differences b e t w e e n the blocks.

digital TV a n d HDTV. Scalability is a c h i e v e d b y m u l t i r e s o l u t i o n r e p r e s e n t a ­ tion t h a t p e r m i t s d o w n s c a l i n g a n i n p u t video signal into a l o w e r r e s o l u t i o n video. Motion c o m p e n s a t i o n in MPEG is illustrated in Figure 5.36. T h e figure s h o w s two successive frames of a video. T h e a l g o r i t h m d e c o m p o s e s frame η into blocks. T h e algorithm t h e n e x a m i n e s frame η + 1 a n d finds t h e t r a n s l a t i o n of t h e blocks of t h a t frame t h a t m a k e s t h e m m o s t similar to t h e c o r r e s p o n d i n g blocks of t h e p r e v i o u s frame. I n s t e a d of t r a n s m i t t i n g all t h e blocks, t h e algo­ r i t h m t r a n s m i t s t h e d i s p l a c e m e n t vectors a n d t h e differences b e t w e e n a block in a frame a n d t h e c o r r e s p o n d i n g b l o c k in t h e n e x t frame after translation. T h e frames t h a t c o n t a i n this m o t i o n - c o m p e n s a t i o n i n f o r m a t i o n are called Ρ frames. Finally, a n MPEG a l g o r i t h m c a n u s e i n t e r p o l a t i o n frames, called Β frames. T h e Β frames p e r f o r m a n i n t e r p o l a t i o n b e t w e e n Ρ frames. Figure 5.37 s h o w s t h e s e q u e n c e of frames p r o d u c e d b y MPEG-2. Refresh frames are s e n t twice e v e r y s e c o n d to avoid e r r o r propagation. T h e n u m b e r of Β a n d Ρ frames c a n b e selected to p r o d u c e different qualities a n d resulting bit rates. MPEG-4 a n d its p r e d e c e s s o r H . 2 6 3 / L assist low bit rate (64 Kbps) trans­ mission w i t h m u l t i m e d i a data access tools to facilitate indexing, d o w n l o a d i n g , a n d q u e r y i n g ; to efficiently c o m b i n e s y n t h e t i c s c e n e s ; a n d to h a n d l e m u l t i p l e c o n c u r r e n t data s t r e a m s . H.263 is u s e d in video s t r e a m i n g .

Circuit-Switched Networks

252 I

Ρ

Ρ

I I

Β

Ρ

ΙΙΠΙΙΠΙΙΠΙΙΠΙΙΙΙΙ

I = Intraframe compression: DCT + run-length encoding of DCT coefficients Ρ = Predictive compression: motion compensation Β = Forward/backward compression: interpolation between previous and future frames 537 FIGURE

5.8

S e q u e n c e p r o d u c e d b y a n MPEG-2 algorithm. T h i s s e q u e n c e c o n s i s t s o f refresh f r a m e s (I), m o t i o n - c o m p e n s a t i o n f r a m e s (P), a n d i n t e r p o l a t i o n f r a m e s (B).

SUMMARY T h e b e a r e r service offered b y circuit-switched n e t w o r k s is fixed bit r a t e endto-end c o n n e c t i o n s w i t h a delay e q u a l to t h e p r o p a g a t i o n delay. T h i s b e a r e r service is well suited for c o n s t a n t bit rate traffic. T h e service c a n of c o u r s e b e u s e d to s u p p o r t b o t h variable bit rate a n d m e s s a g e traffic. However, u s e d in t h a t w a y t h e utilization of t h e n e t w o r k link capacities m a y b e v e r y low. T h e t e l e p h o n e a n d CATV n e t w o r k s are t h e m o s t i m p o r t a n t n e t w o r k s in t e r m s of n u m b e r of c u s t o m e r s a n d ability to finance n e w i n v e s t m e n t . Recent i n n o v a t i o n s will e n a b l e t h e s e n e t w o r k s to provide n e w services t h a t c a n p r o p e l t h e m well b e y o n d t h e i r traditional m a r k e t s . I n n o v a t i o n s in t h e t e l e p h o n e n e t w o r k h a v e led to (1) h i g h e r b a n d w i d t h links in t h e b a c k b o n e n e t w o r k t h r o u g h SONET a n d WDM; (2) h i g h e r s p e e d local loop u s i n g optical fiber or DSL over twisted pairs; (3) i n t e g r a t e d services, providing a c o m m o n interface for m e s s a g e t r a n s m i s s i o n over packet-switched n e t w o r k s a n d c o n s t a n t bit rate circuit-switched c o n n e c t i o n s ; a n d (4) Intelligent N e t w o r k Architecture, p e r m i t t i n g a v e r y flexible m e a n s to c r e a t e n e w u s e r services. SONET i m p l e m e n t s t h e bit w a y l a y e r of t h e O p e n Data N e t w o r k m o d e l , taking a d v a n t a g e of t h e e c o n o m i e s of scale offered b y optical c o m m u n i c a t i o n . T h e optical local loop, p e r h a p s m o r e expensive t h a n today's c o p p e r wire, will b r i n g to t h e u s e r t h e high-speed access n e c e s s a r y for video applications. Lastly, DSL a n d INA exploit t h e e c o n o m i e s of scope b y i n t e g r a t i n g packet-

5.9

Notes

253 a n d circuit-switched services a n d o t h e r c o m m u n i c a t i o n - r e l a t e d services s u c h as call forwarding. I n n o v a t i o n s in CATV h a v e m a d e possible (1) a m o r e efficient u s e of b a n d ­ w i d t h b y digital c o m p r e s s i o n ; (2) i n c r e a s e d b a n d w i d t h b y u s e of optical fiber for t r a n s m i s s i o n to t h e curb; a n d (3) t r a n s m i s s i o n of i n f o r m a t i o n b y t h e sub­ scriber. With t h e s e i n n o v a t i o n s CATV c a n b e g i n to c o m p e t e for I n t e r n e t access a n d local t e l e p h o n e services. Despite t h e i r innovations, packet-switched n e t w o r k s are b e s t suited for m e s s a g e traffic, a n d circuit-switched n e t w o r k s are b e s t s u i t e d for c o n s t a n t bit rate traffic. Both t y p e s of n e t w o r k are still n o t well s u i t e d for variable bit rate traffic. T h a t traffic is b e s t carried b y ATM n e t w o r k s , w h i c h w e s t u d y in t h e next chapter.

5.9

NOTES T h e S O N E T / S D H framing c o n v e n t i o n s are d e s c r i b e d in [BC89, S92, B95]. A discussion of SONET rings a n d o t h e r a r c h i t e c t u r e s t h a t i m p r o v e n e t w o r k sur­ vivability a p p e a r s in [WL92, W95]. ATM over SONET is specified b y t h e ATM F o r u m [A93]. PPP over SONET is discussed in RFC 1619. PPP p r o v i d e s a stan­ d a r d m e t h o d for t r a n s p o r t i n g multiprotocol d a t a g r a m s over point-to-point links (RFC 1661). I P / P P P / S O N E T is d i s c u s s e d in [MADD98] DWDM is discussed in [CM98a, JSAC98]. T h e AT&T SLC 5 is briefly d e s c r i b e d in [C89]. T h e British T P O N s y s t e m is d e s c r i b e d in [R91]. ΑΡΟΝ is d i s c u s s e d in [ W M 9 3 , ALFV98]. T h e passive p h o t o n i c loop is described in [WL89]. P r o m i s i n g n e w local l o o p t e c h n o l o g i e s t h a t c a n provide high-speed access are d e s c r i b e d in [CM91]. A n informative c o m p a r i s o n of t h e costs of different b r o a d b a n d access t e c h n o l o g i e s (CATV, ADSL, h y b r i d s y s t e m s ) is given in [PB95]. ISDN s t a n d a r d s are d e s c r i b e d in [S92]. See [F95, P95] for a r e c e n t appraisal of t h e m a r k e t p e n e t r a t i o n of ISDN. ANSI s t a n d a r d s for xDSL are available at t h e Web site of t h e ADSL Forum, www.adsl.com. A historical o v e r v i e w of xDSL is p r o v i d e d in [H97]. T h e m o d u ­ lation technologies for xDSL are d i s c u s s e d in [JSAC95]. A n issue of Communications Magazine [CM92] c o n t a i n s good discussions of IN in t h e U n i t e d States a n d several o t h e r c o u n t r i e s . T h e CATV i n d u s t r y is d e v e l o p i n g s t a n d a r d s called Data O v e r Cable Ser­ vice Interface Specification (DOCSIS). MPEG s t a n d a r d s are discussed in [S96]. Physical l a y e r a n d MAC protocols are also b e i n g d e v e l o p e d b y t h e IEEE 802.14

Circuit-Switched Networks

254

w o r k i n g group. For a description of video dial tone, see [CM94] a n d [IN95]. ADSL a n d cable m o d e m s are s e e n as c o m p e t i t i v e technologies, a n d b o t h in­ d u s t r y g r o u p s m a k e v e r y optimistic projections.

5.10

PROBLEMS 1. For a n e t w o r k like t h e o n e in Figure 5.1, let {Cy} d e n o t e link capacities. Suppose CR = {C } is a set of r o u t e calls, a n d say t h a t CR is feasible if (5.2) holds. If t h e r e are Ν r o u t e s in R , t h e n CR is a n N - d i m e n s i o n a l vector. r

(a) Show t h a t t h e set of feasible vectors is convex. (b) Suppose t h a t a c o n n e c t i o n over r o u t e r b r i n g s a r e v e n u e of p p e r u n i t of t i m e . Show t h a t t h e set of r o u t e calls C£, w h i c h m a x i m i z e s r e v e n u e , c a n b e obtained as a solution of a l i n e a r p r o g r a m m i n g p r o b l e m . (c) As e a c h call r e q u e s t arrives, t h e call a d m i s s i o n a l g o r i t h m decides w h e t h e r to a d m i t or reject t h e call. We w o u l d like to design a n algo­ r i t h m t h a t m a x i m i z e s r e v e n u e . T h i s p r o b l e m illustrates t h e difficulties in designing s u c h a n algorithm. C o n s i d e r t h e n e t w o r k s h o w n in Fig­ u r e 5.38 w i t h t h r e e links, w i t h capacities as s h o w n , a n d w i t h two r o u t e s . Show t h a t t h e set of feasible calls is given b y t h e convex region o n t h e right. Design a n a d m i s s i o n p r o c e d u r e so t h a t t h e s y s t e m will o p e r a t e n e a r t h e desired p o i n t (4,6). C o n s i d e r two cases. I n t h e first case, exist­ ing calls c a n b e t e r m i n a t e d (so-called p r e e m p t i v e case); in t h e s e c o n d case, existing calls c a n n o t b e t e r m i n a t e d ( n o n p r e e m p t i v e case). (d) P l a n n i n g p r o b l e m s are c o n c e r n e d w i t h e x p a n d i n g t h e n e t w o r k capac­ ity to m e e t growing d e m a n d . Suppose t h a t this g r o w t h is d e s c r i b e d as r

5.38 FIGURE

The panel on the left shows a network with three links and two routes. The set of feasible calls is the convex region shown on the right.

5.10

Problems

255 a n i n c r e a s e of Δ calls along r o u t e r, r e R. Suppose this i n c r e a s e is to b e m e t b y e x p a n d i n g t h e capacity of e a c h l i n k b y a m o u n t 8y > 0 at a u n i t cost ofpy. F o r m u l a t e t h e m i n i m u m cost e x p a n s i o n as a l i n e a r programming problem. Γ

2. In t h e p r e c e d i n g question, it is a s s u m e d t h a t t h e set of r o u t e s h a s a l r e a d y b e e n selected a n d t h e n u m b e r of s i m u l t a n e o u s calls C o n e a c h r o u t e r e R is given. But t h e r o u t e selection a n d a s s i g n m e n t m a y b e c h a n g e d . Suppose t h e l i n k capacities Qj are given. Suppose w e w i s h to r o u t e simultaneous calls from e v e r y originating switch χ to d e s t i n a t i o n switch y. F o r m u l a t e a l i n e a r p r o g r a m m i n g p r o b l e m w h o s e solution gives a r o u t i n g a s s i g n m e n t . r

3. A cellular t e l e p h o n e n e t w o r k consists of a n u m b e r of b a s e stations (switches) c o n n e c t e d b y w i r e d links. A u s e r w i t h i n a cell s e r v e d b y a p a r t i c u l a r b a s e station gains access to t h a t station u s i n g a n idle radio or wireless link. If all t h e access links are b u s y , t h e u s e r ' s r e q u e s t is blocked. (See Figure 5.39.) T h e n u m b e r of access links to e a c h b a s e station is fixed b y g o v e r n m e n t regulations. T h e n u m b e r of m o b i l e u s e r s varies r a n d o m l y . (a) F o r m u l a t e a m o d e l to d e t e r m i n e t h e b l o c k i n g probability. (b) Suppose a call is placed from a mobile in o n e cell to a n o t h e r m o b i l e in a n o t h e r cell. T h e r o u t e for s u c h a call w o u l d h a v e t h r e e parts: a radio access link in t h e first cell, a r o u t e over t h e w i r e d l i n k s from t h e first b a s e station to t h e second, a n d a radio access link in t h e s e c o n d cell.

α

Wired links

Base station

ο

5.39 FIGURE

The base stations or switches in a cellular network are connected b y wired links. Mobile stations in a cell share wireless access links.

Circuit-Switched Networks

256

W h a t is t h e blocking probability of t h e call in t e r m s of t h e b l o c k i n g probabilities for e a c h p a r t of t h e r o u t e ? 4. T h e textbook time-division multiplex s c h e m e a s s u m e s t h a t all t h e multi­ plexed signals h a v e identical bit r a t e s or f r e q u e n c y (see Figure 2.5). Since e a c h signal c o m e s from a s e p a r a t e source, this a s s u m e s t h a t t h e clocks at all those sources are perfectly s y n c h r o n i z e d . T h i s is impossible in practice, a n d so, over time, t h e clocks will drift apart. T h e m o r e a c c u r a t e t h e clocks, t h e s m a l l e r will b e t h e drift. If w e a s s u m e t h a t t h e clock a c c u r a c y is o n t h e o r d e r of 1 0 ~ (i.e., t h e y drift a p a r t b y 1 o u t of e v e r y 1 0 seconds), a n d if w e a s s u m e a bit rate of 1 0 bps, t h e n t h e s e signals will drift b y 1 bit ev­ ery 1 0 seconds. For example, if w e a s s u m e a clock a c c u r a c y of 1 0 ~ (very high) a n d bit rate of 1 0 bps, t h e r e is a drift of 1 bit e v e r y 1 0 s e c o n d s (about 1 bit p e r h o u r ) . If w e a s s u m e a clock a c c u r a c y of 1 0 ~ a n d bit rate of 10 , t h e r e is a drift of 100 bits e v e r y second. Use t h o s e c o n c e p t s to ana­ lyze h o w frequently it will b e n e c e s s a r y to u s e t h e f r e q u e n c y justification p r o c e d u r e s of Figure 5.11. n

n

m

n - m

12

9

3

4

6

5. T h e frame s t r u c t u r e in X.25 u s e s a n 8-bit flag. Suppose it is 01111110. T h i s 8-bit p a t t e r n m a y b e p r e s e n t in t h e data, in w h i c h case it m a y b e i n t e r p r e t e d as t h e flag. Tb avoid this confusion, w h e n e v e r a p a t t e r n of six I s a p p e a r s in t h e data, it is followed b y a s e v e n t h stuffed bit of 1. Show t h a t t h e data will n e v e r c a r r y t h e flag p a t t e r n . Also explain h o w t h e original data c a n b e recovered. 6. Design a caller ID service u s i n g t h e functional c o m p o n e n t s of section 5.6.3.

6

Asynchronous

CHAPTER

Transfer Mode

T n C h a p t e r s 3 t h r o u g h 5 w e s t u d i e d packet- a n d circuit-switched n e t w o r k s . T h o s e n e t w o r k s are suitable for m e s s a g e a n d c o n s t a n t bit rate traffic, r e s p e c ­ tively. In this c h a p t e r w e e x a m i n e t h e A s y n c h r o n o u s Transfer Mode or ATM n e t w o r k s . ATM n e t w o r k s c o m b i n e t h e good features of b o t h t y p e s of n e t w o r k s , m a k i n g t h e m suitable for b o t h c o n s t a n t bit rate (CBR) a n d variable bit rate (VBR) traffic. ATM n e t w o r k s p o t e n t i a l l y c a n provide b e a r e r services w i t h a specified quality of service to m e e t t h e n e e d s of all traffic t y p e s . W h e t h e r this potential c a n b e realized d e p e n d s o n h o w well t h e p r o b l e m s of m a n a g e m e n t a n d control of ATM n e t w o r k s c a n b e solved. T h o s e p r o b l e m s are discussed in C h a p t e r s 8 a n d 9. T h i s c h a p t e r p r e s e n t s t h e c o n c e p t s of ATM. It offers a s i m p l e m o d e l to calculate t h e delay of a n ATM n e t w o r k . By t h e e n d of this c h a p t e r y o u will u n ­ d e r s t a n d t h e ATM l a y e r e d architecture, t h e a d d r e s s i n g a n d r o u t i n g standards, a n d t h e formats for different services. You will also k n o w t h e i m p o r t a n t pro­ posals for ATM LANs a n d for IP service over ATM. With t h e s e proposals, ATM b e c o m e s b a c k w a r d - c o m p a t i b l e w i t h existing LAN e q u i p m e n t a n d IP software. We saw in section 2.2 t h a t different applications i m p o s e different perfor­ m a n c e r e q u i r e m e n t s o n t h e n e t w o r k b e a r e r services in t e r m s of delay, b a n d ­ width, a n d loss. If all t h e s e applications are to s h a r e t h e s a m e n e t w o r k r e s o u r c e s (links, buffers, switches)—and this is v e r y desirable to gain t h e e c o n o m i e s of scale a n d service integration—the n e t w o r k m u s t b e able to allocate its re­ sources differently to different applications. Because switches or r o u t e r s in d a t a g r a m n e t w o r k s do n o t h a v e c o n n e c t i o n state information, t h e y c a n n o t dif­ ferentiate a m o n g p a c k e t s b y application. Therefore, d a t a g r a m n e t w o r k s c a n n o t

Asynchronous Transfer Mode

258

discriminate a m o n g applications b y quality of service. Circuit-switched net­ w o r k s do m a i n t a i n c o n n e c t i o n information. But since t h e y provide a fixed set of r e s o u r c e s to e v e r y c o n n e c t i o n or call, they, too, c a n n o t differentiate a m o n g connections. In t h e mid-1980s s o m e e n g i n e e r s a r g u e d t h a t virtual circuits w e r e ideal for t h e efficient utilization of n e t w o r k r e s o u r c e s w h e n applications h a v e widely different p e r f o r m a n c e r e q u i r e m e n t s . In a virtual circuit n e t w o r k , t h e n o d e s c a n set aside r e s o u r c e s for specific c o n n e c t i o n s , a n d t h e y c a n also discrimi­ n a t e a m o n g c o n n e c t i o n s in o r d e r to m e e t t h e i r different r e q u i r e m e n t s . T h o s e a r g u m e n t s c u l m i n a t e d in t h e d e v e l o p m e n t of a n e w set of s t a n d a r d s for a class of virtual circuit n e t w o r k s called ATM n e t w o r k s . ATM n e t w o r k s s e e k to provide t h e end-to-end transfer of fixed-size p a c k e t s or cells over a virtual circuit a n d w i t h specified quality of service (in t e r m s of delay, speed, a n d e r r o r rate). We describe t h e m a i n features of ATM n e t w o r k s in section 6.1. In section 6.2 w e discuss addressing, signaling, a n d r o u t i n g in ATM n e t w o r k s . In section 6.3 we p r e s e n t t h e s t r u c t u r e of t h e ATM header. In section 6.4 w e discuss t h e ATM adaptation layer, w h i c h c o n v e r t s i n f o r m a t i o n s t r e a m s in a variety of forms into a s e q u e n c e of ATM cells. In o r d e r to satisfactorily serve a w i d e r a n g e of applica­ tions, ATM n e t w o r k s m u s t b e a p p r o p r i a t e l y controlled. Standards relating to t h e m a n a g e m e n t a n d control of ATM n e t w o r k s are discussed in section 6.5. ATM n e t w o r k s are d e s i g n e d to s u p p o r t b o t h real-time applications s u c h as video con­ n e c t i o n s a n d t e l e p h o n e services a n d non-real-time applications s u c h as e-mail a n d file transfers. In section 6.6 w e s t u d y h o w ATM m a y s u p p o r t Broadband Integrated Services Digital N e t w o r k s (BISDNs). In section 6.7 w e discuss in­ t e r n e t w o r k i n g w i t h ATM. O u r objective is to explain h o w t h e familiar T C P / I P applications c a n b e s u p p o r t e d b y a n ATM n e t w o r k . Section 6.7 also covers LAN e m u l a t i o n b y ATM n e t w o r k s . We provide a s u m m a r y in section 6.8.

6.1

MAIN FEATURES OF ATM Four features distinguish t h e ATM b e a r e r service from o t h e r b e a r e r services: 1. t h e service is c o n n e c t i o n - o r i e n t e d , w i t h data t r a n s f e r r e d over a virtual circuit (VC); 2. t h e data is transferred in 53-byte p a c k e t s called cells; 3. cells from different VCs t h a t o c c u p y t h e s a m e c h a n n e l or l i n k are statisti­ cally multiplexed;

6.1

Main Features of ATM

259

4. ATM switches m a y treat t h e cell s t r e a m s in different VC c o n n e c t i o n s u n e q u a l l y over t h e s a m e c h a n n e l in o r d e r to provide different qualities of service (QpS). We e m p h a s i z e t h a t ATM is a bearer service a n d not a bit w a y in t e r m s of t h e O p e n Data N e t w o r k m o d e l of section 2.8. T h e actual transfer of t h e bits t h a t constitute t h e cell m a y b e carried o u t in v e r y different bit ways, r a n g i n g from SONET to DS-3 c h a n n e l s to p r o p r i e t a r y bit ways. We n o w discuss t h e a d v a n t a g e s a n d disadvantages of t h e four features listed above.

6.1.1

Connection-Oriented Service In a c o n n e c t i o n - o r i e n t e d service over a virtual circuit, t h e data s t r e a m from origin to d e s t i n a t i o n follows t h e s a m e p a t h . (See Figure 6.1.) Data from different c o n n e c t i o n s is d i s t i n g u i s h e d b y m e a n s of a virtual p a t h identifier (VPI) a n d virtual c h a n n e l identifier (VCI). A c o n n e c t i o n over a virtual circuit is called a virtual channel in t h e ATM t e r m i n o l o g y . T h u s e a c h cell i n c u r s a n o v e r h e a d c o r r e s p o n d i n g to t h e l e n g t h ( n u m b e r of bits) of t h e VPI/VCI, w h i c h is g e n e r a l l y m u c h s m a l l e r t h a n t h e l e n g t h of a full s o u r c e / d e s t i n a t i o n a d d r e s s n e e d e d in a d a t a g r a m service. ( T h e r e is additional o v e r h e a d as well. Tbgether w i t h t h e small payload, t h e overall o v e r h e a d / p a y l o a d ratio m a y b e h i g h e r for ATM t h a n , say, w i t h IP.) Second, cells in t h e s a m e c o n n e c t i o n r e a c h t h e d e s t i n a t i o n in t h e o r d e r t h e y are s e n t from t h e source, t h u s e l i m i n a t i n g t h e n e e d for s e q u e n c e n u m b e r s a n d for buffering p a c k e t s at t h e d e s t i n a t i o n if t h e y arrive out of order. (However, s e q u e n c e n u m b e r s are n e c e s s a r y if t h e application l a y e r m u s t d e t e c t cell loss. We see this in o u r discussion of t h e a d a p t a t i o n

Asynchronous Transfer Mode

260

layer, in section 6.4.) More i m p o r t a n t l y , ATM switches c a n identify different c o n n e c t i o n s b y t h e i r VPI/VCI. C o n s e q u e n t l y , t h e switches potentially c a n d i s c r i m i n a t e a m o n g different c o n n e c t i o n s . This potential c a n b e u s e d in m a n y ways: a d m i s s i o n control (refusing cer­ tain c o n n e c t i o n s if sufficient n e t w o r k r e s o u r c e s are unavailable), congestion control (limiting t h e a m o u n t of traffic a c c e p t e d from a c o n n e c t i o n ) , r e s o u r c e al­ location (negotiating t h e b a n d w i d t h a n d buffers allocated to a c o n n e c t i o n ) , a n d policing ( m o n i t o r i n g t h e b u r s t i n e s s a n d average rate of traffic in a c o n n e c t i o n ) . We will s t u d y t h e s e different m o d e s of control in C h a p t e r s 8 a n d 9. T h e m a i n disadvantage of c o n n e c t i o n - o r i e n t e d service is t h a t t h e n e t w o r k m u s t i n c u r t h e o v e r h e a d of c o n n e c t i o n s e t u p e v e n w h e n o n l y a few cells are to b e transferred, w h i c h could b e d o n e m o r e efficiently b y a d a t a g r a m service. A n o t h e r disadvantage is t h a t a link or n o d e failure t e r m i n a t e s t h e virtual chan­ nel, w h e r e a s s u c h a failure affects only a few p a c k e t s in a d a t a g r a m n e t w o r k . (However, t h e ATM F o r u m will b e releasing specifications of edge-based rout­ ing t h a t automatically r e r o u t e s a c o n n e c t i o n in case of failure, m i n i m i z i n g cell loss w i t h o u t r e q u i r i n g t h e u s e r to reestablish t h e call.) T h e ATM F o r u m c u r r e n t l y specifies five categories of services t h a t a n ATM n e t w o r k c a n provide: c o n s t a n t bit rate (CBR), variable bit r a t e - r e a l t i m e (VBRRT), variable bit r a t e - n o n - r e a l t i m e (VBR-NRT), available bit rate (ABR), a n d unspecified bit rate (UBR). A sixth service category, g u a r a n t e e d frame rate (GFR), w a s r e c e n t l y p r o p o s e d in t h e ATM F o r u m to e n s u r e m i n i m u m rate g u a r a n t e e s for UBR services. T h e s e services differ in t h e p a r a m e t e r s of t h e traffic a n d of t h e quality of service t h a t t h e y specify. T h e p a r a m e t e r s of t h e traffic are defined b y a n algorithm—called t h e generalized cell rate algorithm (GCRA)—that controls t h e arrival t i m e s of cells. T h e s e p a r a m e t e r s are t h e following: •

p e a k cell rate (PCR),

•

s u s t a i n e d cell rate (SCR),

•

initial cell rate (ICR),

•

cell delay variation t o l e r a n c e (CDVT),

• b u r s t tolerance (BT), •

m i n i m u m cell rate (MCR).

We e x a m i n e t h e s e p a r a m e t e r s in C h a p t e r 8. For now, h e r e is a brief discus­ sion of t h e i r m e a n i n g . PCR is t h e reciprocal of t h e m i n i m u m t i m e b e t w e e n two cells. SCR is t h e l o n g - t e r m average cell rate. ICR is t h e r a t e at w h i c h a source s h o u l d s e n d after a n idle period. CDVT m e a s u r e s t h e p e r m i s s i b l e d e p a r t u r e

6.1

Main Features of ATM

from periodicity of t h e traffic. FT m e a s u r e s t h e m a x i m u m n u m b e r of cells in a b u r s t of back-to-back cells. MCR is t h e reciprocal of t h e m a x i m u m t i m e b e t w e e n two cells. For CBR traffic, PCR m e a s u r e s t h e m a x i m u m rate, a n d CDVT specifies t h e acceptable jitter, t h a t is, t h e d e p a r t u r e from strict periodicity at rate PCR. For VBR traffic, t h e k e y p a r a m e t e r s are SCR a n d BT, w h i c h m e a s u r e t h e l o n g - t e r m rate of t h e traffic a n d its b u r s t i n e s s . For ABR traffic, t h e cell r a t e is b e t w e e n MCR a n d PCR. For UBR, n o p a r a m e t e r s are controlled, b u t n o QpS is g u a r a n t e e d . T h u s , CBR is a n essentially periodic s t r e a m of cells w i t h s o m e acceptable jitter t h a t m a y b e u n a v o i d a b l e b e c a u s e of framing or packetization. VBR is a b u r s t y traffic t h a t m a y b e g e n e r a t e d b y a variable bit r a t e codec. ABR a n d UBR are for irregular data traffic. GFR is p r o p o s e d as a n e n h a n c e m e n t to UBR service. T h e quality of service (QoS) p a r a m e t e r s (attributes) are t h e following: •

cell loss ratio (CLR),

•

cell delay variation (CDV),

•

peak-to-peak cell delay variation (peak-to-peak CDV)

•

m a x i m u m cell transfer delay (Max CTD),

•

m e a n cell transfer delay ( M e a n CTD).

T h e s e service p a r a m e t e r s m a y b e n e g o t i a t e d b e t w e e n e n d s y s t e m s a n d t h e n e t w o r k w h e n a c o n n e c t i o n of a p a r t i c u l a r service category is b e i n g set u p . T h e r e also are "non-negotiated" QoS attributes i n c l u d i n g cell e r r o r ratio (CER). T h e service categories specify t h e traffic a n d QoS p a r a m e t e r s according t o T i b l e 6.1. We s h o u l d t h i n k of t h e traffic p a r a m e t e r s for a p a r t i c u l a r service as placing restrictions o n t h e traffic g e n e r a t e d b y a user, a n d t h e QoS p a r a m e t e r s for t h a t service as obligations o n t h e n e t w o r k t h a t p r o v i d e s t h e service. T h e UBR service is b e s t effort a n d r e q u i r e s o n l y t h e selection of o n e p a t h from t h e source to t h e destination. Tb provide CBR, VBR, a n d ABR services, t h e ATM switches m u s t r e s e r v e r e s o u r c e s w h e n t h e call is set u p . In t h e case of ABR, as w e discuss in C h a p t e r 8, t h e flow of cells is r e g u l a t e d b a s e d o n feedback information a b o u t t h e n e t w o r k congestion. CBR a n d VBR traffic are r e g u l a t e d at t h e source (by t h e GCRA) w i t h o u t a n y n e t w o r k feedback. T h e p r o p o s e d GFR service is n o t i n c l u d e d in Tkble 6.1. GFR is i n t e n d e d for applications t h a t c a n n e i t h e r specify t h e SCR or BT n e e d e d for VBR service, n o r b e subject to t h e rate-based control r u l e s of ABR. C u r r e n t i n t e r n e t w o r k i n g ap­ plications (based o n T C P / I P ) fall into this category. GFR p r o v i d e s a m i n i m u m

261

Asynchronous Transfer Mode

262 ATM Layer Service Characteristics Attribute

CBR

VBR(RT)

VBR(NKT)

ABR

UBR

Parameter

CLR

Specified

Specified

Specified

Specified

Unspecified

QpS

CDT, CDV

CDV and Max CTD

CDV and Max CTD

Mean CTD only

Unspecified

Unspecified

QpS

PCR, CDVT

Specified

Specified

Specified

Specified

Specified

Traffic

SCR, BT

n/a

Specified

Specified

n/a

n/a

Traffic Traffic

MCR

n/a

n/a

n/a

Specified

n/a

Congestion control

No

No

No

Yes

No

6.1

ATM service classes and applicable parameters. Source: [AL95, A96d].

TABLE

frame (not cell) rate g u a r a n t e e . T h e traffic p a r m e t e r s for GFR specify a maxi­ m u m frame size for t h e CPCS-SDU p a c k e t s (CPCS is discussed in section 6.4). If t h e u s e r s e n d s p a c k e t s of s m a l l e r size at a rate s m a l l e r t h a n t h e m i n i m u m g u a r a t e e d rate, t h e y will b e delivered b y t h e n e t w o r k .

6.1.2

Fixed Cell Size In o r d e r to recognize p a c k e t or cell b o u n d a r i e s in t h e bit s t r e a m t r a n s m i t t e d b y t h e physical link, it is c u s t o m a r y to delimit p a c k e t s b y a d i s t i n g u i s h e d bit p a t t e r n . However, if cells are of fixed l e n g t h a n d if t h e y c o n t a i n a fixed-length error-checking s e q u e n c e s u c h as a CRC ( w h i c h ATM cells do), t h e n t h e n o d e c a n u s e t h e s e features to d e t e r m i n e cell b o u n d a r i e s implicitly. T h e basic idea is illustrated in Figure 6.2. Suppose t h a t e a c h cell of l e n g t h Ν c o n t a i n s a CRC s e q u e n c e of l e n g t h η c o m p u t e d over t h e p r e c e d i n g m = N —n bits. In ATM cells, this CRC field is t h e h e a d e r error-control field (HEC). T h e HEC field is t h e last h e a d e r b y t e , a n d it is calculated over t h e p r e v i o u s 4 b y t e s . T h e a l g o r i t h m starts w i t h a n y n-bit "window" as a tentative CRC a n d m a t c h e s it w i t h t h e CRC s e q u e n c e c o m p u t e d over t h e p r e c e d i n g m bits. If a m a t c h occurs, t h e cell b o u n d a r y h a s b e e n correctly identified; otherwise, t h e a l g o r i t h m shifts t h e w i n d o w b y 1 bit a n d r e p e a t s t h e p r o c e d u r e . A m a t c h occurs in at m o s t Ν steps if t h e cell c o n t a i n s n o errors. O n c e t h e cell b o u n d a r y is identified, s u b s e q u e n t b o u n d a r i e s are found b y c o u n t i n g bits. Conversely, if CRC errors are d e t e c t e d in several c o n s e c u t i v e cells, this c a n b e t a k e n to indicate loss of s y n c h r o n i z a t i o n (loss of cell b o u n d a r y location).

6.1

Main Features of ATM

A( 0,

(6.1)

w h e r e w e u s e t h e n o t a t i o n t h a t for a n y n u m b e r z, z = max{z, 0} a n d 1 (·) is t h e indicator function, so l ( z > 0) = 1 if ζ > 0, a n d 0 o t h e r w i s e . T h e t e r m (X — 1 ) a c c o u n t s for t h e fact t h a t if X > 0, t h e n o n e cell will b e t r a n s m i t t e d , leaving (X — 1) cells in t h e buffer. A s s u m e t h a t w e h a v e r e a c h e d statistical s t e a d y state, so t h a t E{X(n + 1)} = E{X(n)} and E{X(n + l ) } = E{X(n) }. Taking expectations o n b o t h sides of (6.1) w e get +

+

n

n

n

2

2

E(X) = E(X) + E(A) - P(X > 0), so P(X>0)

=

E(A)=:p.

Next w e s q u a r e b o t h sides of (6.1) a n d take expectations to get E(A )

+ ρ + 2pE(X)

2

Since E(A ) 2

- 2E(X) - 2p = 0. 2

= p 4- p, w e find 2

2p - p E(X) = - f — £ - - . 2(1 -p) 2

6.13

(6.2)

Statistical Multiplexing A virtual circuit specifies a p a t h from source to d e s t i n a t i o n going t h r o u g h several links a n d switches. M a n y virtual circuits o c c u p y t h e s a m e link. A switch h a s ports t e r m i n a t i n g several i n c o m i n g a n d outgoing links. (A large switch s u c h as in a t e l e p h o n e c e n t r a l office m a y t e r m i n a t e t h o u s a n d s of links; s u c h a switch is m o s t likely built in m o d u l a r fashion following t h e p r i n c i p l e s p r e s e n t e d in C h a p t e r 12. A local area ATM switch, w h i c h i n t e r c o n n e c t s t e r m i n a l e q u i p m e n t

6.1

Main Features of ATM

267

w i t h i n a local area, m a y t e r m i n a t e t e n s of links; t h a t k i n d of switch is likely to b e built in a single stage.) We n o w describe t h e five tasks t h a t a switch carries out. T h e tasks are demultiplexing, r o u t i n g t h r o u g h t h e switch, multiplexing, buffering, a n d dis­ carding. T h e switch d e m u l t i p l e x e s t h e cell s t r e a m arriving over e a c h i n c o m i n g link into "tributary" s t r e a m s b e l o n g i n g to different virtual c h a n n e l s . T h e switch does this o n a cell-by-cell basis, u s i n g t h e VCI in e a c h cell. T h e switch t h e n r o u t e s t h e cell s t r e a m in e a c h virtual c h a n n e l to t h e a p p r o p r i a t e o u t p u t port. For r e a s o n s e x p l a i n e d in t h e n e x t section, t h e switch m a y c h a n g e t h e VCI assigned to a particular c h a n n e l . Routing is carried o u t w i t h t h e h e l p of a table w i t h e n t r i e s of t h e form (VCIi , i n p u t port, V C I n

0Ui

, output port).

(Routers in d a t a g r a m n e t w o r k s do n o t h a v e s u c h c o n n e c t i o n state information. T h o s e r o u t e r s o n l y h a v e tables w i t h e n t r i e s [destination address, o u t p u t port].) A n e n t r y in t h e r o u t i n g table is c r e a t e d at t h e t i m e t h e virtual c h a n n e l is set u p , a n d it is d e l e t e d w h e n t h a t virtual c h a n n e l is t o r n d o w n . T h i s r o u t i n g of t h e cell s t r e a m is i n t e r n a l to t h e switch; t h e m e c h a n i s m t h a t i m p l e m e n t s r o u t i n g d e p e n d s o n t h e switch architecture, as w e will see in C h a p t e r 12. Next, t h e switch multiplexes t h e cell s t r e a m s d i r e c t e d to t h e s a m e o u t p u t port. T h i s is statistical multiplexing. ( T h e cell s t r e a m in a virtual c h a n n e l m a y c a r r y c o n s t a n t bit rate (CBR), variable bit rate (VBR), or m e s s a g e traffic.) T h e capacity of t h e o u t p u t t r a n s m i s s i o n link exceeds t h e average bit r a t e of t h e i n c o m i n g virtual c h a n n e l s , b u t it m a y n o t exceed t h e i r p e a k rate. H e n c e , t h e switch h a s to buffer excess cells. T h i s n e e d arises w h e n for a s h o r t t i m e interval t h e n u m b e r of cells to b e t r a n s f e r r e d over a n o u t p u t link exceeds t h e capacity of t h a t link. Finally, if t h e buffer is full, t h e switch m u s t discard cells of l o w e r priority. (ATM specifies two priorities; see section 6.3.)

6.1.4

Allocating Resources ATM n e t w o r k s offer to transfer cell s t r e a m s from source to destination, u n d e r a r a n g e of quality of service, to m e e t t h e varying n e e d s of applications, as w e saw in section 6.1.1. T h e service a n d traffic p a r a m e t e r s establish a contract for a particular class of service. T h e n e t w o r k g u a r a n t e e s t h e service p a r a m e t e r s ( s u c h as b o u n d s o n delay CTD a n d loss rate CLR), p r o v i d e d t h e cell s t r e a m e m i t t e d b y t h e u s e r conforms to t h e traffic p a r a m e t e r s ( s u c h as average rate

Asynchronous Transfer Mode

268

SCR a n d b u r s t i n e s s BT). T h e s e p a r a m e t e r s are n e g o t i a t e d u s i n g established signaling p r o c e d u r e s . In o r d e r to m e e t its obligations u n d e r s u c h contracts, t h e n e t w o r k takes certain actions. It 1. exercises a d m i s s i o n control, 2. selects t h e r o u t e (virtual c h a n n e l p a t h ) of a d m i t t e d c o n n e c t i o n s , 3. allocates b a n d w i d t h a n d buffers s e p a r a t e l y to e a c h c o n n e c t i o n , 4. selectively drops low priority cells, a n d 5. asks sources to limit t h e cell s t r e a m rate (for ABR service). T h e ATM F o r u m h a s g r o u p e d t h e s e actions u n d e r CAC a n d UPC proce­ dures. UNI call admissions control (CAC) allows t h e n e t w o r k to grant or d e n y a c o n n e c t i o n to a user. Usage parameter control or UPC d e n o t e s t h o s e m o n i t o r i n g a n d control actions u s e d b y t h e n e t w o r k for c o m p l i a n c e to a g r e e d - u p o n traffic p a r a m e t e r s after a c o n n e c t i o n is granted. T h e s e n e t w o r k decisions are c o n s i d e r e d in d e p t h in C h a p t e r s 8 a n d 9. Admission control r e q u i r e s t h e n e t w o r k to d e t e r m i n e w h e t h e r , given existing c o n n e c t i o n s , t h e r e are sufficient idle r e s o u r c e s to m e e t t h e QpS r e q u i r e m e n t s of a n e w c o n n e c t i o n r e q u e s t . O n c e a n e w c o n n e c t i o n is admitted, t h e n e t w o r k m u s t assign to it a virtual circuit w i t h a r o u t e t h a t h a s sufficient r e s o u r c e s . (Route selection is discussed in section 6.2.3.) T h e n e t w o r k m u s t t h e n allocate those r e s o u r c e s to t h e c o n n e c t i o n so t h a t it c a n m e e t t h e QoS. T h e first t h r e e decisions are t a k e n d u r i n g t h e c o n n e c t i o n s e t u p p h a s e . T h e last two decisions are t a k e n d u r i n g t h e data transfer p h a s e . It h a p p e n s o n occasion t h a t buffers at a switch b e c o m e full, a n d s o m e cells m u s t b e d r o p p e d . If t h e r e are cells of different priorities in t h e buffer, it is preferable to d r o p t h o s e w i t h low priority. Finally, t h e switch m a y n e e d to signal to a source to r e d u c e or i n c r e a s e its rate (flow control) d e p e n d i n g o n h o w well it conforms to t h e QoS contract. In order to i m p l e m e n t t h e r o u t i n g a n d r e s o u r c e allocation decisions, t h e n e t w o r k n e e d s to distinguish b e t w e e n cell s t r e a m s of different c o n n e c t i o n s . T h a t distinction c a n b e m a d e o n t h e basis of e a c h cell's VCI. Of course, e a c h switch h a s to create a table of t h e relation a m o n g VCI, its QoS p a r a m e t e r s , its route, a n d its allocated r e s o u r c e s a n d consult this table as n e e d e d . T h e information in t h e table is said to b e implicit in t h e VCI. (Since r e f e r e n c i n g t h e table takes time, w e shall see t h a t s o m e of t h e implicit i n f o r m a t i o n is, in fact, explicitly p r e s e n t in e a c h cell of a c o n n e c t i o n . ) By contrast, different cells w i t h t h e s a m e VCI m a y h a v e different priorities, a n d so priority i n f o r m a t i o n m u s t b e indicated explicitly in e a c h cell. Similarly, t h e switch m a y w i s h to

6.2

Addressing, Signaling, and Routing

signal to t h e source t h e n e c e s s i t y for flow control, a n d so explicit provision m u s t b e m a d e w i t h i n t h e cell for r e p r e s e n t i n g this signal. We see h o w t h e ATM cell provides this information in section 6.3.

6.2

ADDRESSING, SIGNALING, AND ROUTING In this section w e discuss ATM addressing, signaling, a n d r o u t i n g in ATM n e t w o r k s . ATM addressing is b a s e d o n t h e E.164 i n t e r n a t i o n a l s t a n d a r d for t h e 15-digit n u m b e r i n g p l a n u s e d in t h e public I S D N / t e l e p h o n e s y s t e m . Since t h o s e n u m b e r s are too l i m i t e d to a c c o m m o d a t e t h e g r o w t h of private n e t w o r k s , t h e ATM addressing s c h e m e e x t e n d s t h a t n a m e space. ATM signaling is a n e x t e n s i o n of t h e Q.931 ISDN protocol for signaling b e t w e e n e n d s y s t e m s a n d t h e public n e t w o r k . ATM r o u t i n g d r a w s u p o n c o n c e p t s u s e d in packet-switched a n d circuit-switched n e t w o r k s .

6.2.1

ATM Addressing T h e ATM F o r u m h a s defined ATM addressing. A n ATM a d d r e s s indicates t h e location of a n ATM interface in t h e n e t w o r k topology. T h i s m e a n s t h a t ATM a d d r e s s e s are n o t portable. (Anycast service, discussed below, is a n exception.) T h e prefix of a n a d d r e s s is associated w i t h a g r o u p of interfaces w i t h t h e s a m e prefix. Prefixes are u s e d in call r o u t i n g tables. Each ATM s y s t e m is assigned a n a d d r e s s i n d e p e n d e n t of t h e h i g h e r proto­ col a d d r e s s e s ( s u c h as IP addresses) t h a t t h e s y s t e m s u p p o r t s . T h e d e c o u p l i n g afforded b y this overlay model allows t h e higher-level protocols to b e devel­ o p e d i n d e p e n d e n t l y of t h e ATM protocols. H e n c e all protocols o p e r a t i n g over a n ATM s u b n e t r e q u i r e a n address-resolution protocol t h a t m a p s t h e h i g h e r protocol a d d r e s s into t h e c o r r e s p o n d i n g ATM address. For example, to s e n d a n IP p a c k e t to a given IP a d d r e s s over a n ATM n e t w o r k , r o u t e r s u s e t h e address-resolution protocol to d e t e r m i n e t h e ATM a d d r e s s of t h e destination. T h e r o u t i n g a l g o r i t h m t h e n sets u p t h e c o n n e c t i o n to t h e destination. We dis­ cuss IP over ATM in m o r e detail later. A n ATM address is 20 b y t e s long. T h e last 7 b y t e s c o n t a i n a 48-bit MAC address, followed b y a 1-byte select field t h a t h a s n o n e t w o r k significance. T h e prefix m a y c o n t a i n a 15-digit E.164 n u m b e r ( s p e c i f i e d b y ITU-T for establishing ISDN a n d t e l e p h o n y calls) or a n o t h e r hierarchical address. T h e ATM F o r u m defines a n a d d r e s s registration m e c h a n i s m u s i n g t h e ILMI ( I n t e g r a t e d Local

269

Asynchronous Transfer Mode

270

M a n a g e m e n t Interface) protocol, discussed in section 6.5.3. T h i s allows a n e n d s y s t e m to inform a n ATM switch to w h i c h it is a t t a c h e d of its MAC a d d r e s s a n d get b a c k its full ATM address. A n ATM G r o u p Address r e p r e s e n t s a collection of ATM e n d s y s t e m s . A n e n d s y s t e m m a y j o i n or leave a g r o u p b y registering a g r o u p a d d r e s s u s i n g t h e ILMI. T h e only ATM Forum-defined service u s i n g g r o u p addressing is t h e anycast service. A n y c a s t service, defined in UNI 4.0 a n d PNNI 1.0, allows m u l t i p l e s y s t e m s to register as a server for s o m e service (e.g., p r i n t e r server, file server). A well-known a n y c a s t a d d r e s s c a n b e u s e d to r o u t e a r e q u e s t to a n o d e providing a particular service w i t h o u t explicitly identifying t h e n o d e . A call m a d e to a n a n y c a s t address is r o u t e d to t h e "nearest" e n d s y s t e m t h a t registered itself w i t h t h e n e t w o r k to provide t h e associated service. A n y c a s t is a useful m e c h a n i s m for n e t w o r k configuration a n d o p e r a t i o n since it p r e c l u d e s t h e n e e d for m a n u a l configuration or service locations protocols.

6.2.2

Signaling A n ATM n e t w o r k consists of ATM switches c o n n e c t e d b y point-to-point links or interfaces. T h e ATM F o r u m specifies several interfaces, as s e e n in Figure 6.5. T h e two m o s t i m p o r t a n t are t h e user-network interface or UNI a n d t h e networknode or network-network interface or NNI. A UNI c o n n e c t s ATM e n d s y s t e m s (hosts, routers) to a n ATM switch, a n d a n NNI c o n n e c t two ATM switches. T h e formats of t h e cells flowing across t h e s e interfaces are described in section 6.3.

ATM network

ATM end system

non-ATM end system

NNI

FUNI

UNI

(PNNI) FUNI to ATM conversion function

6.5 FIGURE

UNI is the interface between an ATM end system and an ATM switch, NNI is the interface between two ATM switches, and FUNI is the frame interface between a non-ATM end system and an ATM switch.

6.2

Addressing, Signaling, and Routing

UNI is further distinguished b y private or public UNI, d e p e n d i n g o n w h e t h e r t h e e n d s y s t e m is c o n n e c t e d to a private or public n e t w o r k , a n d similarly for NNI. PNNI stands for private network-network or node-network interface. T h e public n e t w o r k - n e t w o r k interface is specified in B-ICI or BISDN Inter Carrier Interface. It c u r r e n t l y provides o n l y for PVC ( p e r m a n e n t virtual con­ nections), a l t h o u g h w o r k h a s started for s w i t c h e d virtual c o n n e c t i o n s (SVCs). It is possible t h a t t h e powerful facilities of r o u t e discovery a n d aggregation in PNNI m a y b e i m p o r t e d into t h e public n e t w o r k s . T h e frame-based u s e r - n e t w o r k interface or FUNI p e r m i t s a non-ATM legacy s y s t e m to c o n n e c t to a n ATM, exchanging frames r a t h e r t h a n ATM cells. F r a m e s of size u p to 64 KB m a y b e s u p p o r t e d at a s p e e d of η χ 64 Kbps u p to 1.5 Mbps, b u t ongoing w o r k m a y i n c r e a s e this s p e e d . T h e ATM switch carries out t h e FUNI to UNI c o n v e r s i o n function via AAL5. FUNI b r i n g s m o s t of t h e UNI control signaling to non-ATM e n d s y s t e m s . T h e ATM F o r u m a n d ITU-T h a v e also specified o t h e r interfaces b e t w e e n non-ATM legacy s y s t e m s a n d ATM switches. UNI 4.0 a n d PNNI 1.0 specify t h e signaling protocols u s e d across UNI a n d NNI. T h e PNNI signaling protocols are b a s e d o n UNI 4.0, a n d so o n l y t h e latter is described h e r e . PNNI signaling m a k e s u s e of t h e i n f o r m a t i o n g a t h e r e d b y PNNI routing. UNI specifies t h e signaling p r o c e d u r e s for d y n a m i c a l l y establishing, m a i n ­ taining, a n d clearing ATM c o n n e c t i o n s at t h e UNI interface. T h e r e are two basic t y p e s of c o n n e c t i o n s : point-to-point a n d point-tom u l t i p o i n t . Point-to-point c o n n e c t i o n s m a y b e u n i d i r e c t i o n a l or bidirectional. Point-to-multipoint c o n n e c t i o n s c o n n e c t a single root n o d e to m u l t i p l e destina­ tions called leaves. T h e s e c o n n e c t i o n s are u n i d i r e c t i o n a l : a root c a n t r a n s m i t to leaves, b u t leaves c a n n o t t r a n s m i t to t h e root. Cell replication w i t h i n t h e n e t w o r k is d o n e b y t h e ATM switches w h e r e t h e c o n n e c t i o n splits into t w o or more branches. Signaling p r o c e d u r e s are defined in t e r m s of m e s s a g e types, i n f o r m a t i o n e l e m e n t s (IE) t h a t c a r r y data s u c h as a d d r e s s e s a n d QpS r e q u i r e m e n t s of a n ATM c o n n e c t i o n , a n d state m a c h i n e s t h a t describe t h e p r o c e d u r e s . (QpS r e q u i r e m e n t s are expressed as values of t h e p a r a m e t e r s in Tkble 6.1.) Signaling r e q u e s t s are carried across t h e UNI in a w e l l - k n o w n default c o n n e c t i o n (VPI = 0, VCI = 5). T h e specification is b a s e d o n Q.2931, a public n e t w o r k signaling protocol d e v e l o p e d b y ITU-T, which, in t u r n , w a s b a s e d o n t h e Q..931 protocol d e v e l o p e d for ISDN. A source e n d s y s t e m a t t e m p t i n g to set u p a point-to-point c o n n e c t i o n cre­ ates a n d s e n d s into t h e n e t w o r k , across its UNI, a S e t u p m e s s a g e , c o n t a i n i n g t h e d e s t i n a t i o n address, desired traffic a n d QpS p a r a m e t e r s , IEs defining desired

271

Asynchronous Transfer Mode

272

higher-layer protocol bindings, a n d so on. T h i s Setup m e s s a g e is s e n t to t h e ingress switch, w h i c h r e s p o n d s w i t h a local Call Proceeding a c k n o w l e d g m e n t . T h e switch t h e n invokes t h e PNNI r o u t i n g protocol to p r o p a g a t e t h e signaling r e q u e s t t h r o u g h switches across t h e n e t w o r k , setting u p c o n n e c t i o n identifiers (VPI/VCI) along t h e way, to t h e egress switch to w h i c h t h e d e s t i n a t i o n e n d s y s t e m is attached. T h e egress switch forwards t h e Setup m e s s a g e to t h e d e s t i n a t i o n e n d station. T h e latter m a y e i t h e r accept or reject t h e c o n n e c t i o n r e q u e s t . In t h e former case, it r e t u r n s a C o n n e c t message, b a c k t h r o u g h t h e n e t w o r k , along t h e s a m e path, to t h e r e q u e s t i n g source e n d s y s t e m . O n c e t h e source e n d s y s t e m receives a n d a c k n o w l e d g e s t h e C o n n e c t message, e i t h e r n o d e c a n t h e n start t r a n s m i t t i n g data o n t h e c o n n e c t i o n . If t h e d e s t i n a t i o n e n d s y s t e m rejects t h e c o n n e c t i o n request, it r e t u r n s a Release message, w h i c h is also s e n t b a c k to t h e source e n d s y s t e m , clearing t h e c o n n e c t i o n (e.g., a n y allocated c o n n e c t i o n identifiers) as it p r o c e e d s . Release m e s s a g e s are also u s e d b y e i t h e r of t h e e n d systems, or b y t h e network, to clear a n established c o n n e c t i o n . T h e data flows along t h e s a m e p a t h traveled b y t h e c o n n e c t i o n r e q u e s t . It is possible t h a t a switch along t h e p a t h rejects a r e q u e s t . T h i s m a y h a p p e n , for instance, b e c a u s e t h e switch realizes t h a t t h e r e s o u r c e s specified in t h e r e q u e s t are n o t available. T h e switch t h e n s e n d s a reject m e s s a g e along t h e reverse p a t h to t h e ingress switch. T h a t switch m a y t h e n t r y to r e p e a t t h e p r o c e d u r e to find a n a l t e r n a t e p a t h . T h i s p r o c e d u r e is called crankback b y t h e ATM F o r u m . It is similar to t h e d y n a m i c a l t e r n a t e routing of circuit-switched n e t w o r k s (see section 8.2.2). T h e ingress switch m a y e v e n t u a l l y reject t h e c o n n e c t i o n r e q u e s t from t h e source. T h e p r o c e d u r e followed b y a switch to grant or reject a r e q u e s t is called c o n n e c t i o n a d m i s s i o n control or CAC. A m u l t i p o i n t c o n n e c t i o n is initiated b y a Setup m e s s a g e from t h e root n o d e to a single leaf n o d e . T h e root c a n t h e n s e n d Add Party or D r o p Party m e s s a g e s to add or delete leaves in a n existing c o n n e c t i o n . T h e r e is also a Leaf-Initiated J o i n capability in w h i c h a leaf n o d e c a n j o i n a n existing c o n n e c t i o n w i t h or w i t h o u t t h e i n t e r v e n t i o n of t h e root n o d e .

6.2.3

PNNI Routing T h e ATM F o r u m ' s Private Network-Network Interface (PNNI) version 1.0 in­ cludes two classes of protocols. T h e r e is a protocol for distributing topology information b e t w e e n switches a n d clusters of switches. T h i s i n f o r m a t i o n is u s e d to c o m p u t e p a t h s or r o u t e s t h r o u g h t h e specification n e t w o r k . A hier­ a r c h y m e c h a n i s m e n s u r e s t h a t this protocol scales well for large w o r l d w i d e ATM n e t w o r k s . T h e s e c o n d protocol is for signaling. As n o t e d above, PNNI

6.2

Addressing, Signaling, and Routing

signaling is b a s e d o n UNI signaling, w i t h m e c h a n i s m s to s u p p o r t source rout­ ing, crankback, a n d a l t e r n a t e r o u t i n g of call s e t u p in case of c o n n e c t i o n s e t u p failure. We will first discuss topology i n f o r m a t i o n a n d r o u t i n g a s s u m i n g a flat or n o n h i e r a r c h i c a l n e t w o r k , a n d t h e n discuss t h e h i e r a r c h y m e c h a n i s m . PNNI r o u t i n g u s e s a generalization of t h e link state (Dijkstra's) a l g o r i t h m discussed in section 4.3. T h e generalization consists in t h e fact t h a t a link's state is n o t a single m e t r i c r e p r e s e n t i n g , say, delay, b u t a vector of "topology state parameters," including "link state parameters," w h i c h describe t h e char­ acteristics of logical links, a n d "nodal state parameters," w h i c h describe t h e characteristics of n o d e s . Tbpology state p a r a m e t e r s are classified as attributes or m e t r i c s . A n at­ tribute is c o n s i d e r e d individually w h e n m a k i n g r o u t i n g decisions. For example, a s e c u r i t y "nodal attribute" could c a u s e a p r o p o s e d p a t h to b e refused. T h e capacity of a link is also a n attribute: a c o n n e c t i o n r e q u e s t for a specified b a n d ­ w i d t h (SCR or s u s t a i n e d cell rate) m u s t b e m e t b y e a c h l i n k along t h e route. A metric, o n t h e o t h e r h a n d , is a p a r a m e t e r w h o s e effect is c u m u l a t i v e along a p a t h . For example, a d e l a y m e t r i c a d d s u p as o n e p r o g r e s s e s along a given path. T h e PNNI specification d e s i g n a t e s Routing Control C h a n n e l s (particular VPI/VCI) for t h e exchange of PNNI r o u t i n g p a c k e t s (Helios, PTSP, PTSE acks, etc.) b e t w e e n switches. N o d e s in t h e n e t w o r k exchange Hello p a c k e t s w i t h t h e i r i m m e d i a t e neigh­ b o r s a n d t h e r e b y d e t e r m i n e t h e i r local state information. T h i s state i n f o r m a t i o n i n c l u d e s t h e identity a n d p e e r g r o u p m e m b e r s h i p (described b e l o w ) of t h e n o d e ' s i m m e d i a t e neighbors, a n d t h e status of its links to t h e neighbors. Each n o d e t h e n b u n d l e s its state i n f o r m a t i o n in PNNI Tbpology State Elements (PTSEs) reliably flooded t h r o u g h o u t t h e p e e r g r o u p . (PTSEs are e n c a p s u l a t e d in PNNI Tbpology State Packets or PTSPs.) In this way, e a c h n o d e obtains t h e link a n d n o d e state p a r a m e t e r s of t h e e n t i r e n e t w o r k . T h i s i n f o r m a t i o n is u s e d b y t h e ingress switch to c o m p u t e a p a t h t h a t c a n satisfy a c o n n e c t i o n r e q u e s t r e c e i v e d from a source e n d station. W h e n t h e ingress switch d e t e r m i n e s t h e c o m p l e t e path, t h e p a t h is in­ cluded in t h e c o n n e c t i o n r e q u e s t as a Designated Transit List or DTL. T h u s PNNI u s e s source routing. T h e PNNI d o e s n o t specify h o w to c o m p u t e t h e source route. We formulate a m a t h e m a t i c a l p r o b l e m t h a t s u c h a c o m p u t a t i o n s h o u l d solve. T h e topology information at a switch c a n b e r e p r e s e n t e d as a g r a p h w h o s e n o d e s are t h e ATM switches a n d w h o s e edges are t h e links. Each n o d e i is labeled w i t h its n o d e attribute p a r a m e t e r 0(i), w h i c h d e t e r m i n e s w h e t h e r or

Asynchronous Transfer Mode

274

n o t a r e q u e s t e d c o n n e c t i o n c a n traverse t h e n o d e . Each link (i, ;) is labeled w i t h a vector d(i,j) = (d\ · · · , djt(i,;)) of t h e Κ link state m e t r i c s . (Link attributes are ignored: t h e y c a n b e t r e a t e d like n o d e attributes.) A n o d e p a r a m e t e r θ is a predicate o n c o n n e c t i o n requests, so t h a t a partic­ ular r e q u e s t c a n b e r o u t e d t h r o u g h t h e n o d e o n l y if it satisfies t h e predicate. T h e link m e t r i c s r e p r e s e n t quality characteristics s u c h as delay, or jitter, so t h a t t h e characteristics of t h e p a t h or r o u t e are t h e s u m of t h e characteristics of t h e links along t h e path. A c o n n e c t i o n s e t u p r e q u e s t is a triple R =

(i,j,D=(D .. D )), lr

t

K

w h e r e i is t h e source n o d e , ; is t h e d e s t i n a t i o n node, a n d (D\, · · · , D#) specify t h e c o n n e c t i o n ' s QpS r e q u i r e m e n t s . A feasible path is a s e q u e n c e of n o d e s i = i , · · · , i = ; h that s u c

0

R satisfies θ(ΐ ), η

n = 0, · · · , Ν a n d

N

(6.3)

Ν

(6.4) n=l

R e q u i r e m e n t (6.3) states t h a t t h e n o d e s along t h e p a t h m u s t a c c e p t t h e c o n n e c ­ tion, so this r e q u i r e m e n t effectively d e l e t e s n o d e s from t h e g r a p h t h a t c a n n o t accept t h e request. We a s s u m e h e n c e f o r t h t h a t t h o s e n o d e s h a v e b e e n d e l e t e d a n d so w e focus o n (6.4). It states t h a t t h e quality characteristics along t h e p a t h m u s t m e e t t h e r e q u i r e m e n t s of t h e r e q u e s t . So t h e r o u t i n g task is to find a p a t h t h a t satisfies e a c h of t h e Κ QpS r e q u i r e m e n t s (6.4). Observe t h a t if t h e link state is a single n u m b e r (K = 1), o n e c a n u s e t h e shortest-path algorithms of section 4.3. But w h e n Κ > 1, t h e c o m p u t a t i o n of feasible p a t h s is m o r e complex. We explain w h y this is so in t h e e x a m p l e of t h e five-link n e t w o r k of Figure 6.6. T h e r e are n o n o d e p a r a m e t e r s . Each link h a s two m e t r i c s as s h o w n . T h u s for link a t h e values of t h e link state p a r a m e t e r s are (d\(a) = 5, d2(a) = 1), a n d similarly for t h e o t h e r links. Consider t h r e e r e q u e s t s for a c o n n e c t i o n from A to B. T h e first r e q u e s t is (Di, Dz) = (10, 2). T h e only p a t h t h a t c a n satisfy this r e q u e s t c o m p r i s e s t h e links a, b. T h e r e q u e s t (2, 10) is m e t o n l y b y t h e p a t h c, d. Lastly, t h e r e q u e s t (8, 8) is m e t only b y t h e p a t h a, e, d. In t h e case of a single m e t r i c , t h e r e is a shortest-path t r e e r o o t e d at t h e des­ tination n o d e B. This tree c a n b e obtained using, for example, t h e Bellman-Ford

6.2

Addressing, Signaling, and Routing

275 c

{(0.0)}

{(10,2), (8,8), (2,10)}

Β

A

D 6.6

A five-link network with two link state metrics.

FIGURE

algorithm of section 4.3. In t h a t a l g o r i t h m o n e p r o c e e d s u p s t r e a m , e s t i m a t i n g t h e l e n g t h of t h e shortest p a t h to t h e d e s t i n a t i o n from n o d e i b y L(i) =

mm[d(i,j)+L(j)].

(6.5)

T h e algorithm starts w i t h t h e initial e s t i m a t e L(B) = 0 a n d L(i) = oo, for all o t h e r nodes. For o u r case w h e r e d is a vector of metrics, this algorithm c a n b e modified as follows. We e s t i m a t e a set of m e t r i c vectors L(i) c R by: K

L(i) = n{[d{i j) t

4-

I; = 1, · · ·, n;

e L(j)}),

(6.6)

w h e r e for a set L C R , μ(Χ) is t h e subset of n o n i n f e r i o r m e t r i c vectors: /* e L is noninferior if t h e r e is n o o t h e r I e L w i t h < Z£, k = 1, · · · , K. T h e a l g o r i t h m (6.6) starts w i t h t h e initial e s t i m a t e L(B) = {(0, · · ·, 0)}, a n d L(i) = {(oo, · · ·, oo)} for all o t h e r n o d e s i. Figure 6.6 shows t h e final e s t i m a t e s obtained from this algorithm. F r o m t h e lists displayed at e a c h n o d e w e c a n d e t e r m i n e w h e t h e r t h e r e is a feasible p a t h from a n y source to t h e d e s t i n a t i o n B. For instance, t h e r e q u e s t for a c o n n e c t i o n from C to Β w i t h characteristics (4, 7) is feasible since (3, 7) < (4, 7) a n d (3, 7) e L(C). O n t h e o t h e r h a n d , t h e r e q u e s t for a c o n n e c t i o n from A to Β w i t h characteristics (7, 9) is not feasible. In t h e case of a single metric, t h e r o u t i n g table at a n y n o d e h a s a s i m p l e structure. It contains, for e a c h destination, t h e e s t i m a t e of t h e m i n i m u m K

276

Asynchronous Transfer Mode

distance a n d t h e next n o d e along t h e shortest p a t h . In t h e case of m u l t i p l e metrics, t h e r o u t i n g table m u s t n o w store, for e a c h destination, t h e set of noninferior m e t r i c vectors, a n d for e a c h s u c h vector t h e next n o d e along t h e corresponding path. T h e algorithm suggested above is n o t scalable. I b b e scalable, a r o u t i n g al­ g o r i t h m m u s t b e hierarchical. T h e PNNI r o u t i n g algorithm defines a h i e r a r c h y of n o d e s . At e a c h level, n o d e s are clustered in p e e r groups. T h e n o d e s of t h e s a m e p e e r g r o u p elect a p e e r g r o u p leader, w h i c h m a i n t a i n s a n aggregate de­ scription of t h e group. This aggregate description indicates t h e characteristics of t h e service across t h e g r o u p . T h e precise p r o c e d u r e for deriving s u c h a n aggregate description is n o t fully specified. T h e g r o u p m e m b e r s also m a i n t a i n routing tables to r e a c h e a c h o t h e r a n d t h e i r p e e r g r o u p leader. T h e p e e r g r o u p l e a d e r s exchange information to identify t h e i r n e i g h b o r s a n d to calculate rout­ ing tables to r e a c h o n e another. At t h e next level u p , t h e s e p e e r g r o u p l e a d e r s are t h e n clustered into a p a r e n t p e e r group, a n d so on. T h e a d d r e s s e s of t h e n o d e s are d e s i g n e d to identify t h e i r g r o u p m e m b e r s h i p s , as t h e t e l e p h o n e n u m ­ b e r i n g does ( c o u n t r y code, area code, zone, n u m b e r in zone). Note t h a t w h e n a g r o u p of n o d e s is aggregated, t h e r e is a loss of topology information. T h e feasible r o u t e s c o n s t r u c t e d o n t h e basis of t h e aggregated i n f o r m a t i o n m i g h t t u r n out to b e infeasible. T h e c o n n e c t i o n r e q u e s t will t h e n b e rejected, a n d t h e ingress switch will resort to crankback. I b r o u t e from A to B, t h e routing algorithm c a n t h e n look at t h e a d d r e s s e s to find out t h e g r o u p l e a d e r s t h a t s h o u l d b e involved. T h u s , if t h e a d d r e s s of A is XYZ a n d t h a t of Β is XVW, t h e n A a n d Β b e l o n g to t h e same group or groups of nodes. T h e routing will t h e n go from A to its p e e r g r o u p leader, say C, to t h e p a r e n t p e e r g r o u p l e a d e r of C, say D, to t h e p e e r g r o u p l e a d e r of B, say E, to Β itself. T h e routing table at A specifies h o w to get to C. T h e table at C specifies h o w to get to E, a n d t h e table at Ε indicates h o w to r e a c h B. At e a c h level, t h e routing is decided b y a n algorithm, possibly of t h e k i n d suggested above, t h a t i n c o r p o r a t e s QpS characteristics. If t h e c o n n e c t i o n m u s t go t h r o u g h a c o n n e c t i o n l e s s public service, s u c h as F r a m e Relay or SMDS, t h e n it b e c o m e s a l m o s t impossible to g u a r a n t e e t h e quality of service of t h e c o n n e c t i o n . This is t h e case e v e n t h o u g h o n e c a n t r a n s m i t t h e desired characteristics of t h e c o n n e c t i o n across t h e public service to o t h e r private ATM n e t w o r k s . It is interesting to n o t e h o w earlier c o n c e p t s are b o r r o w e d a n d i m p r o v e d . PNNI e x t e n d s t h e shortest-path protocols to i n c l u d e QpS a n d source routing. In t u r n PNNI i n n o v a t i o n s of source r o u t i n g ( s u c h as DTL) are a d o p t e d b y MPLS (see section 4.6).

6.3

ATM Header Structure

6.3

277

ATM HEADER STRUCTURE In this section w e will e x a m i n e t h e ATM cell s t r u c t u r e , a n d w e will see h o w a n ATM switch c a n obtain t h e i n f o r m a t i o n it n e e d s from t h e cell header. As indicated in Figure 6.7 t h e AAL (ATM a d a p t a t i o n l a y e r ) p r o d u c e s a data s t r e a m of 48-byte cells or PDUs (protocol data u n i t s ) . (We will see l a t e r t h e function of t h e AAL layer.) T h e ATM l a y e r a d d s a 5-byte h e a d e r a n d forwards t h e 53b y t e cell to t h e physical layer. T h e 53-byte cell is c o n v e r t e d into a serial bit s t r e a m b y r e a d i n g t h e cell from left to right ( m o s t significant bit first) a n d top to b o t t o m (byte 1 first). T h e figure s h o w s t h e h e a d e r s t r u c t u r e for b o t h t h e u s e r - n e t w o r k interface (UNI) a n d t h e n e t w o r k - n e t w o r k interface (NNI). T h e abbreviations u s e d in t h e figure are •

GFC, g e n e r i c flow control,

•

VPI, virtual p a t h identifier,

•

VCI, virtual c h a n n e l identifier,

•

PT, payload type,

Information stream

AAL

8

7

6

5

4

3

2

GFC

VPI

VPI

VCI

SAR convergence

1

VCI VCI

PT

CLP

HEC

48-byte cells

UNI header ATM layer 8

7

6

5

4

3

53-byte ATM cells VPI

VCI

VCI

6.7 FIGURE

1

VCI

Physical layer Signals that encode bit stream

2

VPI

PT

CLP

HEC NNI header

The figure shows the headers of ATM cells across the user-network interface (UNI) and across a network-network interface (NNI).

Asynchronous Transfer Mode

278 •

CLP, cell loss priority, a n d

•

HEC, h e a d e r error control.

T h e only difference b e t w e e n t h e two h e a d e r s is t h a t t h e UNI h e a d e r h a s a 4-bit field t h a t c a n b e u s e d to signal to t h e u s e r t h e n e e d for flow control. T h e NNI u s e s t h o s e bits to e x p a n d t h e VPI field.

63.1

VCI and VPI Consider first t h e 8-bit VPI a n d t h e 16-bit VCI. T h e VPI/VCI c o m b i n a t i o n is t h e c o n n e c t i o n identifier. A virtual p a t h identifier (VPI) g r o u p s m a n y virtual c h a n n e l identifiers (VCI). ATM allows two levels of multiplexing: virtual chan­ n e l s m a y b e multiplexed into t h e s a m e virtual path, a n d virtual p a t h s m a y b e multiplexed into t h e s a m e link. T h e m o s t i m p o r t a n t feature is t h a t t h e VCI is local to e a c h VPI, a n d t h e VPI is local to e a c h link. More precisely, different s i m u l t a n e o u s c o n n e c t i o n s t h a t s h a r e t h e s a m e VPI m u s t h a v e different VCIs, a n d virtual p a t h s t h a t s h a r e t h e s a m e link m u s t h a v e different VPIs. But c o n n e c t i o n s o n different VPIs m a y s h a r e t h e same VCI, a n d virtual p a t h s o n different links m a y h a v e t h e same VPI. In t h e e x a m p l e below, s u p p o s e t h a t t h e entire link is d e v o t e d to a single VPI, so VCIs are local to a link. Because VCIs are local to a link, c o n n e c t i o n s c o m i n g into a switch from different switches (or sources) m a y h a v e t h e s a m e VCI. However, if t h e s e c o n n e c t i o n s s h a r e t h e s a m e outgoing link, t h e y m u s t b e assigned different VCIs o n t h a t link. T h i s w o r k s as in Figure 6.8, w h i c h s h o w s four u s e r n o d e s (A, B, C, D) a n d two switches. Initially t h e r e is a c o n n e c t i o n over VCI #1 from A to C. At s o m e later t i m e t h e r e is a r e q u e s t to establish a c o n n e c t i o n from Β to D. Since Β is n o t c u r r e n t l y u s i n g VCI # 1 , it assigns t h a t VCI to t h e c o n n e c t i o n . W h e n t h e p a c k e t establishing this c o n n e c t i o n r e a c h e s t h e first switch, t h a t switch k n o w s t h a t VCI #1 is assigned to a n o t h e r c o n n e c t i o n t h a t s h a r e s a link w i t h t h e p r o p o s e d c o n n e c t i o n . T h e switch therefore c h a n g e s t h e VCI from #1 to #2, n o t i n g t h e c h a n g e in a table. ( D u r i n g t h e data transfer phase, this switch m u s t c h a n g e t h e VCI o n e a c h cell o n t h e B - D c o n n e c t i o n from #1 to #2.) Strictly speaking, t h e VPI/VCI is assigned b y t h e n e t w o r k a n d n o t b y t h e u s e r as suggested in t h e discussion. T h e 16-bit VCI field p e r m i t s 64,000 s i m u l t a n e o u s c o n n e c t i o n s t h r o u g h e a c h link, a n d since t h e s a m e identifier m a y b e r e u s e d b y c o n n e c t i o n s w i t h disjoint paths, t h e n e t w o r k c a n s u p p o r t orders of m a g n i t u d e of m o r e s i m u l t a n e o u s connections.

6.3

ATM Header Structure

279

VCI#

vci

in

vci

#1 (2)

(1)

6.8 FIGURE

out

#2 (3)

Virtual channel identifiers m u s t be unique per connection on any given link. When the network sets u p the second virtual channel, it assigns it a different VCI on the link that the second virtual channel shares with the first. (1) VCI #1 from A to C has b e e n set up. (2) Request from Β for connection to D is initially assigned VCI #1. (3) Because of the c o m m o n link, the n e w connection is given VCI #2. The switch creates and updates the VCI translation table.

Some p u r p o s e s m a y b e b e t t e r s e r v e d b y t r e a t i n g several virtual c h a n n e l s t o g e t h e r as a group. Suppose, for example, as s h o w n in Figure 6.9, t h a t a u s e r w i s h e s to establish t h r e e virtual c h a n n e l s from A to B. It w o u l d t h e n m a k e s e n s e to give t h e s e c h a n n e l s t h e s a m e p a t h a n d p e r m i t t h e u s e r to assign VCIs to t h o s e c h a n n e l s arbitrarily. T h a t is t h e function of t h e VPI. Virtual c h a n n e l s

VPI 2; VCI 1,2,3 VPI 1; VCI 1,2,3

VPI 2; VCI 1,2

VPI

in

1 6.9 FIGURE

VPI 2

out

VPI

in

1

VPI

out

v rA

2

A virtual path is a group of virtual channels that the network routes together.

Asynchronous Transfer Mode

280

w i t h t h e s a m e VPI form a g r o u p . T h e y are assigned t h e s a m e p a t h a n d are switched together, t h a t is, r o u t i n g a n d switching decisions are b a s e d only o n t h e VPI. More interestingly, t h e b a n d w i d t h a n d buffer r e s o u r c e s allocated to a VPI m a y b e assigned statically (at t h e t i m e of service subscription) a n d s h a r e d only b y virtual c h a n n e l s w i t h t h a t VPI. This u s e of VPIs p e r m i t s t h e creation of virtual -private networks: a multilocation firm c a n r e n t several virtual p a t h s to form its o w n private n e t w o r k w h o s e r e s o u r c e s are t h e n s h a r e d b y its virtual channels. F r o m Figure 6.7 w e n o t e t h a t t h e UNI h e a d e r h a s a n 8-bit VPI field, while t h e NNI h a s a n additional 4 bits. T h e n e t w o r k m a y u s e t h e s e additional bits to create certain fixed routes, similar to w h a t is d o n e in today's long-distance t e l e p h o n e n e t w o r k s . This m a y allow r e s o u r c e s allocated to virtual p a t h s to b e statically assigned a n d s h a r e d d y n a m i c a l l y a m o n g c o m p o n e n t virtual c h a n n e l s . T h e u s e of VPIs m a y s p e e d u p processing, since switches only n e e d to consult a table w i t h e n t r i e s indexed b y s h o r t e r VPIs.

6.3.2

Other Fields T h e 4-bit GFC (generic flow control) m a y b e u s e d b y t h e n e t w o r k to signal to a u s e r t h e n e e d for m o m e n t a r y c h a n g e s in t h e i n s t a n t a n e o u s cell s t r e a m rate. T h e functionality of t h e GFC h a s n o t y e t b e e n established. Since GFC is n o t p r e s e n t in t h e NNI header, it c a n b e u s e d only to control t h e flow at t h e UNI interface. T h e 3-bit PT (payload t y p e ) p e r m i t s n e t w o r k s to distinguish a m o n g dif­ ferent t y p e s of information. If t h e first bit is 0 (PT = Oxx), it indicates u s e r information. In t h a t case t h e s e c o n d bit indicates congestion n o t e x p e r i e n c e d (PT = OOx) or congestion e x p e r i e n c e d (PT = Olx). T h e t h i r d bit is u s e d to dis­ tinguish two t y p e s of payload, d e p e n d i n g o n t h e context. For example, it is u s e d to indicate t h e e n d of a p a c k e t in AAL 5 (see section 6.4). PT = 100 or 101 is u s e d for m a i n t e n a n c e a n d for n e t w o r k e q u i p m e n t to i n t r o d u c e a n d r e m o v e special cells t h a t are r o u t e d as o r d i n a r y cells b u t t h a t c a r r y special information for control p u r p o s e s . PT = 110 is for r e s o u r c e m a n a g e m e n t cells. T h e 1-bit CLP (cell loss priority) distinguishes b e t w e e n cells t h a t t h e net­ w o r k m a y n o t discard (CLP = 0) a n d cells t h a t it m a y discard (CLP = 1) if n e c e s s a r y . Of course, t h e n e t w o r k m a y i n t r o d u c e a low-priority service (as a possible QpS) at t h e level of a connection—all cells in s u c h a c o n n e c t i o n are subject to discard. Since t h a t information is implicit in t h e VCI, t h e CLP field is n o t n e e d e d , a l t h o u g h it m a y b e u s e d to m a k e t h e information explicit. T h e CLP field is n e e d e d , however, if different loss priority cells are p r e s e n t

6.3

ATM Header Structure

in t h e s a m e c o n n e c t i o n . O n e e x a m p l e w o u l d b e if voice w e r e e n c o d e d into e q u a l n u m b e r s of h i g h a n d low o r d e r bits a n d s e n t over cells t h a t a l t e r n a t e l y carry t h e high a n d low o r d e r bits. T h e low o r d e r bit cells w o u l d b e assigned a CLP = 1, a n d w o u l d b e subject to discard, w h i l e t h e cells w i t h h i g h o r d e r bits w o u l d b e assigned CLP = 0 a n d w o u l d n o t b e discarded. By e n c o d i n g voice in this way, t h e statistical m u l t i p l e x i n g gain c a n b e i n c r e a s e d consid­ erably. Also, if a b o r d e r switch n o t i c e s t h a t cells do n o t c o n f o r m to t h e service p a r a m e t e r s , t h e n t h a t switch c a n set CLP = 1 in t h o s e cells so t h a t t h e s e n o n c o n f o r m a n t cells are eligible for discard. T h e b o r d e r switch d e t e c t s s u c h n o n c o n f o r m a n t cells b y u s i n g t h e GCRA w i t h t h e traffic p a r a m e t e r s , as w e explain in section 8.4.2. Finally, t h e 8-bit HEC ( h e a d e r e r r o r control) field is e q u a l to t h e s u m of t h e b y t e 0101Ό101 a n d t h e CRC (cyclic r e d u n d a n c y c h e c k ) calculated over t h e rest of t h e h e a d e r w i t h t h e g e n e r a t o r p o l y n o m i a l 1Ό000Ό111. ( T h e functioning of t h e CRC is discussed in section 2.6.3.) As e x p l a i n e d in section 6.1.2, t h e HEC is also u s e d to d e l i n e a t e t h e cell b o u n d a r y . T h e HEC c a n correct single bit errors a n d detect m u l t i p l e bit errors. T h e error-control a l g o r i t h m h a s t w o states: error d e t e c t i o n (D) a n d error correction (C). T h e a l g o r i t h m is initially in state C. If t h e algorithm is in state C a n d d e t e c t s a single bit error, it corrects t h e e r r o r a n d m o v e s to state D. If it d e t e c t s a m u l t i p l e bit error w h e n in state C, t h e a l g o r i t h m discards t h e cell a n d m o v e s to state D. T h e a l g o r i t h m discards cells t h a t c o n t a i n errors w h e n in state D a n d r e m a i n s in state D. T h e a l g o r i t h m m o v e s from state D to state C w h e n it gets a cell w i t h o u t error.

6.3.3

Reserved VCI/VPI S o m e VCI/VPI c o m b i n a t i o n s are r e s e r v e d . O n e set of c o m b i n a t i o n s t r a n s p o r t s so-called u n a s s i g n e d cells. T h e s e m a y b e i n t r o d u c e d b y t h e t r a n s p o r t l a y e r at t h e source, b u t t h e y do n o t c o n t a i n u s e r information. T h e s e cells are r e m o v e d b y t h e d e s t i n a t i o n t r a n s p o r t layer; t h e y are n o t forwarded to t h e application layer. O t h e r c o m b i n a t i o n s are r e s e r v e d for UNI a n d PNNI signaling a n d for ILMI messages. Sets of VCI/VPI c o m b i n a t i o n s are r e s e r v e d for cells t h a t c a n b e i n t r o d u c e d a n d r e m o v e d b y t h e physical layer. T h e s e cells are invisible to t h e ATM layer. O n e set is assigned to "idle" cells t h a t m a y b e i n t r o d u c e d periodically to h e l p s y n c h r o n i z a t i o n (cell b o u n d a r y r e c o v e r y ) . If t h e bit s t r e a m g e n e r a t e d b y t h e physical l a y e r is s y n c h r o n o u s (as in SONET), t h e n idle cells m a y b e u s e d to fill u p a frame w h e n e v e r t h e r e is n o u s e r traffic to b e sent. A n o t h e r set of VCI/VPI c o m b i n a t i o n s is assigned for OAM (organization, a d m i n i s t r a t i o n ,

281

Asynchronous Transfer Mode

282

a n d m a n a g e m e n t ) functions at t h e physical layer. In essence, t h e s e form c o n n e c t i o n s dedicated to assist in various m o n i t o r i n g a n d physical l a y e r control functions.

6.4

ATM ADAPTATION LAYER As s h o w n o n Figure 6.10, t h e n e t w o r k c o n v e r t s t h e information s t r e a m into a s t r e a m of 48-byte data cells. T h i s c o n v e r s i o n is p e r f o r m e d b y t h e ATM adaptation layer or AAL. T h e AAL is divided into two sublayers: t h e CS or c o n v e r g e n c e sublayer a n d t h e SAR or s e g m e n t a t i o n a n d r e a s s e m b l y sublayer. T h e CS c o n v e r t s t h e information s t r e a m into four t y p e s of p a c k e t s t r e a m s , called AAL Type 1, Type 2, Type 3/4, a n d Type 5. T h e p a c k e t formats m a t c h t h e r e q u i r e m e n t s of t h e information s t r e a m , classified into five t y p e s of trafficc o n s t a n t bit r a t e - r e a l time, variable bit r a t e - r e a l time, c o n n e c t i o n - o r i e n t e d packet s t r e a m s , datagrams, a n d IP packets, as indicated in Tkble 6.1. Five AAL t y p e s of p a c k e t s w e r e originally envisioned. AAL Type 2 w a s r e c e n t l y specified for u s e in n a r r o w b a n d services. T h e individual p a c k e t s are called SDUs (service data units), following OSI terminology. Since a u s e r information s t r e a m m a y b e e n c o d e d into p a c k e t s t r e a m s in m a n y different w a y s (e.g., video m a y b e e n c o d e d into a c o n s t a n t or variable bit rate s t r e a m ) , a n d since s o m e of this e n c o d i n g i n f o r m a t i o n m u s t b e i n c l u d e d in t h e SDU, CS tasks are likely to b e a p p l i c a t i o n - d e p e n d e n t . H e n c e

Information stream

5 types 1: SN SNP

Convergence sublayer Packets or - CS-PDUs in many formats Segmentation/reassembly (SAR) 48-byte cells

610 FIGURE

1 0DD :

2:

SF

3,4:

2 ST

5:

24 Η 4 SN

SAR-SDU

SAR-SDU 10 RES/MID

PAD

6 10 SAR-SDU LI CRC

SAR-SDU

The ATM adaptation layer (AAL) converts the information stream into 48byte cells. The AAL is decomposed into the convergence sublayer and the segmentation/reassembly sublayer.

6.4

ATM Adaptation Layer

283

CS is further subdivided into t h e upper, service-specific or SSCS s u b l a y e r a n d t h e lower, c o m m o n p a r t or CPCS sublayer. By contrast, SAR tasks are q u i t e standard, d e p e n d i n g only o n traffic type. T h e SAR m u s t s e g m e n t t h e SDUs r e c e i v e d from t h e CS, a d d t h e n e c e s s a r y overhead, a n d c o n v e r t t h e result into 48-byte cells or PDUs (protocol data units), w h i c h are t h e n h a n d e d over to t h e ATM layer. As w e will explain, t h e o v e r h e a d is n e e d e d b y t h e SAR at t h e d e s t i n a t i o n to r e a s s e m b l e t h e SDUs from t h e 48-byte cells. We describe e a c h type, its typical i n t e n d e d applications, a n d t h e corre­ s p o n d i n g SAR functions.

6.4.1

Type 1 This traffic is g e n e r a t e d at a c o n s t a n t bit rate, a n d it is r e q u i r e d to b e delivered at t h e s a m e rate (with a fixed delay). I n t e n d e d applications are voice a n d c o n s t a n t bit rate video or audio. Since t h e cells m a y suffer variable delays, t h e CS at t h e d e s t i n a t i o n m u s t c o m p e n s a t e for t h o s e delays. T h i s is d o n e in o n e of t w o ways. I n c o m i n g cells are buffered at t h e d e s t i n a t i o n a n d t h e buffer is r e a d o u t at a fixed rate. Alternatively, t h e CS at t h e s o u r c e i n s e r t s a n explicit t i m e s t a m p into t h e p a c k e t s t r e a m . T h e d e s t i n a t i o n CS extracts this t i m e s t a m p a n d a t t e m p t s to r e a d cells w i t h a c o n s t a n t l a t e n c y . (A cell t h a t arrives w i t h a d e l a y t h a t exceeds this l a t e n c y m a y b e discarded.) T h e source a n d d e s t i n a t i o n CS s u b l a y e r s m u s t agree o n t h e s c h e m e to b e u s e d for t i m i n g information. T h a t a g r e e m e n t m u s t be reached during the connection setup phase. Figure 6.10 displays t h e s t r u c t u r e of t h e 48-byte PDU for Type 1 traffic. T h e SAR s u b l a y e r takes t h e (periodic) p a c k e t s t r e a m g e n e r a t e d b y t h e CS sublayer, s e g m e n t s it into 47-byte SDUs, a n d p r e p e n d s to e a c h SDU a 4-bit s e q u e n c e n u m b e r (SN) p r o t e c t e d b y a 4-bit s e q u e n c e n u m b e r p r o t e c t i o n (SNP) field. T h e SNP corrects single bit errors a n d d e t e c t s m u l t i p l e bit errors in t h e s e q u e n c e n u m b e r . T h e eight possible s e q u e n c e n u m b e r s p e r m i t t h e SAR s u b l a y e r at t h e d e s t i n a t i o n to d e t e r m i n e if fewer t h a n eight c o n s e c u t i v e SDUs are lost. ( T h e source a n d d e s t i n a t i o n m u s t agree o n w h a t to do w h e n cell loss is discovered.)

6.4.2

Type 2 AAL Type 2 w a s r e c e n t l y specified. It is i n t e n d e d to serve t h e n e e d to t r a n s p o r t low bit rate traffic s u c h as c o m p r e s s e d or u n c o m p r e s s e d voice, c o m p r e s s e d video, a n d fax. It allows m u l t i p l e x i n g m o r e t h a n o n e AAL 2 traffic s t r e a m over t h e s a m e c o n n e c t i o n . AAL 2 p e r m i t s a u s e r p a c k e t to b e split across two cells.

Asynchronous Transfer Mode

284

T h e PDU h a s t h r e e parts. T h e 8-bit start field (SF) i n c l u d e s a 6-bit off­ set field (OSF), w h i c h indicates t h e n u m b e r of b y t e s b e t w e e n t h e start field a n d t h e payload, or b e t w e e n t h e start field a n d t h e next p a c k e t in case t h e payload c o n t a i n s t h e e n d of o n e u s e r p a c k e t a n d t h e b e g i n n i n g of t h e next packet. T h e next field is t h e 24-bit h e a d e r (H) of t h e SAR-SDU. T h e h e a d e r i n c l u d e s t h e c h a n n e l ID (CID) for u s e in multiplexing several s t r e a m s , a n d t h e l e n g t h indicator (LI) for variable-length payload. T h e SAR-SDU payload c a n h a v e variable l e n g t h a n d m a y h a v e m u l t i p l e SDUs from different AAL 2 s t r e a m s . T h e final field c o m p r i s e s p a d d i n g bytes, if n e e d e d , to fill u p t h e 48 b y t e s . AAL 2 m a y b e u s e d to s u p p o r t voice over ATM. T h e specification d o e s n o t define t h e t i m i n g relationship b e t w e e n s e n d i n g a n d receiving applications, or h o w to c o m p e n s a t e for jitter. T h a t is left to applications. For instance, t h e Real T i m e Protocol (see C h a p t e r 4) provides a t i m e s t a m p .

6.4.3

Type 3/4 T h e original s t a n d a r d specified two t y p e s : AAL 3 for c o n n e c t i o n - o r i e n t e d streams, a n d AAL 4 for c o n n e c t i o n l e s s m e s s a g e s . T h e two t y p e s are n o w com­ b i n e d into a single type, AAL 3/4, i n t e n d e d for i n t e r n e t w o r k i n g SMDS a n d MAN n e t w o r k s over ATM. A A L 3 / 4 accepts CS-PDUs of u p to 64 KB, w h i c h are s e g m e n t e d into 53-byte cells for t r a n s m i s s i o n . Two t y p e s of data trans­ fer are s u p p o r t e d : a m e s s a g e - m o d e service w h i c h allows a single SDU a n d a s t r e a m - m o d e in w h i c h o n e or m o r e fixed-size SDUs are t r a n s p o r t e d . Two k i n d s of error quality are envisaged. In t h e first k i n d t h e CS s u b l a y e r g u a r a n t e e s error-free delivery. In t h e s e c o n d k i n d n o s u c h g u a r a n t e e is m a d e . Error-free delivery is e n s u r e d b y t h e d e s t i n a t i o n CS r e q u e s t i n g r e t r a n s m i s s i o n w h e n a n error is detected. Tb e n s u r e r e c o n s t r u c t i o n of t h e p a c k e t s a n d to d e t e c t errors, t h e o v e r h e a d fields a p p e n d e d b y t h e SAR s u b l a y e r are as in Figure 6.10. T h e 2-bit ST or s e g m e n t t y p e is set as follows: ST = 10 for BOM, ST = 00 for COM, ST = 01 for EOM, a n d ST = 11 if t h e p a c k e t fits inside a single cell. T h e 4-bit SN ( s e q u e n c e n u m b e r ) is u s e d as before to d e t e c t loss or cells i n s e r t e d b y n e t w o r k routing errors. T h e 10-bit MID field is explained below. T h e 6-bit LI field is n e e d e d w h e n t h e cell is only partially filled. Finally t h e 10-bit CRC is a c h e c k over t h e SAR-SDU u s e d to d e t e c t errors in t h e CS SDUs. It w o u l d b e c o m m o n for a u s e r to multiplex several different d a t a g r a m s over t h e s a m e virtual c h a n n e l c o n n e c t i o n . T h e MID (multiplexing identifier) field c a n b e u s e d to distinguish a m o n g t h e cells originating from different d a t a g r a m s . (This is similar to t h e u s e of t h e MID or m e s s a g e identifier field in SMDS.)

6.5

Management and Control

6.4.4

285

Type 5 This traffic carries IP packets. T h e frame s t r u c t u r e of Type 5 e l i m i n a t e s t h e o v e r h e a d p r e s e n t in Type 3 / 4 . T h e IP packet, u p to 6 4 K b y t e s , is p a c k a g e d into a CS p a c k e t t h a t c o n t a i n s a l e n g t h indicator a n d a 32-bit CRC calculated over t h e c o m p l e t e CS p a c k e t w i t h t h e g e n e r a t o r

lOOOOOlOO'llOO'OOOlOOOl'llOl'lOll'Olll. T h e CS p a c k e t c o n t a i n s a p a d d i n g field so t h a t t h e l e n g t h of this CS p a c k e t is a n exact m u l t i p l e of 48 b y t e s . T h e CS p a c k e t is s e n t as a s e q u e n c e of SAR p a c k e t s of AAL Type 5. T h e PT of t h e ATM cell h e a d e r indicates w h e t h e r t h e cell is t h e last o n e of t h e CS p a c k e t or not.

6.5

MANAGEMENT AND CONTROL A v e r y i m p o r t a n t feature of ATM n e t w o r k s is t h a t t h e y c a n m a k e a n u m b e r of m a n a g e m e n t a n d control decisions to d i s c r i m i n a t e a m o n g c o n n e c t i o n s a n d to provide t h e variety of QpS t h a t different applications n e e d . T h e decisions are divided into t h r e e groups. W h e n a r e q u e s t is m a d e for a c o n n e c t i o n w i t h a particular QpS, t h e n e t w o r k m u s t d e t e r m i n e w h e t h e r to a c c e p t or reject t h e request, d e p e n d i n g o n t h e r e s o u r c e s t h e n available. (Recall t h a t QpS involves t h r e e sets of p a r a m e t e r s : delay, cell loss, a n d s o u r c e traffic rate.) If t h e r e s o u r c e s are insufficient to m e e t t h e request, t h e n e t w o r k m a y n e g o t i a t e w i t h t h e u s e r t h e traffic p a r a m e t e r s in t h e r e q u e s t e d service class. O n c e t h e c o n n e c t i o n is a d m i t t e d , t h e n e t w o r k m u s t assign a r o u t e or p a t h to t h e virtual c h a n n e l t h a t carries t h e c o n n e c t i o n . It m u s t inform t h e switches a n d o t h e r n e t w o r k e l e m e n t s along t h e p a t h t h a t this virtual c h a n n e l m u s t b e allocated c e r t a i n r e s o u r c e s so t h a t t h e agreed-on QpS is m e t . Lastly, t h e n e t w o r k m u s t m o n i t o r t h e data transfer to m a k e s u r e t h a t t h e source also conforms to t h e QpS specification a n d to d r o p its cells as appropriate. (This is called traffic policing.) T h e n e t w o r k m a y also ask a source to slow d o w n its t r a n s m i s s i o n s . In addition, t h e n e t w o r k carries a n u m b e r of i n f o r m a t i o n flows to m o n i t o r its o p e r a t i o n s a n d to d e t e c t a n d identify t h e location of c o n g e s t e d or failed devices. T h e BISDN s t a n d a r d so far is silent a b o u t h o w t h e s e decisions are to b e carried out. (However, r e c e n t ITU r e c o m m e n d a t i o n s deal w i t h OAM, perfor­ m a n c e m o n i t o r i n g a n d p r o t e c t i o n switching at t h e ATM l a y e r to c o m p l e m e n t

Asynchronous Transfer Mode

286 Layer management

Plane management

Control plane

User plane

I = Necessary for SVC

6.11 FIGURE

The BISDN model layer arrangement of network functions, including the operation and m a n a g e m e n t functions.

SONET protection switching.) We shall discuss potential solutions in C h a p t e r 8. T h e ATM F o r u m specifies frame formats t h a t t h e n e t w o r k s h o u l d u s e to c a r r y its m o n i t o r i n g information a n d to interact w i t h u s e r s . We r e v i e w t h e s e next. T h e n e t w o r k u s e s o p e r a t i o n a n d m a i n t e n a n c e information flows for t h e following functions: •

fault m a n a g e m e n t ,

•

traffic a n d congestion control,

•

n e t w o r k status m o n i t o r i n g a n d configuration, a n d

•

u s e r / n e t w o r k signaling.

T h e s e functions, like t h e o t h e r n e t w o r k functions, are organized into layers, called t h e BISDN reference model. Figure 6.11 shows t h e l a y e r a r r a n g e m e n t of all n e t w o r k functions, i n c l u d i n g those of o p e r a t i o n a n d m a n a g e m e n t . T h e l a y e r s in t h e user plane c o m p r i s e t h e functions r e q u i r e d for t h e t r a n s m i s s i o n of u s e r information. For instance, for a n I n t e r n e t Protocol over ATM, t h e s e layers could b e H T T P / T C P / I P / A A L 5 . T h e layers in t h e control plane are t h e functions n e e d e d to set u p , super­ vise, a n d release a virtual circuit c o n n e c t i o n . T h e s e functions, i m p l e m e n t e d b y signaling protocols s u c h as PNNI, are n e e d e d only for s w i t c h e d virtual con­ n e c t i o n s a n d are a b s e n t in a n e t w o r k t h a t i m p l e m e n t s o n l y p e r m a n e n t virtual c o n n e c t i o n s . (In a p e r m a n e n t virtual circuit c o n n e c t i o n , t h e p a t h or r o u t e as­ signed to a source a n d d e s t i n a t i o n a n d t h e VCI for t h a t r o u t e are fixed.) T h e layer management plane c o n t a i n s m a n a g e m e n t functions specific to individual layers. Layer m a n g e m e n t also h a n d l e s t h e o p e r a t i o n s a n d m a i n -

6.5

Management and Control

287

t e n a n c e flows specific to e a c h layer. T h e protocols u s e d for t h e s e functions i n c l u d e ILMI a n d SNMP. Finally, plane management consists of t h e functions t h a t s u p e r v i s e t h e o p e r a t i o n s of t h e w h o l e n e t w o r k . P l a n e m a n a g e m e n t h a s n o l a y e r e d s t r u c t u r e .

6.5.1

Fault Management C o n s i d e r a virtual circuit c o n n e c t i o n over a n ATM n e t w o r k a n d a s s u m e t h a t t h e c o n n e c t i o n is i m p l e m e n t e d b y a SONET n e t w o r k . We k n o w from section 5.2 t h a t SONET establishes t r a n s m i s s i o n paths for t h e ATM layer. T h e t r a n s m i s s i o n is over optical fibers. T h e t r a n s m i t t e r s in SONET are all s y n c h r o n i z e d to t h e s a m e m a s t e r clock. T h i s s y n c h r o n i z a t i o n e n a b l e s t h e time-division multiplex­ ing of different bit s t r e a m s . T h i s m u l t i p l e x i n g is d o n e byte-by-byte. T h e physical l a y e r (SONET) is d e c o m p o s e d into t h r e e sublayers: section, line, a n d p a t h . T h e section l a y e r t r a n s m i t s bits b e t w e e n a n y two devices w h e r e light is c o n v e r t e d b a c k into electronic signals or conversely. For instance, t h e r e is a section b e t w e e n two successive r e g e n e r a t o r s or b e t w e e n a r e g e n e r a t o r a n d a multiplexer. T h e line l a y e r t r a n s p o r t s bits b e t w e e n m u l t i p l e x e r s w h e r e SONET signals are a d d e d to or d r o p p e d from t h e t r a n s m i s s i o n . Finally, t h e p a t h l a y e r t r a n s p o r t s u s e r information. T h u s , a p a t h goes across a n u m b e r of l i n e s (or links) t h a t are switched b y t h e SONET d e m u l t i p l e x e r s a n d multiplexers, a n d a line consists of a n u m b e r of sections. Each l a y e r i n s e r t s a n d strips its o w n o v e r h e a d information, w h i c h it u s e s to m o n i t o r t h e t r a n s m i s s i o n functions for w h i c h it is responsible. (See Figure 6.12.) Each of t h e t h r e e sublayers u s e s o v e r h e a d b y t e s in t h e SONET frames to s u p e r v i s e its operations. T h e o v e r h e a d b y t e s are said to c a r r y a flow of o p e r a t i o n a n d m a i n t e n a n c e information. T h e flow carried b y t h e section o v e r h e a d b y t e s is called F l . T h e flow carried b y t h e l i n e a n d p a t h o v e r h e a d b y t e s are F2

vcc

F5

VPC

F4

Path

F3 F2

Line

F2

Section

Fl

6.12 FIGURE

Fl

Fl

Fl

Operation and maintenance flows for a virtual circuit connection over SONET.

Asynchronous Transfer Mode

288 End-to-end

Segment

Terminal or router

6.13 FIGURE

Private ATM switch

Public ATM switch

Private ATM switch

Public ATM switch

Terminal or router

A segment indicates a connection across the user-network interface. An end-to-end connection is b e t w e e n the source and destination user equipment.

a n d F3, respectively. T h e virtual circuit c o n n e c t i o n is c a r r i e d b y a virtual p a t h c o n n e c t i o n . Accordingly, t h e n e t w o r k u s e s a flow of cells to s u p e r v i s e t h e virtual p a t h c o n n e c t i o n a n d a flow of cells to s u p e r v i s e t h e virtual circuit c o n n e c t i o n . T h e s e two flows are called F4 a n d F5, respectively. T h e format of t h e F4 a n d F5 cells d e p e n d s o n w h e t h e r t h e cells m o n i t o r t h e s e g m e n t across t h e u s e r - n e t w o r k interface or t h e end-to-end c o n n e c t i o n (see Figure 6.13). T h e cell formats are s h o w n in Figure 6.14. Note t h a t t h e F5 cells h a v e t h e s a m e VPI/VCI as t h e u s e r cells of t h e c o n n e c t i o n t h e y monitor.

Same as user's cells 3: Segment 4: End-to-end

VPC: F4

GFC

VCC: F5

6.14 FIGURE

VPI

VCI

PT

0001 = Fault management

CLP HEC

100 = segment 101 = End-to-end Same as user's cells

Format of OAM cells.

OAM cell type

FT

Functionspecific field

CRC-10

Function type: 0000 = AIS 0001=FERF 1000 = Loopback

6.5

Management and Control

289 35

bytes Failure type bytes

Unused

1

4

12

12

16

Loopback indication

Correlation tag

Loopback location ID

Source ID

Unused

000000j 6.15 ™ G ™ R ™

Failure location

If all Is, then end point

Function-specific fields in AIS and FERF cells (above) and in loopback cells (below).

T h e F5 cells are d i s t i n g u i s h e d from t h e u s e r cells b y t h e PT field. Similarly, t h e F4 cells h a v e t h e s a m e VPI as t h e u s e r cells a n d are d i s t i n g u i s h e d b y t h e i r VCI. T h e m a i n function of t h e OAM cells is to d e t e c t a n d m a n a g e faults. Faultm a n a g e m e n t OAM cells h a v e t h e l e a d i n g 4 bits of t h e cell p a y l o a d set to 0001. T h e next 4 bits, t h e function t y p e (FT) field, indicate t h e t y p e of function p e r f o r m e d b y t h e cell: a l a r m i n d i c a t i o n signal (AIS), signaled b y FT = 0000; far e n d receive failure (FERF), signaled b y F T = 0001; a n d l o o p b a c k cell, signaled b y FT = 1000. T h e AIS cells are s e n t along t h e VPC (virtual p a t h c o n n e c t i o n ) or VCC (virtual circuit c o n n e c t i o n ) b y a n e t w o r k device t h a t detects a n e r r o r condition along t h e c o n n e c t i o n . T h o s e cells are t h e n s e n t along to t h e d e s t i n a t i o n of t h e c o n n e c t i o n . W h e n t h e e q u i p m e n t at t h e e n d of t h a t c o n n e c t i o n receives t h e AIS, it s e n d s b a c k FERF cells to t h e o t h e r e n d of t h e c o n n e c t i o n . As s h o w n in Figure 6.15, t h e AIS a n d FERF cells specify t h e t y p e of failure as well as t h e failure location. A loopback cell c o n t a i n s a field t h a t specifies w h e t h e r t h e cell s h o u l d b e l o o p e d back, a correlation tag, a l o o p b a c k location identification, a n d a source identification. T h e s e loopback cells are u s e d as s h o w n in Figure 6.16. T h e device t h a t r e q u e s t s a loopback (we call it t h e source) i n s e r t s a l o o p b a c k cell a n d selects a value for t h e correlation tag. T h e device c a n specify w h e r e t h e loopback s h o u l d take place. T h e device sets t h e l o o p b a c k i n d i c a t i o n field of t h e cell to 1 to indicate t h a t t h e cell m u s t b e l o o p e d b a c k . W h e n t h e device w h e r e t h e loopback m u s t o c c u r receives t h e cell, it sets its l o o p b a c k indication field to 0 a n d s e n d s t h e cell b a c k to t h e source. T h e s o u r c e c o m p a r e s t h e correlation tag of t h e cell it receives w i t h t h e value it selected. T h i s correlation tag p r e v e n t s a device from getting confused b y o t h e r l o o p b a c k cells.

Asynchronous Transfer Mode

290

End-to-end

Loopback location = end Correlation tag = X Loopback indication = 1

End-to-end

Terminal or router

6.16

Private ATM switch

Public ATM switch

Loopback location = end Correlation tag = X Loopback indication = 0

Public ATM switch j

Segment

Loopback location = end Correlation tag = X Loopback indication = 1

Segment

Loopback location = end Correlation tag = X Loopback indication = 0

Private ATM switch

Terminal or router

Loopback at the end of connection (above) and at the segment (below).

FIGURE

O n e o t h e r OAM function m a y b e m e n t i o n e d . P e r f o r m a n c e m a n a g e m e n t , indicated b y OAM cell t y p e 0010, consists of forward m o n i t o r i n g , b a c k w a r d monitoring, a n d reporting. In forward m o n i t o r i n g , for example, a block of cells of o n e c o n n e c t i o n is b o u n d e d b y OAM cells, w h i c h include, a m o n g o t h e r information, t h e size of t h e block. T h e receiving n o d e c a n c o m p a r e t h e n u m b e r of received cells w i t h t h e block size to d e t e c t m i s s i n g or i n s e r t e d cells.

6.5.2

Traffic and Congestion Control T h e objectives of traffic a n d congestion control are to g u a r a n t e e t h e c o n t r a c t e d quality of service to virtual c o n n e c t i o n s . T h e o p e r a t i o n s t h a t t h e n e t w o r k p e r f o r m s are t h e subject of C h a p t e r s 8 a n d 9.

6.5

Management and Control

6.53

291

Network Status Monitoring and Configuration T h e OAM functions d e s c r i b e d above do n o t provide diagnostic, m o n i t o r i n g , a n d configuration services across t h e u s e r - n e t w o r k interfaces. T h a t is t h e p u r p o s e of t h e Integrated Local Management Interface or ILMI protocol, Version 4.0. ILMI u s e s t h e Simple N e t w o r k M a n a g e m e n t Protocol (SNMP) a n d a m a n a g e m e n t i n f o r m a t i o n b a s e (MIB). T h e situation is illustrated in Figure 6.17. T h e figure s h o w s a private ATM n e t w o r k c o n n e c t e d b y a private ATM switch to a public ATM n e t w o r k . E a c h c o n n e c t i o n across t w o interfaces is s u p e r v i s e d b y two ATM Interface M a n a g e m e n t Entities (IMEs): o n e for e a c h of t h e ATM devices. Two s u c h IMEs are said to b e adjacent, ana t h e ILMI specifies t h e s t r u c t u r e of t h e M a n a g e m e n t I n f o r m a t i o n Base (MIB) t h a t c o n t a i n s t h e attributes of t h e c o n n e c t i o n s u p e r v i s e d b y t h e adjacent entities. T h e ATM devices m a y b e w o r k s t a t i o n s w i t h ATM interfaces t h a t s e n d ATM cells to a n ATM switch, or ATM switches, or IP r o u t e r s t h a t transfer t h e i r p a c k e t s w i t h i n ATM cells to a n ATM switch. ILMI 4.0 describes four MIB m o d u l e s . T h e Tfextual C o n v e n t i o n s MIB de­ fines c o m m o n textual c o n v e n t i o n s . T h e Link M a n a g e m e n t M o d u l e defines t h e objects for e a c h ATM interface a n d t h e m e t h o d s to d e t e c t ILMI c o n n e c t i v i t y b e t w e e n IMEs. It is further d e s c r i b e d below. T h e Address Registration MIB s u p p o r t s p r o c e d u r e s for t h e u s e r a n d t h e n e t w o r k to k n o w e a c h other's ATM address. T h e Service Registry MIB h e l p s to locate ATM n e t w o r k services s u c h as LAN E m u l a t i o n Configuration Servers (LECS).

NMS r- SNMP/UDP/IP/AAL I

Terminal or router Β

6.17 GURΕ

SNMP/AAL

Private ATM switch

Public ATM switch

User-network interface management entity

The Integrated Local Management Interface (ILMI) protocol is designed to supervise the connections across user-network interfaces.

Asynchronous Transfer Mode

292 I—

ILMI

Status, configuration, control information The ILMI attributes are organized as one MIB per IME.

ι—

MIB Cells received/dropped/sent

I Physical ATM VPC In/out loop

VCC

ATM statistics

Traffic descriptors/QoS; status (up, down)

ABR-VPC ABR-VCC

I

I

Operational parameters

Max. no. VPCs, max. no. VCCs, no. VPCs, no. VCCs, no. active VPI bits,... -

SNMP Get, Get-Next, Set, Trap

6.18 Fl GU RE

Structure of the ILMI link m a n a g e m e n t MIB that contains the attributes of ^ t i ° supervised by adjacent interface m a n a g e m e n t entities. e

c o n n e c

n

I m p o r t a n t objects of t h e link m a n a g e m e n t MIB are s u m m a r i z e d in Fig­ u r e 6.18. As t h e figure indicates, o n e MIB is defined p e r IME. T h e c o n t e n t s of a n IME MIB are t h e attributes of t h e physical l a y e r ( w h i c h i m p l e m e n t s t h e bit way), t h e ATM traffic, t h e VPCs, a n d VCCs t h a t go across t h a t UNI. T h e fig­ u r e indicates r e p r e s e n t a t i v e attributes. T h e ATM statistics attributes are n o w d e e m p h a s i z e d . T h e MIB also i n c l u d e s ABR attributes. T h e ABR virtual p a t h a n d virtual c h a n n e l are t u n e d o n a p e r - c o n n e c t i o n basis via t h e s e attributes. Examples of ABR attributes are ICR (initial cell rate), w h i c h is a n u p p e r b o u n d o n t h e source's t r a n s m i s s i o n rate; RIF (rate i n c r e m e n t factor), w h i c h controls t h e allowed increase in t h e source t r a n s m i s s i o n rate; a n d RDF (rate d e c r e m e n t factor), w h i c h controls t h e r e q u i r e d d e c r e a s e in t h a t rate. T h e s e a t t r i b u t e s are set u s i n g ABR r e s o u r c e m a n a g e m e n t (RM) cells, as discussed in section 8.4.2.

6.5.4

User/Network Signaling T h e basic signaling functions b e t w e e n t h e n e t w o r k a n d a u s e r are as follows: •

t h e u s e r r e q u e s t s a switched virtual c o n n e c t i o n ,

•

t h e n e t w o r k indicates w h e t h e r t h e r e q u e s t is a c c e p t e d or not, a n d

•

t h e n e t w o r k indicates error conditions w i t h a c o n n e c t i o n .

We h a v e discussed t h e UNI signaling protocol above.

6.6

BISDN

293

6.6

BISDN T h e ATM b e a r e r service is t h e t r a n s p o r t of cells w i t h a variety of quality of service. Tbgether w i t h t h e formats defined for t h e ATM a d a p t a t i o n layer, this b e a r e r service c a n b e u s e d to s u p p o r t a w i d e r a n g e of applications. In this sec­ tion w e explain h o w t h e ATM b e a r e r service in t u r n c a n b e i m p l e m e n t e d u s i n g SONET n e t w o r k s . T h e result is a n i m p l e m e n t a t i o n of a B r o a d b a n d I n t e g r a t e d Services Digital Network, or BISDN. Figure 6.11 illustrates t h e BISDN r e f e r e n c e m o d e l . T h e s t a n d a r d specifies b o t h t r a n s p o r t a n d physical layers. T h e n e t w o r k a n d data link l a y e r s are t h e s a m e as in t h e ATM s t a n d a r d s d e s c r i b e d above. T h e physical l a y e r s t a n d a r d h a s b e e n specified for SONET STS-12C (622.08 Mbps) a n d STS-3C (155.52 Mbps) signals a n d for DS3 (44.736 Mbps) signals. We illustrate t h e ideas for STS-3C. T h e STS-3C frame, s h o w n in Figure 6.19, is a r r a n g e d in a 9 χ 270 b y t e m a t r i x . T h e first n i n e c o l u m n s are d e v o t e d to section a n d line o v e r h e a d (SOH, LOH). T h i s leaves 9 χ 261 b y t e s , of w h i c h o n e c o l u m n is d e v o t e d to p a t h over­ h e a d (ΡΟΗ). T h e resulting 9 χ 260 b y t e SPE ( s y n c h r o n o u s p a y l o a d e n v e l o p e ) c o n t a i n s consecutively a r r a n g e d 53-byte ATM cells. (Details of SONET c a n b e found in section 5.2.) T h e m o s t i m p o r t a n t features of this frame s t r u c t u r e are t h e following. T h e ATM bit rate is 155.52 χ 260/270 = 149.76 Mbps. T h e 9 χ 260 = 2,340 b y t e SPE h o l d s a b o u t 44.15 53-byte ATM cells. T h u s ATM cell b o u n d a r i e s b e a r n o

Information stream AAL 48-byte cells ATM layer 53-byte ATM cells SONET layer Bit stream

6.19 FIGURE

155.52 Mbps

BISDN over STS-3C SONET frames.

Asynchronous Transfer Mode

294

relationship to t h e SPE or STS-3C frame b o u n d a r i e s . ( T h e H4 b y t e in Ρ Ο Η is a p o i n t e r to t h e n e x t o c c u r r e n c e of t h e cell b o u n d a r y a n d m a y b e u s e d b y t h e destination to recover cell b o u n d a r i e s . However, t h e s t a n d a r d n o w r e q u i r e s r e c o v e r y of t h e cell b o u n d a r y from t h e h e a d e r CRC, as explained before.) If t h e t r a n s p o r t layer does n o t provide a sufficient n u m b e r of ATM cells, t h e physical layer inserts idle ATM cells into t h e frame, w h i c h are r e m o v e d b y t h e physical l a y e r at t h e destination. As w i t h SONET traffic generally, t h e SPE is n o t aligned w i t h t h e STS-3C frame. T h e SPE location is obtained from t h e AU-4 p o i n t e r in t h e LOH. F u t u r e subscribers to BISDN m i g h t receive a n STS-12 (622.08-Mbps) signal, a n d t h e y will b e able to s e n d a n STS-3C signal. T h e subscriber c a n s h a r e this b a n d w i d t h a m o n g m a n y s i m u l t a n e o u s c o n n e c t i o n s of v a r y i n g QpS: c o n s t a n t a n d variable bit rate video a n d audio traffic w i t h real-time constraints, data traffic w i t h g u a r a n t e e d r e t r a n s m i s s i o n in case of e r r o r for file transfer traffic, d a t a g r a m s traffic for s h o r t t r a n s a c t i o n s s u c h as database q u e r i e s or r e m o t e p r o c e d u r e calls, a n d so on. T h e ATM/BISDN s t a n d a r d is sufficiently flexible to a c c o m m o d a t e this variety of traffic service. BISDN is a p r o d u c t of a t e l e p h o n e - c e n t r i c view of n e t w o r k i n g . T h e stan­ dards p a y m u c h a t t e n t i o n to i n t e r c o n n e c t i n g public t e l e p h o n e carriers, a n d t h e y a s s u m e d t h a t ATM w a s flexible e n o u g h to s u p p o r t all services. Over time, however, t h e situation s e e m s to h a v e reversed, a n d ATM h a s t a k e n over t h e b u r d e n of i n t e r n e t w o r k i n g . BISDN h a s b e e n relegated into t h e b a c k g r o u n d .

6.7

INTERNETWORKING WITH ATM A n ATM n e t w o r k c a n b e u s e d to c a r r y i n t e r n e t w o r k traffic. For instance, a n ATM n e t w o r k c a n b e u s e d to i n t e r c o n n e c t various LANs or IP s u b n e t w o r k s . Such i n t e r n e t w o r k i n g c a n take place at t h e data link l a y e r (bridging) or at t h e n e t w o r k l a y e r (routing). For m o r e details o n t h e m a t e r i a l of this section, w e refer t h e r e a d e r to [AL95]. I n t e r n e t w o r k i n g is a major challenge facing ATM. TWO basic tasks are re­ q u i r e d for i n t e r n e t w o r k i n g over ATM. T h e first is e n c a p s u l a t i o n of t h e protocol data units, a n d t h e s e c o n d is t h e r o u t i n g or bridging of t h e s e PDUs. T h e r o u t i n g itself consists of t h e r o u t e calculation a n d t h e switching of packets. T h e r o u t e calculation necessitates t h e a d d r e s s resolution p l u s a r o u t i n g algorithm. T h e address resolution m a p s t h e protocol address ( s u c h as IP, MAC, FR, or SMDS) into a n ATM address. T h e r o u t i n g a l g o r i t h m calculates t h e r o u t e s t h r o u g h t h e

6.7

Internetworking with ATM

n e t w o r k . T h a t r o u t i n g a l g o r i t h m for t h e private ATM n e t w o r k is PNNI, as w e l e a r n e d in section 6.2.2. T h e non-ATM n e t w o r k c a n u s e e i t h e r its o w n r o u t i n g a l g o r i t h m or a n e x t e n s i o n of PNNI. I n t h e following s u b s e c t i o n s w e explain e n c a p s u l a t i o n , LAN e m u l a t i o n , IP over ATM, a m o r e g e n e r a l multiprotocol over ATM, a n d FR or SMDS over ATM. It s h o u l d b e n o t e d t h a t i n t e r n e t w o r k i n g involves o t h e r tasks n o t c o n s i d e r e d h e r e . T h e s e are signaling i n t e r w o r k i n g a n d u s e r data transfer i n t e r n e t w o r k i n g b e t w e e n different protocols.

67.1

Multiprotocol Encapsulation over AAL5 T h e e n c a p s u l a t i o n of i n t e r n e t w o r k traffic is defined in RFC 1483. T h e RFC specifies two m e t h o d s d e p e n d i n g o n w h e t h e r o n e ATM VCC carries a single or m u l t i p l e protocols. In e i t h e r case, t h e protocol data u n i t s are c a r r i e d b y t h e p a y l o a d of t h e c o n v e r g e n c e s u b l a y e r of AAL5. T h a t is, a PDU of u p to 64 KB is first p a d d e d so that, t o g e t h e r w i t h t h e 8-byte c o n v e r g e n c e s u b l a y e r trailer t h a t specifies t h e l e n g t h of t h e PDU a n d a CRC, it a d d s u p to a m u l t i p l e of 48 b y t e s . If t h e VCC carries a single protocol, this protocol is identified implicitly b y t h e VPI/VCI. Otherwise, t h e c o n v e r g e n c e s u b l a y e r PDU starts w i t h a h e a d e r t h a t specifies t h e protocol (e.g., r o u t e d IP or b r i d g e d 802.3-6). If t h e protocol is bridged, t h e n t h e MAC a d d r e s s of t h e d e s t i n a t i o n m u s t b e specified in t h e c o n v e r g e n c e s u b l a y e r PDU. Note t h a t t h e ATM interface m u s t p e r f o r m t h e u s u a l functions of a b r i d g e w i t h d y n a m i c l e a r n i n g b y looking into t h e MAC a d d r e s s e s of t h e e n c a p s u l a t e d PDUs. We explain h o w s u c h a bridging function is i m p l e m e n t e d in LAN i n t e r c o n ­ n e c t i o n s in t h e next section.

67.2

LAN Emulation over ATM LAN e m u l a t i o n (LANE) is a glue t h a t e n a b l e s o r d i n a r y LAN software to o p e r a t e over ATM a n d also to i n t e r c o n n e c t LANs a n d ATM. (See [A95b].) T h i s e m u l a t i o n e n a b l e s t h e c o n n e c t i o n of E t h e r n e t s t h r o u g h a n ATM n e t w o r k or of t o k e n rings t h r o u g h a n ATM n e t w o r k (not a m i x t u r e ) . Protocols, s u c h as IP's Address Resolution Protocol (ARP), t h a t are d e p e n ­ d e n t o n t h e availability of a b r o a d c a s t function, are s u p p o r t e d b y LANE over ATM, which, d u e to its c o n n e c t i o n - o r i e n t e d n a t u r e , is n o n b r o a d c a s t . ARP u s e s b r o a d c a s t to resolve i n t e r n e t w o r k l a y e r a d d r e s s e s to MAC addresses. O n a n Em­ u l a t e d LAN (ELAN), LANE s u p p o r t s this b r o a d c a s t w i t h a B r o a d c a s t / U n k n o w n

295

Asynchronous Transfer Mode

296 Node Β netBIOS, IP, etc. LLC MAC Physical

netBIOS, IP, etc.

1

—I

LAN emulation AAL ATM Physical Node A

6.20 FIGURE

A LAN emulation layer is inserted between the network layer and the AAL layer in ATM nodes.

Server (BUS). However, to effect t h e actual data transfer over a n ATM n e t w o r k , a further m a p p i n g from MAC a d d r e s s to ATM address is n e c e s s a r y . T h e LANE server provides this m a p p i n g . T h e host t h e n transfers t h e data b y setting u p a n ATM VCC to t h e target ATM address. Figure 6.20 illustrates this a p p r o a c h . A p a c k e t d e s t i n e d for Β invokes t h e LAN e m u l a t i o n l a y e r t h a t contacts t h e server to find t h e ATM a d d r e s s of t h e bridge. T h e packet is s e g m e n t e d a n d s e n t to t h e bridge, w h o s e LAN e m u l a t i o n software r e c o n s t r u c t s t h e p a c k e t before s e n d i n g it o n t h e LAN. T h e r e v e r s e transfer, from Β to A, is similar. Broadcast p a c k e t s a n d p a c k e t s w i t h u n k n o w n d e s t i n a t i o n a d d r e s s e s are flooded b y t h e server. T h i s solution e n a b l e s ATM n e t w o r k s to i n t e r o p e r a t e w i t h existing "legacy" LANs.

6.7

Internetworking with ATM

6.7.3

IP over ATM IP is a powerful a n d widely u s e d i n t e r n e t w o r k i n g protocol. With IP w e c a n i n t e r c o n n e c t IEEE 802 n e t w o r k s easily. IP is a d a t a g r a m n e t w o r k layer. In this c h a p t e r w e discussed t h e ATM t e c h n o l o g y a n d h o w it c a n b e u s e d to b u i l d local a r e a n e t w o r k s a n d w i d e area n e t w o r k s w i t h a good control o n t h e quality of service it provides to applications. For t h e ATM t e c h n o l o g y to b e w i d e l y i m p l e m e n t e d , it m u s t i n t e r o p e r a t e w i t h t h e IP protocol suite. In this section w e explain h o w ATM n e t w o r k s c a n t r a n s p o r t IP packets. T h i s possibility e n a b l e s a progressive u p g r a d e of t h e I n t e r n e t to t h e ATM technology. T h e a d v a n t a g e s u c h a n u p g r a d e w o u l d provide is t h a t applications r e q u i r i n g t h e tight control of QpS c a n b e s u p p o r t e d b y ATM a n d n o t easily b y t h e T C P / I P protocols. T h u s , progressively, t h e I n t e r n e t w o u l d evolve into a BISDN n e t w o r k while r e m a i n i n g c o m p a t i b l e w i t h t h e installed b a s e of best-effort services. We explain t h r e e strategies: t h e classical IP m o d e l , t h e s h o r t c u t models, a n d t h e integrated m o d e l s . We t h e n explain m u l t i c a s t IP over ATM. T h e s e strategies are b e i n g evolved b y t h e IETF w o r k i n g g r o u p IP over ATM. (See RFC 1754.)

Classical IP T h e u s e of LANE for IP transfer r e q u i r e s two levels of a d d r e s s resolution: t h e IP address m u s t b e resolved to a MAC address, a n d t h e MAC a d d r e s s m u s t b e resolved to a n ATM a d d r e s s u s i n g LANE. Classical IP directly provides IPto-ATM a d d r e s s resolution u s i n g a n ARP server, t h e r e b y r e d u c i n g b r o a d c a s t traffic. Consider t h e situation s h o w n in Figure 6.21. In t h e classical IP m o d e l , t h e n o d e s a t t a c h e d to a n ATM n e t w o r k are g r o u p e d into logical IP s u b d o m a i n s (LIS). Routing b e t w e e n logical IP s u b d o m a i n s is via routers, as s h o w n in t h e figure. Note t h a t AAL5 is u s e d so t h a t t h e r o u t e r r e a s s e m b l e s t h e p a c k e t before forwarding. Within o n e given logical IP s u b d o m a i n , a n o d e u s e s a n address-resolution protocol (ARP) server. T h e stations all k n o w t h e ATM a d d r e s s of t h e i r ARP server. T h u s , to find a particular destination, i n s t e a d of b r o a d c a s t i n g a r e q u e s t are you node IP.address? to find t h e physical address, h e r e a n o d e s e n d s a r e q u e s t to t h e ARP server of t h e s u b d o m a i n asking, what is the VCI of a particular IP.address? In t h e case of SVCs (switched virtual circuits), t h e n o d e s n e e d to register w i t h t h e ARP server. T h e y do so b y calling t h e ARP s e r v e r (using t h e ATM

297

Asynchronous Transfer Mode

298

Logical IP subnet 1

Router Logical IP subnet 2

6.21 GURE

\ Router

Nodes attached to an ATM network are grouped into logical IP subdomains (LIS). The routing is via routers between subdomains. Within one domain, the routing uses ATM ARP with an ARP server.

addresses). T h e server t h e n asks, what is your IP.address? and e n t e r s t h a t information in its table [IP -> ATM]. T h e IP a n d ARP p a c k e t s are e n c a p s u l a t e d over AAL5. Two alternatives exist: e i t h e r o n e VC is allocated p e r protocol (one for IP, o n e for ARP), or m u l t i p l e protocols are multiplexed over a single VC p e r s u b n e t w o r k a t t a c h m e n t point (SNAP). T h e m a x i m u m t r a n s m i s s i o n u n i t in IP over ATM is fixed to 9,180 b y t e s . O t h e r sizes (up to 64 KB) c a n b e agreed o n b y configuration (in t h e case of PVC or p e r m a n e n t virtual circuit) or b y signaling (for SVC). T h e t y p e of ATM c o n n e c t i o n is e i t h e r CBR or VBR w i t h specified p e a k r a t e s forward a n d b a c k w a r d . O t h e r e n c a p s u l a t i o n s are b e i n g p r o p o s e d to e l i m i n a t e or r e d u c e t h e r e d u n d a n t IP header. (See RFC 1483, 1755, 2225.)

Shortcut Models I n s t e a d of r e t r a n s m i t t i n g via r o u t e r s as in t h e classical IP m o d e l , t h e idea of t h e shortcut m o d e l s is to go directly from source to d e s t i n a t i o n ATM n o d e s . In t h e a c c e p t e d terminology, t h e ATM n e t w o r k is called a nonbroadcast multiaccess (NBMA) link layer. In Figure 6.22 w e indicate h o w a n o d e S finds t h e NBMA address. T h e NBMA Next H o p Resolution Protocol (NHRP) e x t e n d s t h e idea above. It allows a source wishing to c o m m u n i c a t e over a NBMA s u b n e t w o r k to de-

6.7

Internetworking with ATM

299

IP address "A" I

IP address "S'

NBMAARP server

6.22 ββ™αα

I NBMAARP server

Tb transmit to node A, node S finds out the nonbroadcast multiaccess (NBMA) address of the router R according to the following steps: (1) What is the NBMA address of A? (2) forwards request; (3) knows router of network of A; (4) ARP to locate A; then, in reverse, (3), (2), (1) gives NBMA address R; a connection is t h e n made as shown in (5).

t e r m i n e t h e i n t e r n e t w o r k i n g l a y e r a d d r e s s e s a n d NBMA a d d r e s s e s of suitable "NBMA next h o p s " toward a d e s t i n a t i o n station. Routers are r e q u i r e d to inter­ c o n n e c t t h e s e subnets, b u t N H R P allows i n t e r m e d i a t e r o u t e r s to b e b y p a s s e d o n t h e data p a t h . N H R P provides a n e x t e n d e d a d d r e s s r e s o l u t i o n protocol, w h i c h p e r m i t s Next H o p Clients to s e n d q u e r i e s b e t w e e n different s u b n e t s . Q u e r i e s are p r o p a g a t e d b y Next H o p Servers along t h e r o u t e d p a t h d e t e r m i n e d b y a s t a n d a r d r o u t i n g protocol. T h i s e n a b l e s t h e e s t a b l i s h m e n t of ATM VCCs across s u b n e t b o u n d a r i e s w i t h o u t r o u t e r s in t h e data p a t h . Note t h a t this m e t h o d c a n also b e u s e d for F r a m e Relay, ISDN, a n d X.25 n e t w o r k s t h a t do n o t s u p p o r t b r o a d c a s t i n g . (See RFC 1735, 2332.)

Integrated Model T h e i n t e g r a t e d m o d e l proposal a i m s to simplify t h e r o u t i n g b y integrating addressing a n d r o u t i n g of IP a n d ATM. In this m o d e l , t h e ATM a d d r e s s could b e a s u p e r s e t of t h e IP address. (This a p p r o a c h d o e s n o t w o r k w i t h n e t w o r k s t h a t u s e t h e E.164 addresses.) T h e IP r o u t e r t h e n m a p s t h e d e s t i n a t i o n IP a d d r e s s into t h e ATM a d d r e s s of t h e d e s t i n a t i o n if it is directly r e a c h a b l e or of t h e b e s t r o u t e r o t h e r w i s e . T h e selection of t h e b e s t r o u t e r m a y b e load d e p e n d e n t .

Asynchronous Transfer Mode

300 Multicast IP over ATM

T h e m a i n difference b e t w e e n ATM m u l t i c a s t VCs a n d m u l t i c a s t IP is t h a t t h e source m u s t add n e w d e s t i n a t i o n s in t h e case of ATM. T h e m e c h a n i s m u s e s a Multicast Address Resolution Server (MARS) t h a t m a p s a n IP multicast address to e i t h e r t h e list of all t h e individual ATM addresses or to t h e ATM address of a Multicast Server (MCS). A cluster is a set of h o s t s t h a t u s e t h e s a m e MARS. T h e c o m m u n i c a t i o n b e t w e e n clusters w o r k s as t h e regular IP multicast routing. We describe t h e c o m m u n i c a t i o n w i t h i n o n e cluster. T h e r e are two a p p r o a c h e s to m u l t i c a s t c o n n e c t i o n s w i t h i n o n e cluster. In t h e direct approach, e a c h s e n d e r sets u p o n e VC p e r m e m b e r of t h e m u l t i c a s t group. In t h e indirect approach, t h e s e n d e r s e n d s to a n MCS, w h i c h t h e n sets u p VCs to t h e g r o u p m e m b e r s . In t h e direct approach, t h e hosts s e n d j o i n a n d leave r e q u e s t s to t h e MARS. T h e MARS m a i n t a i n s a point-to-multipoint VC to all t h e s o u r c e s (called t h e ClusterControlVC) to inform t h e m of g r o u p c h a n g e s . T h e j o i n a n d leave m e s s a g e s are r e t r a n s m i t t e d over t h e ClusterControlVC. Before sending, t h e source asks MARS for a list of ATM addresses. T h e n t h e s e n d e r sets u p t h e VCs. In t h e indirect approach, t h e s e n d e r k n o w s o n l y t h e MCS as a m e m b e r of a group. T h e h o s t s register w i t h MCS, a n d e v e r y MCS registers w i t h MARS. (See RFC 2022.)

67.4

Multiprotocol over ATM (MPOA) Tb i n c r e a s e t h e p e n e t r a t i o n of ATM technology, t h e ATM F o r u m issued MPOA version 1.0. T h e objective of MPOA is to s u p p o r t t h e efficient transfer of inters u b n e t u n i c a s t traffic in a LANE e n v i r o n m e n t . MPOA integrates LANE a n d NHRP, allowing t h e transfer of layer 3 p a c k e t s (typically IP) over fast ATM VCCs w i t h o u t r e q u i r i n g r o u t e r s in t h e data p a t h . MPOA allows t h e physical s e p a r a t i o n of t h e i n t e r n e t w o r k (IP) l a y e r r o u t e calculation a n d t h e forwarding of t h e data. T h e t e c h n i q u e is called virtual routing. Virtual r o u t i n g c a n d e c r e a s e t h e n u m b e r of devices participating in i n t e r n e t w o r k l a y e r r o u t e calculations, increasing scalability. MPOA provides MPOA clients (MPC) a n d MPOA servers (MPS) a n d defines t h e protocols u s e d b y MPCs to issue q u e r i e s for s h o r t c u t ATM a d d r e s s e s a n d receive replies from MPSs. T h e s h o r t c u t is a n ATM VCC from a n ingress MPC source to a n egress MPC sink t h r o u g h a n ATM cloud (MPOA s y s t e m ) . Until a

6.8

Summary

301

shortcut is obtained, a default r o u t e d p a t h is used. T h e t e c h n i q u e is similar to tag switching in w h i c h l a y e r 3 r o u t i n g table l o o k u p s are b y p a s s e d . T h e m a i n function of t h e MPC is to serve as a s h o r t c u t source a n d sink. T h e MPC p e r f o r m s forwarding b u t does n o t r u n r o u t i n g protocols. In its ingress role, a n MPC d e t e c t s p a c k e t flows t h a t are b e i n g forwarded over a n e m u l a t e d LAN to a r o u t e r t h a t c o n t a i n s a n MPS. W h e n t h e MPC recognizes a flow (for example, b y c o u n t i n g t h e n u m b e r of p a c k e t s in t h e flow over a c e r t a i n d u r a t i o n ) t h a t could benefit from a s h o r t c u t t h a t b y p a s s e s t h e default route, it u s e s a n NHRPb a s e d q u e r y to r e q u e s t a s h o r t c u t to t h e destination. If t h e s h o r t c u t is available, t h e MPC c a c h e s this information, sets u p a s h o r t c u t VCC, a n d forwards t h e p a c k e t s over t h e shortcut. T h e s h o r t c u t i n f o r m a t i o n m a y b e in t h e form of a 32-bit tag t h a t t h e egress MPC provides to t h e ingress MPC. T h e tag is i n c l u d e d b y t h e ingress MPC in t h e MPOA p a c k e t header. In its egress role t h e MPC receives t h e p a c k e t s from o t h e r MPCs to b e forwarded to its local users. For frames r e c e i v e d over a shortcut, t h e egress MPC adds t h e a p p r o p r i a t e data link l a y e r (MAC) e n c a p s u l a t i o n . T h e MPOA server is t h e logical p a r t of a r o u t e r t h a t p r o v i d e s forwarding information to MPCs u s i n g NHRP.

6.7.5

FR and SMDS over ATM T h e t r a n s p o r t of F r a m e Relay p a c k e t s over ATM r e q u i r e s t h e c o n v e r s i o n of t h e FR data link c o n n e c t i o n identifier into a VPI/VCI for t h e ATM c o n n e c t i o n a n d t h e s e g m e n t a t i o n of t h e FR p a c k e t p a y l o a d into AAL5 packets. T h e FR congestion a n d discard eligibility fields are m a p p e d into t h e ATM EFCI bit a n d t h e ATM CLP bit, respectively. Moreover, t h e FR c o m m i t t e d information rate is m a p p e d into VRB traffic p a r a m e t e r s . SMDS p a c k e t s are m a p p e d into A A L 3 / 4 cells a n d t r a n s p o r t e d over a wellk n o w n VPI/VCI. A c o n n e c t i o n l e s s s e r v e r w i t h i n t h e ATM n e t w o r k receives t h e cells a n d forwards t h e packet.

6.8

SUMMARY I n t h e two p r e v i o u s c h a p t e r s w e studied packet- a n d circuit-switched n e t w o r k s . Packet-switched n e t w o r k s h a n d l e well m e s s a g e traffic t h a t i m p o s e s v e r y loose delay constraints. Circuit-switched n e t w o r k s are well suited for c o n s t a n t bit rate traffic w i t h h a r d delay constraints. Of course, b o t h n e t w o r k s c a n a c c o m m o d a t e all t y p e s of traffic, b u t t h e y m a y do this inefficiently.

Asynchronous Transfer Mode

302

E n g i n e e r s i n v e n t e d ATM n e t w o r k s w i t h t h e objective t h a t t h e y w o u l d provide a highly flexible b e a r e r service capable of s u p p o r t i n g in a n efficient m a n n e r applications t h a t r a n g e from those t h a t n e e d g u a r a n t e e d d e l a y a n d loss b o u n d s to t h o s e t h a t n e e d o n l y best-effort service p r o v i d e d b y d a t a g r a m n e t w o r k s s u c h as I n t e r n e t . At t h e e n d of t h e 1980s, ATM n e t w o r k s w e r e exotic topics of discussion at r e s e a r c h conferences. Tbday, m a n y v e n d o r s provide ATM e q u i p m e n t , a l t h o u g h w i t h l i m i t e d functionality. Some ATM switches could already b e classified as carrier grade a n d satisfy r e q u i r e m e n t s in t e r m s of s p e e d (2.5 Gbps ports, 50 Gbps b a c k p l a n e ) , reliability (PNNI r e r o u t i n g ) a n d overall functionality. T h i s c h a p t e r w a s d e v o t e d to a discussion of t h e functionality of ATM. In principle ATM c a n offer t h e full r a n g e of services n e e d e d to m a k e t h e claim t h a t ATM n e t w o r k s c a n function as t h e "universal network." T h e great interest a m o n g t e l e p h o n e c o m p a n i e s in providing ATM over SONET suggests t h a t ATM will b e a serious c o n t e n d e r for this title, in t h e n a m e of BISDN. However, BISDN is still a considerable distance in t h e future. T h e m o r e i m m e d i a t e goal is t h e migration p a t h for IP over ATM. T h e migration c a n p e r m i t a n "upgrade" of IP in t h e s e n s e t h a t IP/ATM m a y b e able to provide service g u a r a n t e e s t h a t IP c a n n o t . If this t u r n s out to b e t h e case, IP a n d ATM m a y gain t h r o u g h e c o n o m i e s of scope a n d service integration. T r e m e n d o u s progress in ATM h a s b e e n m a d e over t h e 1990s. But t h e potential of ATM will r e m a i n unfulfilled until t h e difficult issues of r e s o u r c e allocation a n d control are satisfactorily resolved s u c h t h a t t h e s a m e ATM n e t w o r k c a n efficiently provide a variety of service qualities. In C h a p t e r s 8 a n d 9 w e discuss t h e s e issues. At t h e p r e s e n t time, however, t h e challenge facing ATM is attracting IP.

6.9

NOTES Specifications of t h e ATM F o r u m are n o w available at www.atmforum.com. T h e p a r a m e t e r s a n d p r o c e d u r e s related to traffic m a n a g e m e n t a n d QpS are specified in [A96d]. GFR is likely to a p p e a r in ATM F o r u m ' s Traffic M a n a g e m e n t Specification version 4.1. GFR is discussed in [GJF98]. ATM a d d r e s s i n g is specified in [A99]. T h e basic ATM formats a n d protocols are d i s c u s s e d in [B99, P94]. FUNI is specified in [A97a]. T h e BISDN m o d e l is d e s c r i b e d in [A95a], w h i c h also specifies i n t e r w o r k i n g w i t h FR a n d SMDS. T h e all-important m a n a g e m e n t functions t h a t g u a r a n t e e t h e service qual­ ity level, however, are n o t y e t sufficiently defined to c o n s t i t u t e a proposal. Nevertheless, t h e r e is i m p o r t a n t a n d c o n t i n u i n g discussion w i t h i n t h e ATM

6.10

Problems

303

F o r u m t h a t s e e k s to d e v e l o p t h o s e proposals a n d standards. PNNI 1.0, UNI 4.0, ILMI 4.0 are specified in [A96a, A96b, A96c]. LAN e m u l a t i o n is specified in [A95b], a n d MPOA in [A97b]. Proposals for IP over ATM are discussed in various IETF RFC, n o t a b l y RFC 1483, 1755, 2225, 2332. MPOA is c o m p a r e d w i t h Multiprotocol Label Switching (MPLS) in [DDR98].

6.10

PROBLEMS 1. Suppose t h e t r a n s p o r t l a y e r transfers fixed-size p a c k e t s of Ν bits e i t h e r as d a t a g r a m s or over a virtual circuit. In t h e former case bits are u s e d to e n c o d e t h e d e s t i n a t i o n address. In t h e latter case n < rid bits are u s e d to e n c o d e t h e VCI. In addition, in t h e latter case, t h e r e is a fixed d e l a y i n c u r r e d to set u p t h e c o n n e c t i o n . We m e a s u r e this d e l a y in t e r m s of t h e t i m e n e e d e d to t r a n s m i t D bits. Suppose t h e s o u r c e w a n t s to t r a n s m i t a m e s s a g e of Μ bits. Will t h e d a t a g r a m or t h e c o n n e c t i o n - o r i e n t e d service i n c u r greater delay? c

2. H o w m a n y s e q u e n c e n u m b e r s are n e e d e d to d e t e c t cell loss? H o w m a n y are u s e d in SMDS? 3. T h e packetization delay d e p e n d s o n t h e s p e e d of t h e i n f o r m a t i o n transfer. Calculate t h e packetization d e l a y for (a) 53-byte ATM cells a n d (b) a 1,000b y t e p a c k e t transfer service for (1) voice s a m p l e s t h a t are s a m p l e d 8,000 t i m e s p e r s e c o n d a n d e n c o d e d into a 64-Kbps s t r e a m a n d (2) MPEG1, w h i c h takes 30 video frames p e r s e c o n d a n d e n c o d e s t h e m into a 1-Mbps s t r e a m . 4. T h e size of t h e ATM cell affects cell error rate. If t h e t r a n s m i s s i o n s y s t e m h a s a bit e r r o r rate (BER) of p, a n d t h e r e are Ν bits p e r cell, s h o w t h a t t h e cell error rate (CER) is CER = 1 - ( 1 -p)

N

^Np.

Suppose t h a t a cell is r e t r a n s m i t t e d w h e n e v e r it c o n t a i n s a n error. T h e n t h e average n u m b e r of cells t r a n s m i t t e d p e r error-free cell r e c e p t i o n is (1 — CER)~ . Suppose t h e o v e r h e a d p e r cell is fixed at η bits, i n d e p e n d e n t of cell size. Show t h a t t h e average n u m b e r of bits t r a n s m i t t e d p e r error-free r e c e p t i o n of o n e bit of i n f o r m a t i o n is l

N + n N(l

-p) + ' N

n

Asynchronous Transfer Mode

304

For a given BER, p, a n d o v e r h e a d n, t h e o p t i m a l cell size Ν m i n i m i z e s this n u m b e r . Taking η = 40 (5-byte o v e r h e a d ) , find t h e b e s t Ν for two e x t r e m e cases: ρ — 10~~ a n d ρ = 10~ . 9

3

5. H o w m a n y s i m u l t a n e o u s c o n n e c t i o n s are n e e d e d to s u p p o r t all foreseeable n e e d s , including t e l e p h o n e , video, a n d text? Will t h e 16-bit VCI b e e n o u g h ? 6. C o m p a r e t h e n u m b e r of bits d e v o t e d for IP addressing w i t h t h e r e q u i r e ­ m e n t s for ATM. Suppose y o u w a n t to t r a n s m i t voice in small IP p a c k e t s so t h a t t h e packetization delay is n o t m o r e t h a n X m s . H o w large a p a c k e t c a n y o u tolerate, a n d w h a t is t h e o v e r h e a d rate (ratio of n u m b e r of o v e r h e a d bits to p a c k e t size)? 7. Suppose voice is s a m p l e d a n d t r a n s p o r t e d over ATM. Suppose t h e cell s t r e a m is subject to r a n d o m delay. H o w w o u l d y o u characterize t h e re­ sulting distortion? 8. Using LAN e m u l a t i o n , h o s t s a t t a c h e d to E t h e r n e t switches w i t h a n ATM backbone (a) (b) (c) (d)

get a g u a r a n t e e d quality of service. c a n n o t select t h e p a r a m e t e r s of t h e virtual circuits. m a y benefit from a well-provisioned b a c k b o n e . m u s t i m p l e m e n t AAL-5.

9. Please a n s w e r t h e following q u e s t i o n s in o n e or two s e n t e n c e s . (a) W h y is "out-of-band signaling" preferable to "in-band signaling"? (b) W h y do ATM n e t w o r k s u s e cell switching? (c) W h y is virtual circuit switching critical for ATM n e t w o r k s ? (Please list t h r e e reasons.) (d) W h a t are t h e "adaptation layers"? Are t h e y in t h e switches? Are t h e y in t h e end-hosts? (e) W h a t are t h e relative advantages a n d disadvantages of u s i n g ATM in t h e "native m o d e " vs. "LAN Emulation"? (f) Is it practical to b u i l d a large (i.e., w i t h lots of ports) input-buffered ATM switch w i t h fast links? W h y or w h y not? (g) Is it practical to b u i l d a large (i.e., w i t h lots of ports) output-buffered ATM switch w i t h fast links? W h y or w h y not? (h) A s s u m e y o u h a v e a choice b e t w e e n a n input-buffered a n d a n outputbuffered ATM switch, e a c h w i t h t h e s a m e n u m b e r of p o r t s a n d s a m e s p e e d links, w h i c h is b e t t e r ? Explain w h y .

7

Wireless

CHAPTER

Networks

J7 u t u r e wireless n e t w o r k s will e n a b l e p e o p l e o n t h e m o v e to c o m m u n i c a t e w i t h a n y o n e , a n y w h e r e , at a n y time, u s i n g a r a n g e of m u l t i m e d i a services. T h e e x p o n e n t i a l g r o w t h of cellular t e l e p h o n e a n d paging s y s t e m s c o u p l e d w i t h t h e proliferation of laptop a n d p a l m t o p c o m p u t e r s indicate a b r i g h t future for s u c h n e t w o r k s , b o t h as s t a n d a l o n e s y s t e m s a n d as p a r t of t h e larger n e t w o r k i n g infrastructure. T h i s c h a p t e r describes wireless n e t w o r k s . Section 7.1 p r o v i d e s a n introduc­ tion to t h e s e n e t w o r k s , i n c l u d i n g t h e i r history a n d prospects, a n d t h e t e c h n i c a l challenges of design a n d o p e r a t i o n p o s e d b y t h e u n d e r l y i n g wireless c h a n n e l . Section 7.2 p r e s e n t s t h e m a i n characteristics of t h e wireless c h a n n e l a n d t h e i r i m p a c t o n t h e link a n d n e t w o r k l a y e r design. Link l a y e r design t e c h n i q u e s d e v e l o p e d to o v e r c o m e wireless c h a n n e l i m p a i r m e n t s to delivering h i g h data rates w i t h low distortion are d e s c r i b e d in Section 7.3. T h e wireless c h a n n e l is a l i m i t e d r e s o u r c e t h a t is s h a r e d a m o n g m a n y users. Section 7.4 is d e v o t e d to c h a n n e l access protocols. Section 7.5 o u t l i n e s de­ sign issues for wireless n e t w o r k s , i n c l u d i n g n e t w o r k architecture, u s e r location a n d r o u t i n g protocols, n e t w o r k reliability a n d QpS, i n t e r n e t w o r k i n g b e t w e e n wireless a n d w i r e d n e t w o r k s , a n d security. C u r r e n t wireless n e t w o r k t e c h n o l o g y is d e s c r i b e d in Section 7.6 i n c l u d i n g cellular a n d cordless t e l e p h o n e s , w i r e l e s s LANs a n d w i d e a r e a data services, paging s y s t e m s , a n d global satellite s y s t e m s . Section 7.7 gives a n o v e r v i e w of e m e r g i n g s y s t e m s a n d s t a n d a r d s for future wireless n e t w o r k s . Section 7.8 c o n t a i n s a s u m m a r y a n d a discussion of future t r e n d s .

Wireless Networks

306

7.1

INTRODUCTION Wireless c o m m u n i c a t i o n s is t h e fastest growing s e g m e n t of t h e c o m m u n i c a ­ tions industry. Cellular p h o n e s , cordless p h o n e s , a n d paging services h a v e e x p e r i e n c e d e x p o n e n t i a l growth over t h e last decade, a n d this g r o w t h con­ t i n u e s u n a b a t e d worldwide. Wireless c o m m u n i c a t i o n s h a s b e c o m e a critical b u s i n e s s tool a n d p a r t of e v e r y d a y life in m o s t d e v e l o p e d c o u n t r i e s . Wireless c o m m u n i c a t i o n s y s t e m s are replacing a n t i q u a t e d wireline s y s t e m s in m a n y developing countries. Will future wireless n e t w o r k s live u p to t h e i r p r o m i s e of m u l t i m e d i a c o m m u n i c a t i o n s a n y w h e r e a n d a n y t i m e ? In this i n t r o d u c t i o n w e r e v i e w t h e history of wireless n e t w o r k s . We t h e n discuss t h e wireless vision in m o r e detail, including t h e t e c h n i c a l c h a l l e n g e s t h a t m u s t b e o v e r c o m e to m a k e this vision a reality.

7.1.1

History of Wireless Networks T h e first wireless n e t w o r k s w e r e d e v e l o p e d in t h e p r e i n d u s t r i a l age. T h e s e s y s t e m s t r a n s m i t t e d information over line-of-sight distances (later e x t e n d e d b y telescopes) u s i n g s m o k e signals, torch signaling, flashing mirrors, signal flares, a n d s e m a p h o r e flags. A n elaborate set of signal c o m b i n a t i o n s w a s d e v e l o p e d to c o n v e y c o m p l e x m e s s a g e s w i t h t h e s e r u d i m e n t a r y signals. Observation stations w e r e built o n hilltops a n d along roads to relay t h e s e m e s s a g e s over large distances. T h e s e early c o m m u n i c a t i o n n e t w o r k s w e r e r e p l a c e d first b y t h e t e l e g r a p h n e t w o r k ( i n v e n t e d b y S a m u e l Morse in 1838) a n d l a t e r b y t h e t e l e p h o n e . In 1895, t w e n t y y e a r s after t h e t e l e p h o n e w a s i n v e n t e d , M a r c o n i d e m o n s t r a t e d t h e first radio t r a n s m i s s i o n from t h e Isle of Wight to a tugboat 18 m i l e s away, a n d radio c o m m u n i c a t i o n s w a s b o r n . Radio t e c h n o l o g y a d v a n c e d rapidly to e n a b l e t r a n s m i s s i o n s over larger distances w i t h b e t t e r quality, less power, a n d smaller, c h e a p e r devices, t h e r e b y e n a b l i n g public a n d private radio c o m m u n i c a t i o n s , television, a n d wireless n e t w o r k i n g . Early radio s y s t e m s t r a n s m i t t e d analog signals. Tbday m o s t radio s y s t e m s t r a n s m i t digital signals c o m p o s e d of b i n a r y bits, w h e r e t h e bits are obtained directly from a data signal or b y digitizing a n analog voice or m u s i c signal. A digital radio c a n t r a n s m i t a c o n t i n u o u s bit s t r e a m or it c a n g r o u p t h e bits into packets. T h e latter t y p e of radio is called a packet radio a n d is c h a r a c t e r i z e d b y b u r s t y t r a n s m i s s i o n s : t h e radio is idle except w h e n it t r a n s m i t s a packet. T h e first packet radio n e t w o r k , Alohanet, was d e v e l o p e d at t h e University of Hawaii in 1971. This n e t w o r k e n a b l e d c o m p u t e r sites at s e v e n c a m p u s e s s p r e a d

7.1

Introduction

o u t over four islands to c o m m u n i c a t e w i t h a central c o m p u t e r o n O a h u via radio t r a n s m i s s i o n . T h e n e t w o r k a r c h i t e c t u r e u s e d a star topology w i t h t h e central c o m p u t e r at its h u b . A n y two c o m p u t e r s could establish a bidirectional c o m m u n i c a t i o n s link b e t w e e n t h e m b y going t h r o u g h t h e c e n t r a l h u b . A l o h a n e t i n c o r p o r a t e d t h e first set of protocols for c h a n n e l access a n d r o u t i n g in p a c k e t radio s y s t e m s , a n d p r i n c i p l e s u n d e r l y i n g t h e s e protocols are still in u s e today. Activity in p a c k e t radio, p r o m o t e d b y DARPA, p e a k e d in t h e m i d 1980s, b u t t h e resulting n e t w o r k s fell far short of expectations in t e r m s of s p e e d a n d p e r f o r m a n c e . Packet radio n e t w o r k s t o d a y are m o s t l y u s e d b y c o m m e r c i a l providers of w i d e a r e a wireless data services. T h e s e services, first i n t r o d u c e d in t h e early 1990s, e n a b l e wireless data access (including e-mail, file transfer, a n d Web b r o w s i n g ) at fairly low speeds, o n t h e o r d e r of 20 Kbps. T h e m a r k e t for t h e s e data services h a s n o t g r o w n significantly. In t h e 1970s E t h e r n e t t e c h n o l o g y s t e e r e d c o m p a n i e s a w a y from radio-based n e t w o r k i n g . E t h e r n e t ' s 10 Mbps data rate far e x c e e d e d a n y t h i n g available u s i n g radio. In 1985 t h e Federal C o m m u n i c a t i o n s C o m m i s s i o n (FCC) e n a b l e d t h e c o m m e r c i a l d e v e l o p m e n t of wireless LANs b y a u t h o r i z i n g t h e public u s e of t h e Industrial, Scientific, a n d Medical (ISM) f r e q u e n c y b a n d s for wireless LAN products. T h e ISM b a n d was v e r y attractive to wireless LAN v e n d o r s since t h e y did n o t n e e d to obtain a n FCC l i c e n s e to o p e r a t e in this b a n d . However, t h e wireless LAN s y s t e m s could n o t interfere w i t h t h e p r i m a r y ISM b a n d users, w h i c h forced t h e m to u s e a low p o w e r profile a n d a n inefficient signaling s c h e m e . Moreover, t h e i n t e r f e r e n c e from p r i m a r y u s e r s w i t h i n this f r e q u e n c y b a n d w a s quite high. As a result t h e s e initial LAN s y s t e m s h a d v e r y poor p e r f o r m a n c e in t e r m s of data r a t e s a n d coverage. T h e p o o r p e r f o r m a n c e , c o u p l e d w i t h c o n c e r n s a b o u t security, lack of standardization, a n d high cost (the first n e t w o r k a d a p t o r s listed for $1,400 as c o m p a r e d w i t h a few h u n d r e d dollars for a w i r e d E t h e r n e t card) r e s u l t e d in w e a k sales for t h e initial wireless LAN s y s t e m s . T h e c u r r e n t g e n e r a t i o n of wireless LANs, b a s e d o n t h e IEEE 802.11 standard, h a s b e t t e r p e r f o r m a n c e , a l t h o u g h t h e data rates are still l o w (on t h e o r d e r of 2 Mbps) a n d t h e coverage area is still small ( a r o u n d 500 feet). E t h e r n e t s today offer data r a t e s of 100 Mbps, a n d t h e p e r f o r m a n c e gap b e t w e e n w i r e d a n d wireless LANs is likely to i n c r e a s e over t i m e w i t h o u t additional s p e c t r u m allocation. T h u s , it is n o t clear if wireless LANs will b e c o m p e t i t i v e except w h e r e u s e r s sacrifice p e r f o r m a n c e for mobility or w h e n a w i r e d infrastructure is n o t available. By far t h e m o s t successful application of wireless n e t w o r k i n g h a s b e e n t h e cellular t e l e p h o n e s y s t e m . Cellular t e l e p h o n e s h a v e a p p r o x i m a t e l y 200 million subscribers worldwide, a n d t h e i r g r o w t h c o n t i n u e s at a n e x p o n e n t i a l pace. T h e c o n v e r g e n c e of radio a n d t e l e p h o n y b e g a n in 1915, w h e n wireless

307

308

Wireless Networks

voice t r a n s m i s s i o n b e t w e e n N e w York a n d San Francisco w a s first established. In 1946 public mobile t e l e p h o n e service w a s i n t r o d u c e d in 25 cities across t h e U n i t e d States. T h e s e initial s y s t e m s u s e d a c e n t r a l t r a n s m i t t e r to cover a n entire m e t r o p o l i t a n area. This inefficient u s e of t h e radio s p e c t r u m c o u p l e d w i t h t h e state of radio t e c h n o l o g y at t h a t t i m e severely l i m i t e d t h e s y s t e m capacity: thirty y e a r s after t h e i n t r o d u c t i o n of mobile t e l e p h o n e service t h e N e w York s y s t e m could only s u p p o r t 543 users. A solution to this capacity p r o b l e m e m e r g e d d u r i n g t h e fifties a n d sixties w h e n r e s e a r c h e r s at AT&T Bell Laboratories d e v e l o p e d t h e cellular concept. Cellular s y s t e m s exploit t h e fact t h a t t h e p o w e r of a t r a n s m i t t e d signal falls off w i t h distance, so t h e s a m e f r e q u e n c y c h a n n e l c a n b e allocated to u s e r s at spatially s e p a r a t e locations w i t h m i n i m a l i n t e r f e r e n c e . A cellular s y s t e m divides a geographical area into adjacent, n o n o v e r l a p p i n g "cells." Cells assigned t h e s a m e c h a n n e l set are s p a c e d a p a r t so t h a t i n t e r f e r e n c e b e t w e e n t h e m is small. Each cell h a s a centralized t r a n s m i t t e r a n d receiver (called a base station) t h a t c o m m u n i c a t e s w i t h t h e mobile u n i t s in t h a t cell, b o t h for control p u r p o s e s a n d as a call relay. All b a s e stations h a v e h i g h - b a n d w i d t h c o n n e c t i o n s to a mobile t e l e p h o n e switching office (MTSO), w h i c h is itself c o n n e c t e d to t h e public-switched t e l e p h o n e n e t w o r k (PSTN). T h e handoff of mobile u n i t s crossing cell b o u n d a r i e s is typically h a n d l e d b y t h e MTSO, a l t h o u g h in c u r r e n t s y s t e m s s o m e of this functionality is h a n d l e d b y t h e b a s e stations a n d / o r mobile units. T h e original cellular s y s t e m design w a s finalized in t h e late 1960s a n d d e p l o y e d in t h e early 1980s. T h e large a n d u n e x p e c t e d g r o w t h led to t h e d e v e l o p m e n t of digital cellular t e c h n o l o g y to i n c r e a s e capacity a n d i m p r o v e performance. T h e c u r r e n t g e n e r a t i o n of cellular s y s t e m s are all digital. In addition to voice c o m m u n i c a t i o n , t h e s e s y s t e m s provide e-mail, voice mail, a n d paging services. Unfortunately, t h e great m a r k e t potential for cellular p h o n e s led to a proliferation of digital cellular standards. I b d a y t h e r e are t h r e e different digital cellular p h o n e s t a n d a r d s in t h e U.S. alone, a n d o t h e r s t a n d a r d s in E u r o p e a n d J a p a n , n o n e of w h i c h are compatible. T h e i n c o m p a t i b l e s t a n d a r d s m a k e r o a m i n g t h r o u g h o u t t h e U.S. u s i n g o n e digital cellular p h o n e impossible. Most cellular p h o n e s today are dual-mode: t h e y i n c o r p o r a t e o n e of t h e digital s t a n d a r d s along w i t h t h e old analog s t a n d a r d t h a t provides coverage t h r o u g h o u t t h e U.S. Radio paging s y s t e m s r e p r e s e n t a n o t h e r e x a m p l e of a successful wireless data network, w i t h 50 million subscribers in t h e U.S. T h e i r p o p u l a r i t y is start­ ing to w a n e w i t h t h e w i d e s p r e a d p e n e t r a t i o n a n d c o m p e t i t i v e cost of cellular t e l e p h o n e s y s t e m s . Paging s y s t e m s allow coverage over v e r y w i d e areas b y

7.1

Introduction

309 s i m u l t a n e o u s l y b r o a d c a s t i n g t h e p a g e r m e s s a g e at h i g h p o w e r from m u l t i p l e b a s e stations or satellites. Early radio paging s y s t e m s w e r e analog 1-bit m e s ­ sages signaling a u s e r t h a t s o m e o n e w a s t r y i n g to r e a c h h i m or her. T h e s e s y s t e m s r e q u i r e d callback over t h e r e g u l a r t e l e p h o n e s y s t e m to obtain t h e p h o n e n u m b e r of t h e paging party. Paging s y s t e m s n o w allow a short digital m e s s a g e , i n c l u d i n g a p h o n e n u m ­ b e r a n d brief text, to b e s e n t to t h e pagee. In paging s y s t e m s m o s t of t h e complexity is built into t h e t r a n s m i t t e r s , so t h a t p a g e r receivers are small, lightweight, a n d h a v e a long b a t t e r y life. T h e n e t w o r k protocols are also v e r y s i m p l e since b r o a d c a s t i n g a m e s s a g e over all b a s e stations r e q u i r e s n o r o u t i n g or handoff. T h e spectral inefficiency of t h e s e s i m u l t a n e o u s b r o a d c a s t s is com­ p e n s a t e d b y limiting e a c h m e s s a g e to b e v e r y short. Paging s y s t e m s c o n t i n u e to evolve to e x p a n d t h e i r capabilities b e y o n d v e r y low-rate o n e - w a y c o m m u n i ­ cation. C u r r e n t s y s t e m s are a t t e m p t i n g to i m p l e m e n t two-way, "answer-back" capability. T h i s r e q u i r e s a major c h a n g e in p a g e r design, since it m u s t n o w t r a n s m i t signals in addition to receiving t h e m , a n d t h e t r a n s m i s s i o n distances c a n b e quite large. C o m m e r c i a l satellite c o m m u n i c a t i o n s y s t e m s form a n o t h e r major c o m p o ­ n e n t of t h e wireless c o m m u n i c a t i o n s infrastructure. T h e y provide b r o a d c a s t services over v e r y w i d e areas a n d h e l p fill t h e coverage gap b e t w e e n highd e n s i t y u s e r locations. Satellite m o b i l e c o m m u n i c a t i o n s y s t e m s follow t h e s a m e basic principle as cellular s y s t e m s , except t h a t t h e cell b a s e stations are n o w satellites orbiting t h e e a r t h . Satellite s y s t e m s are typically c h a r a c t e r i z e d b y t h e h e i g h t of t h e satellite orbit, low-earth orbit (LEO), m e d i u m - e a r t h orbit (MEO), or g e o s y n c h r o n o u s orbit (GEO). T h e idea of u s i n g g e o s y n c h r o n o u s satellites for c o m m u n i c a t i o n s w a s first suggested b y t h e science fiction w r i t e r A r t h u r C. Clarke in 1945. However, t h e first d e p l o y e d satellites, t h e Soviet U n i o n ' s Sputnik in 1957 a n d t h e NASA-Bell Laboratories Echo-1 in 1960, w e r e n o t g e o s y n c h r o n o u s d u e to t h e difficulty of lifting a satellite into s u c h a h i g h orbit. T h e first GEO satellite w a s l a u n c h e d b y H u g h e s a n d NASA in 1963, a n d from t h e n until v e r y r e c e n t l y GEOs d o m i n a t e d b o t h c o m m e r c i a l a n d g o v e r n m e n t satellite s y s t e m s . T h e c u r r e n t t r e n d is to u s e lower orbits so t h a t lightweight h a n d h e l d devices c a n c o m m u n i c a t e w i t h t h e satellite. Services p r o v i d e d b y satellite s y s t e m s i n c l u d e voice, paging, a n d m e s s a g i n g services, all at fairly low data rates.

7.1.2

Wireless Data Vision Wireless c o m m u n i c a t i o n providing high-speed, high-quality i n f o r m a t i o n ex­ c h a n g e b e t w e e n portable devices located a n y w h e r e in t h e world is t h e vision

Wireless Networks

310

for t h e next c e n t u r y . People will o p e r a t e a virtual office a n y w h e r e in t h e world u s i n g a small h a n d h e l d device—with s e a m l e s s t e l e p h o n e , m o d e m , fax, a n d c o m p u t e r c o m m u n i c a t i o n s . Wireless LANs will c o n n e c t t o g e t h e r p a l m t o p , lap­ top, a n d d e s k t o p c o m p u t e r s a n y w h e r e w i t h i n a n office b u i l d i n g or c a m p u s , as well as from t h e c o r n e r cafe. In t h e h o m e t h e s e LANs will e n a b l e intelligent h o m e a p p l i a n c e s to i n t e r a c t w i t h o n e a n o t h e r a n d w i t h t h e I n t e r n e t . Video t e l e c o n f e r e n c i n g will take place b e t w e e n buildings t h a t are b l o c k s or conti­ n e n t s apart, a n d t h e s e c o n f e r e n c e s m a y i n c l u d e travelers. Wireless video will b e u s e d to create r e m o t e classrooms, r e m o t e t r a i n i n g facilities, a n d r e m o t e hospitals a n y w h e r e in t h e world. Wireless c o m m u n i c a t i o n s m a y b e classified in t e r m s of applications, sys­ t e m s , or coverage regions, as indicated in Figure 7.1. T h e applications i n c l u d e voice, I n t e r n e t access, Web browsing, paging a n d short messaging, subscriber information services, file transfer, a n d video t e l e c o n f e r e n c i n g . S y s t e m s i n c l u d e cellular t e l e p h o n e s y s t e m s , wireless LANs, w i d e a r e a wireless data s y s t e m s , a n d satellite s y s t e m s . Coverage regions i n c l u d e in-building, c a m p u s , city, re­ gional, a n d global. Wireless applications h a v e different r e q u i r e m e n t s , as w e saw in C h a p t e r 2. Voice s y s t e m s h a v e low data rate r e q u i r e m e n t s ( a r o u n d 20 Kbps) a n d c a n tolerate a high bit e r r o r rate (BER) of 10~ , b u t t h e total d e l a y m u s t b e less t h a n 100 m s . Data s y s t e m s m a y r e q u i r e h i g h e r data r a t e s (1-100 Mbps) a n d v e r y small BER, b u t do n o t h a v e a fixed delay r e q u i r e m e n t . Real-time video s y s t e m s h a v e h i g h data rate r e q u i r e m e n t s c o u p l e d w i t h t h e s a m e d e l a y c o n s t r a i n t s as voice systems, while paging a n d s h o r t m e s s a g i n g h a v e v e r y l o w data rate r e q u i r e m e n t s a n d n o delay constraints. T h e s e diverse r e q u i r e m e n t s m a k e it difficult to b u i l d a single wireless s y s t e m t h a t c a n satisfy t h e m all. As w e h a v e seen, IP a n d ATM c a n s u p p o r t diverse r e q u i r e m e n t s o n w i r e d n e t w o r k s , w i t h data rates o n t h e o r d e r of Gbps a n d BERs o n t h e o r d e r of 1 0 ~ . But this is n o t possible o n wireless n e t w o r k s , w h i c h h a v e m u c h l o w e r data r a t e s a n d h i g h e r BERs. T h e r e f o r e wireless s y s t e m s will c o n t i n u e to b e fragmented, w i t h different protocols tailored to s u p p o r t different r e q u i r e m e n t s . It is u n c e r t a i n w h e t h e r t h e r e will b e a large d e m a n d for all wireless applica­ tions. C o m p a n i e s are investing h e a v i l y to b u i l d m u l t i m e d i a wireless s y s t e m s , y e t t h e only highly profitable wireless application so far is voice. Despite opti­ mistic predictions, t h e m a r k e t for t h e s e p r o d u c t s r e m a i n s relatively small w i t h sluggish growth. Tb e x a m i n e t h e future of wireless data, it is useful to see t h e g r o w t h of various c o m m u n i c a t i o n services over t h e last five years, as s h o w n in Figure 7.2. T h e e x p o n e n t i a l g r o w t h of cellular a n d paging subscribers is ex­ c e e d e d only b y t h e d e m a n d for I n t e r n e t access. T h e n u m b e r of l a p t o p a n d 3

12

7.1

Introduction

7.1

311

Wireless applications, systems, and coverage regions.

FIGURE

p a l m t o p c o m p u t e r s is also growing steadily. T h e s e t r e n d s indicate t h a t peo­ ple w a n t to c o m m u n i c a t e w h i l e o n t h e m o v e . T h e y also w a n t to take t h e i r c o m p u t e r s w h e r e v e r t h e y go. It is therefore r e a s o n a b l e to a s s u m e t h a t p e o p l e w a n t t h e s a m e data c o m m u n i c a t i o n s capabilities o n t h e m o v e as t h e y enjoy in t h e i r h o m e or office. Yet t h e n u m b e r of wireless data subscribers r e m a i n s relatively flat, m o s t likely b e c a u s e of t h e h i g h cost a n d p o o r p e r f o r m a n c e of today's s y s t e m s .

Wireless Networks

312 Millions 100

1995

7.2

2000

Growth of wireless communication services.

FIGURE

Figure 7.3 displays t h e large a n d growing p e r f o r m a n c e gap b e t w e e n w i r e d a n d wireless n e t w o r k s . T h u s , t h e m o s t formidable obstacle to t h e g r o w t h of wireless data s y s t e m s is t h e i r p e r f o r m a n c e . M a n y t e c h n i c a l c h a l l e n g e s m u s t b e o v e r c o m e to i m p r o v e wireless n e t w o r k p e r f o r m a n c e so t h a t u s e r s will accept this p e r f o r m a n c e in exchange for mobility.

7.1.3

Technical Challenges Technical p r o b l e m s m u s t b e solved across all levels of t h e s y s t e m design to realize t h e wireless vision. At t h e h a r d w a r e level t h e t e r m i n a l m u s t h a v e m u l ­ tiple m o d e s of o p e r a t i o n to s u p p o r t t h e different applications a n d different m e d i a . Desktop c o m p u t e r s process voice, image, text, a n d video data, b u t break­ t h r o u g h s in circuit design are r e q u i r e d to i m p l e m e n t m u l t i m o d e o p e r a t i o n in a small, lightweight, h a n d h e l d device. Since m o s t p e o p l e d o n ' t w a n t to c a r r y a h e a v y b a t t e r y , t h e signal processing a n d c o m m u n i c a t i o n s h a r d w a r e of t h e portable t e r m i n a l m u s t c o n s u m e v e r y little power, w h i c h will i m p a c t h i g h e r levels of t h e s y s t e m design.

7.1

Introduction

313 User bit rate (Kbps) 100,000

100 Μ Ethernet

r

ATM

10,000 |Ethernet 1000 100

1st gen. WLAN

Polling

10

2nd gen. WLAN

Packet radio

1 .1 .01

1970

1980

Year

2000

1990

Local area packet switching

ATM

User bit rate (Kbps) 100,000

Wired-wireless bit rate "gap"

10,000 1000 100 10 L

2.4 modem

9.6 modem 9.6cellular 2.4 cellular

1

*7. , digital cellular 4

32 Kbps PCS

.1 .01

1970

1980

Year

1990

Wide area circuit switching 73 FIGURE

Performance gap b e t w e e n wireless and wired networks.

2000

314

Wireless Networks

M a n y signal processing t e c h n i q u e s for efficient spectral utilization a n d n e t w o r k i n g d e m a n d m u c h processing power, p r e c l u d i n g t h e u s e of low-power devices. H a r d w a r e a d v a n c e s for low-power circuits w i t h high processing ability will relieve s o m e of t h e s e limitations. However, placing t h e processing b u r d e n o n fixed location sites w i t h large p o w e r r e s o u r c e s will as in t h e past c o n t i n u e to d o m i n a t e wireless s y s t e m designs. But t h e associated b o t t l e n e c k s a n d single points-of-failure are undesirable. T h e finite b a n d w i d t h a n d r a n d o m variations of t h e c o m m u n i c a t i o n c h a n n e l will also r e q u i r e robust c o m p r e s s i o n s c h e m e s t h a t degrade gracefully w i t h t h e c h a n n e l . T h e wireless c h a n n e l is a n u n p r e d i c t a b l e a n d difficult c o m m u n i c a t i o n s m e d i u m . T h e scarce radio s p e c t r u m is regulated. In t h e U.S. s p e c t r u m is allo­ cated b y t h e FCC. In E u r o p e t h e e q u i v a l e n t b o d y is t h e E u r o p e a n T e l e c o m m u n i ­ cations Standards Institute (ETSI), a n d globally s p e c t r u m is controlled b y t h e I n t e r n a t i o n a l T e l e c o m m u n i c a t i o n s U n i o n (ITU). A regional or global s y s t e m operating in a given f r e q u e n c y b a n d m u s t o b e y t h e regulatory restrictions. S p e c t r u m h a s b e c o m e expensive, especially in t h e U.S., since t h e FCC b e ­ gan a u c t i o n i n g spectral allocations. In t h e r e c e n t spectral a u c t i o n s at 2 GHz, c o m p a n i e s s p e n t over $9 billion for licenses. T h e s p e c t r u m o b t a i n e d t h r o u g h t h e s e a u c t i o n s m u s t b e u s e d e x t r e m e l y efficiently to get a r e a s o n a b l e r e t u r n on i n v e s t m e n t , a n d it m u s t also b e r e u s e d over a n d over in t h e s a m e geograph­ ical area, t h u s r e q u i r i n g cellular s y s t e m designs w i t h h i g h capacity a n d good p e r f o r m a n c e . At frequencies a r o u n d several GHz, wireless radio c o m p o n e n t s w i t h r e a s o n a b l e size, p o w e r c o n s u m p t i o n , a n d cost are available. However, t h e s p e c t r u m in this f r e q u e n c y r a n g e is e x t r e m e l y crowded. Technological b r e a k t h r o u g h s t h a t e n a b l e h i g h e r f r e q u e n c y s y s t e m s w i t h t h e s a m e cost a n d p e r f o r m a n c e w o u l d greatly r e d u c e t h e s p e c t r u m shortage, a l t h o u g h p a t h loss at t h e s e h i g h e r frequencies increases, t h e r e b y limiting range. As a signal p r o p a g a t e s t h r o u g h a wireless c h a n n e l , it e x p e r i e n c e s r a n d o m fluctuations in t i m e if t h e t r a n s m i t t e r or receiver is moving, d u e to c h a n g i n g reflections a n d a t t e n u a t i o n . T h i s m a k e s it difficult to design reliable s y s t e m s w i t h g u a r a n t e e d p e r f o r m a n c e . Security is also m o r e difficult to i m p l e m e n t in wireless systems, since t h e airwaves are susceptible to s n o o p i n g from a n y o n e w i t h a n RF a n t e n n a . Analog cellular s y s t e m s h a v e n o security, a n d y o u c a n eas­ ily listen in o n conversations b y s c a n n i n g t h e analog cellular f r e q u e n c y b a n d . I b s u p p o r t applications like electronic c o m m e r c e a n d credit card transactions, t h e wireless n e t w o r k m u s t b e s e c u r e against s u c h listeners. Wireless n e t w o r k i n g p r e s e n t s a n o t h e r set of p r o b l e m s . T h e n e t w o r k m u s t locate a u s e r a m o n g millions of globally distributed mobile t e r m i n a l s . It m u s t t h e n r o u t e a call to t h a t u s e r m o v i n g at s p e e d s of u p to 100 m p h . T h e finite r e s o u r c e s of t h e n e t w o r k m u s t b e allocated in a fair a n d efficient m a n n e r

7.2

The Wireless Channel

relative to c h a n g i n g u s e r d e m a n d s a n d locations. Different p r o b l e m s are p o s e d b y t h e n e e d for protocols to interface b e t w e e n wireless a n d w i r e d n e t w o r k s w i t h vastly different p e r f o r m a n c e capabilities. P e r h a p s t h e m o s t significant t e c h n i c a l challenge in wireless n e t w o r k de­ sign is a n o v e r h a u l of t h e design process itself. Wired n e t w o r k s are m o s t l y d e s i g n e d according to t h e layers of t h e OSI m o d e l : e a c h l a y e r is d e s i g n e d i n d e ­ p e n d e n t l y from t h e o t h e r l a y e r s w i t h m e c h a n i s m s to interface b e t w e e n layers. This greatly simplifies n e t w o r k design, a l t h o u g h it l e a d s to s o m e inefficiency a n d p e r f o r m a n c e loss. T h e situation is v e r y different in a wireless n e t w o r k . Wireless link p e r f o r m a n c e , u s e r connectivity, a n d n e t w o r k topology c h a n g e over t i m e . Tb achieve good p e r f o r m a n c e wireless n e t w o r k s m u s t a d a p t to t h e s e changes, r e q u i r i n g a n i n t e g r a t e d a n d adaptive protocol stack across all layers, from t h e link l a y e r to t h e application layer.

7.2

THE WIRELESS CHANNEL T h e wireless radio c h a n n e l is a difficult m e d i u m , susceptible to noise, interfer­ ence, blockage, a n d m u l t i p a t h , a n d t h e s e c h a n n e l i m p e d i m e n t s c h a n g e over t i m e b e c a u s e of u s e r m o v e m e n t . T h e s e characteristics i m p o s e f u n d a m e n t a l limits o n t h e range, data rate, a n d reliability of c o m m u n i c a t i o n over wireless links. T h e s e limits are d e t e r m i n e d b y several factors, m o s t significantly t h e p r o p a g a t i o n e n v i r o n m e n t a n d t h e u s e r mobility. For example, t h e radio c h a n ­ n e l for a n i n d o o r u s e r at walking s p e e d s typically s u p p o r t s h i g h e r data r a t e s w i t h b e t t e r reliability t h a n t h e c h a n n e l of a n o u t d o o r u s e r s u r r o u n d e d b y tall buildings a n d m o v i n g at h i g h s p e e d . Wireless s y s t e m s u s e t h e a t m o s p h e r e as t h e i r t r a n s m i s s i o n m e d i u m . Ra­ dio signals are s e n t across this m e d i u m b y i n d u c i n g a c u r r e n t of sufficient a m p l i t u d e in a n a n t e n n a w h o s e d i m e n s i o n s are a p p r o x i m a t e l y t h e s a m e as t h e w a v e l e n g t h of t h e g e n e r a t e d signal. ( T h e signal w a v e l e n g t h λ = c/f, w h e r e / is t h e signal's carrier f r e q u e n c y a n d c is t h e s p e e d of light.) We focus o n radio w a v e s in t h e U H F a n d SHF f r e q u e n c y b a n d s , w h i c h o c c u p y t h e 0.3- to 3-GHz a n d 3- to 30-GHz p o r t i o n s of t h e s p e c t r u m , respectively. Most terrestrial m o ­ bile s y s t e m s u s e t h e U H F b a n d , w h i l e satellite s y s t e m s typically o p e r a t e in t h e SHF b a n d . At t h e s e frequencies t h e earth's c u r v a t u r e a n d t h e i o n o s p h e r e do n o t affect signal propagation. Figure 7.4 depicts a typical situation. T h e t r a n s m i t t e d signal h a s a directp a t h c o m p o n e n t b e t w e e n t h e t r a n s m i t t e r a n d t h e receiver t h a t is a t t e n u a t e d

315

Wireless Networks

316

Ψ

7.4

A wireless propagation scenario: the received signal has a direct-path component which m a y be attenuated or blocked, and other reflected, components.

or obstructed. O t h e r c o m p o n e n t s of t h e t r a n s m i t t e d signal, referred to as multipath components, are reflected, scattered, or diffracted b y s u r r o u n d i n g objects a n d arrive at t h e receiver shifted in a m p l i t u d e , phase, a n d t i m e relative to t h e direct-path signal. T h e received signal m a y also e x p e r i e n c e i n t e r f e r e n c e from o t h e r u s e r s in t h e s a m e f r e q u e n c y b a n d . Based o n this m o d e l t h e wireless radio c h a n n e l h a s four m a i n characteristics: p a t h loss, shadowing, m u l t i p a t h , a n d interference. Path loss d e t e r m i n e s h o w t h e average received signal p o w e r d e c r e a s e s w i t h t h e distance b e t w e e n t h e t r a n s m i t t e r a n d t h e receiver. Shadowing character­ izes t h e signal a t t e n u a t i o n d u e to obstructions from buildings or o t h e r objects. Multipath fading is c a u s e d b y constructive a n d destructive c o m b i n i n g of t h e m u l t i p a t h signal c o m p o n e n t s , w h i c h c a u s e s r a n d o m fluctuations in t h e re­ ceived signal a m p l i t u d e (flat-fading) as well as self-interference ( i n t e r s y m b o l i n t e r f e r e n c e or frequency-selective fading). I n t e r f e r e n c e characterizes t h e ef­ fects of o t h e r u s e r s o p e r a t i n g in t h e s a m e f r e q u e n c y b a n d e i t h e r in t h e s a m e or a n o t h e r s y s t e m .

7.2.1

Path Loss Path loss is t h e ratio of received p o w e r to t h e t r a n s m i t t e d p o w e r for a given propagation p a t h a n d is a function of p r o p a g a t i o n distance. Free-space is t h e

7.2

The Wireless Channel

317

s i m p l e s t p r o p a g a t i o n m o d e l for p a t h loss. In this m o d e l t h e r e is a directp a t h signal c o m p o n e n t b e t w e e n t h e t r a n s m i t t e r a n d t h e receiver, w i t h n o a t t e n u a t i n g objects or m u l t i p a t h reflections. If PR is t h e r e c e i v e d signal p o w e r a n d PT is t h e t r a n s m i t t e d power, t h e n in free-space propagation, GxP 72—^'f χ d T

Pr(X

l

a

w h e r e f is t h e carrier frequency, d is t h e p r o p a g a t i o n distance, G is t h e p o w e r gain from t h e t r a n s m i t a n d receive a n t e n n a s , a n d a — 2. Radio w a v e s in m o s t wireless s y s t e m s p r o p a g a t e t h r o u g h e n v i r o n m e n t s m o r e c o m p l e x t h a n free space, w h e r e t h e y are reflected, scattered, a n d diffracted b y walls, terrain, buildings, a n d o t h e r objects. T h e full details of p r o p a g a t i o n in c o m p l e x e n v i r o n m e n t s c a n b e o b t a i n e d b y solving Maxwell's e q u a t i o n s w i t h b o u n d a r y c o n d i t i o n s t h a t express t h e physical characteristics of t h e obstructing objects. Since t h e s e calculations are difficult, a n d often t h e n e c e s s a r y p a r a m e t e r s are n o t available, a p p r o x i m a t i o n s h a v e b e e n d e v e l o p e d to characterize p a t h loss. F r e q u e n t l y t h e s i m p l e e x p o n e n t i a l m o d e l for p a t h loss above is used. In free-space p r o p a g a t i o n a = 2, w h e r e a s in typical p r o p a g a t i o n e n v i r o n m e n t s a r a n g e s b e t w e e n 2 a n d 4. For a n y p a t h loss m o d e l , t h e r e c e i v e d signal-to-noise ratio is S N R = P R / N w h e r e Ν is t h e noise power. (Noise is u s u a l l y m o d e l e d as G a u s s i a n a n d w h i t e w i t h c o n s t a n t p o w e r spectral density, N . ) T h e BER of a w i r e l e s s c h a n n e l is a function of its SNR. T h e SNR r e q u i r e d to m e e t a given BER target d e p e n d s o n t h e data rate of t h e c h a n n e l , t h e c o m m u n i c a t i o n t e c h n i q u e s used, a n d t h e c h a n n e l characteristics. Since p a t h loss r e d u c e s SNR, it limits e i t h e r t h e data rate or t h e signal r a n g e of a given c o m m u n i c a t i o n s y s t e m . Moreover, since t h e p a t h loss e x p o n e n t d e t e r m i n e s h o w quickly t h e signal p o w e r falls off w i t h r e s p e c t to distance, wireless c h a n n e l s w i t h small p a t h loss e x p o n e n t s will typically h a v e larger coverage areas t h a n t h o s e w i t h large p a t h loss e x p o n e n t s . Note t h a t t h e p a t h loss is inversely p r o p o r t i o n a l to t h e s q u a r e of t h e signal f r e q u e n c y so, for example, i n c r e a s i n g t h e signal f r e q u e n c y b y a factor of 10 r e d u c e s t h e r e c e i v e d p o w e r a n d c o r r e s p o n d i n g SNR b y a factor of 100.

7.2.2

Shadow Fading T h e t r a n s m i s s i o n p a t h b e t w e e n a t r a n s m i t t e r a n d a r e c e i v e r is often b l o c k e d b y hills or buildings outdoors a n d b y furniture or walls indoors. R a n d o m signal variations d u e to t h e s e obstructing objects are referred to as shadow fading. M e a s u r e m e n t s in m a n y e n v i r o n m e n t s indicate t h a t t h e power, m e a s u r e d in decibels (dB), of a received signal subject to s h a d o w fading follows a Gaussian

Wireless Networks

318

( n o r m a l ) distribution, w i t h t h e m e a n d e t e r m i n e d b y p a t h loss a n d t h e s t a n d a r d deviation r a n g i n g from 4 to 12 dB, d e p e n d i n g o n t h e e n v i r o n m e n t . T h e r a n d o m value of t h e s h a d o w fading changes, or decorrelates, as t h e mobile m o v e s past or a r o u n d t h e obstruction. Based o n p a t h loss alone, t h e received signal p o w e r at a fixed distance from t h e t r a n s m i t t e r s h o u l d b e constant. However, s h a d o w fading c a u s e s t h e received signal p o w e r at e q u a l distances from t h e t r a n s m i t t e r to b e different, since s o m e locations h a v e m o r e s e v e r e s h a d o w fading t h a n others. T h u s , to e n s u r e t h a t t h e received SNR r e q u i r e m e n t s are m e t at a given distance from t h e transmitter, t h e t r a n s m i t p o w e r m u s t b e i n c r e a s e d to c o m p e n s a t e for s e v e r e s h a d o w fading at s o m e locations. T h i s p o w e r i n c r e a s e i m p o s e s additional b u r d e n s o n t h e t r a n s m i t t e r b a t t e r y a n d c a u s e s additional i n t e r f e r e n c e to o t h e r u s e r s in t h e s a m e f r e q u e n c y b a n d .

7.23

Multipath Flat-Fading and Intersymbol Interference Multipath c a u s e s two significant c h a n n e l i m p a i r m e n t s : flat-fading and inter­ symbol interference. Flat-fading describes t h e rapid fluctuations of t h e r e c e i v e d signal p o w e r over short t i m e p e r i o d s or over short distances. Such fading is c a u s e d b y t h e i n t e r f e r e n c e b e t w e e n different m u l t i p a t h signal c o m p o n e n t s t h a t arrive at t h e receiver at different t i m e s a n d h e n c e are subject to constructive a n d destructive interference. T h i s constructive a n d destructive i n t e r f e r e n c e g e n e r a t e s a s t a n d i n g wave p a t t e r n of t h e received signal p o w e r relative to dis­ t a n c e or, for a m o v i n g receiver, relative to t i m e . Figure 7.5 s h o w s a plot of t h e fading exhibited b y t h e r e c e i v e d signal p o w e r in dB as a function of t i m e or distance. T h e destructive i n t e r f e r e n c e c a n cause t h e received signal p o w e r to fall m o r e t h a n 30 dB ( t h r e e orders of m a g n i t u d e ) b e l o w its average value. A c h a n n e l is said to b e in a deep fade w h e n e v e r its received signal p o w e r falls b e l o w t h a t r e q u i r e d to m e e t t h e l i n k p e r f o r m a n c e specifications. Since c o m m u n i c a t i o n links are d e s i g n e d w i t h a n extra p o w e r m a r g i n (link m a r g i n ) of 10 to 20 dB to c o m p e n s a t e for fading a n d o t h e r c h a n n e l i m p a i r m e n t s , a c h a n n e l is in a d e e p fade if its received p o w e r falls 10 to 20 dB b e l o w its average r e c e i v e d power. Figure 7.5 s h o w s t h a t flat-fading c h a n n e l s often e x p e r i e n c e d e e p fades. In addition, t h e signal p o w e r c h a n g e s drastically over distances of a p p r o x i m a t e l y half a signal w a v e l e n g t h . At a signal f r e q u e n c y of 900 MHz, this c o r r e s p o n d s to e v e r y 0.3 m or e v e r y millisecond for t e r m i n a l s m o v i n g at 30 m p h .

7.2

The Wireless Channel

319

Received signal power (dB)

dB with respect to RMS value _ 10

o——

l=

_l_

_i_

0.5

1.5

1

f, in seconds

ι 0

10

20

30

JC, in wavelength

7.5

Signal fading over time and distance.

FIGURE

T h e variation in t h e r e c e i v e d signal e n v e l o p e of a flat-fading signal typically follows a Rayleigh distribution if t h e signal p a t h b e t w e e n t h e t r a n s m i t t e r a n d t h e receiver is o b s t r u c t e d a n d a Ricean distribution if this signal p a t h is n o t obstructed. (If x, y are zero-mean, G a u s s i a n r a n d o m variables w i t h t h e s a m e variance, t h e n z = x +y ) / h a s a Rayleigh distribution. If m is a constant, t h e n ζ = ((χ Λ-ρ) + # ) h a s a Ricean distibution. 2

2

2 1 2

2

1 / / 2

T h e c o m b i n a t i o n of p a t h loss, s h a d o w i n g , a n d flat-fading is s h o w n in Figure 7.6. T h e p o w e r fall-off w i t h distance d u e to p a t h loss is fairly slow, while t h e signal variation d u e to s h a d o w i n g c h a n g e s m o r e quickly, a n d t h e variation d u e to flat-fading is v e r y fast. Flat-fading h a s two m a i n i m p l i c a t i o n s for wireless link design. First, in flat-fading t h e r e c e i v e d signal p o w e r falls well b e l o w its average value. T h i s c a u s e s a large i n c r e a s e in BER, w h i c h c a n b e r e d u c e d b y i n c r e a s i n g t h e t r a n s m i t t e d p o w e r of t h e signal. However, it is v e r y wasteful of p o w e r to c o m p e n s a t e for flat-fading in this m a n n e r , since d e e p fades o c c u r rarely a n d over v e r y short t i m e periods. So m o s t s y s t e m s do n o t i n c r e a s e t h e i r t r a n s m i t p o w e r sufficiently to r e m o v e d e e p signal fades. For typical u s e r s p e e d s a n d

Wireless Networks

320 Received signal power (dB)

log (distance)

7.6

Received signal power with path loss, shadow fading, and

flat-fading.

FIGURE

data rates, t h e s e fades affect m a n y bits, causing long strings of bit errors called error bursts. Error b u r s t s are difficult to correct u s i n g error-correction codes, w h i c h typically correct only a few s i m u l t a n e o u s bit errors. O t h e r m e t h o d s to c o m p e n s a t e for error b u r s t s d u e to flat-fading are discussed in section 7.3.3. T h e o t h e r m a i n i m p a i r m e n t i n t r o d u c e d b y m u l t i p a t h is i n t e r s y m b o l inter­ ference (ISI). ISI b e c o m e s a significant p r o b l e m w h e n t h e m a x i m u m difference in t h e p a t h delays of t h e different m u l t i p a t h c o m p o n e n t s , called t h e multipath delay spread, exceeds a significant fraction of a bit t i m e . T h e result is selfinterference, since a m u l t i p a t h reflection carrying a given bit t r a n s m i s s i o n will arrive at t h e receiver s i m u l t a n e o u s l y w i t h a different (delayed) multip a t h reflection carrying a p r e v i o u s bit t r a n s m i s s i o n . In t h e f r e q u e n c y d o m a i n this self-interference c o r r e s p o n d s to a nonflat f r e q u e n c y s p e c t r u m , so signal c o m p o n e n t s at different frequencies are multiplied b y different c o m p l e x scale factors, t h e r e b y distorting t h e t r a n s m i t t e d signal. For this r e a s o n ISI is also referred to as frequency-selective fading. A c h a n n e l exhibits frequency-selective fading if its coherence bandwidth, defined as t h e inverse of t h e c h a n n e l ' s m u l t i p a t h delay spread, is less t h a n t h e b a n d w i d t h of t h e t r a n s m i t t e d signal. ISI c a u s e s a high BER t h a t c a n n o t b e r e d u c e d b y increasing t h e signal power, since t h a t also i n c r e a s e s t h e p o w e r

7.2

The Wireless Channel

321

of t h e self-interference. T h u s , w i t h o u t c o m p e n s a t i o n ISI forces a r e d u c t i o n in data rate so t h a t t h e delay s p r e a d associated w i t h t h e m u l t i p a t h c o m p o n e n t s is less t h a n o n e bit t i m e . This i m p o s e s a s t r i n g e n t limit o n data rates, o n t h e order of 100 Kbps for o u t d o o r e n v i r o n m e n t s a n d 1 Mbps for i n d o o r e n v i r o n m e n t s . T h u s , s o m e form of ISI c o m p e n s a t i o n is n e e d e d to a c h i e v e h i g h data rates. T h e s e c o m p e n s a t i o n t e c h n i q u e s are discussed in section 7.3.4.

7.2.4

Doppler Frequency Shift Relative m o t i o n b e t w e e n t h e t r a n s m i t t e r a n d t h e r e c e i v e r c a u s e s a shift in t h e f r e q u e n c y of t h e t r a n s m i t t e d signal called t h e Doppler shift. T h e D o p p l e r shift, /b, is given b y /b = ν/λ, w h e r e ν is t h e relative velocity b e t w e e n t h e t r a n s m i t t e r a n d t h e receiver a n d λ is t h e w a v e l e n g t h of t h e t r a n s m i t t e d signal. Since t h e mobile velocity varies w i t h time, so does t h e D o p p l e r shift. T h i s variable D o p p l e r shift i n t r o d u c e s a n FM m o d u l a t i o n into t h e signal, c a u s i n g t h e signal b a n d w i d t h to i n c r e a s e b y r o u g h l y fr>. For a t r a n s m i t f r e q u e n c y of 900 MHz a n d a u s e r s p e e d of 60 m p h , t h e D o p p l e r shift is a b o u t 80 Hz. Since typical signal b a n d w i d t h s are o n t h e o r d e r of t e n s of kilohertz or m o r e , b a n d w i d t h s p r e a d i n g d u e to D o p p l e r is n o t a significant p r o b l e m in m o s t applications. However, D o p p l e r does cause t h e signal to d e c o r r e l a t e over a t i m e p e r i o d r o u g h l y e q u a l to l//b. For differential signal detection, t h e D o p p l e r i m p o s e s a l o w e r b o u n d o n t h e c h a n n e l BER t h a t c a n n o t b e r e d u c e d b y i n c r e a s i n g t h e signal power. More details o n t h e i m p a c t of c h a n n e l D o p p l e r for differential d e t e c t i o n are provided in section 7.3.1.

7.2.5

Interference Wireless c o m m u n i c a t i o n c h a n n e l s e x p e r i e n c e i n t e r f e r e n c e from various sources. T h e m a i n source of i n t e r f e r e n c e in cellular s y s t e m s is f r e q u e n c y reuse, w h e r e frequencies are r e u s e d at spatially s e p a r a t e d locations to i n c r e a s e spec­ tral efficiency. I n t e r f e r e n c e from f r e q u e n c y r e u s e c a n b e r e d u c e d b y m u l t i u s e r detection, directional a n t e n n a s , a n d d y n a m i c c h a n n e l allocation, all of w h i c h increase s y s t e m complexity. O t h e r sources of i n t e r f e r e n c e in wireless s y s t e m s i n c l u d e adjacent c h a n n e l interference, c a u s e d b y signals in adjacent c h a n n e l s w i t h signal c o m p o n e n t s outside t h e i r allocated f r e q u e n c y range, a n d n a r r o w b a n d interference, c a u s e d b y u s e r s in o t h e r s y s t e m s o p e r a t i n g in t h e s a m e f r e q u e n c y b a n d . Adjacent chan­ nel interference can be mostly removed by introducing guard bands between c h a n n e l s , b u t this is wasteful of b a n d w i d t h . N a r r o w b a n d i n t e r f e r e n c e c a n b e r e m o v e d t h r o u g h n o t c h filters or s p r e a d s p e c t r u m t e c h n i q u e s . Notch filters are

Wireless Networks

322

simple devices b u t t h e y r e q u i r e k n o w l e d g e of t h e exact location of t h e n a r r o w ­ b a n d interference. Spread s p e c t r u m is v e r y effective at r e m o v i n g n a r r o w b a n d interference, b u t it r e q u i r e s significant s p r e a d i n g of t h e signal b a n d w i d t h as well as a n increase in s y s t e m complexity. For t h e s e r e a s o n s s p r e a d s p e c t r u m in not typically u s e d j u s t to r e m o v e n a r r o w b a n d interference. Because s p r e a d s p e c t r u m allows m u l t i p l e u s e r s to s h a r e t h e s a m e b a n d w i d t h , it is u s e d for m u l t i p l e access.

7.2.6

Infrared versus Radio In infrared c o m m u n i c a t i o n t h e f r e q u e n c y of t h e t r a n s m i t t e d signal is m u c h h i g h e r t h a n typical radio frequencies, a r o u n d 100 GHz. Because t h e r e c e i v e d signal p o w e r is inversely proportional to t h e s q u a r e of t h e signal frequency, infrared t r a n s m i s s i o n over large distances r e q u i r e s a v e r y h i g h t r a n s m i t p o w e r or highly directional a n t e n n a s . T h e r e are two m a i n forms of infrared t r a n s m i s ­ sion: directive a n d nondirective. In directive t r a n s m i s s i o n t h e t r a n s m i t a n t e n n a is p o i n t e d directly at t h e receiver, w h e r e a s in n o n d i r e c t i v e t r a n s m i s s i o n t h e signal is t r a n s m i t t e d u n i f o r m l y in all directions. Since directive t r a n s m i s s i o n c o n c e n t r a t e s all its p o w e r in o n e direction, it achieves m u c h h i g h e r data rates t h a n nondirective t r a n s m i s s i o n . However, t h e s e s y s t e m s are severely d e g r a d e d b y obstructing objects a n d t h e c o r r e s p o n d i n g s h a d o w fading, w h i c h is difficult to avoid in m o s t i n d o o r e n v i r o n m e n t s w i t h m o b i l e users. N o n d i r e c t e d links h a v e limited range d u e to p a t h loss, typically in t h e t e n s of m e t e r s . Infrared t r a n s m i s s i o n enjoys a n u m b e r of a d v a n t a g e s over radio, m o s t sig­ nificantly t h e fact t h a t s p e c t r u m in this f r e q u e n c y r a n g e is u n r e g u l a t e d . T h u s , infrared s y s t e m s n e e d n o t obtain a n FCC license for operation. In addition, infrared s y s t e m s are i m m u n e to radio interference. Infrared radiation will n o t p e n e t r a t e walls a n d o t h e r o p a q u e materials, so a n infrared signal is confined to t h e r o o m in w h i c h it originates. T h i s m a k e s infrared m o r e s e c u r e against eavesdropping, a n d it allows n e i g h b o r i n g r o o m s to u s e t h e s a m e infrared links w i t h o u t interference. T h e s e signals are n o t subject to flat-fading since varia­ tions in t h e received signal p o w e r are i n t e g r a t e d out b y t h e detector. ISI is a major p r o b l e m for high-speed infrared s y s t e m s , as it is for radio s y s t e m s . T h u s , high-speed infrared s y s t e m s m u s t u s e s o m e form of ISI mitiga­ tion, typically equalization. Infrared s y s t e m s are also significantly d e g r a d e d b y a m b i e n t light, w h i c h radiates at r o u g h l y t h e s a m e frequency, c a u s i n g substan­ tial noise. In s u m m a r y , infrared s y s t e m s are u n l i k e l y to b e u s e d for low-cost outdoor s y s t e m s b e c a u s e of t h e i r l i m i t e d range. T h e y h a v e s o m e a d v a n t a g e s over radio s y s t e m s in i n d o o r e n v i r o n m e n t s b u t m u s t o v e r c o m e t h e a m b i e n t light a n d r a n g e limitations to b e successful in t h e s e applications. D u e to t h e s e

12

The Wireless Channel

323

p r o b l e m s radio is still t h e d o m i n a n t t e c h n o l o g y for b o t h i n d o o r a n d outdoor systems.

7.27

Capacity Limits of Wireless Channels In 1949 Claude S h a n n o n d e t e r m i n e d t h e capacity limits of c o m m u n i c a t i o n c h a n n e l s w i t h additive w h i t e G a u s s i a n noise. For a c h a n n e l w i t h o u t shadowing, fading, or ISI, S h a n n o n p r o v e d t h a t t h e m a x i m u m possible data rate o n a given c h a n n e l of b a n d w i d t h JB is £ = £log (l +SNR)bps, 2

w h e r e SNRis t h e received signal-to-noise p o w e r ratio. T h e S h a n n o n capacity is a theoretical limit t h a t c a n n o t b e a c h i e v e d in practice, b u t as link level design t e c h n i q u e s improve, data r a t e s for this additive w h i t e noise c h a n n e l a p p r o a c h this theoretical b o u n d . T h e S h a n n o n capacity for m a n y wireless c h a n n e l s of i n t e r e s t is u n k n o w n a n d d e p e n d s n o t o n l y o n t h e c h a n n e l b u t also o n w h e t h e r t h e t r a n s m i t t e r a n d / o r t h e receiver c a n track t h e c h a n n e l variations. For cases w h e r e t h e capacity is k n o w n , t h e r e is typically a large gap b e t w e e n actual p e r f o r m a n c e a n d t h e capacity. For example, o n e digital cellular s t a n d a r d h a s a 30-kHz b a n d w i d t h a n d a received SNR of a p p r o x i m a t e l y 20 dB (or m o r e ) after a t t e n u a t i o n from s h a d o w fading. T h e S h a n n o n capacity of this c h a n n e l w i t h Rayleigh fading, a s s u m i n g t h a t t h e c h a n n e l variation c a n b e tracked, is o n t h e o r d e r of 200 Kbps. However, cellular s y s t e m s only achieve data r a t e s o n t h e o r d e r of 20 Kbps p e r c h a n n e l . While s o m e of this p e r f o r m a n c e gap is d u e to c h a n n e l i m p a i r m e n t s t h a t are n o t i n c o r p o r a t e d into t h e theoretical m o d e l , m o s t of t h e gap is d u e to t h e relatively inefficient signaling m e t h o d s u s e d o n today's wireless c h a n n e l s . In w i r e d c h a n n e l s , w h e r e t e c h n o l o g y is quite m a t u r e a n d t h e c h a n n e l characteristics fairly b e n i g n , t h e data rates a c h i e v e d in practice are quite close to t h e S h a n n o n b o u n d . For example, t e l e p h o n e loops h a v e a capacity limit of b e t w e e n 30 a n d 60 Kbps, d e p e n d i n g o n t h e l i n e quality, a n d m o d e m rates today are close to this limit. T h e s i m p l e formula given above for S h a n n o n capacity is applicable to static c h a n n e l s w i t h w h i t e Gaussian noise. S h a n n o n capacity is also k n o w n for static c h a n n e l s w i t h n o n w h i t e noise a n d i n t e r s y m b o l i n t e r f e r e n c e . However, d e t e r m i n i n g t h e capacity limits of time-varying wireless c h a n n e l s w i t h shadow­ ing, m u l t i p a t h fading, a n d i n t e r s y m b o l i n t e r f e r e n c e is quite challenging, a n d

Wireless Networks

324

d e p e n d s o n t h e c h a n n e l characteristics, t h e c h a n n e l rate of change, a n d t h e ability to track t h e c h a n n e l variations. A relatively s i m p l e l o w e r b o u n d for t h e capacity of a n y c h a n n e l is t h e S h a n n o n capacity u n d e r t h e worst-case propagation conditions. T h i s c a n b e a good b o u n d to apply in practice, since m a n y c o m m u n i c a t i o n links are d e s i g n e d to h a v e acceptable p e r f o r m a n c e e v e n u n d e r t h e worst-case conditions. But worst-case s y s t e m design a n d capacity e v a l u a t i o n c a n b e overly pessimistic. For c h a n n e l s with severe m u l t i p a t h , t h e c h a n n e l capacity u n d e r worst-case fading conditions c a n b e e q u a l or close to zero. Fading c o m p e n s a t i o n t e c h n i q u e s c a n b e u s e d o n t h e s e c h a n n e l s to increase b o t h t h e i r S h a n n o n capacity a n d t h e i r achievable data rates in practice. Some of t h e s e c o m p e n s a t i o n t e c h n i q u e s are discussed in t h e next section.

7.3

LINK LEVEL DESIGN T h e goal of link level design is to provide high data rates w i t h low delays a n d BERs while u s i n g m i n i m u m b a n d w i d t h a n d t r a n s m i t power. T h e link l a y e r design m u s t p e r f o r m well in radio e n v i r o n m e n t s w i t h fading, shadowing, m u l ­ tipath, a n d interference. H a r d w a r e constraints, s u c h as i m p e r f e c t t i m i n g a n d n o n l i n e a r amplifiers, m u s t also b e t a k e n into consideration. Low-power i m p l e ­ m e n t a t i o n s are n e e d e d , particularly for t h e mobile units, w h i c h h a v e l i m i t e d b a t t e r y power. In addition, low-cost i m p l e m e n t a t i o n s for b o t h t r a n s m i t t e r a n d receiver are clearly desirable. M a n y of t h e s e p r o p e r t i e s are m u t u a l l y exclusive a n d i n d u c e trade-offs in t h e choice of link level design t e c h n i q u e s .

7.3.1

Modulation Techniques Digital m o d u l a t i o n is t h e process of e n c o d i n g a digital i n f o r m a t i o n signal into t h e amplitude, phase, or f r e q u e n c y of t h e t r a n s m i t t e d signal. T h e e n c o d i n g process affects t h e b a n d w i d t h of t h e t r a n s m i t t e d signal a n d its r o b u s t n e s s to c h a n n e l i m p a i r m e n t s . In general, a m o d u l a t i o n t e c h n i q u e e n c o d e s several bits into o n e symbol, a n d t h e rate of s y m b o l t r a n s m i s s i o n d e t e r m i n e s t h e b a n d w i d t h of t h e t r a n s m i t t e d signal. Since t h e signal b a n d w i d t h is d e t e r m i n e d b y t h e symbol rate, h a v i n g a large n u m b e r of bits p e r s y m b o l g e n e r a l l y yields a h i g h e r data rate for a given signal b a n d w i d t h . However, t h e larger t h e n u m b e r of bits p e r symbol, t h e g r e a t e r t h e r e q u i r e d received SNR for a given target BER. Digital m o d u l a t i o n t e c h n i q u e s m a y b e l i n e a r or nonlinear. In l i n e a r m o d ­ ulation t h e a m p l i t u d e a n d / o r p h a s e of t h e t r a n s m i t t e d signal varies l i n e a r l y

73

Link Level Design

w i t h t h e digital m o d u l a t i n g signal, w h e r e a s t h e t r a n s m i t t e d signal a m p l i t u d e is c o n s t a n t for n o n l i n e a r t e c h n i q u e s . Linear m o d u l a t i o n t e c h n i q u e s , i n c l u d i n g all forms of q u a d r a t u r e - a m p l i t u d e m o d u l a t i o n (QAM) a n d phase-shift-keying (PSK), u s e less b a n d w i d t h t h a n n o n l i n e a r t e c h n i q u e s , including v a r i o u s forms of f r e q u e n c y / m i n i m u m - s h i f t k e y i n g (FSK a n d MSK). Since l i n e a r t e c h n i q u e s e n c o d e i n f o r m a t i o n into t h e a m p l i t u d e a n d p h a s e of l i n e a r m o d u l a t i o n , this t y p e of m o d u l a t i o n is m o r e sus­ ceptible to a m p l i t u d e a n d p h a s e fluctuations c a u s e d b y m u l t i p a t h flat-fading. In addition, t h e amplifiers u s e d for l i n e a r m o d u l a t i o n m u s t b e linear, a n d t h e s e amplifiers are m o r e expensive a n d less efficient t h a n n o n l i n e a r amplifiers. T h u s , t h e b a n d w i d t h efficiency of l i n e a r m o d u l a t i o n is g e n e r a l l y o b t a i n e d at t h e e x p e n s e of h a r d w a r e cost, power, a n d h i g h e r BERs in fading. L i n e a r m o d u ­ lation t e c h n i q u e s are u s e d in m o s t wireless LAN products, w h e r e a s n o n l i n e a r t e c h n i q u e s are u s e d in m o s t cellular a n d w i d e a r e a wireless data s y s t e m s . Linear m o d u l a t i o n t e c h n i q u e s c a n b e d e t e c t e d c o h e r e n t l y or differentially. C o h e r e n t d e t e c t i o n r e q u i r e s t h e r e c e i v e r to obtain a c o h e r e n t p h a s e r e f e r e n c e for t h e t r a n s m i t t e d signal. T h i s is difficult to do in a rapidly fading environ­ m e n t , a n d also i n c r e a s e s t h e c o m p l e x i t y of t h e receiver. Differential d e t e c t i o n u s e s t h e p r e v i o u s l y d e t e c t e d s y m b o l as a p h a s e r e f e r e n c e for t h e c u r r e n t s y m ­ bol. Because this d e t e c t e d s y m b o l is a n o i s y reference, differential d e t e c t i o n r e q u i r e s r o u g h l y twice t h e p o w e r of c o h e r e n t d e t e c t i o n for t h e s a m e BER. More­ over, if t h e c h a n n e l is c h a n g i n g rapidly, t h e n differential d e t e c t i o n is n o t v e r y accurate, since t h e c h a n n e l p h a s e m a y c h a n g e c o n s i d e r a b l y over o n e s y m b o l t i m e . As a result, rapidly c h a n g i n g c h a n n e l s w i t h differential d e t e c t i o n h a v e a n irreducible error floor, t h a t is, t h e BER of t h e c h a n n e l h a s a l o w e r b o u n d (error floor) t h a t c a n n o t b e r e d u c e d b y i n c r e a s i n g t h e r e c e i v e d SNR. T h i s e r r o r floor i n c r e a s e s as t h e rate of c h a n n e l variation (the c h a n n e l D o p p l e r ) i n c r e a s e s a n d d e c r e a s e s as t h e data rate i n c r e a s e s (since a h i g h e r data r a t e c o r r e s p o n d s to a s h o r t e r bit time, so t h e c h a n n e l p h a s e h a s less t i m e to d e c o r r e l a t e b e t w e e n bits). For high-speed wireless data (above 1 Mbps), t h e e r r o r floor is q u i t e low at u s e r s p e e d s b e l o w 60 m p h , b u t at l o w e r data r a t e s t h e e r r o r floor b e c o m e s significant, t h e r e b y p r e v e n t i n g t h e u s e of differential d e t e c t i o n .

73.2

Channel Coding and Link Layer Retransmission C h a n n e l coding adds r e d u n d a n t bits to t h e t r a n s m i t t e d bit s t r e a m , w h i c h are u s e d b y t h e receiver to correct e r r o r s i n t r o d u c e d b y t h e c h a n n e l . T h i s allows for a r e d u c t i o n in t r a n s m i t p o w e r to achieve a given target BER a n d

325

Wireless Networks

326

also p r e v e n t s p a c k e t r e t r a n s m i s s i o n s if all t h e bit errors in a p a c k e t c a n b e corrected. C o n v e n t i o n a l forward error-correction (FEC) codes u s e block or convolutional code designs to p r o d u c e t h e r e d u n d a n t bits for FEC: t h e errorcorrection capabilities of t h e s e code designs are obtained at t h e e x p e n s e of i n c r e a s e d signal b a n d w i d t h or a l o w e r data rate. Trellis codes, i n v e n t e d in t h e early 1980s, u s e a j o i n t design of t h e c h a n n e l code a n d t h e m o d u l a t i o n to provide FEC w i t h o u t b a n d w i d t h e x p a n s i o n or rate r e d u c t i o n . T h e latest a d v a n c e in coding t e c h n o l o g y is t h e family of TUrbo codes, in­ v e n t e d in 1993. TUrbo codes, w h i c h achieve w i t h i n a fraction of a dB of t h e S h a n n o n capacity o n certain c h a n n e l s , are c o m p l e x code designs t h a t c o m b i n e a n e n c o d e d version of t h e original bit s t r e a m w i t h a n e n c o d e d version of o n e or m o r e p e r m u t a t i o n s of this original bit s t r e a m . T h e o p t i m a l d e c o d e r for t h e resulting code is v e r y complex, b u t TUrbo codes u s e a n iterative t e c h n i q u e to ap­ proximate t h e o p t i m a l d e c o d e r w i t h r e a s o n a b l e complexity. While TUrbo codes exhibit dramatically i m p r o v e d p e r f o r m a n c e over p r e v i o u s coding t e c h n i q u e s a n d c a n b e u s e d w i t h e i t h e r b i n a r y or multilevel m o d u l a t i o n , t h e y g e n e r a l l y h a v e high complexity a n d large delays, w h i c h m a k e s t h e m u n s u i t a b l e for m a n y wireless applications. A n o t h e r w a y to c o m p e n s a t e for t h e c h a n n e l errors p r e v a l e n t in wireless s y s t e m s is to i m p l e m e n t link l a y e r r e t r a n s m i s s i o n u s i n g t h e A u t o m a t i c Repeat Request or A R Q p r o t o c o l . In A R Q d a t a is collected into packets, a n d e a c h p a c k e t is e n c o d e d w i t h a c h e c k s u m . T h e receiver u s e s t h e c h e c k s u m to d e t e r m i n e if o n e or m o r e bits in t h e p a c k e t w e r e r e c e i v e d in error. If a n e r r o r is detected, t h e receiver r e q u e s t s a r e t r a n s m i s s i o n of t h e e n t i r e packet. Link l a y e r r e t r a n s ­ mission is wasteful, since e a c h r e t r a n s m i s s i o n r e q u i r e s additional p o w e r a n d b a n d w i d t h a n d also interferes w i t h o t h e r users. In addition, p a c k e t r e t r a n s ­ missions c a n c a u s e data to b e delivered to t h e receiver o u t of o r d e r as well as triggering duplicate a c k n o w l e d g m e n t s or end-to-end r e t r a n s m i s s i o n s at t h e t r a n s p o r t layer, further b u r d e n i n g t h e n e t w o r k . While A R Q h a s disadvantages, in applications t h a t r e q u i r e error-free p a c k e t delivery at t h e link layer, FEC is n o t sufficient a n d so A R Q i s t h e o n l y alternative.

73.3

Flat-Fading Countermeasures T h e r a n d o m variation in received signal p o w e r resulting from m u l t i p a t h flatfading c a u s e s a v e r y large i n c r e a s e in t h e average BER o n t h e link. For ex­ a m p l e , in o r d e r to m a i n t a i n a n average BER of 1 0 (a typical r e q u i r e m e n t for t h e link design of voice s y s t e m s ) , 60 t i m e s m o r e p o w e r is r e q u i r e d if flat- 3

73

Link Level Design

fading is p r e s e n t . (This calculation a s s u m e s binary-phase-shift-keying or BPSK m o d u l a t i o n . Higher-level m o d u l a t i o n s r e q u i r e s a n e v e n larger t r a n s m i t p o w e r increase.) T h i s r e q u i r e d i n c r e a s e in p o w e r is e v e n larger at t h e m u c h lower BERs r e q u i r e d for data t r a n s m i s s i o n . T h u s , c o u n t e r m e a s u r e s to c o m b a t t h e ef­ fects of flat-fading c a n significantly r e d u c e t h e t r a n s m i t p o w e r r e q u i r e d o n t h e link to achieve a target BER. T h e m o s t c o m m o n flat-fading c o u n t e r m e a s u r e s are diversity, coding a n d interleaving, a n d adaptive m o d u l a t i o n . In diversity, several separate, i n d e p e n d e n t l y fading signal p a t h s are es­ tablished b e t w e e n t h e t r a n s m i t t e r a n d t h e receiver, a n d t h e r e c e i v e d signals o b t a i n e d from e a c h of t h e s e p a t h s are c o m b i n e d . Because t h e r e is a low prob­ ability of i n d e p e n d e n t fading p a t h s e x p e r i e n c i n g d e e p fades s i m u l t a n e o u s l y , t h e signal obtained b y c o m b i n i n g several s u c h fading p a t h s is u n l i k e l y to ex­ p e r i e n c e large p o w e r variations, especially w i t h four or m o r e diversity p a t h s . I n d e p e n d e n t fading p a t h s c a n b e a c h i e v e d b y s e p a r a t i n g t h e signal in time, frequency, space, or polarization. T i m e a n d f r e q u e n c y diversity are inefficient, b e c a u s e i n f o r m a t i o n is du­ plicated. Polarization diversity is of l i m i t e d effectiveness b e c a u s e o n l y two i n d e p e n d e n t fading p a t h s ( c o r r e s p o n d i n g to horizontal a n d vertical polariza­ tion) c a n b e created, a n d t h e t r a n s m i s s i o n p o w e r is u s u a l l y divided b e t w e e n t h e s e t w o p a t h s . T h a t leaves space diversity as t h e m o s t efficient diversity t e c h n i q u e . In space diversity, i n d e p e n d e n t fading p a t h s are o b t a i n e d u s i n g a n a n t e n n a array, w h e r e e a c h a n t e n n a e l e m e n t receives a n i n d e p e n d e n t fading p a t h . In o r d e r to obtain i n d e p e n d e n t fading paths, t h e a n t e n n a e l e m e n t s m u s t b e s p a c e d at least o n e half signal w a v e l e n g t h apart, w h i c h m a y b e difficult o n small h a n d h e l d devices, especially in t h e l o w e r f r e q u e n c y b a n d s (f < 1 GHz or, equivalently, λ > .3 m ) . A n o t h e r t e c h n i q u e to c o m b a t flat-fading effects is coding a n d interleav­ ing. In general, flat-fading c a u s e s bit e r r o r s to o c c u r in b u r s t s c o r r e s p o n d i n g to t h e t i m e s w h e n t h e c h a n n e l is in a d e e p fade. Low-complexity c h a n n e l codes c a n correct at m o s t a few s i m u l t a n e o u s bit errors, a n d t h e code per­ f o r m a n c e d e t e r i o r a t e s rapidly w h e n e r r o r s o c c u r in large b u r s t s . T h e basic idea of coding a n d i n t e r l e a v i n g is to s p r e a d b u r s t e r r o r s over m a n y codewords. Specifically, in t h e i n t e r l e a v i n g p r o c e s s adjacent bits in a single c o d e w o r d are s e p a r a t e d b y bits from o t h e r c o d e w o r d s a n d t h e n t h e s e s c r a m b l e d bits are t r a n s m i t t e d over t h e c h a n n e l . Since c h a n n e l e r r o r s o c c u r in bursts, t h e scrambling p r e v e n t s adjacent bits in t h e s a m e c o d e w o r d from b e i n g affected b y t h e s a m e error burst. At t h e r e c e i v e r t h e bits are d e i n t e r l e a v e d ( d e s c r a m b l e d ) b a c k to t h e i r original o r d e r a n d t h e n p a s s e d to t h e decoder. If t h e in­ terleaving process s p r e a d s o u t t h e b u r s t errors, a n d b u r s t e r r o r s do n o t o c c u r too frequently, t h e n t h e c o d e w o r d s p a s s e d to t h e d e c o d e r will h a v e at m o s t

327

Wireless Networks

328

o n e bit error, a n d t h e s e errors are easily c o r r e c t e d w i t h m o s t FEC c h a n n e l codes. T h e cost of coding a n d i n t e r l e a v i n g (increased delay a n d complexity) m a y b e large if t h e fading rate relative to t h e data rate is slow, w h i c h is typically t h e case for high-speed data. For example, at a c h a n n e l D o p p l e r of 10 Hz a n d a data rate of 10 Mbps, a n error b u r s t will last a r o u n d 300,000 bits. In this case t h e i n t e r l e a v e r m u s t b e large e n o u g h to h a n d l e at least t h a t m u c h data, a n d t h e application m u s t b e able to tolerate a n i n t e r l e a v e r delay of at least 30 m s . W h e n t h e c h a n n e l can b e e s t i m a t e d a n d this e s t i m a t e is s e n t b a c k to t h e transmitter, t h e t r a n s m i s s i o n s c h e m e c a n b e a d a p t e d relative to t h e c h a n n e l conditions. In particular, t h e data rate, power, a n d coding s c h e m e c a n b e a d a p t e d relative to t h e c h a n n e l fading to m a x i m i z e t h e average data rate or to m i n i m i z e t h e average t r a n s m i t p o w e r or BER. This a d a p t a t i o n allows t h e c h a n n e l to b e u s e d m o r e efficiently, since t h e t r a n s m i s s i o n p a r a m e t e r s are optimized to take advantage of favorable c h a n n e l conditions. Several practical constraints d e t e r m i n e w h e n adaptive t e c h n i q u e s s h o u l d b e used. If t h e c h a n n e l is c h a n g i n g so fast t h a t it c a n n o t b e accurately e s t i m a t e d or t h e e s t i m a t e c a n n o t b e fed b a c k to t h e t r a n s m i t t e r before t h e c h a n n e l c h a n g e s significantly, t h e n adaptive t e c h n i q u e s will p e r f o r m poorly. Adaptive tech­ n i q u e s also i n c r e a s e t h e complexity of b o t h t h e t r a n s m i t t e r a n d t h e receiver to a c c o u n t for t h e c h a n n e l e s t i m a t i o n a n d t h e adaptive t r a n s m i s s i o n . Finally, a feedback p a t h is r e q u i r e d to relay t h e c h a n n e l e s t i m a t e b a c k to t h e transmitter, w h i c h occupies a small a m o u n t of additional b a n d w i d t h o n t h e r e t u r n c h a n n e l .

7.3.4

Intersymbol Interference Countermeasures T e c h n i q u e s to c o m b a t ISI fall into two categories: signal processing a n d a n t e n n a solutions. Signal processing t e c h n i q u e s , including equalization, m u l t i c a r r i e r m o d u l a t i o n , a n d s p r e a d s p e c t r u m , c a n e i t h e r c o m p e n s a t e for ISI at t h e receiver or m a k e t h e t r a n s m i t t e d signal less sensitive to ISI. A n t e n n a solutions, includ­ ing directive b e a m s a n d s m a r t a n t e n n a s , c h a n g e t h e p r o p a g a t i o n e n v i r o n m e n t so t h a t t h e delay b e t w e e n m u l t i p a t h c o m p o n e n t s , a n d t h e c o r r e s p o n d i n g ISI resulting from t h e s e delays, is r e d u c e d . T h e goal of equalization is to cancel t h e ISI or, equivalently, to i n v e r t t h e effects of t h e c h a n n e l . C h a n n e l inversion c a n b e a c h i e v e d b y passing t h e received signal t h r o u g h a l i n e a r equalizing filter w i t h a f r e q u e n c y r e s p o n s e t h a t is t h e inverse of t h e c h a n n e l f r e q u e n c y r e s p o n s e . T h i s m e t h o d of c h a n n e l inversion is called zero forcing, since t h e ISI is forced to zero. In l i n e a r zero-

73 Link Level Design 1 » » ^ ^

MtmmimmimmmmmmmMm**

forcing equalization, t h e noise is also p a s s e d t h r o u g h t h e i n v e r s e c h a n n e l filter, a n d t h e noise is amplified over frequencies w h e r e t h e c h a n n e l h a s l o w gain. T h i s noise e n h a n c e m e n t c a n significantly d e g r a d e t h e r e c e i v e d SNR of s y s t e m s w i t h zero-forcing equalization. A b e t t e r l i n e a r equalization t e c h n i q u e u s e s a n equalizing filter t h a t m i n ­ imize t h e average m e a n s q u a r e e r r o r b e t w e e n t h e equalizer o u t p u t a n d t h e t r a n s m i t t e d bit s t r e a m . T h i s t y p e of l i n e a r equalizer is called a minimum mean square error equalizer. Both t y p e s of l i n e a r equalizers c a n b e i m p l e m e n t e d in relatively simple h a r d w a r e . A l t h o u g h l i n e a r equalizers w o r k well o n s o m e c h a n n e l s , t h e i r p e r f o r m a n c e c a n b e quite poor o n c h a n n e l s w i t h a long delay s p r e a d or w i t h large variations in t h e c h a n n e l f r e q u e n c y r e s p o n s e . For t h e s e c h a n n e l s t h e n o n l i n e a r decisionfeedback equalizer (DFE) t e n d s to w o r k m u c h b e t t e r t h a n t h e l i n e a r t e c h n i q u e s . A DFE d e t e r m i n e s t h e ISI from p r e v i o u s l y d e t e c t e d s y m b o l s a n d subtracts it from t h e i n c o m i n g symbols. T h e DFE does n o t suffer from noise e n h a n c e m e n t b e c a u s e it e s t i m a t e s t h e c h a n n e l r a t h e r t h a n i n v e r t i n g it. O n m o s t ISI c h a n n e l s t h e DFE h a s a m u c h l o w e r BER t h a n a l i n e a r equalizer w i t h a slightly h i g h e r complexity. O t h e r forms of equalization i n c l u d e m a x i m u m - l i k e l i h o o d s e q u e n c e esti­ m a t i o n a n d TUrbo equalization, b o t h of w h i c h u s u a l l y o u t p e r f o r m t h e DFE, b u t also h a v e significantly h i g h e r complexity. All equalizer t e c h n i q u e s r e q u i r e a n a c c u r a t e c h a n n e l estimate, w h i c h is u s u a l l y o b t a i n e d b y s e n d i n g a training s e q u e n c e over t h e c h a n n e l . T h e equalizer m u s t also track variations in t h e c h a n n e l b y periodic r e t r a i n i n g a n d b y adjusting t h e filter coefficients d u r i n g data t r a n s m i s s i o n b a s e d o n t h e equalizer o u t p u t s . For this r e a s o n equalizers do n o t w o r k well o n c h a n n e l s t h a t c h a n g e so rapidly t h a t a n a c c u r a t e c h a n n e l e s t i m a t e c a n n o t b e m a i n t a i n e d w i t h o u t significant t r a i n i n g o v e r h e a d . Multicarrier m o d u l a t i o n is a n o t h e r t e c h n i q u e to mitigate ISI. In multicarrier m o d u l a t i o n t h e t r a n s m i s s i o n b a n d w i d t h is divided into n a r r o w s u b c h a n ­ nels, a n d t h e i n f o r m a t i o n bits are divided into a n e q u a l n u m b e r of parallel s t r e a m s . Each s t r e a m is u s e d to m o d u l a t e o n e of t h e s u b c h a n n e l s , w h i c h are t r a n s m i t t e d in parallel. Ideally t h e s u b c h a n n e l b a n d w i d t h s are less t h a n t h e co­ h e r e n c e b a n d w i d t h of t h e c h a n n e l , so t h a t t h e fading o n e a c h s u b c h a n n e l is flat, n o t frequency-selective, t h e r e b y e l i m i n a t i n g ISI. T h e s i m p l e s t m e t h o d for im­ p l e m e n t i n g m u l t i c a r r i e r m o d u l a t i o n is o r t h o g o n a l ( n o n o v e r l a p p i n g ) s u b c h a n ­ nels, b u t t h e spectral efficiency of m u l t i c a r r i e r m o d u l a t i o n c a n b e i n c r e a s e d b y o v e r l a p p i n g t h e s u b c h a n n e l s . T h i s is called orthogonal frequency division multi­ plexing (OFDM). A big advantage of O F D M is t h a t it c a n b e efficiently i m p l e m e n t e d us­ ing t h e fast Fourier t r a n s f o r m at b o t h t h e t r a n s m i t t e r a n d t h e receiver. Since

329

330

Wireless Networks

t h e entire b a n d w i d t h of a n OFDM signal e x p e r i e n c e s frequency-selective fad­ ing, s o m e of t h e OFDM s u b c h a n n e l s will h a v e w e a k SNRs. T h e p e r f o r m a n c e over t h e w e a k s u b c h a n n e l s can b e i m p r o v e d b y coding across s u b c h a n n e l s , frequency equalization, or adaptive loading ( t r a n s m i t t i n g at a h i g h e r data rate o n t h e s u b c h a n n e l s w i t h high SNRs). T h e advantage of m u l t i c a r r i e r m o d u ­ lation over equalization is t h a t f r e q u e n c y equalization r e q u i r e s less training t h a n t i m e equalization. However, flat-fading, f r e q u e n c y offset, a n d t i m i n g mis­ m a t c h i m p a i r t h e orthogonality of t h e m u l t i c a r r i e r s u b c h a n n e l s , resulting in self-interference t h a t c a n significantly degrade p e r f o r m a n c e . Spread s p e c t r u m is a t e c h n i q u e t h a t s p r e a d s t h e t r a n s m i t signal over a w i d e signal b a n d w i d t h in o r d e r to r e d u c e t h e effects of flat-fading, ISI, a n d nar­ r o w b a n d interference. In s p r e a d s p e c t r u m t h e information signal is m o d u l a t e d b y a w i d e b a n d pseudo-noise (PN) signal, resulting in a m u c h larger t r a n s m i t signal b a n d w i d t h t h a n in t h e original signal. Spread s p e c t r u m first a c h i e v e d w i d e s p r e a d u s e in military applications b e c a u s e of its ability to a h i d e a sig­ nal b e l o w t h e noise b y s p r e a d i n g o u t its p o w e r over a w i d e b a n d w i d t h , its resistance to n a r r o w b a n d j a m m i n g , a n d its i n h e r e n t ability to r e d u c e multip a t h flat-fading a n d ISI. However, t h e s e a d v a n t a g e s c o m e w i t h a significant complexity increase, especially in t h e receiver. T h e r e are two c o m m o n forms of s p r e a d s p e c t r u m : direct s e q u e n c e , in w h i c h t h e data s e q u e n c e is multiplied b y t h e PN s e q u e n c e , a n d f r e q u e n c y hopping, in w h i c h t h e n a r r o w b a n d signal is "hopped" over different carrier frequencies b a s e d o n t h e PN s e q u e n c e . Both t e c h n i q u e s result in a t r a n s m i t signal b a n d w i d t h t h a t is m u c h larger t h a n t h e original signal b a n d w i d t h , h e n c e t h e n a m e spread spectrum. Spread s p e c t r u m d e m o d u l a t i o n occurs in two stages: first t h e received signal is d e s p r e a d b y r e m o v i n g t h e PN s e q u e n c e m o d u l a t i o n , a n d t h e n t h e original information signal is d e m o d u l a t e d to get t h e information bits. In direct s e q u e n c e , d e s p r e a d i n g is a c c o m p l i s h e d b y m u l t i p l y i n g t h e received signal with a n exact copy of t h e PN s e q u e n c e , perfectly s y n c h r o n i z e d in t i m e . T h e s y n c h r o n i z a t i o n process entails a great deal of receiver c o m p l e x i t y a n d c a n b e d e g r a d e d b y interference, fading, a n d noise. F r e q u e n c y - h o p p e d signals are d e s p r e a d b y s y n c h r o n i z i n g t h e carrier f r e q u e n c y at t h e r e c e i v e r to t h e h o p p i n g p a t t e r n of t h e PN s e q u e n c e . After d e s p r e a d i n g t h e data signal is d e m o d u l a t e d at b a s e b a n d in t h e con­ v e n t i o n a l way. In t h e d e s p r e a d i n g process, n a r r o w b a n d i n t e r f e r e n c e a n d de­ l a y e d m u l t i p a t h signal c o m p o n e n t s are m o d u l a t e d b y t h e PN s e q u e n c e , t h e r e b y spreading t h e i r signal p o w e r over t h e wide b a n d w i d t h of t h e PN s e q u e n c e . T h e n a r r o w b a n d filter in t h e b a s e b a n d d e m o d u l a t o r t h e n r e m o v e s m o s t of this power, w h i c h yields t h e n a r r o w b a n d i n t e r f e r e n c e a n d m u l t i p a t h rejection prop-

7.4

Channel Access

331

erties of s p r e a d s p e c t r u m . A RAKE r e c e i v e r c a n also b e u s e d to c o h e r e n t l y c o m b i n e all m u l t i p a t h c o m p o n e n t s , t h e r e b y providing i m p r o v e d p e r f o r m a n c e t h r o u g h receiver diversity. A n t e n n a design c a n also r e d u c e t h e effects of flat-fading a n d ISI. T h e m o s t c o m m o n wireless a n t e n n a is a n o m n i d i r e c t i o n a l a n t e n n a , w h e r e t h e p o w e r gain in all a n g u l a r directions is t h e s a m e . Directional a n t e n n a s a t t e n u a t e signals in all b u t a n a r r o w a n g u l a r range, a n d in this r a n g e signals are greatly magnified. Multipath signal c o m p o n e n t s typically c o m e from a large r a n g e of a n g u l a r directions. T h u s , b y u s i n g directional a n t e n n a s at t h e t r a n s m i t t e r a n d / o r t h e receiver, t h e p o w e r in m o s t of t h e m u l t i p a t h c o m p o n e n t s c a n b e greatly r e d u c e d , t h e r e b y e l i m i n a t i n g m o s t flat-fading a n d ISI. ISI r e d u c t i o n directly t r a n s l a t e s into i n c r e a s e d data rates. For example, data r a t e s exceeding 600 Mbps h a v e b e e n o b t a i n e d e x p e r i m e n t a l l y in a n i n d o o r e n v i r o n m e n t w i t h directional a n t e n n a s at b o t h t h e t r a n s m i t t e r a n d t h e receiver. Directional a n t e n n a s c a n also greatly r e d u c e i n t e r f e r e n c e from o t h e r u s e r s in a cellular s y s t e m if t h e s e u s e r s are l o c a t e d outside t h e a n t e n n a ' s a n g u l a r r a n g e of h i g h p o w e r gain. However, directional a n t e n n a s m u s t b e carefully positioned to p o i n t toward t h e u s e r of interest, a n d this direction c h a n g e s as u s e r s m o v e a r o u n d . T h i s is a k e y m o t i v a t i o n b e h i n d t h e d e v e l o p m e n t of steerable a n t e n n a s , also called smart or adaptive antennas. T h e s e a n t e n n a s c h a n g e t h e s h a p e a n d direction of t h e i r t r a n s m i s s i o n b e a m s to a c c o m m o d a t e c h a n g e s in m o b i l e position. A steerable t r a n s m i t t i n g a n t e n n a w o r k s b y controlling t h e p h a s e s of t h e signals at e a c h of its e l e m e n t s , w h i c h c h a n g e s t h e a n g u l a r locations of t h e a n t e n n a b e a m s (angles w i t h large gain) a n d nulls (angles w i t h small gain). Using feedback control, a n a n t e n n a b e a m c a n b e s t e e r e d to follow t h e m o v e m e n t of a mobile, greatly r e d u c i n g b o t h flat-fading a n d i n t e r f e r e n c e from o t h e r users. Smart a n t e n n a s c a n also provide diversity gain.

7.4

CHANNEL ACCESS D u e to t h e scarcity of wireless s p e c t r u m , efficient t e c h n i q u e s to s h a r e b a n d ­ w i d t h a m o n g m a n y h e t e r o g e n e o u s u s e r s are n e e d e d . Applications r e q u i r i n g c o n t i n u o u s t r a n s m i s s i o n (e.g., voice a n d video) g e n e r a l l y allocate d e d i c a t e d c h a n n e l s for t h e d u r a t i o n of t h e call. Sharing b a n d w i d t h t h r o u g h d e d i c a t e d c h a n n e l allocation is called multiple access. In c o n t r a s t to voice or video t r a n s ­ mission, t r a n s m i s s i o n of data t e n d s to o c c u r in b u r s t s . B a n d w i d t h s h a r i n g for u s e r s w i t h b u r s t y t r a n s m i s s i o n s g e n e r a l l y u s e s o m e form of r a n d o m c h a n n e l

Wireless Networks

332

allocation t h a t does n o t g u a r a n t e e c h a n n e l access. B a n d w i d t h s h a r i n g u s i n g r a n d o m c h a n n e l allocation is called random access. W h e t h e r to u s e m u l t i p l e access or r a n d o m access, a n d w h i c h t e c h n i q u e to u s e w i t h i n e a c h access type, will d e p e n d o n t h e traffic characteristics of t h e system, t h e state of c u r r e n t access technology, a n d compatibility w i t h o t h e r systems. This choice is further c o m p l i c a t e d w h e n frequencies are r e u s e d to increase spectral efficiency, as in cellular s y s t e m designs. We n o w describe t h e different m u l t i p l e access a n d r a n d o m access t e c h n i q u e s along w i t h t h e i r c o r r e s p o n d i n g trade-offs.

7.4.1

Multiple Access Multiple access t e c h n i q u e s assign d e d i c a t e d c h a n n e l s to m u l t i p l e u s e r s t h r o u g h b a n d w i d t h division. M e t h o d s to divide t h e s p e c t r u m i n c l u d e frequency-division (FDMA), time-division (TDMA), code-division (CDMA), a n d c o m b i n a t i o n s of t h e s e m e t h o d s . In FDMA t h e total s y s t e m b a n d w i d t h is divided into orthog­ onal c h a n n e l s t h a t are n o n o v e r l a p p i n g in f r e q u e n c y a n d are allocated to t h e different users. I n TDMA t i m e is divided into n o n o v e r l a p p i n g t i m e slots t h a t are allocated to different users. In CDMA t i m e a n d b a n d w i d t h are u s e d si­ m u l t a n e o u s l y b y different users, m o d u l a t e d b y o r t h o g o n a l or semi-orthogonal spreading codes. With orthogonal s p r e a d i n g codes t h e receiver c a n s e p a r a t e out t h e signal of i n t e r e s t from t h e o t h e r CDMA u s e r s w i t h n o residual interfer­ e n c e b e t w e e n users. However, o n l y a finite n u m b e r of o r t h o g o n a l s p r e a d i n g codes exist for a n y given signal b a n d w i d t h . With semi-orthogonal s p r e a d i n g codes t h e receiver c a n n o t c o m p l e t e l y sep­ arate out signals from different users, so t h a t after receiver processing t h e r e is s o m e residual i n t e r f e r e n c e b e t w e e n users. However, t h e r e is n o h a r d limit o n h o w m a n y semi-orthogonal codes exist w i t h i n a given signal b a n d w i d t h . This property, k n o w n as soft capacity, h a s a d v a n t a g e s a n d disadvantages in t h e over­ all s y s t e m design, w h i c h will b e o u t l i n e d in m o r e detail below. D i r e c t - s e q u e n c e s p r e a d s p e c t r u m is often u s e d to g e n e r a t e t h e semi-orthogonal CDMA signals. As discussed in section 7.3.4, d i r e c t - s e q u e n c e s p r e a d s p e c t r u m h a s i n h e r e n t benefits of m u l t i p a t h mitigation a n d n a r r o w b a n d i n t e r f e r e n c e rejection t h a t are n o t i n h e r e n t to e i t h e r FDMA or TDMA, at a cost of s o m e w h a t i n c r e a s e d complexity in t h e t r a n s m i t t e r a n d t h e receiver. FDMA is t h e least c o m p l e x of t h e s e m u l t i p l e access t e c h n i q u e s . T D M A is s o m e w h a t m o r e complex, since it r e q u i r e s t i m i n g s y n c h r o n i z a t i o n a m o n g all users. In addition, t h e orthogonality of t h e u s e r s in T D M A is significantly d e g r a d e d b y ISI, since a signal t r a n s m i t t e d in o n e t i m e slot will interfere w i t h s u b s e q u e n t t i m e slots d u e to t h e m u l t i p a t h delay spread. Semi-orthogonal

7.4

Channel Access

CDMA is t h e m o s t c o m p l e x of t h e m u l t i p l e access s c h e m e s d u e to t h e i n h e r e n t complexity of s p r e a d s p e c t r u m in g e n e r a l a n d t h e additional complexity of s e p a r a t i n g o u t different semi-orthogonal CDMA u s e r s . Semi-orthogonal CDMA also r e q u i r e s s t r i n g e n t p o w e r control to p r e v e n t t h e near-far p r o b l e m . T h e near-far p r o b l e m arises from t h e non-orthogonality of t h e s p r e a d i n g codes, so t h a t e v e r y u s e r c a u s e s i n t e r f e r e n c e to all o t h e r users. D u e to t h e p o w e r falloff w i t h distance d e s c r i b e d in section 7.2.1, t h e received signal p o w e r of a mobile u n i t located close to a r e c e i v e r (or b a s e station) is typically m u c h larger t h a n t h e r e c e i v e d signal p o w e r of a m o b i l e u n i t farther away. T h u s , t h e i n t e r f e r e n c e p o w e r of this "near" m o b i l e u n i t c a n b e large in c o m p a r i s o n w i t h t h e r e c e i v e d signal p o w e r of t h e "far" m o b i l e u n i t . As a result t h e mobile u n i t s located far from t h e r e c e i v e r typically e x p e r i e n c e high i n t e r f e r e n c e from o t h e r u s e r s a n d c o r r e s p o n d i n g l y p o o r p e r f o r m a n c e . Power control mitigates t h e near-far p r o b l e m b y equalizing t h e received p o w e r ( a n d t h e c o r r e s p o n d i n g i n t e r f e r e n c e p o w e r ) of all m o b i l e u n i t s regardless of t h e i r distance from t h e receiver. However, this equalized p o w e r is difficult to m a i n t a i n in a flat-fading e n v i r o n m e n t a n d is o n e of t h e m a j o r design c h a l l e n g e s of semi-orthogonal CDMA. A n o t h e r interesting trade-off in t h e s e m u l t i p l e access m e t h o d s is h a r d ver­ s u s soft s y s t e m capacity. T D M A a n d FDMA place a h a r d limit o n t h e n u m b e r of u s e r s s h a r i n g a given b a n d w i d t h , since t h e c h a n n e l is divided into orthogonal t i m e or f r e q u e n c y slots, e a c h of w h i c h c a n o n l y s u p p o r t o n e user. O r t h o g o n a l CDMA also h a s this h a r d limit. By contrast, s e m i - o r t h o g o n a l CDMA h a s t h e advantage of soft capacity: t h e r e is n o absolute limit o n t h e n u m b e r of users. However, since t h e semi-orthogonal codes interfere w i t h o n e another, t h e per­ f o r m a n c e of all u s e r s d e g r a d e s as t h e n u m b e r of u s e r s in t h e s y s t e m increases; if t h e n u m b e r of u s e r s is too large, t h e n n o u s e r will h a v e acceptable perfor­ m a n c e . I n t e r f e r e n c e from o t h e r CDMA u s e r s c a n b e r e d u c e d u s i n g a r a n g e of sophisticated t e c h n i q u e s i n c l u d i n g s m a r t a n t e n n a s , i n t e r f e r e n c e cancellation, a n d m u l t i u s e r detection. T h e s e t e c h n i q u e s significantly i n c r e a s e t h e complex­ ity of t h e s y s t e m a n d c a n s o m e t i m e s exhibit poor p e r f o r m a n c e in practice. T h e c o m p e t i n g m u l t i p l e access m e t h o d s in t h e U.S. for cellular a n d PCS services are mixed F D M A / T D M A w i t h t h r e e t i m e slots p e r f r e q u e n c y c h a n n e l (IS-54), semi-orthogonal CDMA (IS-95), a n d a c o m b i n a t i o n of T D M A a n d slow frequency-hopping (GSM). T h e d e b a t e a m o n g cellular a n d p e r s o n a l c o m m u n i ­ cation s t a n d a r d s c o m m i t t e e s a n d e q u i p m e n t p r o v i d e r s over w h i c h a p p r o a c h to u s e h a s led to n u m e r o u s analytical s t u d i e s c l a i m i n g s u p e r i o r i t y of o n e t e c h n i q u e over t h e o t h e r u n d e r different c h a n n e l a s s u m p t i o n s a n d o p e r a t i n g scenarios. A definitive analysis of t h e p e r f o r m a n c e of t h e s e different multi­ ple access t e c h n i q u e s in all real o p e r a t i n g e n v i r o n m e n t s is n o t possible. T h u s

333

Wireless Networks

334

t h e r e is n o c o m m o n a g r e e m e n t as to w h i c h access t e c h n i q u e is s u p e r i o r for a n y given system, especially for s y s t e m s w i t h f r e q u e n c y r e u s e or significant channel impairments.

7.4.2

Random Access In m o s t wireless data n e t w o r k s only a small, unpredictable, a n d c h a n g i n g subset of all t h e u s e r s in t h e n e t w o r k h a s data to s e n d at a n y given t i m e . For t h e s e s y s t e m s it is inefficient to assign e a c h u s e r a d e d i c a t e d c h a n n e l . W h e n dedicated c h a n n e l access is n o t provided a n d access to t h e c h a n n e l is n o t g u a r a n t e e d , a r a n d o m access protocol is r e q u i r e d . R a n d o m access protocols are b a s e d o n packetized data t r a n s m i s s i o n s a n d typically fall in t w o categories: ALOHA t e c h n i q u e s a n d r e s e r v a t i o n or d e m a n d - a s s i g n m e n t protocols. In p u r e ALOHA a t r a n s m i t t e r will s e n d data p a c k e t s over t h e c h a n n e l w h e n e v e r data is available. T h i s leads to a large n u m b e r of data collisions at t h e receiver. A collision occurs w h e n t w o or m o r e p a c k e t s are r e c e i v e d simul­ t a n e o u s l y at t h e receiver a n d therefore n o n e of t h e p a c k e t s c a n b e d e c o d e d correctly. Packets t h a t collide m u s t b e r e s e n t at a later t i m e . T h e t h r o u g h p u t of a n ALOHA c h a n n e l , defined as t h e rate at w h i c h p a c k e t s are correctly received, is low d u e to t h e h i g h probability of collisions. In fact, u n d e r s t a n d a r d m o d e l i n g a s s u m p t i o n s , t h e m a x i m u m t h r o u g h p u t in a n ALOHA c h a n n e l is 18% of t h e c h a n n e l data rate (the rate t h a t a single u s e r could achieve w e r e it n o t s h a r i n g the channel). Fewer collisions o c c u r if t i m e is divided into s e p a r a t e slots a n d p a c k e t trans­ missions are confined to t h e s e p r e d e t e r m i n e d slots, since t h e r e is n o partial overlap of packets. This modification of p u r e ALOHA is called slotted ALOHA. Although slotted ALOHA r o u g h l y d o u b l e s t h e m a x i m u m t h r o u g h p u t relative to p u r e ALOHA, this t h r o u g h p u t is still insufficient to s u p p o r t b a n d w i d t h - s h a r i n g a m o n g m o r e t h a n a handful of high-speed users. T h e n u m b e r of collisions in ALOHA c a n b e r e d u c e d b y t h e c a p t u r e effect, w h i c h is similar to t h e n e a r far effect described above for s p r e a d s p e c t r u m s y s t e m s . Specifically, d u e to t h e p o w e r falloff w i t h distance of t h e t r a n s m i t t e d signal, a p a c k e t t r a n s m i t ­ t e d from a mobile t h a t is far from t h e receiver typically c a u s e s j u s t a small a m o u n t of i n t e r f e r e n c e to a p a c k e t t r a n s m i t t e d from a closer location, so t h a t despite a collision b e t w e e n t h e s e packets, t h e latter p a c k e t c a n b e "captured," t h a t is, received w i t h o u t error. Spread s p e c t r u m c a n also b e c o m b i n e d w i t h ALOHA to r e d u c e collisions, since w h e n p a c k e t s m o d u l a t e d w i t h PN spread­ ing s e q u e n c e s collide, t h e y c a n b e s e p a r a t e d o u t b y a s p r e a d s p e c t r u m receiver. As s p r e a d s p e c t r u m r a n d o m access receivers m u s t b e able to d e m o d u l a t e t h e

7.4

Channel Access

PN s p r e a d i n g s e q u e n c e s for a large n u m b e r of users, t h e s e receivers typically h a v e v e r y h i g h complexity. ALOHA w i t h carrier s e n s i n g is often u s e d in w i r e d n e t w o r k s (e.g., t h e Eth­ e r n e t ) to avoid p a c k e t collisions. I n carrier s e n s i n g a t r a n s m i t t e r s e n s e s t h e c h a n n e l before t r a n s m i s s i o n to d e t e r m i n e if t h e c h a n n e l is b u s y . If so, t h e n t h e t r a n s m i s s i o n is d e l a y e d until t h e c h a n n e l is free. Carrier s e n s i n g is often com­ b i n e d w i t h collision detection, w h e r e t h e c h a n n e l is m o n i t o r e d d u r i n g p a c k e t t r a n s m i s s i o n . If a n o t h e r u s e r accesses t h e c h a n n e l d u r i n g this t r a n s m i s s i o n , t h e r e b y causing a collision, t h e n b y d e t e c t i n g this collision t h e t r a n s m i t t e r c a n r e s e n d t h e p a c k e t w i t h o u t waiting for a negative a c k n o w l e d g m e n t or t i m e o u t . Carrier s e n s i n g a n d collision d e t e c t i o n r e q u i r e a t r a n s m i t t i n g u s e r to detect p a c k e t t r a n s m i s s i o n s from o t h e r u s e r s to its i n t e n d e d receiver. T h i s d e t e c t i o n is often impossible in a wireless e n v i r o n m e n t b e c a u s e of p a t h loss. S u p p o s e two t r a n s m i t t i n g u s e r s are o n opposite sides of a receiver, s u c h t h a t t h e distance b e t w e e n e a c h u s e r a n d t h e r e c e i v e r is half t h e distance b e t w e e n t h e t w o users. In this case, a l t h o u g h t h e receiver m a y correctly d e c o d e a p a c k e t t r a n s m i t t e d from e i t h e r u s e r in t h e a b s e n c e of a collision, e a c h u s e r c a n n o t d e t e c t a t r a n s m i t t e d p a c k e t from t h e o t h e r u s e r b e c a u s e t h e u s e r s are so far apart. Collision d e t e c t i o n is also i m p a i r e d b y s h a d o w fading, since t h e u s e r s m a y h a v e a n object obstructing t h e signal p a t h b e t w e e n t h e m . T h e difficulty of d e t e c t i n g collisions in a wireless e n v i r o n m e n t is called t h e hidden terminal problem since, d u e to p a t h loss a n d s h a d o w fading, t h e signals from o t h e r u s e r s in t h e s y s t e m may be hidden. Since carrier s e n s i n g a n d collision d e t e c t i o n are n o t v e r y effective in wire­ less c h a n n e l s , t h e c u r r e n t g e n e r a t i o n of wireless LANs u s e collision avoidance. In collision avoidance t h e r e c e i v e r notifies all n e a r b y t r a n s m i t t e r s w h e n it is receiving a p a c k e t b y b r o a d c a s t i n g a "busy tone." T r a n s m i t t e r s w i t h p a c k e t s to s e n d wait until s o m e r a n d o m t i m e after t h e b u s y t o n e t e r m i n a t e s to b e g i n s e n d i n g t h e i r packets. T h e r a n d o m backoff p r e v e n t s all u s e r s w i t h p a c k e t s to s e n d from s i m u l t a n e o u s l y t r a n s m i t t i n g as s o o n as t h e b u s y t o n e t e r m i n a t e s . Collision avoidance significantly i n c r e a s e s t h e t h r o u g h p u t of A L O H A a n d is c u r r e n t l y part of several wireless LAN standards. However, t h e efficacy of col­ lision avoidance c a n b e d e g r a d e d b y t h e effects of p a t h loss, shadowing, a n d m u l t i p a t h fading o n t h e b u s y t o n e . Reservation protocols assign c h a n n e l s to u s e r s o n d e m a n d t h r o u g h a dedi­ cated r e s e r v a t i o n c h a n n e l . T h e c h a n n e l a s s i g n m e n t is m a d e b y a c e n t r a l b a s e station or b y a c o m m o n algorithm r u n n i n g in e a c h t e r m i n a l . In s u c h a sys­ t e m t h e total c h a n n e l b a n d w i d t h is divided into data c h a n n e l s a n d r e s e r v a t i o n c h a n n e l s , w h e r e typically t h e r e s e r v a t i o n c h a n n e l s o c c u p y o n l y a small frac­ tion of t h e total b a n d w i d t h . A u s e r w i t h data to t r a n s m i t s e n d s a s h o r t p a c k e t

335

336

Wireless Networks

containing a c h a n n e l r e q u e s t over t h e r e s e r v a t i o n c h a n n e l . If this p a c k e t is correctly received a n d a data c h a n n e l is available, a data c h a n n e l is r e s e r v e d for t h e user's data t r a n s m i s s i o n , a n d this c h a n n e l a s s i g n m e n t is c o n v e y e d b a c k to t h e user. D e m a n d - b a s e d a s s i g n m e n t c a n b e a v e r y efficient m e a n s of ran­ d o m access since c h a n n e l s are only assigned w h e n t h e y are n e e d e d , as long as t h e r e q u i r e d o v e r h e a d traffic to assign c h a n n e l s is a small p e r c e n t a g e of t h e m e s s a g e to b e t r a n s m i t t e d . Several p r o b l e m s c a n arise. T h e s e t u p delay a n d o v e r h e a d associated w i t h t h e c h a n n e l r e s e r v a t i o n r e q u e s t a n d a s s i g n m e n t p r o c e d u r e m a y b e excessive for n e t w o r k s w i t h a considerable a m o u n t of short messaging. Second, for h e a v ­ ily loaded s y s t e m s t h e r e s e r v a t i o n c h a n n e l m a y b e c o m e c o n g e s t e d with reser­ vation requests, shifting t h e m u l t i p l e access p r o b l e m from t h e data c h a n n e l to t h e r e s e r v a t i o n c h a n n e l . For n e t w o r k s serving a w i d e variety of data u s e r s w h e r e a considerable a m o u n t of t h e n e t w o r k traffic consists of small messages, reservation-based r a n d o m access m a y n o t b e t h e b e s t choice of r a n d o m access protocol. Packet-Reservation Multiple Access (PRMA) is a relatively n e w r a n d o m access t e c h n i q u e t h a t c o m b i n e s t h e a d v a n t a g e s of r e s e r v a t i o n protocols a n d ALOHA. In PRMA t i m e is slotted a n d t h e t i m e slots are organized into frames w i t h Ν t i m e slots p e r frame. Active t e r m i n a l s w i t h packets to t r a n s m i t c o n t e n d for free t i m e slots in e a c h frame. O n c e a p a c k e t is successfully t r a n s m i t t e d in a t i m e slot, t h e t i m e slot is r e s e r v e d for t h a t u s e r in e a c h s u b s e q u e n t frame, as long as t h e u s e r h a s p a c k e t s to t r a n s m i t . W h e n t h e u s e r stops t r a n s m i t ­ ting packets in t h e r e s e r v e d slot t h e r e s e r v a t i o n is forfeited, a n d t h e u s e r m u s t again c o n t e n d for free t i m e slots in s u b s e q u e n t p a c k e t t r a n s m i s s i o n s . PRMA is well suited to m u l t i m e d i a traffic w i t h a m i x of voice (or c o n t i n u o u s s t r e a m ) traffic a n d data. O n c e t h e c o n t i n u o u s s t r e a m traffic h a s b e e n suc­ cessfully t r a n s m i t t e d , it m a i n t a i n s a d e d i c a t e d c h a n n e l for t h e d u r a t i o n of its transmission, while data traffic o n l y u s e s t h e c h a n n e l as long as it is n e e d e d . PRMA r e q u i r e s little central control a n d n o r e s e r v a t i o n o v e r h e a d , so it is su­ perior to reservation-based protocols w h e n t h e r e is a m i x of voice a n d data traffic. All r a n d o m access protocols r e q u i r e p a c k e t a c k n o w l e d g m e n t s to g u a r a n t e e successful r e c e p t i o n of packets. If a p a c k e t is n o t a c k n o w l e d g e d w i t h i n s o m e p r e d e t e r m i n e d t i m e window, t h e n t h e p a c k e t is r e t r a n s m i t t e d . Since p a c k e t a c k n o w l e d g m e n t s are also s e n t over a wireless c h a n n e l , t h e y are frequently delayed, lost, or c o r r u p t e d d u e to c h a n n e l i m p a i r m e n t s . T h i s c a n result in u n ­ n e c e s s a r y p a c k e t r e t r a n s m i s s i o n , w h i c h is inefficient, a n d p a c k e t duplication, w h i c h m u s t b e h a n d l e d b y t h e n e t w o r k l a y e r protocol. Wireless n e t w o r k s can i m p r o v e t h e likelihood of t i m e l y a n d u n c o r r u p t e d p a c k e t a c k n o w l e d g m e n t s b y using m o r e powerful link l a y e r t e c h n i q u e s to s e n d t h e s e a c k n o w l e d g m e n t s .

7.5

Network Design

337

However, this r e q u i r e s t h a t t h e link l a y e r differentiate a m o n g different t y p e s of data t r a n s m i s s i o n s , w h i c h adds to t h e link l a y e r complexity.

7.4.3

Spectral Etiquette C h a n n e l access allows m u l t i p l e u s e r s in t h e s a m e s y s t e m to s h a r e a given b a n d w i d t h allocation. However, in s o m e cases m u l t i p l e s y s t e m s will s h a r e t h e s a m e b a n d w i d t h w i t h o u t a n y coordination, a n d this r e q u i r e s interoperability of t h e i r different access t e c h n i q u e s a n d c o m m u n i c a t i o n designs. T h i s coexistence c a n b e a c c o m p l i s h e d t h r o u g h e t i q u e t t e rules, w h i c h are a m i n i m u m set of r u l e s t h a t allows m u l t i p l e s y s t e m s to s h a r e t h e available b a n d w i d t h fairly. T h e s e t e c h n i q u e s offer a n a l t e r n a t i v e to standardization m e t h o d s t h a t r e q u i r e a g r e e m e n t o n c h a n n e l access a n d s y s t e m design before s y s t e m s c a n b e built a n d deployed. WINForum, a n association of c o m p a n i e s d e v e l o p i n g wireless products, h a s defined a set of e t i q u e t t e r u l e s for t h e u n l i c e n s e d 2-GHz PCS b a n d s t h a t h a s b e e n a d o p t e d b y t h e FCC. T h e s a m e set of r u l e s is b e i n g c o n s i d e r e d for t h e 60-GHz s p e c t r u m allocation. T h e k e y e l e m e n t s of t h e s e e t i q u e t t e r u l e s are (1) listen before t r a n s m i t t i n g , to e n s u r e t h a t t h e t r a n s m i t t e r is t h e only u s e r of t h e s p e c t r u m while m i n i m i z i n g t h e possibility of interfering w i t h o t h e r s p e c t r u m users; (2) limit t r a n s m i s s i o n t i m e in o r d e r to m a k e it possible for o t h e r u s e r s to m a k e u s e of t h e s p e c t r u m in a fair m a n n e r ; a n d (3) limit t r a n s m i t t e r power, so as n o t to interfere w i t h u s e r s w h o are in n e a r b y s p e c t r u m s or r e u s i n g t h e s a m e f r e q u e n c y s p e c t r u m s o m e distance away.

7.5

NETWORK DESIGN We first describe different n e t w o r k a r c h i t e c t u r e s a n d t h e i r relative trade-offs a n d associated distributed a n d centralized n e t w o r k control strategies. Next w e c o n s i d e r protocols for mobility m a n a g e m e n t , i n c l u d i n g t h e location of mobile users, u s e r a u t h e n t i c a t i o n , a n d call routing. N e t w o r k reliability a n d QpS g u a r a n t e e s for wireless n e t w o r k s are also discussed. Some pitfalls of w i r e d a n d wireless n e t w o r k interoperability u s i n g u n i v e r s a l n e t w o r k i n g protocols like T C P / I P a n d ATM n e t w o r k s will b e outlined, followed b y a brief discussion of security.

7.5.1

Architecture T h e t h r e e m a i n t y p e s of n e t w o r k a r c h i t e c t u r e s are a star (central h u b ) topology, a n ad h o c or peer-to-peer structure, a n d a h i e r a r c h i c a l or t r e e s t r u c t u r e . T h e s e are illustrated in Figure 7.7.

Wireless Networks

338

Hierarchical 77

Star

Peer-to-peer

Architectures for wireless networks.

FIGURE

Hierarchical n e t w o r k a r c h i t e c t u r e s are usually o n l y u s e d for wireless net­ w o r k s s p a n n i n g a r a n g e of coverage regions, as w a s s h o w n in Figure 7.1. In this figure t h e lowest level of t h e h i e r a r c h y consists of i n d o o r s y s t e m s w i t h small coverage areas; t h e next level of t h e h i e r a r c h y consists of cellular s y s t e m s cov­ ering a city, followed b y s y s t e m s w i t h regional a n d t h e n global coverage. Since t h e coverage regions define a n a t u r a l h i e r a r c h y of t h e overall n e t w o r k , a hier­ archical n e t w o r k a r c h i t e c t u r e along w i t h protocols for r o u t i n g a n d identifying u s e r locations are well suited to this t y p e of s y s t e m . In a peer-to-peer a r c h i t e c t u r e t h e n o d e s self-configure into a n i n t e g r a t e d n e t w o r k u s i n g distributed control, a n d t h e c o n n e c t i o n b e t w e e n a n y t w o n o d e s in t h e n e t w o r k consists of o n e or m o r e peer-to-peer c o m m u n i c a t i o n links. I n a star architecture, c o m m u n i c a t i o n flows from n e t w o r k n o d e s to a c e n t r a l h u b over o n e set of c h a n n e l s , a n d from t h e h u b to t h e n o d e s over a s e p a r a t e set of c h a n n e l s . T h e choice of a peer-to-peer or a star n e t w o r k a r c h i t e c t u r e d e p e n d s o n m a n y factors. Peer-to-peer a r c h i t e c t u r e s r e q u i r e n o existing infrastructure, are easily reconfigurable, a n d h a v e n o single p o i n t s of failure. Peer-to-peer a r c h i t e c t u r e s c a n u s e m u l t i p l e h o p s for t h e end-to-end con­ nection, w h i c h h a s t h e advantage of e x t e n d i n g t h e n e t w o r k r a n g e a n d t h e disadvantage t h a t if o n e of t h e h o p s fails, t h e e n t i r e end-to-end c o n n e c t i o n is lost. T h i s disadvantage is mitigated b y t h e fact t h a t e a c h n o d e m a y h a v e c o n n e c t i o n s to m a n y o t h e r nodes, so t h e r e m a y b e m u l t i p l e w a y s to form a n end-to-end c o n n e c t i o n w i t h a n y o t h e r user. T h e s e a d v a n t a g e s m a k e peer-top e e r a r c h i t e c t u r e s t h e a r c h i t e c t u r e of choice in military s y s t e m s . Since star a r c h i t e c t u r e s h a v e only o n e h o p b e t w e e n a n e t w o r k n o d e a n d t h e central hub, t h e y t e n d to b e m o r e predictable a n d reliable; however, if t h a t con­ n e c t i o n is weak, t h e n t h e r e is n o alternative c o n n e c t i o n . A big advantagetof star architectures is t h a t t h e y c a n u s e centralized control functions at t h e h u b for c h a n n e l estimation, access, routing, a n d r e s o u r c e allocation. T h i s centralized control usually results in a m o r e efficient a n d reliable n e t w o r k , a n d for this rea-

7.5

Network Design

son m a n y c o m m e r c i a l wireless n e t w o r k s u s e t h e star a r c h i t e c t u r e . C o m m o n e x a m p l e s of wireless s y s t e m s w i t h peer-to-peer a r c h i t e c t u r e s i n c l u d e p a c k e t radio n e t w o r k s a n d s o m e wireless LANs. T h e star a r c h i t e c t u r e is e m p l o y e d in cellular a n d paging s y s t e m s .

7.5.2

Mobility Management Mobility m a n a g e m e n t consists of two r e l a t e d functions: location m a n a g e m e n t a n d call routing. Location m a n a g e m e n t is t h e process of identifying t h e p h y s ­ ical location of t h e u s e r so t h a t calls directed to t h a t u s e r c a n b e r o u t e d to t h a t location. Location m a n a g e m e n t is also responsible for verifying t h e a u t h e n ­ ticity of u s e r s accessing t h e n e t w o r k . Routing consists of setting u p a r o u t e t h r o u g h t h e n e t w o r k over w h i c h data directed to a p a r t i c u l a r u s e r is sent, a n d d y n a m i c a l l y reconfiguring t h e r o u t e as t h e u s e r location c h a n g e s . In cellular s y s t e m s location m a n a g e m e n t a n d r o u t i n g are c o o r d i n a t e d b y t h e b a s e sta­ tions or t h e central mobile t e l e p h o n e switching office (MTSO), w h e r e a s o n t h e I n t e r n e t t h e s e functions are h a n d l e d b y t h e Mobile I n t e r n e t w o r k i n g Routing Protocol (Mobile IP). T h e location m a n a g e m e n t a n d r o u t i n g protocols in Mobile IP a n d in cel­ lular s y s t e m s are s o m e w h a t different, b u t t h e y b o t h u s e local a n d r e m o t e databases for u s e r tracking, a u t h e n t i c a t i o n , a n d call routing. In cellular sys­ t e m s location m a n a g e m e n t a n d call r o u t i n g are h a n d l e d b y t h e MTSO in e a c h city. A n MTSO is c o n n e c t e d to all b a s e stations in its city via h i g h - s p e e d com­ m u n i c a t i o n links. T h e MTSO in e a c h city m a i n t a i n s a h o m e location database for local u s e r s a n d a visitor location database for visiting users. Calls directed to a particular mobile u n i t are r o u t e d t h r o u g h t h e public-switched t e l e p h o n e n e t w o r k to t h e MTSO in t h a t mobile's h o m e city. W h e n a m o b i l e u n i t in t h e h o m e city t u r n s t h e h a n d s e t on, t h a t signal is r e l a y e d b y t h e local b a s e station to t h e MTSO. T h e MTSO a u t h e n t i c a t e s t h e ID n u m b e r of t h e mobile a n d t h e n registers t h a t u s e r in its h o m e location database. After registration, a n y calls a d d r e s s e d to t h a t u s e r are s e n t to h i m b y t h e MTSO via o n e of its b a s e stations. If a mobile is r o a m i n g in a different city, t u r n i n g t h e h a n d s e t o n registers t h e mobile w i t h t h e MTSO in t h e visiting city. Specifically, t h e mobile's signal is picked u p b y a local b a s e station in t h e visiting city, w h i c h relays t h e signal to t h e visiting city's MTSO. T h e visiting city's MTSO t h e n s e n d s a m e s s a g e to t h e MTSO in t h e mobile's h o m e city r e q u e s t i n g u s e r a u t h e n t i c a t i o n a n d call forwarding for t h a t user. T h e MTSO in t h e mobile's h o m e city a u t h e n t i c a t e s t h e mobile's ID n u m b e r , adds t h e location of t h e visiting city's MTSO to its h o m e location database e n t r y for t h e visiting mobile, a n d s e n d s a c o n f i r m a t i o n m e s s a g e to t h e visiting city's MTSO. T h e visiting m o b i l e is t h e n registered in

Wireless Networks

340

t h e visitor location database of t h e visiting city's MTSO. After this process is complete, w h e n a call for a visiting mobile arrives at t h a t mobile's h o m e city, t h e h o m e city MTSO sets u p a circuit-switched c o n n e c t i o n w i t h t h e visiting city's MTSO along w h i c h t h e call is routed. T h i s m e t h o d of call r o u t i n g is s o m e w h a t inefficient, since a call m u s t travel from its origin to t h e h o m e city's MTSO a n d t h e n b e r e r o u t e d to t h e visiting city. T h e MTSO also c o o r d i n a t e s handoffs b e t w e e n b a s e stations b y d e t e c t i n g w h e n a mobile signal is b e c o m i n g w e a k at its c u r r e n t b a s e station a n d finding t h e n e i g h b o r i n g b a s e station w i t h t h e b e s t c o n n e c t i o n to t h a t mobile. T h i s handoff process will b e discussed in m o r e detail in section 7.6.1. Location m a n a g e m e n t a n d r o u t i n g o n t h e I n t e r n e t is h a n d l e d b y t h e Mobile IP protocol, described in section 4.3. T h e protocol does n o t s u p p o r t real-time handoff of a mobile b e t w e e n different n e t w o r k s : it is d e s i g n e d m a i n l y for stationary u s e r s w h o occasionally m o v e t h e i r c o m p u t e r from o n e n e t w o r k to another.

7.5.3

Network Reliability A n end-to-end c o n n e c t i o n in a wireless n e t w o r k is c o m p o s e d of o n e or m o r e wireless a n d w i r e d links, w i t h at least o n e wireless link. T h e s e different links h a v e widely varying data rates, BERs, a n d delays. Moreover, u s e r mobility causes o n e or m o r e of t h e s e links to c h a n g e over t i m e . T h e s e characteristics m a k e it difficult to i n s u r e reliability of t h e end-to-end n e t w o r k c o n n e c t i o n . Protocols like TCP t h a t are d e s i g n e d for w i r e d n e t w o r k s do n o t w o r k well in wireless n e t w o r k s . T h e s e protocols a s s u m e t h a t p a c k e t losses are c a u s e d b y congestion, a n d t h e y react b y throttling t h e source. O n wireless n e t w o r k s m o s t packet losses are d u e to p o o r link quality a n d i n t e r m i t t e n t connectivity. Using t h e congestion control m e c h a n i s m s of TCP to correct for t h e s e p r o b l e m s c a n cause large a n d variable end-to-end delays a n d low n e t w o r k t h r o u g h p u t . In addition, wireless c h a n n e l s h a v e low data rates a n d high BERs, a n d t h e r a n d o m characteristics of t h e s e c h a n n e l s m a k e it difficult to g u a r a n t e e or e v e n predict end-to-end data rates, delay statistics, or packet loss probabilities. P e r f o r m a n c e m e t r i c s s u c h as data rates, end-to-end latency, a n d likelihood of packet loss are usually referred to as a c o n n e c t i o n ' s quality of service (QpS). T h e QoS r e q u i r e m e n t s for a c o n n e c t i o n are b a s e d o n t h e k i n d of data b e i n g t r a n s p o r t e d over t h a t c o n n e c t i o n . For example, voice h a s a h i g h t o l e r a n c e to p a c k e t loss b u t a low t o l e r a n c e for delay, w h e r e a s data h a s t h e opposite requirements. T h e i n h e r e n t i m p a i r m e n t s a n d r a n d o m variations of t h e wireless c h a n n e l m a k e it difficult to provide a n y t h i n g o t h e r t h a n best-effort service in wire-

7.5

Network Design

less n e t w o r k s . T h i s difficulty is t h e m a i n c h a l l e n g e in s u p p o r t i n g high-speed real-time applications like video t e l e c o n f e r e n c i n g over t h e s e n e t w o r k s . O n e possible a p p r o a c h to c o m p e n s a t e for t h e lack of QpS g u a r a n t e e s is to a d a p t at t h e application l a y e r to t h e variable QpS offered b y t h e n e t w o r k , as described in m o r e detail in section 7.5.6.

7.5.4

Internetworking In o r d e r to c o n n e c t w i r e d a n d wireless n e t w o r k s together, t h e y m u s t s h a r e a c o m m o n n e t w o r k i n g protocol s u c h as T C P / I P a n d ATM. As w e h a v e s e e n , TCP h a s p r o b l e m s o p e r a t i n g over wireless links, m a i n l y d u e to its u s e of congestion control in r e s p o n s e to p a c k e t delays. However, wireless links c a n e x p e r i e n c e large a n d variable delays, sporadic error bursts, a n d i n t e r m i t t e n t c o n n e c t i v i t y d u e to handoffs. Large a n d vari­ able link delays c a u s e large oscillations in t h e TCP s e n d i n g rate, resulting in large a n d variable end-to-end delays. Error b u r s t s c a n result in u n n e c e s s a r y r e t r a n s m i s s i o n s b y TCP, since t h e s e e r r o r s are u s u a l l y c o r r e c t e d at t h e link layer, a n d also c a u s e significant t h r o u g h p u t degradation, since flow is r e d u c e d in r e s p o n s e to e v e r y error burst. T h e effect of i n t e r m i t t e n t c o n n e c t i v i t y o n TCP is similar to t h a t of e r r o r b u r s t s , resulting in u n n e c e s s a r y r e t r a n s m i s s i o n s a n d t h r o u g h p u t r e d u c t i o n . Various modifications to TCP h a v e b e e n p r o p o s e d to address this issue, b u t n o n e h a s e m e r g e d as a clear solution. ATM provides QpS g u a r a n t e e s , w h i c h are r e q u i r e d for s o m e applications. But it is n o t clear t h a t t h e QpS g u a r a n t e e s of ATM c a n b e a c h i e v e d in a wireless network.

7.5.5

Security Wireless c o m m u n i c a t i o n s y s t e m s are i n h e r e n t l y less private t h a n w i r e l i n e sys­ t e m s b e c a u s e t h e wireless l i n k c a n b e i n t e r c e p t e d w i t h o u t a n y physical tap, a n d this i n t e r c e p t i o n c a n n o t b e d e t e c t e d b y t h e t r a n s m i t t e r or t h e receiver. T h i s lack of link security also m a k e s wireless n e t w o r k s m o r e subject to usage fraud a n d activity m o n i t o r i n g t h a n t h e i r w i r e l i n e c o u n t e r p a r t s . O p p o r t u n i t i e s for fraudulent attacks will i n c r e a s e as services like w i r e l e s s b a n k i n g a n d c o m ­ m e r c e b e c o m e available. T h u s , s e c u r i t y t e c h n o l o g y is a n i m p o r t a n t challenge. Security issues c a n b e b r o k e n d o w n into t h r e e categories: n e t w o r k security, radio link security, a n d h a r d w a r e security. N e t w o r k security i n c l u d e s c o u n t e r m e a s u r e s to fraudulent access a n d m o n ­ itoring of n e t w o r k activity, a n d end-to-end e n c r y p t i o n . Radio link s e c u r i t y

Wireless Networks

342

entails p r e v e n t i n g i n t e r c e p t i o n of t h e radio signal, e n s u r i n g privacy of u s e r lo­ cation information and, for military applications, anti-jam a n d l o w probability of i n t e r c e p t i o n a n d d e t e c t i o n capabilities. H a r d w a r e security s h o u l d p r e v e n t fraudulent u s e of t h e mobile t e r m i n a l in t h e e v e n t of theft or loss, a n d u s e r databases s h o u l d also b e s e c u r e against u n a u t h o r i z e d access.

7.5.6

A New Paradigm for Wireless Network Design N e t w o r k design u s i n g t h e l a y e r e d OSI a r c h i t e c t u r e h a s w o r k e d well for w i r e d n e t w o r k s , especially as t h e c o m m u n i c a t i o n links evolved to provide gigabitper-second data rates a n d BERs of 1 0 ~ . Wireless c h a n n e l s typically h a v e m u c h l o w e r data rates (tens or h u n d r e d s of Kbps for typical c h a n n e l s w i t h high u s e r mobility), h i g h e r BERs ( 1 0 ~ to 10~ ), a n d exhibit sporadic error b u r s t s a n d i n t e r m i t t e n t connectivity. T h e s e p e r f o r m a n c e characteristics also vary over time, as do t h e n e t w o r k topology a n d u s e r traffic. C o n s e q u e n t l y , good end-to-end wireless n e t w o r k p e r f o r m a n c e , will n o t b e possible w i t h o u t a t r u l y optimized, integrated, a n d adaptive n e t w o r k design. 12

2

6

In o r d e r to optimize n e t w o r k p e r f o r m a n c e , e a c h level in t h e protocol stack should a d a p t to wireless link variations in a n a p p r o p r i a t e m a n n e r , taking into a c c o u n t t h e adaptive strategies at t h e o t h e r layers. Therefore, a n i n t e g r a t e d adaptive design across all levels of t h e protocol stack is n e e d e d to b e s t exploit i n t e r d e p e n d e n c i e s a m o n g protocol layers. This i n t e g r a t e d a p p r o a c h to adaptive protocol design entails two r e l a t e d questions: w h a t p e r f o r m a n c e m e t r i c s should b e m e a s u r e d at e a c h l a y e r of t h e protocol stack, a n d w h a t adaptive strategies s h o u l d b e d e v e l o p e d for e a c h l a y e r of t h e protocol stack to b e s t r e s p o n d to variations in t h e s e local p e r f o r m a n c e metrics. N e t w o r k design b a s e d o n t h e OSI m o d e l considers t h e s e two q u e s t i o n s in isolation for e a c h layer. But t h e b e s t a n s w e r to b o t h q u e s t i o n s at a p a r t i c u l a r layer d e p e n d s o n h o w a n d to w h a t t h e n e t w o r k a d a p t s at o t h e r l a y e r s of t h e protocol stack. In o t h e r words, t h e b e s t overall n e t w o r k design r e q u i r e s t h a t t h e s e q u e s t i o n s b e a d d r e s s e d at all l a y e r s of t h e protocol stack simultaneously. T h e i n t e g r a t e d adaptive protocol design s h o u l d still b e b a s e d o n a hierar­ chical approach, since n e t w o r k variations take place o n different t i m e scales. Variations in link SNR (i.e., BER a n d connectivity) c a n b e v e r y fast, o n t h e or­ d e r of m i c r o s e c o n d s for vehicle-based users. N e t w o r k topology c h a n g e s m o r e slowly, o n t h e o r d e r of seconds, while variations of u s e r traffic m a y c h a n g e over t e n s to h u n d r e d s of s e c o n d s (although this m a y c h a n g e as n e t w o r k s sup­ port m o r e applications w i t h short messaging). T h e different t i m e scales of t h e

7.6

Wireless Networks Today

n e t w o r k variations suggest a hierarchical a p p r o a c h for protocol design, since t h e rate at w h i c h a protocol c a n a d a p t to overall n e t w o r k c h a n g e s is, to a large extent, d e t e r m i n e d b y its location in t h e protocol stack. For example, s u p p o s e t h e link c o n n e c t i v i t y (link SNR) in t h e wireless l i n k of a n end-to-end n e t w o r k c o n n e c t i o n is weak. By t h e t i m e this c o n n e c t i v i t y information is r e l a y e d to a h i g h e r level of t h e protocol stack (e.g., t h e n e t w o r k l a y e r for r e r o u t i n g or t h e application l a y e r for r e d u c e d - r a t e c o m p r e s s i o n ) , t h e link SNR m a y change. Therefore, it m a k e s s e n s e for e a c h protocol l a y e r to a d a p t to variations t h a t are local to t h a t layer. If this local a d a p t a t i o n is insufficient to c o m p e n s a t e for t h e local p e r f o r m a n c e degradation, t h e n t h e p e r f o r m a n c e m e t r i c s at t h e n e x t l a y e r of t h e protocol stack will d e g r a d e as a result. Adaptation at this next l a y e r m a y t h e n correct or at least mitigate t h e p r o b l e m t h a t could n o t b e fixed t h r o u g h local adaptation. Consider t h e w e a k link situation. Link c o n n e c t i v i t y c a n b e m e a s u r e d quite a c c u r a t e l y a n d quickly at t h e link level. T h e link protocol c a n therefore r e s p o n d to w e a k c o n n e c t i v i t y b y i n c r e a s i n g its t r a n s m i t p o w e r or its error-correction coding. T h i s will correct for variations in c o n n e c t i v i t y due, for example, to m u l t i p a t h flat-fading. However, if t h e w e a k link is c a u s e d b y s o m e t h i n g difficult to correct at t h e link l a y e r (e.g., t h e mobile u n i t is inside a t u n n e l ) , t h e n it is b e t t e r for a h i g h e r l a y e r of t h e n e t w o r k protocol stack to r e s p o n d (for instance, b y delaying p a c k e t t r a n s m i s s i o n s u n t i l t h e m o b i l e l e a v e s t h e t u n n e l ) . However, real-time applications m a y n o t b e able to tolerate a n i n c r e a s e in p a c k e t delay, in w h i c h case t h e application c a n a d a p t b y r e d u c i n g its rate of p a c k e t t r a n s m i s s i o n . This m a y entail u s i n g a l o w e r rate c o m p r e s s i o n s c h e m e or s e n d i n g o n l y priority data (e.g., t h e voice c o m p o n e n t of a video s t r e a m or t h e low resolution c o m p o n e n t s of a n i m a g e ) . It is this i n t e g r a t e d a p p r o a c h to adaptive n e t w o r k i n g — h o w e a c h l a y e r of t h e protocol stack s h o u l d r e s p o n d to local variations given a d a p t a t i o n at h i g h e r layers—that s h o u l d b e c o n s i d e r e d as a n e w p a r a d i g m in wireless n e t w o r k design.

7.6

WIRELESS NETWORKS TODAY In this section w e give a brief overview of wireless n e t w o r k s in o p e r a t i o n today, including cellular s y s t e m s , cordless p h o n e s , wireless LANs, w i d e a r e a wireless data s y s t e m s , paging s y s t e m s , a n d satellite s y s t e m s . T h e s e s y s t e m s are m a i n l y differentiated b y t h e i r application (voice or data), s u p p o r t for u s e r mobility, a n d coverage areas.

343

Wireless Networks

344

7.6.1

Cellular Telephone Systems Cellular t e l e p h o n e systems, also referred to as Personal C o m m u n i c a t i o n Sys­ t e m s (PCS), are e x t r e m e l y p o p u l a r a n d lucrative worldwide: t h e s e s y s t e m s h a v e s p a r k e d m u c h of t h e o p t i m i s m a b o u t t h e future of wireless n e t w o r k s . Cellular t e l e p h o n e s y s t e m s are d e s i g n e d to provide two-way voice c o m m u n i c a t i o n at vehicle s p e e d s w i t h regional or n a t i o n a l coverage. Cellular s y s t e m s w e r e ini­ tially designed for mobile t e r m i n a l s inside vehicles w i t h a n t e n n a s m o u n t e d o n t h e vehicle roof. Tbday t h e s e s y s t e m s h a v e evolved to s u p p o r t lightweight h a n d ­ h e l d mobile t e r m i n a l s o p e r a t i n g inside a n d outside buildings at b o t h p e d e s t r i a n a n d vehicle s p e e d s . T h e basic feature of t h e cellular s y s t e m is f r e q u e n c y reuse, w h i c h exploits p a t h loss to r e u s e t h e s a m e f r e q u e n c y s p e c t r u m at spatially s e p a r a t e d locations. Specifically, t h e coverage a r e a of a cellular s y s t e m is divided into n o n o v e r l a p ­ ping cells w h e r e s o m e set of c h a n n e l s is assigned to e a c h cell. T h i s s a m e c h a n n e l set is u s e d in a n o t h e r cell s o m e distance away, as s h o w n in Figure 7.8, w h e r e t h e s h a d e d cells u s e t h e s a m e c h a n n e l set. O p e r a t i o n w i t h i n a cell is controlled b y a centralized b a s e station, as de­ scribed in m o r e detail below. T h e i n t e r f e r e n c e c a u s e d b y u s e r s in different cells operating o n t h e s a m e c h a n n e l set is called intercell interference. T h e spatial s e p a r a t i o n of cells t h a t r e u s e t h e s a m e c h a n n e l set, t h e reuse distance, s h o u l d b e as small as possible to m a x i m i z e t h e spectral efficiency o b t a i n e d b y fre­ q u e n c y reuse. However, as t h e r e u s e distance decreases, intercell i n t e r f e r e n c e increases, d u e to t h e s m a l l e r propagation distance b e t w e e n interfering cells. Since intercell i n t e r f e r e n c e m u s t r e m a i n b e l o w a given t h r e s h o l d for acceptable s y s t e m p e r f o r m a n c e , r e u s e distance c a n n o t b e r e d u c e d b e l o w s o m e m i n i m u m value. In practice it is quite difficult to d e t e r m i n e this m i n i m u m value since b o t h t h e t r a n s m i t t i n g a n d interfering signals e x p e r i e n c e r a n d o m p o w e r varia­ tions d u e to p a t h loss, shadowing, a n d m u l t i p a t h . In order to d e t e r m i n e t h e b e s t r e u s e distance a n d b a s e station p l a c e m e n t , a n a c c u r a t e characterization of signal p r o p a g a t i o n w i t h i n t h e cells is n e e d e d . This characterization is usually o b t a i n e d u s i n g detailed analytical m o d e l s , sophisticated c o m p u t e r - a i d e d modeling, or e m p i r i c a l m e a s u r e m e n t s . Initial cellular s y s t e m designs w e r e m a i n l y d r i v e n b y t h e h i g h cost of b a s e stations, about $1 million each. For this r e a s o n early cellular s y s t e m s u s e d a relatively small n u m b e r of cells to cover a n e n t i r e city or region. T h e cell b a s e stations w e r e placed o n tall buildings or m o u n t a i n s a n d t r a n s m i t t e d at veryhigh p o w e r w i t h cell coverage areas of several s q u a r e miles. T h e s e large cells

7.6

Wireless Networks Today

7.8

Cellular systems with frequency reuse.

FIGURE

are called macrocells. Signals p r o p a g a t e d o u t from b a s e stations u n i f o r m l y in all directions, so a mobile m o v i n g in a circle a r o u n d t h e b a s e station w o u l d h a v e a p p r o x i m a t e l y c o n s t a n t r e c e i v e d power. This circular c o n t o u r of c o n s t a n t p o w e r yields a hexagonal cell s h a p e for t h e s y s t e m , since a h e x a g o n is t h e clos­ est s h a p e to a circle t h a t c a n cover a given area w i t h m u l t i p l e n o n o v e r l a p p i n g cells. Cellular t e l e p h o n e s y s t e m s are n o w evolving to s m a l l e r cells w i t h b a s e stations close to street level or inside b u i l d i n g s t r a n s m i t t i n g at m u c h l o w e r power. T h e s e s m a l l e r cells are called microcells or picocells, d e p e n d i n g o n t h e i r size. This evolution is d r i v e n b y two factors: t h e n e e d for h i g h e r capacity in areas w i t h high u s e r density, a n d t h e r e d u c e d size a n d cost of a b a s e station. A cell of a n y size c a n s u p p o r t r o u g h l y t h e s a m e n u m b e r of u s e r s if t h e s y s t e m is scaled accordingly. T h u s , for a given coverage area, a s y s t e m w i t h m a n y

345

346

Wireless Networks

microcells h a s a h i g h e r n u m b e r of u s e r s p e r u n i t a r e a t h a n a s y s t e m w i t h j u s t a few macrocells. Small cells also h a v e b e t t e r p r o p a g a t i o n conditions since t h e lower b a s e stations h a v e r e d u c e d s h a d o w i n g a n d m u l t i p a t h . In addition, less p o w e r is r e q u i r e d at t h e mobile t e r m i n a l s in microcellular systems, since t h e t e r m i n a l s are closer to t h e b a s e stations. However, t h e evolution to s m a l l e r cells h a s c o m p l i c a t e d n e t w o r k design. Mobiles traverse a small cell m o r e quickly t h a n a large cell, a n d therefore handoffs m u s t b e pro­ cessed m o r e quickly. Location m a n a g e m e n t also b e c o m e s m o r e complicated, since t h e r e are m o r e cells w i t h i n a given city w h e r e a m o b i l e m a y b e located. It is also h a r d e r to d e v e l o p g e n e r a l propagation m o d e l s for small cells, since signal propagation in t h e s e cells is highly d e p e n d e n t o n b a s e station p l a c e m e n t a n d t h e g e o m e t r y of t h e s u r r o u n d i n g reflectors. In particular, a hexagonal cell s h a p e is n o t a good a p p r o x i m a t i o n to signal propagation in microcells. Microcellular s y s t e m s are often d e s i g n e d u s i n g s q u a r e or t r i a n g u l a r cell shapes, b u t t h e s e s h a p e s h a v e a large m a r g i n of e r r o r in t h e i r a p p r o x i m a t i o n to microcell signal propagation. All b a s e stations in a city are c o n n e c t e d via a high-speed link to a mobile t e l e p h o n e switching office (MTSO). T h e MTSO acts as a c e n t r a l controller for t h e network, allocating c h a n n e l s w i t h i n e a c h cell, coordinating handoffs b e t w e e n cells w h e n a mobile traverses a cell b o u n d a r y , a n d r o u t i n g calls to a n d from mobile u s e r s in conjunction w i t h t h e public-switched t e l e p h o n e n e t w o r k (PSTN). A n e w u s e r located in a given cell r e q u e s t s a c h a n n e l b y s e n d i n g a call r e q u e s t to t h e cell's b a s e station over a s e p a r a t e control c h a n n e l . T h e r e q u e s t is r e l a y e d to t h e MTSO, w h i c h a c c e p t s t h e call r e q u e s t if a c h a n n e l is available in t h a t cell. If n o c h a n n e l is available, t h e call r e q u e s t is rejected. A call handoff is initiated w h e n t h e b a s e station or t h e m o b i l e in a given cell detects t h a t t h e received signal p o w e r for t h a t call is a p p r o a c h i n g a m i n i m u m threshold. In this case t h e b a s e station informs t h e MTSO t h a t t h e mobile r e q u i r e s a handoff, a n d t h e MTSO t h e n q u e r i e s s u r r o u n d i n g b a s e stations to d e t e r m i n e if o n e of t h e s e stations c a n d e t e c t t h a t mobile's signal. T h e MTSO coordinates a handoff b e t w e e n t h e original b a s e station a n d t h e n e w b a s e station. If n o c h a n n e l is available in t h e cell w i t h t h e n e w b a s e station, t h e handoff fails a n d t h e call is t e r m i n a t e d . False handoffs m a y also b e initiated if a mobile is in a d e e p fade, causing its received signal p o w e r to d r o p b e l o w t h e m i n i m u m t h r e s h o l d e v e n t h o u g h it m a y b e n o w h e r e n e a r a cell b o u n d a r y . Cellular t e l e p h o n e s y s t e m s h a v e m o v e d from analog to digital t e c h n o l ­ ogy. Digital t e c h n o l o g y h a s m a n y advantages. T h e c o m p o n e n t s are cheaper, faster, smaller, a n d r e q u i r e less power. Voice quality is i m p r o v e d d u e to errorcorrection coding. Digital s y s t e m s also h a v e h i g h e r capacity t h a n analog sys­ t e m s since t h e y are n o t l i m i t e d to FDMA m u l t i p l e access, a n d t h e y c a n take

7.6

Wireless Networks Today

WfJWiTO^^

1

—

1

1

a d v a n t a g e of a d v a n c e d c o m p r e s s i o n t e c h n i q u e s a n d voice activity factors. I n addition, e n c r y p t i o n t e c h n i q u e s c a n b e u s e d to s e c u r e digital signals against eavesdropping. All cellular s y s t e m s b e i n g d e p l o y e d today are digital, a n d t h e s e s y s t e m s provide voice mail, paging, a n d e-mail services in addition to voice. D u e to t h e i r l o w e r cost a n d h i g h e r efficiency, service p r o v i d e r s h a v e u s e d aggressive pricing tactics to e n c o u r a g e u s e r m i g r a t i o n from a n a l o g to digital s y s t e m s . Since t h e y are relatively new, digital s y s t e m s do n o t always w o r k as well as t h e old analog ones. Users e x p e r i e n c e p o o r voice quality, f r e q u e n t call dropping, short b a t t e r y life, a n d spotty coverage in certain areas. S y s t e m p e r f o r m a n c e will certainly i m p r o v e as t h e t e c h n o l o g y a n d n e t w o r k s m a t u r e . However, it is u n l i k e l y t h a t cellular p h o n e s will provide t h e s a m e quality as w i r e l i n e service a n y t i m e soon. T h e great p o p u l a r i t y of cellular s y s t e m s i n d i c a t e s t h a t u s e r s are willing to tolerate inferior voice c o m m u n i c a t i o n s in e x c h a n g e for mobility. Digital cellular s y s t e m s c a n u s e a n y of t h e m u l t i p l e access t e c h n i q u e s de­ scribed above to divide u p t h e signal b a n d w i d t h in a given cell. In t h e U.S. t h e s t a n d a r d s activities s u r r o u n d i n g t h e c u r r e n t g e n e r a t i o n of digital cellular sys­ t e m s provoked a raging d e b a t e o n m u l t i p l e access for t h e s e s y s t e m s , resulting in several i n c o m p a t i b l e standards. In particular, t h e r e are t w o s t a n d a r d s in t h e 900-MHz (cellular) f r e q u e n c y b a n d : IS-54, w h i c h u s e s a c o m b i n a t i o n of T D M A a n d FDMA, a n d IS-95, w h i c h u s e s s e m i - o r t h o g o n a l CDMA. T h e s p e c t r u m for digital cellular in t h e 2-GHz (PCS) f r e q u e n c y b a n d w a s a u c t i o n e d off, so ser­ vice providers could u s e a n existing s t a n d a r d or d e v e l o p p r o p r i e t a r y s y s t e m s for t h e i r p u r c h a s e d s p e c t r u m . T h e e n d result h a s b e e n t h r e e different digital cellular s t a n d a r d s for this f r e q u e n c y b a n d : IS-136 ( w h i c h is basically t h e s a m e as IS-54 at a h i g h e r fre­ q u e n c y ) , IS-95, a n d t h e E u r o p e a n digital cellular s t a n d a r d GSM, w h i c h u s e s a c o m b i n a t i o n of TDMA a n d slow frequency-hopping. T h e digital cellular stan­ d a r d in J a p a n in similar to IS-54 a n d IS-136 b u t in a different frequency b a n d , a n d t h e GSM s y s t e m in E u r o p e is at a different frequency t h a n t h e GSM s y s t e m s in t h e U.S. T h i s proliferation of i n c o m p a t i b l e s t a n d a r d s m a k e s it impossible to r o a m b e t w e e n s y s t e m s n a t i o n w i d e or globally w i t h o u t u s i n g m u l t i p l e p h o n e s (and phone numbers). Efficient cellular s y s t e m designs are interference-limited, t h a t is, t h e inter­ ference d o m i n a t e s t h e noise floor since o t h e r w i s e m o r e u s e r s could b e a d d e d to t h e s y s t e m . As a result, a n y t e c h n i q u e to r e d u c e i n t e r f e r e n c e in cellular sys­ t e m s leads directly to a n i n c r e a s e in s y s t e m capacity a n d p e r f o r m a n c e . Some m e t h o d s for i n t e r f e r e n c e r e d u c t i o n in u s e today or p r o p o s e d for future s y s t e m s include cell sectorization, directional a n d s m a r t a n t e n n a s , m u l t i u s e r detection, a n d d y n a m i c c h a n n e l a n d r e s o u r c e allocation.

347

Wireless Networks

348

7.6.2

Cordless Phones Cordless t e l e p h o n e s first a p p e a r e d in t h e late 1970s a n d h a v e b e c o m e v e r y popular. Almost half of t h e p h o n e s in U.S. h o m e s today are cordless. Cordless p h o n e s originally provided a low-cost low-mobility wireless link replacing t h e cord c o n n e c t i n g a t e l e p h o n e b a s e u n i t a n d its h a n d s e t . Initial cordless p h o n e s h a d poor voice quality a n d w e r e quickly discarded b y u s e r s . T h e first cordless s y s t e m s allowed only o n e p h o n e h a n d s e t to c o n n e c t to e a c h b a s e unit, a n d coverage w a s l i m i t e d to a few r o o m s of a h o u s e or office. This is still t h e m a i n p u r p o s e of cordless t e l e p h o n e s in t h e U.S. today, a l t h o u g h t h e y n o w u s e digital technology. In E u r o p e a n d t h e Far East, digital cordless p h o n e s y s t e m s h a v e evolved to provide coverage over m u c h w i d e r areas, b o t h in a n d a w a y from h o m e , a n d are similar in m a n y w a y s to today's cellular t e l e p h o n e s y s t e m s . Digital cordless p h o n e s y s t e m s in t h e U.S. today consist of a wireless h a n d ­ set c o n n e c t e d to a single b a s e unit, w h i c h in t u r n is c o n n e c t e d to t h e t e l e p h o n e n e t w o r k . T h e s e cordless p h o n e s i m p o s e n o a d d e d c o m p l e x i t y o n t h e n e t w o r k , since t h e cordless b a s e u n i t acts like a wireline t e l e p h o n e for n e t w o r k i n g pur­ poses. T h e m o v e m e n t of t h e cordless h a n d s e t is e x t r e m e l y limited: it m u s t r e m a i n w i t h i n r a n g e of its b a s e unit. T h e r e is n o coordination w i t h o t h e r cord­ less p h o n e systems, so a h i g h d e n s i t y of t h e s e s y s t e m s in a small area, s u c h as a n a p a r t m e n t building, can result in significant i n t e r f e r e n c e a m o n g s y s t e m s . For this r e a s o n cordless p h o n e s today h a v e m u l t i p l e voice c h a n n e l s a n d s c a n a m o n g t h e s e c h a n n e l s to find t h e o n e w i t h m i n i m a l i n t e r f e r e n c e . Spread spec­ t r u m cordless p h o n e s h a v e also b e e n i n t r o d u c e d to r e d u c e i n t e r f e r e n c e from o t h e r s y s t e m s a n d to mitigate n a r r o w b a n d i n t e r f e r e n c e . In E u r o p e a n d t h e Far East, t h e s e c o n d g e n e r a t i o n of digital cordless p h o n e s (CT-2, for cordless t e l e p h o n e , s e c o n d g e n e r a t i o n ) h a v e a n e x t e n d e d range of u s e b e y o n d a single r e s i d e n c e or office. Within a h o m e t h e s e s y s t e m s o p e r a t e as c o n v e n t i o n a l cordless p h o n e s . Tb e x t e n d t h e r a n g e b e y o n d t h e h o m e , b a s e stations, also called phone-points or telepoints, are m o u n t e d in places w h e r e p e o p l e congregate, like s h o p p i n g malls, b u s y streets, t r a i n stations, a n d airports. Cordless p h o n e s registered w i t h t h e t e l e p o i n t provider c a n place calls w h e n e v e r t h e y are in r a n g e of a telepoint. Calls c a n n o t b e r e c e i v e d from t h e t e l e p o i n t since t h e n e t w o r k h a s n o r o u t i n g s u p p o r t for m o b i l e users, a l t h o u g h s o m e n e w e r CT-2 h a n d s e t s h a v e built-in p a g e r s to c o m p e n s a t e for this deficiency. T h e s e s y s t e m s also do n o t h a n d off calls so a u s e r m u s t r e m a i n w i t h i n range of t h e t e l e p o i n t for t h e d u r a t i o n of a call. Tblepoint service w a s i n t r o d u c e d twice in t h e U n i t e d K i n g d o m a n d failed b o t h times, b u t t h e s e s y s t e m s grew rapidly in H o n g Kong a n d Singapore t h r o u g h t h e m i d 1990s. T h i s rapid g r o w t h d e t e r i o r a t e d quickly after t h e first

7.6

Wireless Networks Today

few years, as cellular p h o n e o p e r a t o r s cut p r i c e s to c o m p e t e w i t h t e l e p o i n t service. T h e m a i n c o m p l a i n t a b o u t t e l e p o i n t service w a s t h e i n c o m p l e t e radio coverage a n d lack of handoff. Since cellular s y s t e m s avoid t h e s e p r o b l e m s , as long as prices w e r e c o m p e t i t i v e t h e r e w a s little r e a s o n for p e o p l e to u s e t e l e p o i n t services. Most of t h e s e services h a v e n o w d i s a p p e a r e d . A n o t h e r evolution of t h e cordless t e l e p h o n e d e s i g n e d p r i m a r i l y for office buildings is t h e E u r o p e a n DECT s y s t e m . DECT provides local mobility s u p p o r t for u s e r s in a n in-building private b r a n c h exchange (PBX). DECT b a s e u n i t s are m o u n t e d t h r o u g h o u t a building, a n d e a c h b a s e station is a t t a c h e d t h r o u g h a controller to t h e PBX of t h e building. H a n d s e t s c o m m u n i c a t e to t h e n e a r e s t b a s e station in t h e building, a n d calls are h a n d e d off as a u s e r walks b e t w e e n b a s e stations. DECT c a n also r i n g h a n d s e t s from t h e closest b a s e station. T h e DECT s t a n d a r d also s u p p o r t s t e l e p o i n t services, a l t h o u g h this application h a s n o t received m u c h attention, probably d u e to t h e failure of CT-2 services. T h e r e are c u r r e n t l y a r o u n d 7 million DECT u s e r s in Europe, b u t t h e s t a n d a r d h a s n o t y e t s p r e a d to o t h e r c o u n t r i e s . T h e m o s t r e c e n t a d v a n c e in cordless t e l e p h o n e s y s t e m design is t h e Per­ sonal H a n d y p h o n e System (PHS) in J a p a n . T h e PHS s y s t e m is q u i t e similar to a cellular s y s t e m , w i t h w i d e s p r e a d b a s e station d e p l o y m e n t s u p p o r t i n g h a n d off a n d call routing b e t w e e n b a s e stations. With t h e s e capabilities PHS d o e s n o t suffer from t h e m a i n limitations of t h e CT-2 s y s t e m . Initially PHS s y s t e m s enjoyed o n e of t h e fastest growth r a t e s e v e r for a n e w t e c h n o l o g y . In 1997, two y e a r s after its introduction, PHS subscribers p e a k e d at a b o u t 7 million users; this n u m b e r h a s d e c l i n e d slightly s i n c e t h e n d u e m a i n l y to s h a r p price-cutting b y cellular providers. T h e m a i n difference b e t w e e n a PHS s y s t e m a n d a cel­ lular s y s t e m is t h a t PHS c a n n o t s u p p o r t call handoff at vehicle s p e e d s . T h i s deficiency is m a i n l y d u e to t h e d y n a m i c c h a n n e l allocation p r o c e d u r e u s e d in PHS. D y n a m i c c h a n n e l allocation greatly i n c r e a s e s t h e n u m b e r of h a n d ­ sets t h a t c a n b e serviced b y a single b a s e station, t h e r e b y l o w e r i n g t h e s y s t e m cost, b u t it also c o m p l i c a t e s t h e handoff p r o c e d u r e . It is too s o o n to tell if PHS s y s t e m s will go t h e s a m e r o u t e as CT-2. However, it is clear from t h e r e c e n t history of cordless p h o n e s y s t e m s t h a t to e x t e n d t h e r a n g e of t h e s e s y s t e m s b e y o n d t h e h o m e r e q u i r e s e i t h e r t h e s a m e functionality as cellular s y s t e m s or a significantly r e d u c e d cost.

7.63

Wireless LANs Wireless LANs provide h i g h - s p e e d data w i t h i n a small region s u c h as a c a m p u s or small building, as u s e r s m o v e from place to place. Wireless devices t h a t ac­ cess t h e s e LANs are typically stationary or m o v i n g at p e d e s t r i a n s p e e d s . Nearly

349

350

Wireless Networks

all wireless LANs in t h e U n i t e d States u s e o n e of t h e ISM f r e q u e n c y b a n d s . T h e a p p e a l of t h e s e f r e q u e n c y b a n d s , located at 900 MHz, 2.4 GHz, a n d 5.8 GHz, is t h a t a n FCC license is n o t r e q u i r e d to o p e r a t e in t h e s e b a n d s . However, this advantage is a double-edged sword, since m a n y o t h e r s y s t e m s o p e r a t e in t h e s e b a n d s for t h e s a m e reason, causing a great deal of i n t e r f e r e n c e a m o n g sys­ t e m s . T h e FCC mitigates this i n t e r f e r e n c e p r o b l e m b y setting a s t r i n g e n t limit o n t h e p o w e r p e r u n i t b a n d w i d t h for ISM-band s y s t e m s . I b satisfy this require­ m e n t , wireless LANs u s e e i t h e r direct-sequence or f r e q u e n c y - h o p p i n g s p r e a d s p e c t r u m so t h a t t h e i r total p o w e r is s p r e a d over a w i d e b a n d w i d t h . Wireless LANs can h a v e e i t h e r a star architecture, w i t h wireless access p o i n t s or h u b s placed t h r o u g h o u t t h e coverage region, or a peer-to-peer architecture, w h e r e t h e wireless t e r m i n a l s self-configure into a n e t w o r k . D o z e n s of wireless LAN c o m p a n i e s a n d p r o d u c t s a p p e a r e d in t h e early 1990s to m e e t t h e e s t i m a t e d d e m a n d for high-speed wireless data. T h e s e first wireless LANs w e r e b a s e d o n p r o p r i e t a r y a n d i n c o m p a t i b l e protocols, a l t h o u g h m o s t o p e r a t e d in t h e 900-MHz ISM b a n d u s i n g d i r e c t - s e q u e n c e s p r e a d spec­ t r u m w i t h data r a t e s o n t h e o r d e r of 1 to 2 Mbps. Both star a n d peer-to-peer a r c h i t e c t u r e s w e r e used. T h e lack of standardization for t h e s e p r o d u c t s led to high d e v e l o p m e n t costs, low-volume production, a n d small m a r k e t s for e a c h individual product. Of t h e s e original products, only a handful r e m a i n , includ­ ing Proxim's RangeLAN, Lucent's WaveLAN, a n d Windata's FreePort. O n l y o n e of t h e first g e n e r a t i o n wireless LANs, Motorola's Altair, o p e r a t e d outside t h e 900-MHz ISM b a n d . This system, operating in t h e l i c e n s e d 18-GHz b a n d , h a d data rates o n t h e o r d e r of 6 Mbps. However, p e r f o r m a n c e of Altair w a s h a m ­ p e r e d b y t h e high cost of c o m p o n e n t s a n d t h e i n c r e a s e d p a t h loss at 18 GHz. As a result Altair w a s r e c e n t l y d i s c o n t i n u e d . T h e 900-MHz ISM b a n d is n o t available in m o s t p a r t s of t h e world, so t h e n e w g e n e r a t i o n of wireless LANs o p e r a t e in t h e 2.4-GHz ISM b a n d , w h i c h is available worldwide. A wireless LAN s t a n d a r d for this f r e q u e n c y b a n d , t h e IEEE 802.11 standard, w a s r e c e n t l y c o m p l e t e d to avoid s o m e of t h e p r o b l e m s w i t h t h e p r o p r i e t a r y first g e n e r a t i o n s y s t e m s . T h e s t a n d a r d specifies f r e q u e n c y - h o p p e d s p r e a d s p e c t r u m w i t h data rates of 1.6 Mbps a n d a r a n g e of a p p r o x i m a t e l y 500 ft. T h e n e t w o r k a r c h i t e c t u r e c a n b e e i t h e r star or peer-to-peer. M a n y c o m p a ­ nies h a v e d e v e l o p e d p r o d u c t s b a s e d o n t h e 802.11 standard, a n d t h e s e p r o d u c t s are constantly evolving to provide h i g h e r data rates a n d b e t t e r coverage. Cabletron's Freelink is t h e o n l y existing wireless LAN in t h e 5.8-GHz r a n g e : this s y s t e m h a s slightly h i g h e r data rates a n d slightly lower r a n g e t h a n t h e 802.11 s y s t e m s . Because data r a t e s are low a n d coverage is limited, t h e m a r k e t for all wireless LANs h a s r e m a i n e d relatively flat ( a r o u n d $200 million, far b e l o w t h e billion dollar m a r k e t of today's cellular s y s t e m s ) . O p t i m i s m r e m a i n s h i g h t h a t t h e wireless LAN m a r k e t is poised to take off, a l t h o u g h this p r e d i c t i o n h a s b e e n

7.6

Wireless Networks Today

351

m a d e e v e r y y e a r since t h e i n c e p t i o n of wireless LANs y e t t h e m a r k e t h a s so far failed to materialize.

7.6.4

Wide Area Wireless Data Services Wide a r e a wireless data services in t h e U.S. provide l o w r a t e wireless data to high-mobility u s e r s over a v e r y large coverage area. T h e initial two service providers for wireless data w e r e t h e ARDIS n e t w o r k r u n b y Motorola a n d RAM Mobile Data, w h i c h u s e s Ericcson's Mobitex t e c h n o l o g y . ARDIS a n d RAM Mobile Data provide service to m o s t m e t r o p o l i t a n a r e a s w i t h i n t h e U.S. In t h e s e s y s t e m s a large geographical region is serviced b y a few b a s e stations m o u n t e d o n towers, rooftops, or m o u n t a i n s a n d t r a n s m i t t i n g at h i g h power. In ARDIS t h e b a s e stations are c o n n e c t e d to n e t w o r k controllers a t t a c h e d to a b a c k b o n e n e t w o r k , w h e r e a s in RAM Mobile Data t h e b a s e stations are at t h e b o t t o m of a hierarchical n e t w o r k a r c h i t e c t u r e . Both s y s t e m s u s e a form of t h e ALOHA protocol for r a n d o m access w i t h collision r e d u c t i o n t h r o u g h e i t h e r a b u s y t o n e t r a n s m i s s i o n or carrier sensing. Initial data r a t e s for t h e s e s y s t e m s w e r e low, 4.8 Kbps for ARDIS a n d 8 Kbps for RAM, b u t t h e s e rates h a v e since b o t h i n c r e a s e d to 19.2 Kbps. A n o t h e r p r o v i d e r is Metricom, w i t h s y s t e m s o p e r a t i n g in t h e San Francisco Bay Area, Seattle, a n d Washington, D.C. T h e M e t r i c o m a r c h i t e c t u r e is similar to t h a t of microcell s y s t e m s : a large n e t w o r k of small i n e x p e n s i v e b a s e stations w i t h small coverage areas m o u n t e d close to street level. T h e i n c r e a s e d effi­ c i e n c y of microcells allows for h i g h e r data rates in Metricom, 76 Kbps, t h a n in t h e o t h e r wide a r e a wireless data s y s t e m s . M e t r i c o m u s e s f r e q u e n c y - h o p p e d s p r e a d s p e c t r u m in t h e 900-MHz ISM b a n d , w i t h p o w e r control to m i n i m i z e i n t e r f e r e n c e a n d i m p r o v e b a t t e r y life. T h e cellular digital p a c k e t data (CDPD) s y s t e m is a w i d e a r e a wireless data service overlaid o n t h e analog cellular t e l e p h o n e n e t w o r k . CDPD s h a r e s t h e FDMA voice c h a n n e l s of t h e analog s y s t e m s , since m a n y of t h e s e c h a n n e l s are idle d u e to t h e g r o w t h of digital cellular. T h e CDPD service p r o v i d e s p a c k e t data t r a n s m i s s i o n at rates of 19.2 Kbps a n d is available t h r o u g h o u t t h e U.S. T h e s e services h a v e failed to attract as m a n y subscribers as initially pre­ dicted, possibly b e c a u s e of t h e i r l o w data r a t e s a n d h i g h price. T h e services do n o t s e e m essential: voice c o m m u n i c a t i o n o n t h e m o v e s e e m s essential to m a n y , b u t p e o p l e a p p a r e n t l y prefer to wait u n t i l t h e y h a v e access to a p h o n e line or w i r e d n e t w o r k for data exchange. T h i s m a y c h a n g e w i t h t h e prolifera­ tion of laptop a n d p a l m t o p c o m p u t e r s a n d t h e insatiable d e m a n d for c o n s t a n t I n t e r n e t access a n d e-mail exchange.

Wireless Networks

352

7.6.5

Paging Systems Paging s y s t e m s provide v e r y low rate one-way data services to highly mobile u s e r s over a v e r y w i d e coverage area. Paging s y s t e m s c u r r e n t l y serve 56 million c u s t o m e r s in t h e U n i t e d States. However, t h e p o p u l a r i t y of paging s y s t e m s is declining as cellular s y s t e m s b e c o m e c h e a p e r a n d m o r e ubiquitous. In order to r e m a i n competitive, paging c o m p a n i e s h a v e slashed prices, a n d few are c u r r e n t l y profitable. Tb reverse t h e i r declining fortunes, a c o n s o r t i u m of paging service providers h a s r e c e n t l y t e a m e d u p w i t h Microsoft a n d C o m p a q to incorporate paging functionality a n d I n t e r n e t access into p a l m t o p c o m p u t e r s . Paging s y s t e m s b r o a d c a s t a short paging m e s s a g e s i m u l t a n e o u s l y from m a n y tall b a s e stations or satellites t r a n s m i t t i n g at v e r y h i g h p o w e r ( h u n d r e d s of watts to kilowatts). Systems w i t h terrestrial t r a n s m i t t e r s are typically local­ ized to a particular geographic area, s u c h as a city or m e t r o p o l i t a n region, while g e o s y n c h r o n o u s satellite t r a n s m i t t e r s provide n a t i o n a l or i n t e r n a t i o n a l cover­ age. In b o t h t y p e s of s y s t e m s , n o location m a n a g e m e n t or r o u t i n g functions are n e e d e d , since t h e paging m e s s a g e is b r o a d c a s t over t h e e n t i r e coverage area. T h e high complexity a n d p o w e r of t h e paging t r a n s m i t t e r s allows lowcomplexity, low-power, pocket paging receivers w i t h a long usage t i m e from small a n d lightweight b a t t e r i e s . In addition, t h e h i g h t r a n s m i t p o w e r allows pag­ ing signals to easily p e n e t r a t e b u i l d i n g walls. Paging service also costs less t h a n cellular service, b o t h for t h e initial device a n d for t h e m o n t h l y u s a g e charge, a l t h o u g h this price advantage h a s d e c l i n e d considerably in r e c e n t y e a r s . T h e low cost, small a n d lightweight h a n d s e t s , l o n g b a t t e r y life, a n d ability of paging devices to w o r k a l m o s t a n y w h e r e i n d o o r s or outdoors are t h e m a i n r e a s o n s for t h e i r appeal. Some paging services today offer r u d i m e n t a r y (1 bit) answer-back capa­ bilities from t h e h a n d h e l d paging device. But t h e r e q u i r e m e n t for two-way c o m m u n i c a t i o n destroys t h e a s y m m e t r i c a l link advantage so well exploited in paging s y s t e m design. A paging h a n d s e t with answer-back capability r e q u i r e s a m o d u l a t o r a n d t r a n s m i t t e r w i t h sufficient p o w e r to r e a c h t h e distant b a s e station. T h e s e r e q u i r e m e n t s significantly increase t h e size a n d w e i g h t a n d re­ d u c e t h e usage t i m e of t h e h a n d h e l d pager. This is especially t r u e for paging s y s t e m s w i t h satellite b a s e stations, u n l e s s terrestrial relays are used.

7.6.6

Satellite Networks Satellite s y s t e m s provide voice, data, a n d b r o a d c a s t services w i t h w i d e s p r e a d , often global, coverage to high-mobility u s e r s as well as to fixed sites. T h e y h a v e t h e s a m e basic a r c h i t e c t u r e as cellular s y s t e m s , except t h a t t h e b a s e

7.6

Wireless Networks Today

353

• • ^ ^ ^ stations are satellites orbiting t h e e a r t h . Satellites are c h a r a c t e r i z e d b y t h e i r orbit distance from t h e e a r t h : low-earth orbit (LEO) at 500 to 2,000 k m , m e d i u m e a r t h orbit (MEO) at 10,000 k m , a n d g e o s y n c h r o n o u s orbit (GEO) at 35,800 k m . A g e o s y n c h r o n o u s satellite h a s a large coverage a r e a t h a t is stationary over time, since t h e e a r t h a n d satellite orbits are s y n c h r o n o u s . Satellites w i t h l o w e r orbits h a v e s m a l l e r coverage areas, a n d t h e s e coverage a r e a s c h a n g e over t i m e so t h a t satellite handoff is n e e d e d for stationary u s e r s or fixed-point service. Since g e o s y n c h r o n o u s satellites h a v e s u c h large coverage areas, j u s t a handful of satellites are n e e d e d for global coverage. However, g e o s y n c h r o n o u s s y s t e m s h a v e several disadvantages for two-way c o m m u n i c a t i o n . It takes a great deal of p o w e r to r e a c h t h e s e satellites, so h a n d s e t s are typically large a n d b u l k y . T h e large r o u n d - t r i p p r o p a g a t i o n d e l a y is q u i t e noticeable in twow a y voice c o m m u n i c a t i o n . Recall t h a t high-capacity cellular s y s t e m s r e q u i r e small cell sizes. Since g e o s y n c h r o n o u s satellites h a v e v e r y large cells, t h e s e s y s t e m s h a v e small capacity, h i g h cost, a n d low data rates, less t h a n 10 Kbps. T h e m a i n g e o s y n c h r o n o u s s y s t e m s in o p e r a t i o n today are t h e global I n m a r s a t s y s t e m , MSAT in N o r t h A m e r i c a , Mobilesat in Australia, a n d EMS a n d LLM in Europe. T h e t r e n d in c u r r e n t satellite s y s t e m s is to u s e t h e l o w e r LEO orbits so t h a t lightweight h a n d h e l d devices c a n c o m m u n i c a t e w i t h t h e satellites a n d propagation delay does n o t degrade voice quality. T h e b e s t k n o w n of t h e s e n e w LEO s y s t e m s are Globalstar, Iridium, a n d Tfeledesic. Both Globalstar a n d I r i d i u m provide voice a n d data services to globally r o a m i n g m o b i l e u s e r s at data rates u n d e r 10 Kbps. Tbledesic u s e s 288 satellites to provide global coverage to fixed-point u s e r s at data rates u p to 2 Mbps. T h e cell size for e a c h satellite in a LEO s y s t e m is m u c h larger t h a n terrestrial cells, w i t h t h e c o r r e s p o n d i n g d e c r e a s e in capacity associated w i t h large cells. T h e cost to build, l a u n c h , a n d m a i n t a i n t h e s e satellites is m u c h h i g h e r t h a n t h a t of terrestrial b a s e stations, so t h e s e n e w LEO s y s t e m s are u n l i k e l y to b e cost-competitive w i t h terrestrial cellular a n d wireless data services. LEO s y s t e m s c a n c o m p l e m e n t terrestrial s y s t e m s in low-population a r e a s a n d m a y a p p e a l to travelers desiring j u s t o n e h a n d s e t a n d p h o n e n u m b e r for global r o a m i n g .

7.6.7

Other Wireless Systems and Applications M a n y o t h e r c o m m e r c i a l s y s t e m s u s e wireless t e c h n o l o g y . R e m o t e s e n s o r net­ w o r k s t h a t collect data from u n a t t e n d e d s e n s o r s a n d t r a n s m i t t h e data b a c k to a central processing location are u s e d for i n d o o r ( e q u i p m e n t m o n i t o r i n g , climate

Wireless Networks

354

control) a n d outdoor ( e a r t h q u a k e sensing, r e m o t e data collection) applications. Satellite s y s t e m s t h a t provide vehicle t r a c k i n g a n d dispatching (OMNITRACs) are c o m m e r c i a l l y successful. Satellite navigation s y s t e m s (the Global Position­ ing System or GPS) are v e r y widely used. A n e w wireless s y s t e m for Digital Audio Broadcasting (DAB) h a s r e c e n t l y b e e n i n t r o d u c e d in Europe.

JJl

FUTURE SYSTEMS AND STANDARDS We describe s o m e of t h e wireless s y s t e m s a n d s t a n d a r d s t h a t will e m e r g e over t h e next few years. T h e r a n g e of activity indicates t h a t today's s y s t e m s do n o t y e t m e e t t h e e x p e c t e d d e m a n d for wireless a n d t h a t t h e capabilities a n d technologies for future wireless s y s t e m s will c o n t i n u e to evolve. T h e i n t e r e s t a n d activity in wireless n e t w o r k i n g is so i n t e n s e t h a t it is impossible to predict w h a t s y s t e m s will e m e r g e m o r e t h a n a y e a r or t w o into t h e future.

7.7.1

Wireless LANs T h e r e are two major activities in future wireless LAN d e v e l o p m e n t : HIPERLAN l y p e 1 a n d Wireless ATM. HIPERLAN (for high p e r f o r m a n c e radio LAN), b e i n g d e v e l o p e d in Europe, is a family of wireless LAN s t a n d a r d s tailored to different k i n d s of users. T h e four m e m b e r s of t h e HIPERLAN family are classified as Types 1 t h r o u g h 4. HIPERLAN Type 1 is similar in t e r m s of protocol s u p p o r t to t h e IEEE 802.11 wireless LAN standard, while HIPERLAN Types 2 a n d 3 s u p p o r t Wireless ATM. HIPERLAN Type 1 o p e r a t e s in t h e 5-GHz f r e q u e n c y b a n d w i t h data rates of 23 Mbps at a r a n g e of 150 feet. T h e n e t w o r k a r c h i t e c t u r e is p e e r to-peer, a n d t h e c h a n n e l access m e c h a n i s m u s e s a variation of ALOHA w i t h prioritization b a s e d o n t h e lifetime of packets. HIPERLAN Type 1 p r o m i s e s a n order of m a g n i t u d e data rate i m p r o v e m e n t over today's technology, m a k i n g it competitive w i t h 100-Mbps w i r e d E t h e r n e t s . Wireless ATM is a s t a n d a r d to e x t e n d ATM capabilities to wireless local access as well as to wireless b r o a d b a n d services, a significant challenge given t h e difficultly of s u p p o r t i n g ATM's QpS g u a r a n t e e s over a wireless m e d i u m . Work o n t h e s t a n d a r d is ongoing a n d involves, a m o n g others, t h e Wireless ATM w o r k i n g g r o u p of t h e ATM F o r u m a n d t h e Broadband Radio Access N e t w o r k s g r o u p of ETSI. T h e Wireless ATM s t a n d a r d d e v e l o p e d b y this latter g r o u p will b e c o m e t h e HIPERLAN Type 2 standard. D e v e l o p m e n t of HIPERLAN s t a n d a r d s for Types 2 a n d 3 h a s n o t y e t started. T h e wireless ATM s t a n d a r d h a s two

7.7

Future Systems and Standards

c o m p o n e n t s : t h e radio access t e c h n o l o g y a n d e n h a n c e m e n t s to t h e existing ATM protocol for s u p p o r t of mobile t e r m i n a l s . T h e r e are different t e c h n o l o g i e s a n d protocols c u r r e n t l y u n d e r c o n s i d e r a t i o n for b o t h c o m p o n e n t s . T h e success of Wireless ATM d e p e n d s o n t h e d e p l o y m e n t of end-to-end ATM. T h e FCC r e c e n t l y set aside 300 M H z of u n l i c e n s e d s p e c t r u m a r o u n d 5 GHz called t h e National I n f o r m a t i o n Infrastructure (NH) f r e q u e n c y b a n d . This allocation frees u p a large u n l i c e n s e d s p e c t r u m for wireless LAN applications w i t h data r a t e s u p to several t e n s of m e g a b i t s p e r second. T h e N i l b a n d is divided into t h r e e 100-MHz blocks w i t h different p o w e r restrictions ( a n d c o r r e s p o n d i n g coverage areas) in e a c h block. T h e m i d d l e b l o c k is d e s i g n a t e d for c a m p u s - a r e a wireless LANs a n d is c o m p a t i b l e w i t h t h e 5-GHz HIPERLAN spectral allocation in E u r o p e so t h a t p r o d u c t s d e v e l o p e d in b o t h places c a n b e u s e d i n t e r c h a n g e a b l y . T h e l o w e r block is restricted to i n d o o r u s e a n d t h e h i g h e r b l o c k is d e s i g n a t e d for c o m m u n i t y n e t w o r k s . I n all t h r e e b l o c k s t h e FCC h a s i m p o s e d m i n i m a l t e c h n i c a l restrictions to provide m a x i m u m flexibility in s y s t e m design. T h e r e is m u c h activity to d e v e l o p s t a n d a r d s a n d p r o d u c t s for t h e s e f r e q u e n c y b a n d s , b u t n o c o n s e n s u s h a s y e t e m e r g e d o n t h e b e s t choice of s y s t e m design.

7.7.2

Ad Hoc Wireless Networks A n ad h o c wireless n e t w o r k is a collection of wireless m o b i l e h o s t s forming a t e m p o r a r y n e t w o r k w i t h o u t t h e aid of a n y established infrastructure or centralized control. Ad h o c wireless n e t w o r k s w e r e traditionally of i n t e r e s t to t h e military. T h r o u g h o u t t h e 1970s a n d 1980s DARPA funded m u c h w o r k in t h e design of ad h o c p a c k e t radio n e t w o r k s ; however, t h e p e r f o r m a n c e of t h e s e n e t w o r k s w a s s o m e w h a t disappointing. Ad h o c wireless n e t w o r k i n g is e x p e r i e n c i n g a r e s u r g e n c e of i n t e r e s t b e ­ cause of n e w applications a n d i m p r o v e d t e c h n o l o g y . T h e s e n e t w o r k s are n o w b e i n g c o n s i d e r e d for m a n y c o m m e r c i a l applications, i n c l u d i n g i n - h o m e net­ working, wireless LANs, n o m a d i c c o m p u t i n g , a n d s h o r t - t e r m n e t w o r k i n g for disaster relief, public events, a n d t e m p o r a r y offices. Both t h e IEEE 802.11 a n d HIPERLAN Type 1 wireless LAN s t a n d a r d s s u p p o r t ad h o c w i r e l e s s n e t w o r k i n g w i t h i n a small area, a n d w i d e r a r e a n e t w o r k s are c u r r e n t l y u n d e r d e v e l o p m e n t . Ad h o c n e t w o r k s r e q u i r e a peer-to-peer architecture, a n d t h e topology of t h e n e t w o r k d e p e n d s o n t h e location of t h e different users, w h i c h c h a n g e s over t i m e . In addition, since t h e p r o p a g a t i o n r a n g e of a given m o b i l e is limited, t h e mobile m a y n e e d to enlist t h e aid of o t h e r m o b i l e s in forwarding a p a c k e t to its final destination. T h u s t h e end-to-end c o n n e c t i o n b e t w e e n a n y two mobile h o s t s m a y consist of m u l t i p l e wireless h o p s . It is a significant t e c h n i c a l

^

Wireless Networks

356

challenge to provide reliable high-speed end-to-end c o m m u n i c a t i o n s in ad h o c wireless n e t w o r k s given t h e i r d y n a m i c n e t w o r k topology, d e c e n t r a l i z e d control, a n d m u l t i h o p c o n n e c t i o n s . C u r r e n t r e s e a r c h in ad h o c wireless n e t w o r k design is focused o n dis­ tributed routing. Every mobile h o s t in a wireless ad h o c n e t w o r k m u s t o p e r a t e as a r o u t e r in o r d e r to m a i n t a i n c o n n e c t i v i t y information a n d forward p a c k e t s from o t h e r mobiles. Routing protocols d e s i g n e d for w i r e d n e t w o r k s are n o t ap­ propriate for this task, since t h e y e i t h e r lack t h e ability to quickly reflect t h e c h a n g i n g topology, or m a y r e q u i r e excessive o v e r h e a d . Proposed a p p r o a c h e s to distributed r o u t i n g t h a t quickly a d a p t to c h a n g i n g n e t w o r k topology w i t h o u t excessive o v e r h e a d i n c l u d e d y n a m i c source a n d associativity-based routing. O t h e r protocols t h a t a d d r e s s s o m e of t h e difficulties in s u p p o r t i n g m u l t i m e d i a applications over ad h o c wireless n e t w o r k s i n c l u d e rate-adaptive c o m p r e s s i o n , p o w e r control, a n d r e s o u r c e allocation t h r o u g h radio clustering.

7.73

IMT-2000 I n t e r n a t i o n a l Mobile T e l e c o m m u n i c a t i o n s 2000 (IMT-2000) is a w o r l d w i d e stan­ dard s p o n s o r e d b y t h e ITU. I n 1992 t h e ITU set aside 230 M H z of global s p e c t r u m in t h e 2-GHz f r e q u e n c y b a n d to provide wireless access to t h e global t e l e c o m m u n i c a t i o n s infrastructure t h r o u g h b o t h satellite a n d land-based sys­ t e m s . T h e initial goal of this spectral allocation w a s to facilitate a m o v e from t h e worldwide collection of different s e c o n d - g e n e r a t i o n wireless s y s t e m s , e a c h w i t h its o w n set of standards, services, coverage areas, a n d spectral allocations, to a w o r l d w i d e s t a n d a r d serving fixed a n d mobile u s e r s in b o t h public a n d pri­ vate n e t w o r k s . IMT-2000 is d e s i g n e d to s u p p o r t a w i d e r a n g e of services i n c l u d i n g voice, high-rate a n d variable-rate data, a n d m u l t i m e d i a in b o t h i n d o o r a n d o u t d o o r e n v i r o n m e n t s . A family of radio protocols suitable for a r a n g e of e n v i r o n m e n t s a n d applications is b e i n g d e v e l o p e d to s u p p o r t t h e s e r e q u i r e m e n t s . T h e goal for t h e protocol family is to m a x i m i z e c o m m o n a l i t y w i t h i n t h e family while m a i n t a i n i n g flexibility to a d a p t to t h e different e n v i r o n m e n t s a n d applications. T h e IMT-2000 s t a n d a r d specifies data rates of 2 Mbps for local coverage a n d 384 Kbps for wide area coverage, a n d t h e s e rates c a n b e variable d e p e n d i n g o n link a n d n e t w o r k conditions. It also s u p p o r t s b o t h c o n t i n u o u s s t r e a m a n d packet data. T h e r e q u i r e m e n t for p a c k e t data s u p p o r t is t h e m a i n differentiator b e t w e e n IMT-2000 a n d digital cellular s y s t e m s . IMT-2000 d o e s n o t r e q u i r e b a c k w a r d compatibility w i t h c u r r e n t wireless s y s t e m s , so t h e design of t h e radio access a n d n e t w o r k i n g protocols c a n i n c o r p o r a t e n e w t e c h n o l o g i e s a n d innovations.

77

Future Systems and Standards

A majority of t h e proposals for t h e IMT-2000 radio t r a n s m i s s i o n t e c h n o l o g y s u b m i t t e d from Asia, Europe, a n d N o r t h A m e r i c a are b a s e d o n w i d e b a n d CDMA technology; a n d it a p p e a r s t h a t s o m e form of this t e c h n o l o g y will b e a d o p t e d as t h e radio access standard. Advantages of w i d e b a n d CDMA i n c l u d e its capac­ ity a n d coverage gain from f r e q u e n c y diversity, its ease of u s e in p a c k e t data transfer, its flexibility for different services, its a s y n c h r o n o u s operation, a n d its built-in s u p p o r t for adaptive a n t e n n a arrays, m u l t i u s e r detection, hierarchi­ cal cell structures, a n d t r a n s m i t t e r diversity. Standardization of t h e IMT-2000 n e t w o r k i n g protocol is still several y e a r s away.

7.7.4

High-Speed Digital Cellular All digital cellular s t a n d a r d s are u n d e r g o i n g e n h a n c e m e n t s to s u p p o r t high rate p a c k e t data t r a n s m i s s i o n . T h e goal is to s u p p o r t t h e IMT-2000 data rate specifi­ cation of 384 Kbps over w i d e areas. In t h e n e a r t e r m , GSM s y s t e m s will provide data rates of u p to 100 Kbps b y aggregating all t i m e slots t o g e t h e r for a single user. T h e e n h a n c e m e n t to s u p p o r t 384 Kbps, called E n h a n c e d Data Services for GSM Evolution (EDGE), i n c r e a s e s t h e data rate further b y u s i n g a highlevel m o d u l a t i o n format c o m b i n e d w i t h FEC coding. T h i s m o d u l a t i o n is m o r e sensitive to fading effects, a n d EDGE u s e s adaptive m o d u l a t i o n a n d coding to mitigate this p r o b l e m . Specifically, EDGE defines six different m o d u l a t i o n a n d coding c o m b i n a t i o n s , e a c h o p t i m i z e d to a different value of r e c e i v e d SNR. T h e received SNR is m e a s u r e d at t h e r e c e i v e r a n d fed b a c k to t h e transmitter, a n d t h e b e s t m o d u l a t i o n a n d coding c o m b i n a t i o n for this SNR value is used. T h e IS-54 a n d IS-136 s y s t e m s c u r r e n t l y provide data r a t e s of 40 to 60 Kbps b y aggregating t i m e slots a n d u s i n g high-level m o d u l a t i o n . T h e s e T D M A s t a n d a r d s will s u p p o r t 384 Kbps b y m i g r a t i n g to t h e GSM EDGE standard. T h i s n e w TDMA s t a n d a r d is referred to as IS-136HS (high-speed). T h e IS-95 s y s t e m s will s u p p o r t h i g h e r data rates b y evolving to t h e w i d e b a n d CDMA s t a n d a r d in IMT-2000. However, it is n o t y e t clear if t h a t s t a n d a r d will b e b a c k w a r d c o m p a t i b l e w i t h t h e IS-95 s y s t e m s .

7.7.5

Fixed Wireless Access Fixed wireless access provides wireless c o m m u n i c a t i o n s b e t w e e n a fixed ac­ cess point a n d m u l t i p l e t e r m i n a l s . T h e s e s y s t e m s w e r e initially p r o p o s e d to s u p p o r t interactive video service to t h e h o m e , b u t t h e application e m p h a s i s h a s n o w shifted to providing high-speed data access ( t e n s of Mbps) for h o m e s a n d

357

Wireless Networks

358

b u s i n e s s e s . In t h e U.S. t w o f r e q u e n c y b a n d s h a v e b e e n set aside for t h e s e s y s t e m s : p a r t of t h e 28-GHz s p e c t r u m is allocated for local distribution sys­ t e m s (local m u l t i p o i n t distribution s y s t e m s or LMDS), a n d a b a n d in t h e 2-GHz s p e c t r u m is allocated for m e t r o p o l i t a n distribution s y s t e m s ( m u l t i c h a n n e l m u l ­ tipoint distribution services or MMDS). LMDS r e p r e s e n t s a quick m e a n s for n e w service providers to e n t e r t h e already stiff c o m p e t i t i o n a m o n g wireless a n d wireline b r o a d b a n d service providers. MMDS is a television a n d t e l e c o m ­ m u n i c a t i o n delivery s y s t e m w i t h t r a n s m i s s i o n r a n g e s of 30 to 50 k m . MMDS h a s t h e capability to deliver m o r e t h a n 100 digital video TV c h a n n e l s along w i t h t e l e p h o n y a n d access to e m e r g i n g interactive services s u c h as t h e I n t e r n e t . MMDS will m a i n l y c o m p e t e w i t h existing cable a n d satellite s y s t e m s .

77.6

HomeRF and Bluetooth H o m e R F is a n RF s t a n d a r d in t h e 2-GHz f r e q u e n c y b a n d for wireless h o m e n e t w o r k i n g . T h e s t a n d a r d w a s initiated b y Intel, HP, Microsoft, C o m p a q , a n d IBM to e n a b l e c o m m u n i c a t i o n s a n d I n t e r n e t c o n n e c t i v i t y a m o n g different electronic devices in a n d a r o u n d t h e h o m e , i n c l u d i n g PCs, laptops, s m a r t pads, a n d intelligent h o m e appliances. T h e data rate for H o m e R F is specified as 2 Mbps, w i t h s i m u l t a n e o u s s u p p o r t for voice a n d data, at a r a n g e of 50 m e t e r s . T h e H o m e R F s t a n d a r d is e x p e c t e d to b e finalized s o m e t i m e in 1999, w i t h p r o d u c t s incorporating t h e s t a n d a r d i n t r o d u c e d s o m e t i m e in t h e y e a r 2000. Bluetooth is a c a b l e - r e p l a c e m e n t RF t e c h n o l o g y for short-range c o n n e c ­ tions (less t h a n 10 m e t e r s ) b e t w e e n wireless devices. Its m a i n application is to c o n n e c t digital cellular p h o n e s , l a p t o p a n d p a l m t o p c o m p u t e r s , portable print­ ers a n d projectors, n e t w o r k access points, a n d o t h e r portable devices w i t h o u t t h e n e e d to c a r r y or c o n n e c t cables. T h e Bluetooth s t a n d a r d w a s initiated b y Ericsson, IBM, Intel, Nokia, a n d Tbshiba, a n d h a s since b e e n a d o p t e d b y over 200 t e l e c o m m u n i c a t i o n s a n d c o m p u t e r c o m p a n i e s . Products c o m p a t i b l e w i t h Bluetooth s h o u l d a p p e a r in late 1999. T h e s y s t e m o p e r a t e s in t h e 2.4-GHz f r e q u e n c y b a n d w i t h data r a t e s of 700 Kbps for data a n d u p to t h r e e voice con­ n e c t i o n s at 64 Kbps.

7.8

SUMMARY A desire for mobility c o u p l e d w i t h t h e d e m a n d for voice, I n t e r n e t , a n d multi­ m e d i a services indicates a b r i g h t future for wireless n e t w o r k s . Digital cellular a n d paging s y s t e m s h a v e enjoyed e n o r m o u s growth, b u t c u r r e n t p r o d u c t s a n d services for wireless data h a v e n o t lived u p to expectations. T h i s is d u e m a i n l y to t h e i r high cost a n d poor p e r f o r m a n c e . N e w s t a n d a r d s a n d s y s t e m s are e m e r g i n g

7.9

Notes

359 w o r l d w i d e to a d d r e s s t h e s e p e r f o r m a n c e a n d cost issues. T h e s e s y s t e m s sup­ port a w i d e r a n g e of voice, data, a n d m u l t i m e d i a services for fixed a n d mobile u s e r s b o t h indoors a n d out, in cities, rural areas, a n d r e m o t e regions. T h e r e are m a n y t e c h n i c a l c h a l l e n g e s to o v e r c o m e in b u i l d i n g highp e r f o r m a n c e wireless n e t w o r k s . T h e wireless c h a n n e l is a difficult c o m m u ­ nications m e d i u m . Sophisticated t e c h n i q u e s exist to c o m p e n s a t e for m a n y of t h e c h a n n e l i m p a i r m e n t s , b u t t h e s e c a n entail significant cost a n d complex­ ity. T h e s p e c t r u m m u s t also b e u s e d e x t r e m e l y efficiently t h r o u g h a d v a n c e d link layer, access, a n d cellular s y s t e m design. N e t w o r k i n g protocols to s u p p o r t r o a m i n g u s e r s a n d end-to-end QpS g u a r a n t e e s also p o s e a significant t e c h n i c a l challenge. T h e l i m i t e d size a n d b a t t e r y life of m o b i l e t e r m i n a l s i m p o s e signifi­ c a n t complexity constraints, so c o m p l e x i t y m u s t b e distributed t h r o u g h o u t t h e n e t w o r k to c o m p e n s a t e for this limitation. Finally, t h e u n p r e d i c t a b l e n a t u r e of t h e wireless c h a n n e l r e q u i r e s a d a p t a t i o n across all levels of t h e wireless n e t w o r k design: t h e link layer, n e t w o r k layer, t r a n s p o r t layer, a n d application layer. T h i s r e q u i r e s i n t e r a c t i o n a m o n g t h e s e layers, w h i c h violates t h e tra­ ditional n e t w o r k design p a r a d i g m of d e s i g n i n g e a c h l a y e r in t h e OSI m o d e l i n d e p e n d e n t l y from t h e others. While this p a r a d i g m h a s w o r k e d well o n w i r e d n e t w o r k s , especially as w i r e d t e c h n o l o g y h a s evolved to t h e h i g h p e r f o r m a n c e of today's n e t w o r k s , h i g h - p e r f o r m a n c e wireless n e t w o r k s will n o t b e possible w i t h o u t significant t e c h n i c a l b r e a k t h r o u g h s at all levels of t h e s y s t e m design as well as a n i n t e g r a t e d a n d adaptable design for t h e overall n e t w o r k .

7.9

NOTES T h e wireless c h a n n e l h a s b e e n c h a r a c t e r i z e d in m a n y b o o k s a n d articles over t h e last 30 y e a r s : [R96, P92] a n d t h e r e f e r e n c e s t h e r e i n describe t h e basic m o d e l s for b o t h i n d o o r a n d o u t d o o r s y s t e m s . A tutorial o n wireless infrared c o m m u n i c a t i o n s y s t e m s c a n b e found in [KB97]. T e c h n i q u e s for r e d u c i n g i n t e r f e r e n c e from f r e q u e n c y r e u s e are p r e s e n t e d in [V98, Wi98, KN96]. For discussions o n Trellis a n d TUrbo codes see [U82, BGT93, HW99, RW98]. Equalization t e c h n i q u e s are p r e s e n t e d i n [BP79, F72, DJB95]. O r t h o g o n a l frequency-divsion m u l t i p l e x i n g (OFDM) is discussed in [C85, B90]. Recent w o r k o n t h e capacity of wireless c h a n n e l s is s u m m a r i z e d in [GC97, BPS98]. Link level design for wireless c h a n n e l s is t r e a t e d in m a n y textbooks [Pr95, R96, St96]. Spread s p e c t r u m for b o t h ISI mitigation a n d m u l t i p l e access is described in [D94, V95].

Wireless Networks

360

T h e ALOHA protocol is r e v i e w e d in [Ab96, BG92]. A n informative s u r v e y of routing is [RS96]. P e r f o r m a n c e of TCP over wireless links is discussed in [BPSK97]. Early w o r k in p a c k e t radio is r e v i e w e d in [LNT87]. For r o u t i n g in ad h o c n e t w o r k s see [ABB96, JM96]. A special issue of t h e IEEE Personal Communications Magazine is d e v o t e d to s m a r t a n t e n n a s a n d t h e i r u s e in wireless s y s t e m s [PC98]. Several textbooks provide additional details o n m u l t i p l e a n d r a n d o m access for wireless n e t w o r k s [R96, St96, PL95, BG92]. Wireless n e t w o r k design is still a n active a r e a of research, a n d t h e r e is n o definitive r e f e r e n c e for this field. T h e cellular c o n c e p t is explained in [M79]. More details o n t h e cellular p h o n e s y s t e m s a n d wide a r e a wireless data services c a n b e found in [PL95]. E m e r g i n g satellite s y s t e m s are described in [AS96]. F u t u r e s y s t e m s a n d stan­ dards are c o n t i n u a l l y evolving: t h e b e s t source of information in this a r e a are t h e s t a n d a r d s b o d i e s a n d c o m p a n i e s b u i l d i n g t h e s y s t e m s [ALM98].

7.10

PROBLEMS 1. Consider a p a t h loss m o d e l PR

P

T

=

do d'

w h e r e PR is t h e received signal power, P is t h e t r a n s m i t t e d signal power, do = 100 m e t e r s is a propagation constant, d is t h e p r o p a g a t i o n distance, a n d a is t h e p a t h loss e x p o n e n t . If Ρτ = 100 milliwatts (mW) a n d PR = 1 m W is r e q u i r e d for acceptable p e r f o r m a n c e , w h a t is t h e m a x i m u m t r a n s m i s s i o n r a n g e of o u r s y s t e m for a p a t h loss e x p o n e n t a = 2? By h o w m u c h w o u l d t h e t r a n s m i s s i o n r a n g e d e c r e a s e if t h e p a t h loss e x p o n e n t w a s a = 4? T

2. Shadow fading often follows a log-normal distribution, w h i c h m e a n s t h a t t h e dB value of t h e received p o w e r 10 l o g ( P ^ ) follows a G a u s s i a n distri­ b u t i o n . Suppose t h e received signal p o w e r follows this log-normal distri­ b u t i o n . A s s u m e t h e average value of 10 \og (PR/N) e q u a l s 10 dB a n d its variance equals 4 dB, w h e r e Ν is a s s u m e d to b e constant. W h a t is t h e prob­ ability t h a t t h e received SNR is less t h a n 5 dB? 10

L0

3. Consider a c h a n n e l w i t h a m u l t i p a t h delay s p r e a d of 10 μδ. Suppose a voice signal with a signal b a n d w i d t h of 30 k H z is t r a n s m i t t e d over this c h a n n e l . Will t h e c h a n n e l exhibit flat or frequency-selective fading? H o w about for a data signal w i t h a 1-MHz signal b a n d w i d t h ?

7.10

Problems

361

4. T h e BER of b i n a r y phase-shift-keying w i t h differential d e t e c t i o n in w h i t e G a u s s i a n noise (i.e., n o fading, s h a d o w i n g , or ISI) is .5e~ , y

w h e r e γ is t h e

average received SNR. If t h e r e is also Rayleigh fading t h e n t h e B E R b e c o m e s [2(1 + y ) ] . C o n s i d e r a data s y s t e m w i t h a r e q u i r e d BER of 1 0 ~ . W h a t - 1

6

average SNR is r e q u i r e d to a c h i e v e this target BER b o t h w i t h a n d w i t h o u t Rayleigh fading? 5. T h e error floor d u e to c h a n n e l D o p p l e r for b i n a r y phase-shift-keying w i t h differential d e t e c t i o n is .5(7τ/χ)Τ^) , w h e r e fo is t h e c h a n n e l D o p p l e r a n d 2

is t h e bit t i m e (the i n v e r s e of t h e data rate). For fa = 80 Hz a n d a data rate of 20 Kbps, w h a t is t h e e r r o r floor? H o w a b o u t for/b = 80 Hz a n d a data r a t e of 1 Mbps? W h y d o e s t h e e r r o r floor d e c r e a s e as t h e data rate i n c r e a s e s ? 6. H o w m a n y i n d e p e n d e n t diversity p a t h s c a n b e o b t a i n e d u s i n g a n a n t e n n a a r r a y m o u n t e d o n a l a p t o p of l e n g t h .1 meter, a s s u m i n g a c a r r i e r f r e q u e n c y of 5 GHz? 7. C o n s i d e r a wireless s y s t e m w i t h total b a n d w i d t h of 3 M H z . H o w m a n y u s e r s c a n s h a r e this c h a n n e l u s i n g FDMA, w h e r e e a c h u s e r is a s s i g n e d a 30k H z c h a n n e l ? Suppose i n s t e a d w e u s e s e m i - o r t h g o n a l CDMA for m u l t i p l e access. If e a c h u s e r h a s a r e c e i v e d signal p o w e r of Ρ and t h e i n t e r f e r e n c e p o w e r c a u s e d b y t h a t u s e r to o t h e r u s e r s is .01P, for a r e q u i r e d signal-toi n t e r f e r e n c e p o w e r ratio of 10, h o w m a n y u s e r s c a n b e a c c o m m o d a t e d i n this s y s t e m ? 8. Radio signals travel at t h e s p e e d of light, e q u a l to 3 χ 1 0 m e t e r s p e r s e c o n d . What is t h e m a x i m u m orbit distance of a satellite s u c h t h a t t h e r o u n d - t r i p p r o p a g a t i o n delay of a signal d o e s n o t exceed t h e 100-ms d e l a y c o n s t r a i n t of voice s y s t e m s . Based o n this calculation, d e t e r m i n e w h i c h of t h e t h r e e satellite orbit distances, GEO, LEO, a n d MEO, c a n s u p p o r t voice services. 8

9. T h e IEEE 802.11 wireless LAN s t a n d a r d s u p p o r t s b o t h star a n d peer-top e e r a r c h i t e c t u r e s . Describe a wireless LAN application t h a t is well s u i t e d to e a c h t y p e of architecture. 10. Discuss t h e differences b e t w e e n a d h o c wireless n e t w o r k s a n d cellular n e t w o r k s in t e r m s of n e t w o r k a r c h i t e c t u r e a n d mobility m a n a g e m e n t . W h a t are t h e a d v a n t a g e s a n d d i s a d v a n t a g e s of e a c h n e t w o r k design? W h a t applications are b e s t suited to e a c h t y p e of n e t w o r k ? 11. S p e c t r u m for n e w wireless s y s t e m s is b e i n g allocated b y t h e FCC at h i g h e r frequencies (e.g., 5 GHz a n d 28 GHz) t h a n in existing s y s t e m s . W h a t are t h e a d v a n t a g e s a n d d i s a d v a n t a g e s of b u i l d i n g s y s t e m s at t h e s e h i g h e r frequencies?

8

Control of

CHAPTER

Networks

m i i n t h e p r e c e d i n g c h a p t e r s w e d e s c r i b e d t h e t r e n d s in packet- a n d circuitswitched n e t w o r k s that, together, c u l m i n a t e in t h e design of ATM n e t w o r k s a n d in i m p r o v e m e n t s of t h e I n t e r n e t protocols. We s a w in C h a p t e r 6 t h a t a n e t w o r k t h a t c o m b i n e s a n ATM t r a n s p o r t layer, or a n IP l a y e r w i t h s o m e form of quality of service or class of service, w i t h a high-speed physical l a y e r s u c h as SONET c a n potentially provide t h e large r a n g e of quality of service (QpS) n e c e s s a r y to s u p p o r t m o s t applications. However, in o r d e r to provide this r a n g e of QoS, t h e n e t w o r k ' s r e s o u r c e s ( b a n d w i d t h a n d buffers) m u s t b e p r o p e r l y m a n a g e d or controlled. In this c h a p t e r y o u will l e a r n t h e c o n c e p t s a n d f u n d a m e n t a l t e c h n i q u e s u s e d to control circuit-switched, datagram, a n d ATM n e t w o r k s in o r d e r to achieve efficient u s e of n e t w o r k r e s o u r c e s . You will u n d e r s t a n d h o w different control t e c h n i q u e s affect different n e t w o r k p e r f o r m a n c e m e a s u r e s , a n d y o u will acquire t h e skills to evaluate t h o s e p e r f o r m a n c e m e a s u r e s , i n c l u d i n g blocking probability, delay, a n d loss. You will l e a r n t h e d e t e r m i n i s t i c proposals of t h e ATM F o r u m for a d m i t t i n g n e w c o n n e c t i o n s a n d for allocating r e s o u r c e s to those c o n n e c t i o n s . You will see t h a t t h o s e proposals are b a s e d o n worst-case scenarios a n d t h a t m o r e sophisticated proposals b a s e d o n effective b a n d w i d t h c a n realize t h e gains from statistical multiplexing. F r o m C h a p t e r 2 y o u already k n o w t h e r e q u i r e m e n t s i m p o s e d b y v a r i o u s applications; n o w y o u will b e able to calculate h o w well a specific application c a n b e s u p p o r t e d b y t h e b a n d w i d t h a n d buffers allocated b y t h e n e t w o r k for t h a t application. In section 8.1 w e describe t h e objectives a n d t h e m e t h o d s of control. We explain t h a t w i t h good control strategies t h e n e t w o r k c a n c a r r y m o r e traffic w i t h t h e s a m e quality of service. We discuss t h e m e a n i n g of quality of service, a n d w e

Control of Networks

364

c o m p a r e d e t e r m i n i s t i c a n d statistical g u a r a n t e e s . We t h e n classify t h e m e t h o d s available to control t h e o p e r a t i o n s of circuit-switched, packet-switched, a n d virtual circuit-switched n e t w o r k s . T h e r e are four principal m e t h o d s : a d m i s s i o n control, routing, flow a n d congestion control, a n d r e s o u r c e allocation. A fifth m e t h o d , control via pricing, is discussed in C h a p t e r 10. In section 8.2 w e explain a d m i s s i o n control a n d r o u t i n g m e c h a n i s m s for circuit-switched n e t w o r k s s u c h as t h e t e l e p h o n e n e t w o r k . We p o i n t out t h e trade-off b e t w e e n accepting all calls t h a t c a n b e carried a n d t h e negative i m p a c t o n future calls. A c o m p r o m i s e is a c h i e v e d b y t r u n k reservations. We discuss flow- a n d congestion-control p r o c e d u r e s for d a t a g r a m n e t w o r k s in section 8.3. Flow-control p r o c e d u r e s a t t e m p t to p r e v e n t a source from over­ w h e l m i n g a destination. Typically, t h e d e s t i n a t i o n informs t h e source w h e n it is b e c o m i n g congested, a n d t h e source stops t r a n s m i t t i n g . T h e congestioncontrol m e c h a n i s m is u s e d to p r e v e n t s o m e i n t e r n a l n o d e s of t h e n e t w o r k from b e c o m i n g congested. D a t a g r a m n e t w o r k s , s u c h as t h e I n t e r n e t , do n o t guar­ a n t e e delay or t h r o u g h p u t . T h e i r congestion control a t t e m p t s to r e d u c e t h e average delay p e r p a c k e t for a n y given t h r o u g h p u t . Section 8.4 i n t r o d u c e s t h e formulations a n d s o m e of t h e results o n t h e con­ trol of ATM n e t w o r k s . Since ATM n e t w o r k s u s e virtual circuits a n d g u a r a n t e e delays, t h r o u g h p u t , a n d loss r a t e s of individual c o n n e c t i o n s , t h e s e n e t w o r k s m u s t exercise a d m i s s i o n control, a n d t h e y m u s t r e s e r v e b a n d w i d t h a n d buffer capacity for c o n n e c t i o n s . O n e k e y q u e s t i o n therefore is to d e t e r m i n e h o w m u c h of t h e s e r e s o u r c e s t h e n e t w o r k m u s t reserve. Resource-allocation p r o c e d u r e s d e t e r m i n e w h i c h cells t h e n e t w o r k n o d e s s h o u l d buffer or t r a n s m i t . For in­ stance, a n o d e m a y t r a n s m i t cells of a video c o n n e c t i o n before e-mail cells. T h e n o d e m a y also discard audio cells to m a k e r o o m for data cells. C h a p t e r 9 carries out t h e m a t h e m a t i c a l analysis t h a t justifies t h e results w e describe h e r e . Some of t h e m a t e r i a l in this c h a p t e r relies o n c o n c e p t s of probability t h e o r y . A l t h o u g h w e h a v e a t t e m p t e d to provide intuitive discussion, r e a d e r s w i t h n o probability b a c k g r o u n d m a y n o t fully a p p r e c i a t e t h e a r g u m e n t in s o m e places. T h o s e r e a d e r s c a n s k i m t h a t m a t e r i a l .

8.1

OBJECTIVES AND METHODS OF CONTROL We provide a n overview of t h e objectives of control a n d t h e principal m e a n s available or likely to b e available to exercise control. T h e s e m e a n s involve tak-

8.1

Objectives and Methods of Control

ing decisions at v e r y different t i m e scales a n d b a s e d o n different information. S o m e e x a m p l e s are given to illustrate t h e ideas.

8.1.1

Overview Let u s consider a n ATM n e t w o r k t h a t transfers i n f o r m a t i o n b e t w e e n u s e r s in cell s t r e a m s over virtual circuits. A virtual circuit specifies a r o u t e of links a n d switches t h a t c o n n e c t s t h e u s e r s . T h e cells in different virtual circuits s h a r e t h e t r a n s m i s s i o n b a n d w i d t h a n d buffers t h a t t h e i r r o u t e s h a v e in c o m m o n . T h e w a y in w h i c h t h o s e r e s o u r c e s are s h a r e d is d e t e r m i n e d b y t h e n e t w o r k ' s control strategy. Because of fluctuations in t h e cell s t r e a m s , t h e r e are p e r i o d s of t i m e w h e n t h e cells arrive in a buffer faster t h a n t h e t r a n s m i t t e r e m p t i e s t h a t buffer. W h e n this situation occurs, t h e buffer stores t h e excess cells. T h i s t e m p o r a r y storage results in delays and, in case t h e buffer is full, in cell loss. Delays a n d losses affect t h e quality of service p r o v i d e d to t h e user. With b e t t e r control strategies, t h e n e t w o r k c a n c a r r y m o r e virtual circuits while m a i n t a i n i n g t h e quality of service p r o m i s e d to t h e u s e r s . T h i s possibility h a s long b e e n k n o w n in t h e case of t h e t e l e p h o n e n e t w o r k , w h e r e i m p r o v e d r o u t i n g algorithms e n a b l e t h e n e t w o r k to c a r r y m o r e p h o n e calls w i t h t h e s a m e blocking probability. T h u s , b y i m p r o v i n g t h e control algorithms, n e t w o r k m a n a g e r s c a n provide b e t t e r service w i t h o u t additional h a r d w a r e . I n a public n e t w o r k t h a t sells services to users, b e t t e r control l e a d s to g r e a t e r r e v e n u e . In a private n e t w o r k , b e t t e r control results in l o w e r cost for service. T h e r e are four basic m e a n s of control: a d m i s s i o n control, routing, flow a n d congestion control, a n d allocation control. As i n d i c a t e d in Table 8.1, t h e m e a n s available d e p e n d o n w h e t h e r t h e t r a n s p o r t m e t h o d u s e d b y t h e n e t w o r k is circuit, datagram, or virtual circuit switching.

8.12

Control Methods Admission control d e t e r m i n e s w h i c h circuit or virtual circuit c o n n e c t i o n re­ q u e s t s are a c c e p t e d b y t h e n e t w o r k . T h i s is similar to t h e a d m i s s i o n of tele­ p h o n e calls b y t h e t e l e p h o n e n e t w o r k . W h e n a subscriber places a r e q u e s t for a n e w c o n n e c t i o n or call, t h e n e t w o r k c a n notify t h e u s e r t h a t it is too b u s y to accept t h a t r e q u e s t . T h e subscriber c a n t h e n decide to place t h e r e q u e s t at s o m e later t i m e or to u s e a c o m p e t i t o r ' s n e t w o r k , if o n e is available. Normally, a d a t a g r a m n e t w o r k always a c c e p t s t h e p a c k e t s ( d a t a g r a m s ) s u b m i t t e d b y a u s e r a n d does n o t exercise a d m i s s i o n control. A n u m b e r of issues are r e l a t e d to t h e a d m i s s i o n of calls. For instance, a u s e r m a y ask t h e n e t w o r k to s c h e d u l e a c o n n e c t i o n at s o m e later t i m e . A u s e r m a y ask t h e n e t w o r k to call b a c k w h e n

365

Control of Networks 366

ι ι .. . . . ι

.1.

ι..

..ι». .1,1.ι™.ι»i.

Network type

Admission

Routing

Flow

Allocation

Circuitswitched

Free paths and costs

• Static (list) • Dynamic (least full)

None

None

Datagram

None

• Static (random) • Dynamic (shortest)

• Link window • End-to-end window

Virtual circuit

Current VCs

Dynamic (largest spare capacity)

• Window • Rate

8.1

Priority, fairness, and QpS based

Means of control of different types of network.

TABLE

it c a n set u p a less expensive c o n n e c t i o n . T h e p r o b l e m s of m u l t i p a r t y c o n n e c ­ tions, s u c h as c o n f e r e n c e calls, are also r e l a t e d to a d m i s s i o n control. C a n n e w parties j o i n in? W h a t h a p p e n s if o n e p a r t y d r o p s out? W h e n t h e n e t w o r k accepts a circuit or virtual circuit call r e q u e s t , it m u s t decide t h e p a t h or r o u t e to b e followed b y t h e bit s t r e a m of t h e call or t h e packets. T h i s decision is called routing and r e m a i n s fixed for t h e d u r a t i o n of t h e c o n n e c t i o n . In m u l t i p a r t y c o n n e c t i o n s , t h e n e t w o r k m u s t decide w h e r e to copy a s t r e a m it delivers to m u l t i p l e destinations. W h a t if different d e s t i n a t i o n s r e q u i r e different versions of t h e p a c k e t s t r e a m ? W h a t if t h e characteristics of t h e p a t h from t h e source to t h e d e s t i n a t i o n c h a n g e over time, as in mobile wireless n e t w o r k s ? For d a t a g r a m a n d virtual circuit switching, t h e n e t w o r k c a n also decide w h e t h e r bit s t r e a m s s h o u l d b e forwarded along t h e i r r o u t e s quickly in o r d e r to r e d u c e delay or w h e t h e r t h e y s h o u l d b e slowed d o w n (throttled) in o r d e r to p r e v e n t congestion d o w n s t r e a m . T h i s decision is called congestion control. A related m e t h o d , called traffic shaping, is u s e d b y a traffic source to regulate its s t r e a m of p a c k e t s before s e n d i n g t h e m to t h e n e t w o r k . Because a circuitswitched n e t w o r k provides a c o n s t a n t b a n d w i d t h ( a n d n o buffers) to e a c h c o n n e c t i o n , it does n o t exercise congestion control. Lastly, in virtual circuit switching, t h e n e t w o r k c a n control t h e b a n d w i d t h a n d buffers allocated to e a c h virtual circuit. T h i s is called (resource) allocation control. T h e allocation c a n b e static, t h a t is, fixed at t h e b e g i n n i n g of t h e call request, or it c a n b e d y n a m i c a n d c h a n g e d over t h e d u r a t i o n of t h e call. It is this flexibility t h a t p e r m i t s t h e n e t w o r k to provide c o n n e c t i o n s w i t h differ­ e n t qualities of service. Circuit-switched n e t w o r k s do n o t exercise allocation control.

8.1

Objectives and Methods of Control

367

Request (QoS) Accept?

y

Flow control Allocation of buffer and bandwidth

Routing

Notes: No admission control in datagram networks. No flow control or dynamic allocation in circuit-switched networks. 8.1

Sequence of control actions for a given call.

FIGURE

We will r e t u r n to discuss t h e r e m a i n i n g e n t r i e s of T&ble 8.1. Figure 8.1 illustrates t h e o r d e r in w h i c h t h e control decisions are t a k e n . T h e figure is a l m o s t self-explanatory. In t h e case of a virtual circuit n e t w o r k , for example, w h e n a r e q u e s t for a c o n n e c t i o n w i t h a p a r t i c u l a r QpS is received, t h e n e t w o r k m u s t first decide to accept or reject it. If t h e a n s w e r is yes, a r o u t e m u s t t h e n b e assigned. Moreover, d u r i n g t h e lifetime of t h e call, t h e n e t w o r k will exercise flow a n d congestion control a n d decide w h a t r e s o u r c e s to allocate to t h e virtual circuit.

8-13

Time Scales T h e four t y p e s of control decisions are t a k e n at different t i m e scales, a n d t h e y are b a s e d o n different t y p e s of information. We illustrate this for virtual circuit n e t w o r k s u s i n g Figure 8.2. T h e decision to a c c e p t a n e w call is b a s e d o n t h e

Request (QoS) Accept?

y

Typical statistics

Flow control Allocation

Routing

"Real time"

Network Switches Buffers 8.2 FIGURE

The different control actions are taken at different levels and based on statistics that are relevant on different time scales.

Control of Networks

368

n u m b e r of virtual circuits c u r r e n t l y active in t h e n e t w o r k a n d o n t h e typical statistics of t h e bit s t r e a m s carried b y t h e s e virtual circuits. T h e p o i n t is t h a t t h e admission-control decision s h o u l d n o t b e b a s e d o n t h e i n s t a n t a n e o u s traffic conditions; rather, it m u s t b e b a s e d o n t h e typical conditions t h a t are likely to prevail for t h e d u r a t i o n of t h e calls. I n d e e d , t h e decision to a c c e p t a call is irreversible a n d c a n n o t b e r e v o k e d if c u r r e n t l y active calls s u d d e n l y b e c o m e m o r e b u s y t h a n t h e y w e r e w h e n t h e call w a s accepted. T h e s a m e considerations a p p l y to r o u t i n g decisions, a n d so t h e s e m u s t also b e b a s e d o n typical traffic statistics. T h u s , call a d m i s s i o n a n d routing are long-term decisions t h a t h a v e to b e m a i n t a i n e d for t h e d u r a t i o n of t h e calls. By contrast, flow- a n d allocation-control decisions c a n b e b a s e d o n i n s t a n t a n e o u s conditions a n d c a n b e modified as t h e s e conditions c h a n g e ; h e n c e t h e label "real t i m e " in Figure 8.2. Voice calls, for example, typically last several m i n u t e s . If voice is t r a n s m i t t e d over a high-speed ATM n e t w o r k , t h e voice cells m u s t b e switched w i t h i n a few m i c r o s e c o n d s . T h u s t h e t i m e scales involved in t h e different control actions r a n g e over several o r d e r s of m a g n i t u d e .

8.1.4

Examples We n o w r e t u r n to T&ble 8.1, discussing e a c h r o w in t u r n . For a circuit-switched network, t h e decision to a d m i t a n e w call is b a s e d o n t h e p a t h s t h a t are t h e n free or, equivalently, o n t h e circuits t h a t are b u s y (not free) w h e n t h e call r e q u e s t is m a d e . T h u s , t h e n e t w o r k k e e p s track of w h i c h circuits are b u s y a n d u s e s t h a t information to decide w h e t h e r to accept a n e w call. T h e r o u t i n g is also b a s e d o n t h e circuits t h a t are b u s y w h e n t h e n e w call is placed. T h e r e are two t y p e s of routing algorithms: static a n d d y n a m i c . A static a l g o r i t h m u s e s p r e c o m p u t e d decisions w h e r e a s a d y n a m i c algorithm b a s e s its decision o n t h e c u r r e n t state of t h e n e t w o r k . T h e r e is n o congestion a n d flow control a n d n o allocation control in t h e s e n e t w o r k s , as we n o t e d before. A d a t a g r a m n e t w o r k n o r m a l l y accepts all packets, so t h e r e is n o admis­ sion control. T h e r o u t i n g algorithm c a n b e static or d y n a m i c . T h e flow a n d congestion control in a d a t a g r a m n e t w o r k typically u s e w i n d o w s , as explained in C h a p t e r 2. T h e w i n d o w s m a y c o r r e s p o n d to individual links or to end-toe n d p a t h s t h r o u g h t h e n e t w o r k . For instance, t h e HDLC (High-Level Data Link Control) u s e d b y X.25 n e t w o r k s limits t h e n u m b e r of u n a c k n o w l e d g e d p a c k e t s b e t w e e n two successive n o d e s . T h i s is a form of link-level w i n d o w congestion control. Specifically, HDLC u s e s t h e Go Back Ν protocol. T h e Selective Repeat (SRP) a n d Automatic Request (ARQ) protocols are o t h e r w i n d o w congestion control protocols. End-to-end w i n d o w r e t r a n s m i s s i o n protocols provide a sim­ ple w a y to i m p l e m e n t flow control. By limiting t h e w i n d o w size to a value t h a t

8.1

Objectives and Methods of Control

t h e destination advertises, t h e source stops t r a n s m i t t i n g w h e n t h e d e s t i n a t i o n tells it to do so. T h e TCP (Transmission Control Protocol) of t h e I n t e r n e t u s e s a n end-toe n d w i n d o w flow a n d congestion control (called SRP). S o m e d a t a g r a m n e t w o r k s u s e a c r u d e form of allocation control. For instance, TCP h a s a provision for expedited data transfer. A p a c k e t s e n t as expedited data is m a d e to j u m p to t h e h e a d of t h e q u e u e s t h a t it goes t h r o u g h in t h e s e n d i n g a n d receiving c o m p u t e r s (not in t h e routers, since t h e y are u n a w a r e of TCP a n d k n o w o n l y IP). More sophisticated r e s o u r c e allocation m e c h a n i s m s h a v e b e e n d e s i g n e d a n d are i m p l e m e n t e d in s o m e IP routers. We discuss t h e s e m e c h a n i s m s in section 8.3.4. A store-and-forward n e t w o r k c a n also b e u s e d as a virtual circuit n e t w o r k (not s e p a r a t e l y indicated in I k b l e 8.1). IBM's SNA (System N e t w o r k Archi­ t e c t u r e ) is a n e x a m p l e of a virtual circuit store-and-forward n e t w o r k . Such a n e t w o r k accepts all r e q u e s t s for virtual circuits, so t h e r e is n o a d m i s s i o n con­ trol. T h e r o u t i n g decisions are m a d e for individual virtual circuits r a t h e r t h a n for e a c h packet, as in d a t a g r a m n e t w o r k s . T h e n e t w o r k s u s e w i n d o w flowcontrol m e t h o d s , as in d a t a g r a m n e t w o r k s . T h e s e n e t w o r k s also u s e a priority b a n d w i d t h allocation for e x p e d i t e d data packets. ATM is t h e m o s t i m p o r t a n t e x a m p l e of a virtual circuit-switched n e t w o r k . T h e QpS r e q u i r e m e n t s in a n ATM n e t w o r k c a n b e m o r e s t r i n g e n t t h a n in d a t a g r a m n e t w o r k s , a n d so a n ATM n e t w o r k d o e s n o t a c c e p t all virtual circuit r e q u e s t s . Were it to accept all t h e virtual circuit r e q u e s t s , it w o u l d b e u n a b l e to deliver t h e cells w i t h small loss a n d delay. T h u s , ATM n e t w o r k s m u s t control t h e a d m i s s i o n of n e w virtual circuits. M e t h o d s are d e s i g n e d for s u c h a d m i s s i o n control a n d also for routing, c o n g e s t i o n a n d flow control, a n d allocation. We will discuss t h e s e m e t h o d s in section 8.4. A n o t h e r set of q u e s t i o n s arises w h e n different t y p e s of n e t w o r k s are in­ t e r c o n n e c t e d . For instance, c o n s i d e r a d a t a g r a m n e t w o r k a t t a c h e d to a virtual circuit n e t w o r k . If individual p a c k e t s are s e n t from t h e d a t a g r a m n e t w o r k to u s e r s t h r o u g h t h e virtual circuit n e t w o r k , t h e n a virtual circuit n e e d s to b e set u p for t r a n s p o r t i n g t h e p a c k e t s in t h e s e c o n d n e t w o r k , as w e e x p l a i n e d i n o u r discussion of IP over ATM in C h a p t e r 6. If s o m e form of quality of service is provided b y t h e different n e t w o r k s , t h e n t h e y s h o u l d collaborate to provide a n end-to-end quality of service.

8.1.5

Quality of Service We said earlier t h a t w i t h b e t t e r control strategies t h e n e t w o r k c a n provide m o r e c o n n e c t i o n s or calls w i t h t h e s a m e quality of service. We will m a k e t h e c o n c e p t of QoS m o r e precise.

369

Control of Networks

370 Quality of service Best

Guaranteed low loss (e.g., Ι Ο ) and small delay (e.g., 1 ms + propagation time) 10

Guaranteed bounded loss (e.g., Ι Ο ) and delay (e.g., 100 ms + propagation time) 4

+

Best effort

Worst 83 FIGURE

Future networks will be able to offer a wide range of qualities of service from low loss and low delay to best effort.

In t h e b e s t case, t h e QpS g u a r a n t e e s v e r y small cell (or packet) loss rate a n d delay. By v e r y small cell loss rate, w e m e a n a loss rate c o m p a r a b l e to t h e loss rate d u e to u n a v o i d a b l e t r a n s m i s s i o n errors. For example, s u p p o s e t h e bit e r r o r rate along t h e fibers is a b o u t 1 0 ~ a n d t h a t a cell h a s 424 bits (53 b y t e s ) . T h e n t h e fraction of cells lost b e c a u s e of t r a n s m i s s i o n e r r o r s is o n t h e o r d e r of 424 χ 1 0 ~ ^ 1 0 ~ . T h i s is t h e least cell loss rate t h a t could b e p r o m i s e d to users. T h e smallest d e l a y t h a t could b e p r o m i s e d w o u l d b e c o m p a r a b l e to t h e propagation delay. For example, t h e p r o p a g a t i o n t i m e of a cell from San Francisco to Boston is o n t h e o r d e r of 10 m s . T h u s , in t h e b e s t case, t h e n e t w o r k could p r o m i s e a cell loss rate of a b o u t 1 0 a n d a delay o n t h e o r d e r of 10 m s for cells going from San Francisco to Boston. 12

12

10

- 1 0

T h e n e t w o r k could p r o p o s e inferior QpS. T h e worst quality is t h e so-called best-effort service, w h e r e t h e n e t w o r k p r o m i s e s to deliver t h e cells o n l y if it finds t h e r e s o u r c e s to do so. T h e r a n g e of QpS c a n b e a r r a n g e d from b e s t to worst as in Figure 8.3. T h e s e different levels of QpS are similar to t h e different grades of service offered b y t h e postal s y s t e m . T h e u s e r s c a n c h o o s e o v e r n i g h t delivery, express mail, first-class mail, a n d so on, d o w n to t h e l o w e s t grade of b u l k mail. T h e u s e r selects t h e QpS b a s e d o n t h e application. For instance, a u s e r mailing a large n u m b e r of a d v e r t i s e m e n t s m i g h t b e satisfied if 9 5 % of t h e c u s t o m e r s receive t h e m w i t h i n a few w e e k s . J u s t as in t h e postal s y s t e m , u s e r s will select from t h e m e n u of QpS offered b y t h e n e t w o r k t h a t service w h i c h b e s t m e e t s t h e n e e d s of t h e i r application. I n Figure 8.4, t h e n e e d s of several applications are displayed in a QpS p a r a m e t e r space of t h r e e d i m e n s i o n s : rate ( b a n d w i d t h ) , loss, a n d delay. T h e r e are t w o

8.1

Objectives and Methods of Control

371

Rate

Delay Loss 8.4 FIGURE

The quality of service specifies a n u m b e r of parameters for the service. The figure illustrates three of the parameters and sketches their acceptable ranges of values for different types of services.

points to r e m e m b e r . First, quality of service c a n h a v e m a n y aspects: in addition to t h e t h r e e listed in t h e figure, o n e m a y i n c l u d e security, reliability, a n d availability of t h e c o n n e c t i o n . However, rate, loss, a n d d e l a y are t h e critical aspects for m o s t applications. Second, t h e n e t w o r k m u s t b e controlled so t h a t it will provide different QpS to different virtual circuits ( u s e r s or applications) at t h e s a m e t i m e . Providing t h e b e s t QoS for all t h e c o n n e c t i o n s is wasteful, like s e n d i n g all j u n k m a i l b y express delivery, a n d it is therefore i m p o r t a n t t h a t t h e n e t w o r k b e able to h a n d l e different c o n n e c t i o n s differently. T h e benefits of s u p p o r t i n g m a n y services o n o n e c o m m o n n e t w o r k ( d u e to e c o n o m i e s of scale a n d scope a n d n e t w o r k externality) a p p e a r to o u t w e i g h t h e costs of t h e i n c r e a s e d complexity. T h e g u a r a n t e e of quality of service across a set of i n t e r c o n n e c t e d n e t w o r k s is a n i m p o r t a n t b u t c o m p l i c a t e d issue. I n a first a p p r o x i m a t i o n , t h e rate t h r o u g h a series of n e t w o r k s is t h e m i n i m u m of t h e rates, t h e d e l a y is t h e s u m of t h e delays t h r o u g h t h e individual n e t w o r k s , a n d t h e loss r a t e is also a p p r o x i m a t e l y t h e s u m of t h e individual loss rates. T h e actual situation is m o r e c o m p l i c a t e d b e c a u s e of t h e n e e d e d interfaces b e t w e e n n e t w o r k s a n d b e c a u s e t h e delay, loss, a n d rate are n o t m e a s u r e d b y scalar q u a n t i t i e s . It is useful to t h i n k of t h e relation b e t w e e n t h e n e t w o r k a n d a u s e r as b e i n g ideally defined b y a (service) contract. T h e c o n t r a c t obligates t h e n e t w o r k to transfer t h e u s e r ' s i n f o r m a t i o n w i t h a defined QoS (delay, loss, rate, etc.) provided t h a t t h e user's traffic c o n f o r m s to specified limits (bit rate, b u r s t i n e s s , etc.). T h u s t h e contract says t h a t t h e n e t w o r k will provide a p a r t i c u l a r QoS

Control of Networks

372

provided t h e user's traffic conforms to specified conditions. With this view, t h e objective of n e t w o r k control is to fulfill t h e largest set of contracts.

8.2

CIRCUIT-SWITCHED NETWORKS We first explain, u s i n g a v e r y s i m p l e n e t w o r k , t h a t t h e m o s t i m p o r t a n t QoS for circuit-switched n e t w o r k s is t h e blocking probability. We t h e n s h o w h o w t h e blocking probability for circuit-switched n e t w o r k s d e p e n d s o n r o u t i n g a n d a d m i s s i o n control.

8.2.1

Blocking Figure 8.5 illustrates a simple circuit-switched n e t w o r k . T h e n e t w o r k consists of two g r o u p s of t e l e p h o n e sets in two buildings. T h e t e l e p h o n e sets in e a c h building are c o n n e c t e d to a switch (a PBX, or private b r a n c h exchange) a n d t h e two PBXs are c o n n e c t e d b y two lines. You see t h a t it is possible for a call to b e r e q u e s t e d b e t w e e n t h e two b u i l d i n g s w h e n t h e two lines b e t w e e n t h e s e buildings are b u s y . Using a s m a l l e r n u m b e r of lines b e t w e e n t h e buildings t h a n m i g h t b e n e e d e d in t h e worst case l e a d s to t h e possibility of calls b e i n g blocked. Of course, in a n y realistic situation it w o u l d b e impractical to u s e t h e m a x i m u m possible n u m b e r of lines. For

Shared "trunk" lines —• possibility of blocking

8.5 ^ J ^ "

A small telephone network that consists of two switches and a few telephone sets. It m a y h a p p e n that all the lines b e t w e e n the switches are busy w h e n a user attempts to place a n e w call. When this occurs, the n e w call is blocked.

8.2

Circuit-Switched Networks

373

instance, if t h e two buildings h a d 100 t e l e p h o n e sets each, t h e n 100 l i n e s w o u l d b e n e e d e d b e t w e e n t h e two b u i l d i n g s to allow all t h e t e l e p h o n e sets to b e u s e d at t h e s a m e t i m e for calls b e t w e e n t h e buildings. In practice, t h e likelihood of s u c h a situation is exceedingly small, so t h a t o n e m a y r e a s o n a b l y design t h e t e l e p h o n e s y s t e m w i t h a s m a l l e r n u m b e r of lines. T h e p r o b l e m faced b y t h e n e t w o r k d e s i g n e r is to select t h e n u m b e r of lines b e t w e e n t h e buildings so t h a t t h e blocking probability is small e n o u g h . We sketch t h e m a i n features of t h e m e t h o d s t h a t t h e n e t w o r k d e s i g n e r uses. We p r e s e n t t h e u n d e r l y i n g m a t h e m a t i c a l derivations in t h e n e x t chapter. Consider t h e n e t w o r k of Figure 8.6, w h i c h s u m m a r i z e s t h e basic m o d e l t h a t n e t w o r k e n g i n e e r s u s e to a n a l y z e r o u t i n g in circuit-switched n e t w o r k s . Switches are c o n n e c t e d b y g r o u p s of circuits called links (or t r u n k s ) . For instance, link i is c o m p o s e d of n* circuits. A circuit provides t h e b a n d w i d t h to t r a n s m i t o n e p h o n e call. A r o u t e r is a p a t h in t h e n e t w o r k from o n e u s e r to a n o t h e r user. A call in progress along a r o u t e u s e s exactly o n e circuit o n e a c h link along t h e route. T h e r e are R r o u t e s . As c u s t o m e r s place t e l e p h o n e calls, t h e n e t w o r k u s e s s o m e r o u t i n g algo­ r i t h m to select t h e r o u t e s for t h e calls. C o n s e q u e n t l y , calls are p l a c e d along t h e various r o u t e s at r a n d o m t i m e s . T h e s e calls h a v e variable durations. For instance, a s s u m e t h a t t h e n e t w o r k r o u t e s N calls along a p a r t i c u l a r r o u t e r e v e r y m i n u t e , o n average. A s s u m e also t h a t t h e average d u r a t i o n of a call is Τ m i n u t e s . With t h e s e a s s u m p t i o n s , w e w o u l d expect t h a t v :=N χ Τ calls are in progress at a typical t i m e . Indeed, in Τ m i n u t e s , a b o u t v calls are placed along r o u t e r, a n d t h e s e calls t h e n t e r m i n a t e a n d are r e p l a c e d b y o t h e r calls. Because r

r

r

r

Control of Networks

374

of t h e variability in t h e t i m e s w h e n calls are p l a c e d a n d in t h e call d u r a t i o n s , t h e actual n u m b e r X of calls along r o u t e r m a y b e s o m e w h a t larger or s m a l l e r r

t h a n v . Consider n o w a p a r t i c u l a r l i n k i w i t h its ni circuits. T h e n u m b e r Y{ of r

calls carried b y l i n k i is t h e total n u m b e r of calls along r o u t e s t h a t go t h r o u g h link i. D e n o t e b y Ri t h e set of r o u t e s r t h a t go t h r o u g h l i n k i. T h e n ,

reRi

We expect Y{ to b e a p p r o x i m a t e l y e q u a l to

Y^

reRi v , b u t w e k n o w t h a t Y{ h a s r

s o m e probability of b e i n g larger t h a n t h a t value. W h e n Yj exceeds n; for s o m e link i along r o u t e r, t h e call is blocked. T h i s discussion s h o w s t h a t t h e probability t h a t a call p l a c e d along r o u t e

r will b e b l o c k e d b e c a u s e all t h e circuits of o n e of t h e l i n k s along t h e r o u t e are b u s y is s o m e function of t h e r a t e s {v\ . . . , VR}. We d e n o t e t h a t function }

b y B (v\, r

. . . , VR). We explain a l g o r i t h m s for calculating t h e function B

r

in

C h a p t e r 9.

8.2.2

Routing Optimization We n o w formalize t h e b e s t call-routing decision as a solution to a n o p t i m i z a t i o n p r o b l e m . We are given t h e n e t w o r k topology, t h a t is, t h e n u m b e r of circuits ni of e a c h l i n k i b e t w e e n a pair of switches. We are also given t h e r a t e λ β of call Α

r e q u e s t s (per u n i t of t i m e ) b e t w e e n e v e r y pair (A, B) of c u s t o m e r locations. T h e r o u t i n g p r o b l e m is to select a r o u t e for e a c h call so as to m a x i m i z e t h e n e t w o r k r e v e n u e s . We will explain several f o r m u l a t i o n s of t h e r o u t i n g p r o b l e m d e p e n d i n g o n t h e i n f o r m a t i o n available a n d t h e c o m p u t a t i o n a l effort o n e is willing to s p e n d .

Static Routing O n e possible formulation of t h e r o u t i n g p r o b l e m is to decide t h a t a call b e t w e e n u s e r s at A a n d Β is r o u t e d along r o u t e r\ w i t h probabilityp(AB, r\), a l o n g r o u t e T2 w i t h probability p(AB, rz), a n d so o n . T h u s , a r o u t e r is s e l e c t e d w i t h probability p(AB, r) for a call from A to B, a n d t h e call is r o u t e d along t h a t r o u t e if t h e r e is a free p a t h along t h a t r o u t e . O t h e r w i s e , t h e call is blocked. With this specific r o u t i n g a s s i g n m e n t , w e c a n calculate t h e r a t e v of call r e q u e s t s along e v e r y r o u t e r from t h e r a t e s X B- We c a n t h e n calculate t h e rate of n e t w o r k r e v e n u e s r

A

W := Σ

^rV [l - B (Vi, r

r

. . . , VR)],

8.2

Circuit-Switched Networks

375

w h e r e w is t h e cost rate for calls along r o u t e r a n d v [ l - B ] is t h e rate of calls t h a t r o u t e r actually carries b e c a u s e t h e y are n o t blocked. T h u s , given t h e static r o u t i n g decisions—the probabilities p(AB, r ) , for ev­ e r y pair AB a n d r o u t e r—we c a n calculate t h e n e t w o r k r e v e n u e s W. A good static r o u t i n g algorithm m u s t c o m e u p w i t h r o u t i n g decisions t h a t yield large n e t w o r k r e v e n u e s . O n e m e t h o d t h a t c a n b e u s e d to d e v e l o p a good r o u t i n g al­ g o r i t h m is to design a n i m p r o v e m e n t s t e p t h a t specifies h o w t h e probabilities p(AB r) c a n b e modified to i n c r e a s e W. If t h e i m p r o v e m e n t s t e p is well de­ signed, t h e n its successive application s h o u l d l e a d to r o u t i n g probabilities t h a t result in a large W. r

r

r

f

I m p r o v e m e n t steps b a s e d o n p r o p e r t i e s of t h e g r a d i e n t of W w i t h r e s p e c t to t h e r o u t i n g probabilities h a v e b e e n p r o p o s e d in t h e literature.

Dynamic Alternate Routing In static routing, a call is b l o c k e d if t h e r e is n o idle p a t h along t h e assigned route, e v e n if t h e r e is a n idle p a t h along a different route. T h e d y n a m i c a l t e r n a t e r o u t i n g algorithm takes t h e n e t w o r k c o n g e s t i o n into a c c o u n t in deciding h o w a call s h o u l d b e routed. In its basic form, t h e a l t e r n a t e r o u t i n g a l g o r i t h m u s e s a table t h a t p r o p o s e s two routes, r\(A, B) a n d Τ2(Α, Β), for e v e r y pair (A, B) of c u s t o m e r locations. For instance, t h e s e could b e t h e s h o r t e s t t w o r o u t e s b e t w e e n t h e locations. A n i n c o m i n g call of t y p e (A, B) is first assigned r o u t e r\. If t h e r e is n o idle p a t h along this route, t h e call is assigned r o u t e Γ2. If t h e r e is n o idle p a t h along V2, t h e call is blocked. T h e advantage of this a l g o r i t h m over t h e static r o u t i n g a l g o r i t h m is t h a t r o u t e selection d e p e n d s o n n e t w o r k congestion. (Of course, in o r d e r to i m p l e ­ m e n t this algorithm, t h e state of n e t w o r k c o n g e s t i o n m u s t b e k n o w n . ) O n e disadvantage of t h e basic d y n a m i c a l t e r n a t e r o u t i n g a l g o r i t h m is metastability, as w e explain next.

Metastability and Trunk Reservations Metastability is t h e p r o p e r t y of a set of states of a d y n a m i c s y s t e m t h a t c a n persist for a long t i m e . W h e n a circuit-switched n e t w o r k u s e s d y n a m i c a l t e r n a t e routing, a set of states c o r r e s p o n d i n g to a c o n g e s t e d n e t w o r k c a n b e metastable. This possibility is illustrated in Figure 8.7. Consider t h e n e t w o r k s h o w n in t h e left p a r t of t h e figure. It is a s y m m e t r i c n e t w o r k w i t h t h r e e u s e r locations, A, B, C, a n d t h r e e switches c o n n e c t e d b y links w i t h η circuits each. A s s u m e t h a t t h e call r e q u e s t s are also s y m m e t r i c . Specifically, a s s u m e t h a t calls b e t w e e n a n y pair of locations are m a d e w i t h

Control of Networks

376 Probability density Desirable states

Undesirable states Number of busy circuits

Λ —

Infrequent transitions

8.7 FIGURE

Illustration of metastability. The network on the left is symmetric and calls P l d i t h rate λ from each node. Each call is equally likely to be for each of the two other nodes. If the network routes a call w h e n e v e r it finds either a direct path or an indirect (alternate) path, then the probability density of the n u m b e r of busy circuits is as shown in the figure on the right. This figure illustrates the persistence (metastability) of undesirable network states.

a r e

a c e

w

rate λ. T h e n e t w o r k u s e s t h e a l t e r n a t e r o u t i n g algorithm. For e v e r y pair t h e preferred r o u t e r\ is t h e direct r o u t e t h r o u g h o n e link, a n d t h e a l t e r n a t e r o u t e Ύ2 is t h e indirect r o u t e t h r o u g h t w o links. A s s u m i n g t h a t t h e call d u r a t i o n s are i n d e p e n d e n t a n d e x p o n e n t i a l l y dis­ tributed, t h e vector of n u m b e r of circuits b u s y in t h e t h r e e links is a Markov chain. (See C h a p t e r 9 for t h e t h e o r y of Markov chains.) We c a n calculate t h e steady-state probabilities of t h a t Markov c h a i n b y solving t h e b a l a n c e equa­ tions. T h e result of this calculation is s h o w n in t h e right part of t h e figure. T h e plot shows t h e steady-state probability distribution of t h e total n u m b e r of b u s y circuits. T h i s total n u m b e r c a n r a n g e from 0 to 3n. T h e plot s h o w s two groups of values w i t h a large probability: o n e g r o u p of small values a n d o n e g r o u p of large values. T h e analysis (not s h o w n h e r e ) of t h e t i m e evolution of t h e net­ w o r k reveals t h a t t h e n u m b e r of b u s y circuits c a n r e m a i n large for a long t i m e before it gets smaller. T h a t is, t h e state of t h e n e t w o r k c a n r e m a i n for a long t i m e in a set of u n d e s i r a b l e states w h e r e m a n y circuits are b u s y . T h i s set of states is a l m o s t stable a n d is said to b e metastable. We c a n explain t h e metastability of c o n g e s t e d states as follows. A s s u m e t h a t t h e n e t w o r k is v e r y c o n g e s t e d a n d t h a t a n e w call is r e q u e s t e d . Since t h e n e t w o r k is congested, it is unlikely t h a t t h e n e w call c a n b e r o u t e d along t h e shortest route. As a result, it is likely t h a t this n e w call will r e q u i r e two circuits a n d t h e r e b y i n c r e a s e t h e congestion e v e n further. T^unk r e s e r v a t i o n provides a n effective p r o t e c t i o n against t h e metastabil­ ity of u n d e s i r a b l e c o n g e s t e d states. W h e n t h e r o u t i n g a l g o r i t h m u s e s t r u n k

8.2

Circuit-Switched Networks

377

reservations, a n u m b e r of circuits are r e s e r v e d for r o u t i n g calls along s h o r t e s t routes. Let u s describe h o w t r u n k r e s e r v a t i o n w o u l d b e i m p l e m e n t e d in t h e n e t w o r k of t h e figure. First o n e c h o o s e s a n u m b e r m s m a l l e r t h a n n. W h e n a n e w call b e t w e e n A a n d Β is r e q u e s t e d , t h e r o u t i n g a l g o r i t h m first tries to r o u t e t h e call along t h e shortest route, u s i n g t h e direct link. If this is n o t possible, t h e n t h e algorithm a t t e m p t s to r o u t e t h e call along t h e indirect route. T h a t r o u t e will b e u s e d as long as t h e r e are at least m circuits b e t w e e n A a n d C a n d b e t w e e n C a n d Β t h a t are e i t h e r u s e d b y shortest-route calls or t h a t are free. T h u s , m circuits of e a c h link are r e s e r v e d for r o u t i n g calls along t h e i r s h o r t e s t route. This r e s e r v a t i o n avoids t h e n e t w o r k r e a c h i n g a state w h e r e m o s t calls are r o u t e d along indirect routes, t h e r e b y u s i n g t h e n e t w o r k links inefficiently. See t h e r e f e r e n c e s cited in section 8.6 for a m o r e detailed discussion of r o u t i n g in circuit-switched n e t w o r k s .

Separable Routing Separable r o u t i n g is also a d y n a m i c r o u t i n g algorithm, since t h e r o u t e for a n e w call is c h o s e n o n t h e basis of t h e n e t w o r k congestion. T h e separable r o u t i n g a l g o r i t h m calculates its r o u t i n g decisions b y per­ forming o n e policy i m p r o v e m e n t s t e p starting w i t h t h e b e s t static r o u t i n g algorithm. Tb u n d e r s t a n d h o w this s t e p is p e r f o r m e d , c o n s i d e r t h e following situation. Say t h a t t h e n e t w o r k is initially in state N. H e r e Ν is a vector t h a t specifies h o w m a n y calls are b e i n g carried along e v e r y possible r o u t e in t h e n e t w o r k . D e n o t e b y V(N, t) t h e e x p e c t e d n e t w o r k r e v e n u e s b e t w e e n t i m e 0 a n d t i m e t w h e n t h e b e s t static r o u t i n g a l g o r i t h m is used. N o w a s s u m e t h a t a n e w call is r e q u e s t e d a n d t h a t two routes, say 1 a n d 2, c a n b e u s e d for this call. If r o u t e 1 is used, t h e n t h e state will j u m p from Ν to s o m e o t h e r value N' to indicate t h a t t h e r e is a n e w call b e i n g carried along r o u t e 1. If r o u t e 2 is used, t h e state j u m p s from Ν to N". W h i c h is t h e b e t t e r decision? T h e policy i m p r o v e m e n t s t e p d e c i d e s t h a t r o u t e 1 is preferable if V(N', t) > V(N", t) for all large r a n d t h a t r o u t e 2 is preferable o t h e r w i s e . Note t h a t t h e policy i m p r o v e ­ m e n t s t e p a s s u m e s t h a t t h e static r o u t i n g a l g o r i t h m is u s e d at all s u b s e q u e n t t i m e s in o r d e r to c o m p a r e t h e initial decisions. T h e crucial q u e s t i o n is h o w w e c a n c o m p a r e V(N' t) a n d V(N", r). T h a t is, w e h a v e to evaluate t h e i m p a c t of a n additional call o n t h e n e t w o r k r e v e n u e s . T h i s additional call h a s a finite h o l d i n g t i m e . D u r i n g its h o l d i n g time, t h e additional call i n c r e a s e s t h e likelihood t h a t s u b s e q u e n t calls are blocked, w h i c h r e d u c e s t h e total n e t w o r k r e v e n u e s . T h e effect of o n e call o n t h e b l o c k i n g of o t h e r calls c a n b e evaluated. t

Control of Networks

378 Admission Control

So far w e h a v e c o n s i d e r e d t h e r o u t i n g decision, a s s u m i n g a call is accepted. T h a t is, w e h a v e s h o w n h o w t h e blocking probabilities of t h e calls d e p e n d o n t h e routing decisions. H o w s h o u l d t h e n e t w o r k decide w h i c h calls s h o u l d b e a c c e p t e d ? T h e future n e t w o r k s will c a r r y m a n y different t y p e s of calls: audio, voice, data, video, a n d so on. T h e s e calls differ in t e r m s of t h e i r r e s o u r c e n e e d s , r e v e n u e g e n e r a t e d , r e q u e s t rate, a n d call duration. C o n s e q u e n t l y , a good a d m i s s i o n policy m u s t b a s e its a d m i s s i o n decision o n t h e set of calls c u r r e n t l y carried b y t h e n e t w o r k a n d o n t h e t y p e of call b e i n g r e q u e s t e d . We e x a m i n e a s i m p l e m o d e l of this a d m i s s i o n control p r o b l e m in sec­ tion 8.4. We also p r e s e n t a m o r e c o m p l e x m o d e l in C h a p t e r 9.

8.3

DATAGRAM NETWORKS W h e n a n e t w o r k t r a n s p o r t s data as datagrams, it first d e c o m p o s e s t h e data into packets of variable size. T h e p a c k e t s are t h e n s e n t o n e b y o n e from n o d e to n o d e along a p a t h to t h e p a c k e t destination. Each p a c k e t c o n t a i n s a special field t h a t specifies its destination address. T h u s , in a d a t a g r a m n e t w o r k , t h e n o d e s store a n d forward t h e individual packets, o n e at a t i m e . T h e r o u t e s t a k e n b y successive p a c k e t s m a y b e different, e v e n if t h e y go from t h e s a m e source to t h e s a m e destination. Also, b e c a u s e t h e p a c k e t sizes m a y b e different, as suggested in Figure 8.8, t h e t r a n s m i s s i o n t i m e s of t h e p a c k e t s are different, since t h e y are e q u a l to t h e p a c k e t l e n g t h s divided b y t h e t r a n s m i s s i o n rate. In o u r s t u d y of d a t a g r a m n e t w o r k s w e b e g i n b y describing a m o d e l t h a t w e will u s e to formulate t h e control q u e s t i o n s .

8.3.1

Queuing Model T h e q u e u i n g m o d e l of Figure 8.9 c a n b e u s e d b y t h e d e s i g n e r to predict t h e t r a n s m i s s i o n delays a n d to design good routing a n d flow-control algorithms. T h e top p a r t of t h e figure s h o w s o n e n o d e w i t h its c o m p o n e n t s : a r e c e i v e r c o n v e r t s t h e optical signal o n t h e i n c o m i n g fiber into packets. T h e p a c k e t s are stored into m e m o r y a n d are t h e n r e t r a n s m i t t e d o n o n e outgoing fiber. T h e b o t t o m p a r t of t h e figure s h o w s a n abstract r e p r e s e n t a t i o n of t h e s a m e node. T h e packets are v i e w e d as "customers" or "jobs," in t h e l a n g u a g e of q u e u i n g theory, arriving at r a n d o m t i m e s into a q u e u e . T h e c u s t o m e r s wait for t h e i r t u r n to b e served, a n d t h e service t i m e s are r a n d o m . T h e service t i m e is t h e l e n g t h of t h e p a c k e t in bits divided b y t h e t r a n s m i s s i o n bit rate. T h e

83

Datagram Networks

379

Packet-switching node

Packets: random size store and forward 8.8 FIGURE

Datagram networks. Packets of different length are transported in a store-and-forward manner.

service t i m e is r a n d o m w h e n t h e p a c k e t l e n g t h is r a n d o m . If t h e p a c k e t s all h a v e t h e s a m e size, as in ATM, t h e service t i m e is d e t e r m i n i s t i c . T h u s , t h e fluctuations of t h e arrival t i m e s a n d of t h e t r a n s m i s s i o n t i m e s are m o d e l e d b y r a n d o m variables. T h e specific a s s u m p t i o n s a b o u t t h e distributions of t h e s e r a n d o m variables d e p e n d o n t h e precise m o d e l b e i n g u s e d .

Key Queuing Result

8.3.2

T h e m o s t useful result of q u e u i n g t h e o r y for t h e analysis of d a t a g r a m n e t w o r k s c o n c e r n s t h e n e t w o r k s h o w n in Figure 8.10. T h a t result is t h e formula for t h e average delay p e r p a c k e t in s u c h a n e t w o r k .

Receiver Fiber — { ] V

Transmitter System

[ Memory

Random arrivals

[r-j

Fibers

Queue •ΟΟ(

Model

Random service times 8.9 FIGURE

Packet-switching node and queuing model.

Control of Networks

380 Average rate through queue (packets/s) "Ί

8.10 «ββΜΐ

|

Average service rate (packets/s)

Queuing network. The symbol λ; indicates the average rate of packets going through node i (in packets per second) while μ,· is the average transmission rate of that node, also in packets per second, w h e n the node is n o n e m p t y .

T h e n e t w o r k consists of a collection of n o d e s c o n n e c t e d b y links. T h e packets t h a t arrive from outside of t h e n e t w o r k are a s s u m e d to form Poisson processes, a n d t h e p a c k e t t r a n s m i s s i o n t i m e s in t h e various n o d e s are a s s u m e d to b e i n d e p e n d e n t a n d exponentially distributed. We u s e t h e following notation. At e a c h n o d e ;, packets arrive at rate λ p a c k e t s / s , a n d t h e service rate is μ p a c k e t s / s . T h u s μ is t h e t r a n s m i s s i o n rate in bits p e r s e c o n d divided b y t h e average p a c k e t size in bits. It is a s s u m e d t h a t μ > λ for a l l ; . T h e n t h e average delay faced b y a p a c k e t e n t e r i n g t h e n e t w o r k is ;

;

;

;

;

(8.1)

w h e r e γ is t h e total rate at w h i c h p a c k e t s arrive into t h e n e t w o r k . T h e a s s u m p t i o n s are n o t exactly satisfied in actual n e t w o r k s . For instance, t h e packet l e n g t h s are n o t e x p o n e n t i a l l y distributed. Typically, t h e p a c k e t s t h a t travel o n t h e I n t e r n e t t e n d to h a v e a b i m o d a l distribution: m o s t p a c k e t s are e i t h e r short or long, a n d t h e fractions of short a n d long p a c k e t s are n o t consistent w i t h a n e x p o n e n t i a l distribution. Also, since t h e l e n g t h of a p a c k e t does n o t c h a n g e as t h e p a c k e t travels t h r o u g h t h e n e t w o r k , t h e t r a n s m i s s i o n t i m e s of a p a c k e t at t h e different n o d e s are n o t i n d e p e n d e n t . In fact, if o n e k n o w s t h e t r a n s m i s s i o n t i m e of a p a c k e t at o n e node, t h e n o n e c a n d e t e r m i n e t h e l e n g t h of t h a t packet a n d therefore its t r a n s m i s s i o n t i m e s in all t h e o t h e r nodes. A l t h o u g h t h e s e a s s u m p t i o n s are n o t always valid, t h e formula for t h e average delay p e r p a c k e t provides a r e a s o n a b l y good e s t i m a t e of t h e actual

83

Datagram Networks

381

value of t h a t average d e l a y in a real n e t w o r k . T h i s s i m p l e formula is t h e starting p o i n t for t h e c o n s t r u c t i o n of r o u t i n g a n d flow-control algorithms. ( T h e formula is usually conservative, i.e., t h e actual d e l a y is s m a l l e r t h a n t h a t p r e d i c t e d b y t h e formula.) We derive t h e delay formula (8.1) i n section 9.3.1.

83.3

Routing Optimization We explain h o w t h e delay formula is u s e d to design good r o u t i n g strategies. We first consider static routing.

Static Routing Consider t h e s i m p l e n e t w o r k illustrated in Figure 8.11. Packets arrive w i t h rate λ. T h e n e t w o r k c a n s e n d t h e s e p a c k e t s along two different r o u t e s to t h e i r c o m m o n destination. Each r o u t e consists of o n e n o d e , t h a t is, of o n e buffer e q u i p p e d w i t h a transmitter. O u r objective is to choose t h e fraction ρ of p a c k e t s t h a t s h o u l d b e s e n t along t h e first r o u t e so as to m i n i m i z e t h e average d e l a y p e r p a c k e t in t h e n e t w o r k . We a s s u m e t h a t t h e p a c k e t s arrive as a Poisson p r o c e s s a n d t h a t t h e i r l e n g t h s are i n d e p e n d e n t a n d e x p o n e n t i a l l y distributed. T h e t w o links u s e t r a n s m i t t e r s w i t h different rates. C o n s e q u e n t l y , t h e t w o n o d e s 1 a n d 2 are m o d e l e d as q u e u e s w i t h e x p o n e n t i a l service t i m e s w i t h different rates. Tb u s e formula (8.1), w e n e e d to identify t h e p a r a m e t e r s in t h a t formula. T h e p a r a m e t e r γ is t h e total rate of p a c k e t arrivals into t h e n e t w o r k . T h a t rate is λ. T h u s , γ = λ. T h e d e l a y formula c o n t a i n s a s u m over all t h e n e t w o r k n o d e s . For e a c h n o d e ) , λ d e s i g n a t e s t h e average rate of p a c k e t s going t h r o u g h ;

O

8.11 IGURE

p

*

ι

ρ

Static routing optimization. Packets are sent to node 1 with probability ρ independently of one another and to node 2 otherwise. The right part of the figure shows the average delay Τ per packet through the network as a function of p.

Control of Networks

382

t h a t node, a n d μ is t h e average service rate of t h a t n o d e . Here, λχ = λρ and λ = λ(1 — ρ). T h e service rate μ ι of t h e first node, in p a c k e t s p e r second, is e q u a l to t h e rate of t h e t r a n s m i t t e r in t h a t node, in bits p e r second, divided b y t h e average l e n g t h of a packet, in bits. T h e rate μ 2 is o b t a i n e d in a similar m a n n e r . Substituting t h e s e v a l u e s in t h e delay formula gives ;

2

T=-

1 λ

|

1

Hi-P)

μι-λρ

μ -λ(1-ρ)]' 2

T h e right p a r t of Figure 8.11 is a plot of t h e average delay Τ as a function of ρ for typical values of λ, μ ι , a n d μ . T h a t plot s h o w s t h a t t h e average delay p e r p a c k e t Τ is m i n i m i z e d for s o m e value p* of p. T h u s , t h e r e is a n o p t i m a l w a y of splitting t h e traffic b e t w e e n t h e two routes. We n o w consider t h e g e n e r a l case. T h e o p t i m a l static r o u t i n g p r o b l e m for a general d a t a g r a m n e t w o r k is to m i n i m i z e t h e average d e l a y p e r p a c k e t Τ w i t h r e s p e c t to all r o u t i n g probabilities. T h e n e t w o r k s e n d s a p a c k e t leaving n o d e i to n o d e ; w i t h probability py. Given t h e s e r o u t i n g probabilities w e c a n calculate t h e average rates of flow Xi t h r o u g h t h e n o d e s i b y solving t h e following flow-conservation e q u a t i o n s : 2

λ

ΐ = Yi + Σ j

)Pji> for a l i i .

k

In t h e s e equations, γι d e n o t e s t h e rate of arrivals of p a c k e t s from outside t h e n e t w o r k into n o d e i. T h e e q u a t i o n s say that, for e a c h i, t h e rate of flow λ* t h r o u g h n o d e i is e q u a l to t h e external arrival rate into t h a t n o d e γι p l u s t h e s u m over all n o d e s ;' of t h e fraction pji of t h e rate λ of flow leaving t h a t n o d e ; a n d b e i n g s e n t to n o d e i. If t h e n e t w o r k is o p e n , t h a t is, if all t h e p a c k e t s t h a t e n t e r t h e n e t w o r k c a n e v e n t u a l l y leave it, t h e n t h e flow-conservation e q u a t i o n s h a v e a u n i q u e solution {λ*, i = 1 , . . . , / } . ;

T h u s , given t h e external r a t e s {γι, i = 1 , . . . , / } , t h e flow-conservation equa­ tions e n a b l e u s to calculate t h e rates {λ;, i = 1 , . . . , / } as a function of t h e r o u t i n g probabilities {py}. O n c e t h e s e r a t e s are d e t e r m i n e d , w e c a n u s e o u r formula (8.1) to c o m p u t e t h e average delay T. Figure 8.12 s k e t c h e s t h e d e l a y Τ as a function of t h e routing probabilities py. T h e delay Τ is a c o m p l i c a t e d function of t h e r o u t i n g probabilities py. T h e m i n i m i z a t i o n of Τ does n o t result in a closed-form expression for t h e o p t i m u m routing probabilities. Instead, o n e m u s t u s e a n u m e r i c a l m i n i m i z a t i o n algo­ r i t h m . For instance, a g r a d i e n t projection a l g o r i t h m c a n b e u s e d to obtain t h e optimal routing probabilities.

8.3

Datagram Networks

383

Average delay Τ per packet through the network as a function of the routing probabilities

FIGURE

Dynamic Routing I n s t e a d of selecting (fixed) probabilities to r o u t e t h e p a c k e t s w h e n t h e y leave t h e n o d e s , w e c a n devise a d y n a m i c r o u t i n g a l g o r i t h m t h a t b a s e s t h e r o u t i n g decisions o n t h e actual backlogs of t h e n o d e s . We illustrate s u c h a n a l g o r i t h m for t h e s i m p l e n e t w o r k of Figure 8.13. T h e left p a r t of t h e figure s h o w s a n e t w o r k in w h i c h p a c k e t s arriving as a Poisson process w i t h rate λ c a n b e s e n t along o n e of two routes. Each of t h e t w o r o u t e s is m o d e l e d as a single n o d e w i t h e x p o n e n t i a l service t i m e s . We a s s u m e t h a t t h e r o u t i n g p r o b a b i l i t y ^ c a n b e b a s e d o n t h e q u e u e l e n g t h s x a n d x . T h a t is, w h e n a p a c k e t arrives, t h e r o u t i n g controller looks at t h e t w o q u e u e l e n g t h s x a n d x a n d s e n d s t h e p a c k e t along r o u t e 1 w i t h probability p(x , x ) a n d along r o u t e 2 o t h e r w i s e . T h e d e s i g n e r of t h e r o u t i n g a l g o r i t h m m u s t find t h e function pix , x ) that, w h e n u s e d b y t h e r o u t i n g algorithm, m i n i m i z e s t h e average d e l a y p e r p a c k e t 1

2

l

l

2

2

1

8.13

2

Dynamic routing. When the q u e u e lengths are x and x , an arriving packet is sent to queue 1 with probability ρ (χ , χ ). The graph in the right-hand part of the figure shows the function p(., .) that minimizes the average delay per packet through the network. l

1

2

2

Control of Networks

384

in t h e n e t w o r k . T h e solution, t h a t is, t h e b e s t function, is illustrated in t h e right p a r t of Figure 8.13. T h i s function specifies t h a t w h e n x is large a n d x is small, t h e packets s h o u l d b e s e n t along r o u t e 2, a n d vice versa. Moreover, if a p a c k e t s h o u l d b e s e n t along r o u t e 2 w h e n t h e q u e u e l e n g t h s are x a n d x , t h e n t h e s a m e r o u t i n g decision s h o u l d b e t a k e n w h e n x is larger or w h e n x is smaller. l

2

l

1

2

2

T h e proof of t h e s e intuitively obvious s t r u c t u r a l p r o p e r t i e s of t h e function p(x , x ) t u r n s o u t to b e r a t h e r involved, e v e n for s u c h a s i m p l e n e t w o r k . In fact, v e r y few s t r u c t u r a l results of this t y p e are k n o w n for m o r e c o m p l i c a t e d n e t w o r k s . Moreover, s u c h structural results do n o t a p p e a r to yield i m p r o v e d p r o c e d u r e s for calculating t h e o p t i m a l d y n a m i c r o u t i n g algorithm. l

2

We n o w t u r n to t h e case of a g e n e r a l n e t w o r k . T h e derivation of t h e o p t i m a l d y n a m i c r o u t i n g a l g o r i t h m for a n e t w o r k w i t h m a n y n o d e s is a formidable prob­ l e m t h a t is still b e y o n d t h e r e a c h of c u r r e n t a p p r o a c h e s . C o n s e q u e n t l y , approx­ i m a t i o n s are n e c e s s a r y . Moreover, t h e state of t h e n e t w o r k is n o t k n o w n instan­ taneously, so that, e v e n if it could b e derived, t h e o p t i m u m d y n a m i c r o u t i n g algorithm w o u l d n o t b e i m p l e m e n t a b l e . As a result of t h e s e limitations, n e t w o r k e n g i n e e r s h a v e d e v e l o p e d s i m p l e h e u r i s t i c s t h a t c a n b e i m p l e m e n t e d . TWO s u c h heuristics are t h e Bellman-Ford a l g o r i t h m a n d t h e distributed-gradient algorithm. Bellman-Ford Algorithm Figure 8.14 s u m m a r i z e s t h e s e t u p of t h e Bellman-Ford algorithm. T h e m o d e l is a n e t w o r k of n o d e s c o n n e c t e d b y links. T h e average delay o n e a c h link is e s t i m a t e d b y t h e c o r r e s p o n d i n g transmitter. O n e possible e s t i m a t i o n m e t h o d is for t h e t r a n s m i t t e r o n e a c h l i n k to k e e p track of t h e backlog in its buffer a n d to calculate t h e average d e l a y b y divid­ ing t h e total n u m b e r of bits stored i n t h e buffer b y t h e t r a n s m i s s i o n rate. T h e p r o p a g a t i o n t i m e of signals along t h e link c a n b e a d d e d for i m p r o v i n g t h e estimate. Tb explain t h e calculations p e r f o r m e d b y t h e Bellman-Ford algorithm, let u s a s s u m e t h a t t h e d e l a y dy o n t h e link from n o d e i to n o d e ; h a s b e e n e s t i m a t e d for all pairs of n o d e s . Let u s d e n o t e b y xt t h e m i n i m u m d e l a y b e t w e e n n o d e i a n d s o m e fixed destination. T h e m i n i m u m delay X{ m u s t satisfy t h e e q u a t i o n Xi = min{dij

+ Xj).

(8.2)

T h e s e e q u a t i o n s are of t h e form χ = F(x), w h e r e χ d e s i g n a t e s t h e vector w i t h c o m p o n e n t s X{. T h u s , t h e vector χ satisfies fixed-point e q u a t i o n s . T h e s e fixedp o i n t e q u a t i o n s c a n b e solved b y t h e r e c u r s i o n x =F(x ). n+1

n

83

Datagram Networks

Destination

8.14 einewnii

Setup of the Bellman-Ford algorithm. The problem is to find the shortest path from each node to the destination. It is assumed that each node knows the lengths of the links to which it is attached. Here, dy is the length of the link from node i to node ;, and it is assumed to be k n o w n b y node i.

It c a n b e s h o w n t h a t this r e c u r s i o n will converge to t h e vector of m i n i m u m delays for a n y n o n n e g a t i v e initial vector x°. T h e r e s u l t i n g a l g o r i t h m is called t h e Bellman-Ford algorithm. O n c e t h e m i n i m u m delays h a v e b e e n calculated, t h e fastest p a t h to t h e d e s t i n a t i o n is easily identified: a p a c k e t t h a t leaves n o d e ι s h o u l d b e s e n t to n o d e ; * if;* is t h e value of; t h a t a c h i e v e s t h e m i n i m u m in e q u a t i o n (8.2). Distributed-Gradient Algorithm T h e delays t h r o u g h t h e links of a n e t w o r k d e p e n d o n t h e traffic along t h o s e links. C o n s e q u e n t l y , a r o u t i n g algorithm c a n b e i m p r o v e d b y taking into a c c o u n t t h e effect of r o u t i n g decisions o n t h e traffic a n d therefore o n t h e delays. T h e distributed-gradient a l g o r i t h m e s t i m a t e s t h a t effect. T h e situation is illustrated in Figure 8.15. D e n o t e b y γι t h e rate of traffic e n t e r i n g t h e n e t w o r k via n o d e i a n d b y λ* t h e average rate of flow t h r o u g h n o d e i. Also, let dy r e p r e s e n t t h e d e l a y along link ij a n d Xi t h e average delay from n o d e i to a specified destination. By py w e designate t h e fraction of traffic leaving n o d e i t h a t is s e n t to n o d e ; . T h e p r o b l e m is to d e t e r m i n e t h e values of t h e s e r o u t i n g probabilities t h a t m i n i m i z e t h e average d e l a y s from t h e n o d e s to t h e destination. In this algorithm, in c o n t r a s t to t h e Bellman-Ford algorithm, t h e link delays are a s s u m e d to d e p e n d o n t h e r a t e s of traffic. T h e k e y idea of t h e a l g o r i t h m is e x p r e s s e d in t h e following formula:

Control of Networks

386

8.15

iiiiijimiiiii

Setup of the distributed-gradient algorithm. T h e problem is to find the u t i n g probabilities that minimize the average delay from each node to the destination. Packets arrive from outside of the network into node i with rate γ\. T h e average delay from node i to node ;', dy is a function of the rate of packets through that link.

r 0

w h e r e ky designates t h e rate of traffic t h r o u g h link ij. T h i s formula calculates t h e derivative of t h e d e l a y b e t w e e n n o d e i a n d t h e d e s t i n a t i o n w i t h r e s p e c t to pij. This formula m a y b e u n d e r s t o o d b y m u l t i p l y i n g b o t h of its sides b y €. W h e n Pij is i n c r e a s e d b y a small value e, t h e rate t h r o u g h t h e link ij i n c r e a s e s b y kie. This rate i n c r e a s e h a s t w o effects. First, it i n c r e a s e s t h e d e l a y along link ij b y k{€ m u l t i p l i e d b y t h e derivative of dy w i t h r e s p e c t to t h e traffic rate along t h e link ij. Second, it i n c r e a s e s t h e traffic rate t h r o u g h n o d e ;. T h e i n c r e a s e in t h e traffic t h r o u g h n o d e ; also i n c r e a s e s t h e d e l a y from n o d e ; to t h e d e s t i n a t i o n a n d will therefore i n c r e a s e t h e d e l a y from n o d e i to t h e d e s t i n a t i o n b y t h e s a m e a m o u n t . T h e formula expresses t h e s e t w o effects. Tb a p p r e c i a t e h o w t h e formula c a n b e used, let u s a s s u m e t h a t e a c h t r a n s m i t t e r c a n e s t i m a t e t h e derivative of t h e form

day dky' E q u a t i o n (8.3) t h e n provides a recursive p r o c e d u r e for e v a l u a t i n g t h e t e r m s dxi/dyt. Suppose n o d e i is a t t a c h e d to t h e d e s t i n a t i o n s b y o n e link is. T h e c o r r e s p o n d i n g e q u a t i o n (8.3) is t h e n s i m p l y

dpis

1

(since x = 0, X{ = di , b e c a u s e a p a c k e t originating at s a n d d e s t i n e d for s e n c o u n t e r s n o delay), a n d t h a t formula e n a b l e s u s to calculate s

S

8.3

Datagram Networks

387 1

dXi

dXi

In this way, t h e t e r m s o n t h e left-hand side of (8.3) c a n b e e v a l u a t e d for t h e n o d e s t h a t are directly a t t a c h e d to t h e destination. T h e s e v a l u e s c a n t h e n b e u s e d t o g e t h e r w i t h e q u a t i o n (8.3) to e v a l u a t e t h e t e r m s c o r r e s p o n d i n g to n o d e s t h a t are two links a w a y from t h e destination, a n d so on. Tb i m p l e m e n t this algorithm, t h e n o d e s m u s t b r o a d c a s t t h e v a l u e s of t h e i r g r a d i e n t e s t i m a t e s to t h e i r neighbors. Readers s h o u l d c o n s u l t t h e r e f e r e n c e s for details o n this algorithm. It s h o u l d b e n o t e d t h a t this a l g o r i t h m is subject to u n d e s i r a b l e oscillations. R e m e d i e s for t h e s e oscillations h a v e b e e n devised.

83.4

Congestion Control Congestion control is t h e n a m e of control p r o c e d u r e s t h a t throttle t h e flow of p a c k e t s along a p a t h to k e e p p a r t s of t h e n e t w o r k from b e c o m i n g excessively congested. Figure 8.16 illustrates t h e usefulness of c o n g e s t i o n control for a n e l e m e n ­ t a r y example. Consider a n o d e t h a t is u s e d b y two traffic s t r e a m s r e p r e s e n t e d b y t h e differently s h a d e d arrows. A s s u m e t h a t t h e traffic flowing o n t h e two top a r r o w s is g u a r a n t e e d a delay Τ less t h a n s o m e specified v a l u e T . Tb m e e t this g u a r a n t e e d b o u n d o n t h e delay, t h e s o u r c e of t h e traffic r e p r e s e n t e d b y t h e l o w e r a r r o w s m u s t b e p r e v e n t e d from e n t e r i n g t h e n o d e when more than c x Τγηαχ Mb are a l r e a d y buffered in t h a t n o d e . H e r e , c d e n o t e s t h e t r a n s m i s s i o n rate in Mbps. max

Source 8.16 FIGURE

x (Mb) t

Illustration of congestion control. Two types of packets go through the same node. We assume the packets are served in their order of arrival. If the packets of one type are guaranteed a b o u n d e d delay, t h e n the packets of the other type must be stopped w h e n the backlog exceeds a given size.

Control of Networks

388 w\,d t

h x

*1

R

*

3

*2 H> , < / , 3

3

i

3

D

3

Network for study of congestion.

8.17 FIGURE

T h e resulting congestion-control p r o c e d u r e for t h e traffic flowing along t h e l o w e r arrows is called a window congestion control. A m o r e involved e x a m p l e is explained next.

Window Congestion Control Figure 8.17 illustrates a set of i n t e r c o n n e c t e d r o u t e r s a n d hosts. O u r objective is to s t u d y end-to-end congestion-control m e c h a n i s m s . A l t h o u g h w e m a k e simplifying a s s u m p t i o n s a b o u t t h e d y n a m i c s of t h e n e t w o r k , t h e c o n c l u s i o n s of o u r analysis are r e l e v a n t to t h e p e r f o r m a n c e of actual n e t w o r k s . We a s s u m e t h a t t h e links all h a v e t h e s a m e rate of 1. T h r e e c o n n e c t i o n s s h a r e t h e n e t w o r k : from source Si to d e s t i n a t i o n Di, i = 1, 2, 3. T h e figure s h o w s t r a n s m i s s i o n s from e a c h d e s t i n a t i o n Di to t h e source Si. T h e s e t r a n s m i s s i o n s are of a c k n o w l e d g m e n t s . T h e following q u a n t i t i e s are defined for t h e con­ n e c t i o n from Si to Df. t h e propagation t i m e di, t h e t r a n s m i s s i o n rate Xi, t h e round-trip t i m e U, a n d t h e n u m b e r of bits in transit Wi. T h e r o u t e r s c a n store a finite n u m b e r of bits. If bits arrive at a r o u t e r t h a t is full, t h e n t h e bits are dropped. Since t h e s e bits do n o t arrive, t h e y are n o t a c k n o w l e d g e d a n d t h e source realizes, after s o m e delay, t h a t t h e bits w e r e d r o p p e d . T h e s e q u a n t i t i e s h a v e t h e following m e a n i n g . Source Si i m p l e m e n t s a w i n d o w congestion control t h a t limits t h e n u m b e r of bits in transit to Wi. T h e t r a n s m i s s i o n links t h a t t h e p a c k e t s from Si to Di a n d t h e i r a c k n o w l e d g m e n t s from Di to Si go t h r o u g h h a v e a total p r o p a g a t i o n t i m e e q u a l to di. T h a t is, w h e n t h e r e is v e r y little traffic in t h e n e t w o r k , t h e t i m e b e t w e e n t h e t r a n s m i s s i o n of a packet a n d t h e r e c e p t i o n of its a c k n o w l e d g m e n t b y Si is di. T h i s delay di is called t h e propagation t i m e of t h e c o n n e c t i o n . T h e round-trip t i m e U of t h e c o n n e c t i o n is e q u a l to t h e p r o p a g a t i o n t i m e di plus t h e q u e u i n g t i m e of t h e packets in t h e routers. If t h e links of t h e n e t w o r k are u s e d efficiently, t h e n x + χ = 1 a n d x\ + X3 = 1. Also, if t h e t h r e e c o n n e c t i o n s h a v e a fair s h a r e of t h e links, t h e n x\ = X2 = X3 = 0.5. T h e w i n d o w congestion control a l g o r i t h m is a Y

2

83

Datagram Networks

389

»«»^^ m e c h a n i s m to adjust t h e w i n d o w size W{ b a s e d o n t h e o b s e r v e d q u a n t i t i e s X{, t{. We a s s u m e t h a t t h e source gets a n o p p o r t u n i t y to m e a s u r e its p r o p a g a t i o n t i m e di b y k e e p i n g track of its m i n i m u m round-trip t i m e . T h e difficulty in designing a w i n d o w congestion control a l g o r i t h m is t h a t it m u s t w o r k for a source t h a t does n o t k n o w t h e n e t w o r k topology, t h e link rates, n o r t h e set of o t h e r c o n n e c t i o n s w i t h w h i c h it is s h a r i n g t h e n e t w o r k . Before w e e x a m i n e t h e c o m p l e t e situation of Figure 8.17, let u s first a s s u m e t h a t t h e source 53 is silent. In t h a t case, t h e r e are two c o n n e c t i o n s (.Si to D\ a n d S2 to D2) a n d only o n e r o u t e r (R\) c a n b e a b o t t l e n e c k . For this simplified situation, a s i m p l e algorithm h a s b e e n i m p l e m e n t e d . T h e s o u r c e s i n c r e a s e t h e i r t r a n s m i s s i o n rate until t h e y d e t e c t d r o p p e d bits. W h e n a s o u r c e d e t e c t s t h a t s o m e of its bits w e r e dropped, it r e d u c e s its t r a n s m i s s i o n rate b y s o m e factor. Figure 8.18 shows t h e evolution of t h e rates x\ a n d X2 w h e n t h e sources i m p l e m e n t t h a t algorithm. T h e left-hand figure s h o w s t h e evolution of t h e r a t e s if w e a s s u m e t h a t t h e two sources increase t h e i r t r a n s m i s s i o n rate at t h e s a m e rate a n d d e t e c t t h e losses at t h e s a m e time. W h e n t h e s u m of t h e rates x\ + X2 is less t h a n 1, t h e r o u t e r R\ is n o t c o n g e s t e d a n d it d o e s n o t d r o p bits. Accordingly, t h e s o u r c e s Si a n d S2 increase t h e i r t r a n s m i s s i o n rate a n d t h e vector (χι, x ) m o v e s along a 45degree line as b o t h c o m p o n e n t s i n c r e a s e at t h e s a m e rate. Eventually, t h e s u m of t h e rates x\ a n d X2 exceeds 1. At t h a t time, t h e buffer of t h e r o u t e r fills u p a n d starts d r o p p i n g packets. A s s u m i n g t h a t t h e two sources discover t h e losses at t h e s a m e time, t h e y r e d u c e t h e i r t r a n s m i s s i o n rates b y 50%. C o n s e q u e n t l y , t h e vector (*ι, X2) is r e p l a c e d b y {x\/2, X2/2), as Figure 8.18 shows. T h e r a t e s t h e n start increasing again, a n d t h e vector m o v e s as s h o w n in t h e left-hand p a r t of t h e figure. In this ideal case, t h e vector e v e n t u a l l y oscillates a r o u n d t h e value (1/2, 1/2). 2

8.18 FIGURE

Two connections that share a single bottleneck. Shown are an ideal case (left), different adjustment rates (center), and different adjustment delays (right).

Control of Networks

390

This algorithm is called additive increase and multiplicative decrease. T h e TCP congestion-avoidance m e c h a n i s m is of this t y p e . As long as a source does n o t d e t e c t a n y p a c k e t loss, it k e e p s o n i n c r e a s i n g its w i n d o w size linearly. W h e n it detects a loss, t h e source r e d u c e s its w i n d o w size b y a factor 2. T h e actual m e c h a n i s m t h a t t h e s o u r c e s u s e to i n c r e a s e t h e i r w i n d o w size results in a n increase rate t h a t is inversely p r o p o r t i o n a l to t h e round-trip t i m e of t h e c o n n e c t i o n . C o n s e q u e n t l y , this a l g o r i t h m exhibits t h e b e h a v i o r s k e t c h e d in t h e c e n t e r p a r t of Figure 8.18. In t h a t p a r t of t h e figure, o n e c o n n e c t i o n h a s a smaller round-trip t i m e a n d i n c r e a s e s its rate faster t h a n t h e o t h e r c o n n e c t i o n . T h e rates e v e n t u a l l y oscillate a r o u n d v a l u e s s u c h t h a t t h e c o n n e c t i o n w i t h t h e s m a l l e r propagation t i m e h a s a m u c h larger average t r a n s m i s s i o n rate. Moreover, in a real n e t w o r k , t h e s o u r c e s l e a r n of p a c k e t losses at different times, so t h a t t h e u p d a t e s look like t h o s e in t h e right-hand p a r t of t h e figure. T h e analysis of this a l g o r i t h m reveals t h a t e v e n for two c o n n e c t i o n s t h a t s h a r e a single bottleneck, t h e additive i n c r e a s e a n d multiplicative d e c r e a s e w i n d o w congestion algorithm t h a t w e j u s t d e s c r i b e d is b i a s e d in favor of t h e c o n n e c t i o n w i t h a s m a l l e r propagation t i m e . W h e n t h e r e is m o r e t h a n o n e b o t t l e n e c k , t h e situation is e v e n m o r e com­ plex. If w e a s s u m e t h a t e a c h source Si k n o w s Wi, Xi, ti, a n d di, t h e n a n a l g o r i t h m t h a t converges to t h e fair a n d efficient r a t e s h a s b e e n devised. T h i s a l g o r i t h m is as follows: d

di(Wi - Xidi - 1)

dt

t{Wi

—mit) = - a

.

In this expression, α is a p a r a m e t e r t h a t controls t h e s t e p size of t h e algorithm. T h e algorithm converges to a set of c o n n e c t i o n r a t e s t h a t are a "proportionally fair" e q u i l i b r i u m . By definition, a vector of r a t e s (x\,..., XN) is a proportionally fair e q u i l i b r i u m if c h a n g i n g t h e s e rates results in a negative s u m of t h e i r relative increases. T h a t is, if t h e modified r a t e s are (z/i, . . . , IJN), then N

l b prove t h e c o n v e r g e n c e result of t h e algorithm, o n e e x p r e s s e s t h e rela­ tions t h a t exist b e t w e e n t h e q u a n t i t i e s {χι, Wi, U, di}. O n e t h e n s h o w s t h a t t h e algorithm r e d u c e s t h e value of s o m e function of those q u a n t i t i e s t h a t is e q u a l to zero only at t h e limiting point. T h i s function is called a L y a p u n o v function for t h e s y s t e m . (See t h e references, section 8.6, for details.) I n practice, t h e algorithm explained above is n o t i m p l e m e n t e d for two r e a s o n s . First, t h e algo-

8.3

Datagram Networks

rithm assumes that the connections k n o w their round-trip propagation time. A l t h o u g h this t i m e c a n b e e s t i m a t e d , t h e e s t i m a t e t e n d s to b e noisy, b i a s e d against c o n n e c t i o n s t h a t start w h e n t h e n e t w o r k is a l r e a d y congested, a n d c a n b e affected b y r e r o u t i n g of c o n n e c t i o n s . Second, t h e a l g o r i t h m d o e s n o t com­ p e t e well w i t h c o n n e c t i o n s t h a t i m p l e m e n t t h e TCP c o n g e s t i o n control. O n e s i m p l e m e c h a n i s m h a s b e e n d e s i g n e d to correct t h e b i a s of t h e TCP congestion control. T h i s m e c h a n i s m is called random early drop, or RED, routers. A r o u t e r t h a t i m p l e m e n t s RED d r o p s i n c o m i n g p a c k e t s w i t h a proba­ bility t h a t is a function of t h e average r e c e n t buffer o c c u p a n c y . If t h a t average value, c o m p u t e d b y a low pass filter, is b e l o w a low threshold, t h e n t h e r o u t e r accepts t h e packet. If t h e average is larger t h a n a h i g h threshold, t h e n t h e r o u t e r d r o p s t h e packet. If t h e average is b e t w e e n t h e s e t h r e s h o l d s , t h e n t h e r o u t e r d r o p s t h e p a c k e t w i t h a probability t h a t i n c r e a s e s l i n e a r l y w i t h t h e average. T h e effect of this m e c h a n i s m is t h a t s o u r c e s l e a r n of t h e c o n g e s t i o n before t h e r o u t e r buffer is full a n d so get a c h a n c e to slow d o w n before facing multi­ ple successive losses. Moreover, RED is m o r e likely to d r o p p a c k e t s from faster c o n n e c t i o n s (since t h e y s e n d m o r e p a c k e t s ) . C o n s e q u e n t l y , faster c o n n e c t i o n s are m o r e likely to slow d o w n t h a n slow c o n n e c t i o n s , w h i c h s o m e w h a t corrects t h e TCP bias. M a n y variations o n RED h a v e b e e n p r o p o s e d . O n e of t h e s e variations, called weighted RED, classifies p a c k e t s into a n u m b e r of classes. T h e r o u t e r d r o p s a n i n c o m i n g packet as in RED, except t h a t t h e t h r e s h o l d v a l u e s t h a t t h e r o u t e r u s e s to c o m p u t e t h e d r o p probability d e p e n d o n t h e class of t h e packet. In a n o t h e r variation, called flow RED, t h e r o u t e r m a i n t a i n s c o u n t e r s w i t h t h e n u m b e r of p a c k e t s of t h e different classes a n d b a s e s t h e d r o p decision o n t h e n u m b e r of p a c k e t s of t h e class of t h e i n c o m i n g packet. Yet a n o t h e r variation, called explicit congestion notification, m a r k s t h e p a c k e t s i n s t e a d of d r o p p i n g t h e m . T h e m a r k is w r i t t e n in t h e p a c k e t header, a n d t h e d e s t i n a t i o n h o s t e c h o e s t h a t m a r k in t h e a c k n o w l e d g m e n t of t h e packet. T h e s o u r c e of t h e p a c k e t s adjusts its w i n d o w s b a s e d o n t h e m a r k s t h a t it receives in t h e acknowledgments.

Rate Congestion Control W i n d o w congestion control is n o t suitable w h e n t h e capacity of t h e fiber is utilized b y a small n u m b e r of sources. T h i s is b e c a u s e b y t h e t i m e t h e first packet's a c k n o w l e d g m e n t b y t h e d e s t i n a t i o n r e a c h e s t h e source, a v e r y large n u m b e r of p a c k e t s will h a v e b e e n t r a n s m i t t e d , a s s u m i n g t h e source does n o t stop in t h e m e a n t i m e . T h u s , b y t h e t i m e t h e d e s t i n a t i o n c a n signal t h e source

Control of Networks

392

t h a t s o m e congestion is occurring, it is probably too late for t h e source to slow d o w n its t r a n s m i s s i o n s . (See section 8.7, p r o b l e m 10.) In addition, w i n d o w congestion control necessitates t h e e s t a b l i s h m e n t of two-way c o n n e c t i o n s to t r a n s m i t a c k n o w l e d g m e n t s , a n d this c o m p l i c a t e s t h e o p e r a t i o n s of t h e n e t w o r k . I b avoid t h e s e p r o b l e m s , n e t w o r k r e s e a r c h e r s are proposing t h e u s e of rate-based congestion control. I n s t e a d of limiting t h e n u m b e r of p a c k e t s in t h e n e t w o r k s e n t b y e a c h source, a rate-based congestion control limits t h e average rate at w h i c h sources t r a n s m i t packets. This control is easier to i m p l e m e n t , b e c a u s e it only r e q u i r e s e a c h source to m o n i t o r its t r a n s m i s s i o n rate. For t h e I n t e r n e t , a s i m p l e rate-based congestion control c a n b e i m p l e ­ m e n t e d o n top of UDP to replace t h e w i n d o w - b a s e d congestion control of TCP. T h i s rate control o p e r a t e s as follows. T h e receiver s e n d s a m e s s a g e to t h e source t h a t specifies t h e rate at w h i c h it w a n t s t h e source to s e n d t h e packets. T h e source i m p l e m e n t s t h a t rate b y c o m p u t i n g a n d controlling a n i n t e r p a c k e t t i m e . T h e receiver c a n calculate w h e n it s h o u l d receive t h e s e r e q u e s t e d packets. If s o m e packet arrives late, t h e receiver c a n s e n d a modified t r a n s m i s s i o n rate to t h e source, say half t h e c u r r e n t rate. If t h e p a c k e t s arrive n o r m a l l y , t h e n t h e receiver c a n r e q u e s t a n i n c r e a s e d rate, say a l i n e a r increase. Tb protect this m e c h a n i s m against lost requests, t h e source s h o u l d stop s e n d i n g after s o m e n u m b e r of p a c k e t s if it d o e s n o t h e a r from t h e receiver. O n e a d v a n t a g e of this strategy over TCP is t h a t t h e b u r d e n is o n t h e receiver to calculate t h e rate u p d a t e s , t h u s m a k i n g t h e server simpler. A n o t h e r i m p l e m e n t a t i o n of rate-based congestion control, w h i c h is r e c o m ­ m e n d e d for ATM n e t w o r k s , is t h e leaky-bucket controller. T h i s control m e c h a ­ n i s m regulates t h e traffic b y s m o o t h i n g o u t t h e b u r s t s of p a c k e t s t h a t t h e source w o u l d o t h e r w i s e t r a n s m i t . We explain this m e c h a n i s m in t h e next section.

8.4

ATM NETWORKS In this section w e a t t e n d to t h e control of virtual circuit n e t w o r k s in g e n e r a l a n d of ATM n e t w o r k s in particular. We first outline t h e p r o b l e m s of control of virtual circuit n e t w o r k s . We explain t h a t t h e u s e r m u s t describe its traffic in a w a y t h a t c a n b e enforced a n d m o n i t o r e d . Moreover, t h e traffic description m u s t e n a b l e t h e n e t w o r k to control t h e traffic efficiently. We t h e n s t u d y results obtained b y u s i n g d e t e r m i n i s t i c m o d e l s . Such de­ t e r m i n i s t i c m o d e l s form t h e justification of m o s t c u r r e n t r e c o m m e n d a t i o n s

8.4

ATM Networks

393

for t h e control of ATM n e t w o r k s . In o u r view, this a p p r o a c h is u n n e c e s s a r i l y conservative. Its m a i n , a n d i m p o r t a n t , m e r i t is t h a t it is s i m p l e to i m p l e m e n t . We c o n c l u d e t h e c h a p t e r b y discussing statistical p r o c e d u r e s for specifying traffic characteristics a n d for controlling t h e n e t w o r k . We believe t h a t b y adopt­ ing s u c h p r o c e d u r e s , t h e n e t w o r k c a n derive significantly h i g h e r r e v e n u e s . However, t h e s e statistical p r o c e d u r e s r e q u i r e further study. O u r p r e s e n t a t i o n o u t l i n e s a few possibilities. We h o p e t h a t this section will invite n e t w o r k engi­ n e e r s to s t u d y s u c h m e t h o d s further.

8.4.1

Control Problems We explained in section 8.1 t h a t t h e u s e r a n d t h e n e t w o r k e n t e r into a contract in w h i c h t h e n e t w o r k g u a r a n t e e s a quality of service w h i l e t h e u s e r a s s u r e s t h a t t h e traffic will obey specific b o u n d s . Implicit in a n y c o n t r a c t are provisions for verifying t h a t t h e t e r m s of t h e c o n t r a c t are o b s e r v e d b y t h e various parties. T h u s , t h e n e t w o r k m u s t b e able to verify t h a t t h e traffic o b e y s its descriptors a n d t h e u s e r t h a t t h e quality of service is acceptable. T h e objective of t h e control is to m a x i m i z e t h e n e t w o r k r e v e n u e s w h i l e m e e t i n g t h e c o n t r a c t e d quality of service of t h e c o n n e c t i o n s . T h e quality of service specifies t h e loss rate a n d delay of t h e transfer of bits (or cells, or packets). O t h e r descriptors of t h e quality of service i n c l u d e s e c u r i t y a n d reliability. Specific t e m p o r a l characteristics of losses a n d of delays m a y b e i m p o r t a n t in particular applications. I n m u l t i m e d i a applications, for instance, two u s e r s m a y b e c o n n e c t e d b y a collection of virtual circuit c o n n e c t i o n s . T h e quality of service m i g h t specify b o u n d s o n t h e s y n c h r o n i z a t i o n offset b e t w e e n t h e virtual circuit c o n n e c t i o n s . As y o u see, e v e n t h e f o r m u l a t i o n of t h e p r o b l e m l e n d s itself to m a n y variations. In this chapter, w e limit o u r a t t e n t i o n to loss rate a n d delay. We also n o t e d in section 8.1 t h a t a virtual circuit n e t w o r k c a n exercise four t y p e s of control action: admission, routing, flow a n d c o n g e s t i o n control, a n d allocation of buffers a n d b a n d w i d t h . Tb p r o p e r l y design t h e control p r o c e d u r e s of t h e n e t w o r k , t h e n e t w o r k e n g i n e e r m u s t b e able to evaluate t h e p e r f o r m a n c e of specific control proce­ d u r e s . Consider, for example, t h e n e t w o r k s h o w n in Figure 8.19. T h e n e t w o r k e n g i n e e r s h o u l d b e able to d e t e r m i n e w h e t h e r t h e loss r a t e s a n d delays of t h e different virtual circuit c o n n e c t i o n s in this n e t w o r k are acceptable. T h e an­ swer to t h a t q u e s t i o n d e p e n d s o n t h e characteristics of t h e traffic g e n e r a t e d b y t h e different c o n n e c t i o n s , o n t h e flow-control a n d allocation strategies, o n t h e buffer capacities in t h e switches, o n t h e r a t e s of t h e t r a n s m i t t e r s , o n t h e

Control of Networks

394

8.19 IG U R Ε

Network control. The network engineer needs to determine whether a given set of connections can be carried by the network.

propagation delays a n d bit error r a t e s of t h e links, a n d o n t h e error-correction p r o c e d u r e s t h a t t h e n e t w o r k uses. T h e e n g i n e e r faces t h r e e major difficulties in t r y i n g to d e t e r m i n e t h e acceptability of given c o n n e c t i o n s . First, t h e u s e r s are often i n t e r e s t e d in v e r y small loss r a t e s a n d relatively h a r d b o u n d s o n delays or d e l a y jitter. T h e s e loss rates a n d delay b o u n d s are sensitive to t h e details of t h e traffic a n d are difficult to calculate e v e n for a single n o d e w i t h precise m o d e l s of t h e traffic. Second, t h e n e t w o r k o p e r a t o r n e v e r k n o w s precisely t h e characteristics of t h e traffic of c o n n e c t i o n s . For instance, e v e n t h e mean rates of c o m p r e s s e d video s t r e a m s vary substantially across m o v i e s . Third, t h e characteristics of t h e traffic of t h e virtual circuit c o n n e c t i o n s are modified w h e n t h e y go t h r o u g h n o d e s a n d interact with other connections. O n e a p p r o a c h to t h e s e c o m p l e x q u e s t i o n s is to play it safe a n d to a d o p t con­ servative p r o c e d u r e s . This is t h e a p p r o a c h followed to d a t e b y t h e ATM Forum, as w e explain in t h e next section. T h e s e p r o c e d u r e s are b a s e d o n d e t e r m i n i s t i c b o u n d s o n t h e traffic t r a n s m i t t e d b y t h e user. T h e u s e r c a n enforce t h e s e deter­ ministic b o u n d s b y u s i n g a m e c h a n i s m called t h e leaky bucket, w h i c h w e also describe in t h e next section. T h e n e t w o r k c a n verify t h a t t h e b o u n d s a r e m e t b y u s i n g t h e s a m e m e c h a n i s m . Moreover, t h e n e t w o r k c a n g u a r a n t e e d e t e r m i n ­ istic delay b o u n d s a n d also t h a t n o loss occurs w h e n buffers overflow. T h e s e p r o c e d u r e s b a s e d o n d e t e r m i n i s t i c b o u n d s m a y l e a d t h e n e t w o r k to c a r r y o n l y a subset of t h e calls it could c a r r y w i t h statistical p r o c e d u r e s . D e t e r m i n i s t i c p r o c e d u r e s are b a s e d o n a worst-case analysis, a n d t h e y do n o t take into a c c o u n t t h e likelihood of s u c h worst-case o c c u r r e n c e s . Such p r o c e d u r e s c a n n o t u s e t h e fact t h a t m a n y c o n n e c t i o n s are v e r y u n l i k e l y to b e h a v e in t h e worst possible w a y at t h e s a m e t i m e . C o n s e q u e n t l y , d e t e r m i n i s t i c

8.4

ATM Networks

p r o c e d u r e s do n o t take advantage of t h e b e n e f i t s of statistical multiplexing. Tb take advantage of t h e s e benefits, t h e n e t w o r k m u s t a d o p t statistical p r o c e d u r e s . W h e n u s i n g a statistical p r o c e d u r e , t h e u s e r specifies statistical descriptors of t h e traffic. T h r e e q u e s t i o n s arise w i t h s u c h a statistical a p p r o a c h . T h e first q u e s t i o n is h o w t h e u s e r c a n enforce statistical characteristics. T h e s e c o n d q u e s t i o n is h o w t h e n e t w o r k c a n verify t h a t t h e traffic satisfies t h e s e statistical characteristics. T h e third q u e s t i o n is h o w t h e n e t w o r k c a n u s e t h e statistics to decide w h i c h calls to a c c e p t a n d w h i c h to reject. We explain later practical statistical m e t h o d s for specifying a n d verifying traffic characteristics a n d sketch p r o c e d u r e s for controlling t h e n e t w o r k .

8.4.2

Deterministic Approaches In this section w e r e v i e w t h e ATM F o r u m r e c o m m e n d a t i o n s a n d n e t w o r k control p r o c e d u r e s b a s e d o n t h e s e r e c o m m e n d a t i o n s .

ATM Forum Recommendations T h e ATM F o r u m r e c o m m e n d a t i o n s for t h e u s e r - n e t w o r k interface specify a m e c h a n i s m for describing t h e traffic flowing t h r o u g h a virtual circuit c o n n e c ­ tion. T h i s m e c h a n i s m , called t h e generalized cell rate algorithm (GCRA), defines t h e five service categories specified b y t h e ATM F o r u m : c o n s t a n t bit rate (CBR), variable bit rate (VBR) e i t h e r real t i m e or n o n - r e a l t i m e , available bit rate (ABR), a n d unspecified bit rate (UBR) as w e explain below. T h e GCRA h a s two p a r a m e t e r s , Τ a n d τ, a n d it t i m e s t h e arrivals of cells as follows. T h e algorithm defines a theoretical arrival time, tat, of a cell. If t h e next cell arrives before tat — τ, t h e n t h e a l g o r i t h m G C R A ( T , r ) d e c l a r e s t h a t cell to b e nonconformant, a n d tat is u n c h a n g e d . If it arrives at t i m e t>tat — τ, t h e n t h e cell is conformant a n d t h e a l g o r i t h m r e s e t s t h e value of tat to max{i, tat] + T. For instance, c o n s i d e r GCRA(10, 3) w i t h t h e initial value of tat = 0 a n d a s s u m e t h a t t h e arrival t i m e s of cells are 1, 6, 8, 19, 27. T h e first cell is c o n f o r m a n t since 1 > 0 — 3. T h e a l g o r i t h m u p d a t e s tat to m a x { l , 0} + 10 = 11. T h e s e c o n d cell is n o n c o n f o r m a n t since 6 < 11 — 3. T h e t h i r d cell is c o n f o r m a n t since 8 > 11 - 3, a n d t h e a l g o r i t h m sets tat = max{8, 11} + 10 = 21. T h e fourth cell is c o n f o r m a n t since 19 > 21 - 3, a n d t h e a l g o r i t h m sets tat = max{19, 21} + 10 = 31. T h e fifth cell is n o n c o n f o r m a n t since 27 < 31 — 3. S u m m a r i z i n g , t h e decisions of t h e a l g o r i t h m are C, NC, C, C, NC, w h e r e C m e a n s c o n f o r m a n t a n d NC n o n c o n f o r m a n t . T h i s a l g o r i t h m is e q u i v a l e n t to a l e a k y b u c k e t , w h i c h w e describe n e x t a n d w h i c h w e will u s e to s t u d y a d m i s s i o n control. Fluid a c c u m u l a t e s at a given

395

Control of Networks

396

rate in a b u c k e t t h a t c a n store u p to Τ + τ u n i t s of fluid. Fluid t h a t arrives w h e n t h e b u c k e t is full is lost. A cell t h a t arrives w h e n t h e b u c k e t c o n t a i n s less t h a n Τ u n i t s of fluid is n o n c o n f o r m a n t . A cell t h a t arrives w h e n t h e b u c k e t c o n t a i n s at least Τ u n i t s of fluid is conformant, a n d it r e m o v e s Τ u n i t s of fluid from t h e b u c k e t . Let u s d e n o t e b y F ( r - ) t h e a m o u n t of fluid in t h e b u c k e t j u s t before t i m e t a n d b y F ( f + ) t h e a m o u n t of fluid j u s t after t i m e r. T h e leakyb u c k e t algorithm is s u c h t h a t a cell is n o n c o n f o r m a n t if F(r—) < Τ a n d t h e n F(t+) = F ( r - ) . Otherwise, t h e cell is c o n f o r m a n t a n d F(t+) = F ( r - ) - T. If t h e next cell arrives s t i m e u n i t s later, t h e n F(r + s - ) = m i n { F ( r + ) + s, Τ + τ } . In o u r previous example, a s s u m e t h a t t h e b u c k e t c o n t a i n s Τ + τ = 13 u n i t s of fluid at t i m e 0. T h e n , F ( l - ) = 13, F ( l + ) = 3, F ( 6 - ) = 8 = F ( 6 + ) , F ( 8 - ) = 10, F ( 8 + ) = 0, F Q 9 - ) = 11, F ( 1 9 + ) = 1, F ( 2 7 - ) = 9 = F ( 2 7 + ) . C o n s e q u e n t l y , t h e successive decisions are C, NC, C, C, NC. T h e ATM F o r u m r e c o m m e n d s t h a t constant bit rate (CBR) traffic o n a l i n e w i t h rate R s h o u l d specify its p e a k cell rate (PCR) a n d its cell delay variation t o l e r a n c e (CDVT). T h e m e a n i n g of t h e s e p a r a m e t e r s is t h a t t h e cells s h o u l d b e c o n f o r m a n t for t h e GCRA

, CDVT

algorithm. T h u s , if R/PCR = 5 a n d CDVT = 0, t h e n t h e p e a k cell rate is 20% of t h e line rate, a n d t h e cells arrive at m u l t i p l e s of 5 cell t r a n s m i s s i o n t i m e s . If R/PCR = 5 a n d CDVT = 1, t h e n t h e fastest arrival t i m e s are —1, 4, 9, 14, 19, a n d so on. T h i s s e q u e n c e c o r r e s p o n d s to a periodic s t r e a m w i t h rate e q u a l to 20% of t h e line rate: o n e arrival e v e r y fifth cell t r a n s m i s s i o n t i m e . If R/PCR = 4.5 a n d CDVT = 1, t h e n cells c a n arrive at t i m e s - 1 , 4, 8, 13, 17, 22, 26, 31, a n d so on. This s e q u e n c e h a s rate R/4.5. An easy w a y to see w h y t h e m a x i m u m rate of a CBR s t r e a m is i n d e e d PCR is to recall t h a t a cell takes a w a y Τ = R/PCR from t h e l e a k y bucket, w h i c h is filled at a specified rate. T h u s , over a long t i m e interval w i t h d u r a t i o n t, t u n i t s of fluid e n t e r t h e b u c k e t a n d at m o s t t/T cells c a n arrive. T h e m a x i m u m rate is therefore 1/T = PCR/R cells p e r cell t r a n s m i s s i o n t i m e 1 /R second, or PCR cells p e r second. T h e intuitive m e a n i n g of GCKA(R/PCR, CDVT) is t h a t t h e cells c a n arrive at t h e i r p e a k cell rate PCR b u t do n o t h a v e to b e exactly periodic. T h e CDVT m e a ­ s u r e s t h e d e p a r t u r e from exact periodicity. Such d e p a r t u r e m a y b e n e c e s s a r y b e c a u s e of t h e framing s t r u c t u r e t h a t carries t h e cells. For instance, i m a g i n e a CBR s t r e a m t h a t c o r r e s p o n d s to 3.5 cells e v e r y frame t i m e of a physical layer. In o n e i m p l e m e n t a t i o n , t h e successive frames m i g h t c a r r y 3 or 4 cells, a n d s u c h framing i n t r o d u c e s a cell d e l a y variation. Similarly, m u l t i p l e x i n g a n d t h e

8.4

ATM Networks

397

injection of m a i n t e n a n c e cells i n t r o d u c e a cell d e l a y variation. Note t h a t a late cell delays t h e set of all future acceptable arrival t i m e s . T h u s , it is n o t correct to t h i n k of a CBR s t r e a m as h a v i n g arrival t i m e s b e i n g m u l t i p l e s of R/PCR w i t h a shift of CDVT. For variable bit rate (VBR) traffic, t h e ATM F o r u m specifies t h e PCR CDVT, b u r s t t o l e r a n c e (BT), a n d t h e s u s t a i n e d cell rate (SCR). T h e m e a n i n g of t h e s e p a r a m e t e r s is t h a t t h e cells s h o u l d b e c o n f o r m a n t for b o t h t

GCRA

, CDVT

and

algorithms. T h e m o t i v a t i o n for this definition of VBR is t h a t s u c h a s t r e a m m i g h t b e v e r y b u r s t y a n d s e n d a n u m b e r of back-to-back cells s e p a r a t e d b y idle periods. T h e BT p a r a m e t e r m a k e s s u c h b u r s t s acceptable. For instance, consider a VBR s t r e a m w i t h R/PCR = 1, R/SCR = 20, CDVT = 0, BT = 57. If w e t h i n k b a c k a b o u t t h e leaky-bucket i n t e r p r e t a t i o n of t h e GCRA(R/SCR, BT + CDVT) = GCRA(20,57) algorithm, w e see t h a t a confor­ m a n t cell takes a w a y 20 u n i t s of fluid. With a n initial c o n t e n t of Τ + τ = 77 u n i t s of fluid, w e find t h a t four cells c a n arrive b a c k to back, at t i m e s 0 , 1 , 2, 3. Indeed, F ( O - ) = 77, F ( 0 + ) = 57, F(l-) = 58, F ( l + ) = 38, F(2-) = 39, F ( 2 + ) = 19, F(3-) = 20, F ( 3 + ) = 0. A n e w g r o u p of four cells c a n t h e n arrive at t i m e 80 b e c a u s e F(80—) = 77. T h u s , b u r s t s of four cells c a n arrive at t i m e s 0, 80, 160, a n d so on. Such a s t r e a m is also c o n f o r m a n t for t h e GCRA(.R/PGR, CDVT) = GCRA(1, 0) since this controller m a k e s all t h e s t r e a m s c o n f o r m a n t . T h e b u r s t y s t r e a m t h a t w e c o n s t r u c t e d h a s a l o n g - t e r m average r a t e e q u a l to 4 / 8 0 = 1/20, since it h a s b u r s t s of size 4 e v e r y 80 t i m e units. T h u s , t h e s u s t a i n e d cell rate of t h e s t r e a m SCR is i n d e e d s u c h t h a t R/SCR = 20. As a n o t h e r example, c o n s i d e r t h e p a r a m e t e r s R/PCR = 5, R/SCR = 10, CDVT = 1 6 , BT = 20. T h i s s t r e a m c a n h a v e b u r s t s of five c o n s e c u t i v e cells w i t h b u r s t s at t i m e 0, 1, 2, 3, 4, 50, 51, 52, 53, 54, 100, 101, 102, 103, 104, 150, 151, a n d so on. T h e s u s t a i n e d cell rate is 1/10 of t h e l i n e r a t e (five cells e v e r y 50 cell t r a n s m i s s i o n t i m e s ) . T h e ATM F o r u m is c u r r e n t l y d e v e l o p i n g specifications for available bit rate (ABR) t r a n s m i s s i o n s . T h e o p e r a t i n g principle is t h a t a n ABR c o n n e c t i o n m a y h a v e a g u a r a n t e e d m i n i m u m cell r a t e a n d a n i m p o s e d p e a k cell rate. T h e

Control of Networks

398

actual rate available to t h e c o n n e c t i o n varies b e t w e e n t h e s e two b o u n d s o n t h e basis of feedback information a b o u t t h e congestion in t h e n e t w o r k a n d destination. T h e rate a d j u s t m e n t s c h e m e is a time-varying GCRA control t h a t o p e r a t e s e n d to e n d a n d is rate-based. T h e ATM F o r u m d o e s n o t specify t h e ratecontrol algorithm t h a t e n d s y s t e m s a n d switches m u s t u s e b u t defines g e n e r a l m e c h a n i s m s . T h e ATM cells c o n t a i n a flow-control field t h a t t h e switches fill to indicate t h e congestion level t h e y e x p e r i e n c e . Also, t h e s o u r c e s a n d destinations c a n s e n d r e s o u r c e m a n a g e m e n t (RM) cells. In a typical s c h e m e , t h e s o u r c e s periodically s e n d RM cells to indicate their desired rate. T h e switches t h e n c o m p a r e t h e r e q u e s t s of various VCs a n d distribute t h e i r spare b a n d w i d t h accordingly. T h e switches indicate t h e i r b a n d w i d t h allocation in r e t u r n i n g RM cells t h a t go b a c k to t h e source. In addition, t h e switches set congestion indication bits in t h e forward cells a n d in t h e r e t u r n i n g RM cells. T h e d e s t i n a t i o n m o n i t o r s t h e congestion indication bits of t h e arriving cells a n d m a r k s t h e c o r r e s p o n d i n g bit of t h e RM cells accordingly. T h e source u s e s t h e congestion indication bits of t h e RM cells a n d t h e b a n d w i d t h allocation values of t h e s e cells to modify t h e p a r a m e t e r s of its GCRA. M a n y variations are possible a n d are b e i n g explored. For m o r e information, t h e r e a d e r s h o u l d k e e p track of t h e revised v e r s i o n s of t h e ATM F o r u m r e c o m m e n d a t i o n s . T h u s , ABR is a n a t t e m p t to u s e feedback to b e t t e r utilize t h e n e t w o r k . T h e last service b e i n g specified b y t h e ATM F o r u m (in addition to CBR, VBR, a n d ABR) is unspecified bit rate (UBR). UBR is best-effort service w i t h n o g u a r a n t e e o n t h e quality of service.

Admission Control H o w m a n y GCRA(T, r ) c o n n e c t i o n s c a n go t h r o u g h a buffer e q u i p p e d w i t h a t r a n s m i t t e r t h a t s e n d s C cells p e r u n i t of t i m e if t h e d e l a y t h r o u g h t h e buffer m u s t b e less t h a n D u n i t s of t i m e ? H e r e , o n e u n i t of t i m e is a cell t r a n s m i s s i o n t i m e o n e a c h o n e of t h e i n c o m i n g c o n n e c t i o n s . Tb a n s w e r this q u e s t i o n w e first s t u d y t h e following p r o b l e m . We w a n t to find t h e fastest s e q u e n c e of cells t h a t is c o n f o r m a n t to t h e G C R A ( T , r ) . We say t h a t a s e q u e n c e of arrival t i m e s *2, . . . } is faster t h a n a n o t h e r s e q u e n c e of arrival t i m e s {z/i, · · .}if*fc 1. We c o n s t r u c t this fastest s e q u e n c e b y filling u p t h e l e a k y b u c k e t w i t h Τ + τ u n i t s of fluid at t i m e 0 a n d b y s e n d i n g a cell a n d r e m o v i n g Τ u n i t s of fluid as soon as t h e b u c k e t c o n t a i n s Τ u n i t s of fluid.

8.4

ATM Networks

399

0

8.20 1

Ι !!!^^

20

40

60

80

100

120

0

20

40

60

80

100

120

Buffer occupancy for the fastest streams with the parameters indicated. In (1), the system is unstable because C is too small. T h e other cases show the effect of the burstiness of the stream on the backlog.

A s s u m e n o w t h a t a G C R A ( T , T ) s t r e a m goes t h r o u g h a buffer e q u i p p e d w i t h a t r a n s m i t t e r t h a t t r a n s m i t s C cells p e r u n i t of t i m e . We w a n t to a n a l y z e t h e m a x i m u m b a c k l o g of t h e buffer, a s s u m i n g t h a t it is initially e m p t y . T h e m a x i m u m buffer backlog will o c c u r if t h e cell s t r e a m is t h e fastest. If C > 1 / T , t h e n t h e buffer is faster t h a n t h e arrival s t r e a m (since 1/T is t h e m a x i m u m l o n g - t e r m rate of a G C R A ( T , r ) s t r e a m ) a n d t h e m a x i m u m b a c k l o g is finite. Figure 8.20 s h o w s t h e e v o l u t i o n of t h e buffer o c c u p a n c y for t h e fastest s t r e a m s possible w i t h given GCRA p a r a m e t e r s . Case (1) is u n s t a b l e . T h i s insta­ bility is c a u s e d b y C b e i n g s m a l l e r t h a n 1/T. Case (2) is stable since C > 1/T. By c o m p a r i n g cases (3) a n d (4) w e o b s e r v e t h e effect of t h e b u r s t i n e s s of t h e stream. If Ν s t r e a m s t h a t c o n f o r m to GCRA(T, τ ) s h a r e t h e s a m e buffer, t h e n t h e m a x i m u m o c c u p a n c y of t h e buffer o c c u r s again w h e n t h e s t r e a m s are t h e fastest. We p r e s e n t a few e x a m p l e s in Figure 8.21. By c o m p a r i n g t h e s e cases w e observe t h a t m u l t i p l y i n g t h e n u m b e r of s o u r c e s Ν a n d t h e service rate C b y t h e s a m e factor V r e s u l t s i n m u l t i p l y i n g t h e worst-case o c c u p a n c y b y V.

Control of Networks

400

This s h o u l d b e e x p e c t e d since b o t h t h e n u m b e r of arrivals a n d t h e n u m b e r of d e p a r t u r e s in e a c h t i m e step are m u l t i p l i e d b y V. A good a p p r o x i m a t i o n of t h e m a x i m u m backlog in t h e buffer w i t h rate C a n d Ν GCRA(T,r) s t r e a m s c a n b e derived as follows. A s s u m e t h a t τ » T. T h e n Κ + 1 cells c a n b e s e n t back-to-back b y a single s t r e a m if τ + Τ-ΚΤ

+

Κ>Τ.

Indeed, d u r i n g t h e t i m e interval [0, Κ), Κ u n i t s of fluid e n t e r t h e l e a k y b u c k e t a n d KT u n i t s of fluid are r e m o v e d b y t h e first Κ cells. T h e left-hand side of t h e above i n e q u a l i t y is t h e a m o u n t of fluid t h a t is left at t i m e K, and this a m o u n t should b e larger t h a n Τ if t h e (K + l ) t h cell is c o n f o r m a n t . T h e m a x i m u m n u m b e r of back-to-back cells is therefore a p p r o x i m a t e l y e q u a l to τ

M:=

Τ —1

+ 1=

τ + Τ - 1 . Τ-1

_ (8.4) x

After t h e s e back-to-back cells, t h e o t h e r cells follow e a c h o t h e r a p p r o x i m a t e l y e v e r y Τ t i m e units, t h e t i m e r e q u i r e d to collect e n o u g h fluid in t h e l e a k y b u c k e t to s e n d a n e w cell. O u r description neglects round-off effects t h a t are n o t significant for o u r analysis. If e a c h of Ν s t r e a m s p r o d u c e s t h e s e Μ back-

8.4

ATM Networks

401

to-back cells d u r i n g t h e interval [0, Μ — 1], t h e n t h e buffer a c c u m u l a t e s Ν χ Μ cells a n d c a n serve only (M — 1) χ C cells, so t h a t t h e backlog is a p p r o x i m a t e l y Β(Τ,τ, C, N), w h e r e

B(T, τ, C, Ν) := Ν χ Μ - (Μ - 1) χ C « Ν

τ + Τ - 1 Τ

_ 1

CT Τ

- 1

.

If Τ χ C > Ν , as m u s t b e a s s u m e d for stability, t h e n t h e cells stop a c c u m u l a t i n g after t h e b u r s t of back-to-back cells, so t h a t Β is t h e m a x i m u m backlog. For t h e n u m e r i c a l e x a m p l e s of Figures 8.20 a n d 8.21 this formula gives B(3, 23, 0.34, 1) = 8.59, B(3, 25, 0.4, 1) = 8.5, E(3, 40, 0.4, 1) = 13 E(3, 25, 4, 10) = 85, B(3, 40, 4, 10) = 130 B(3 40, 8, 20) = 260, B(3 40, 40, 100) = 1300. t

t

C o m p a r i n g t h e s e n u m b e r s w i t h t h e figures s h o w s t h a t this a p p r o x i m a t i o n is satisfactory. Let u s go b a c k to t h e q u e s t i o n t h a t w e a s k e d at t h e b e g i n n i n g of this section. A s s u m e t h a t t h e m a x i m u m acceptable d e l a y t h r o u g h t h e buffer is D cell t r a n s m i s s i o n t i m e s . We w a n t to e s t i m a t e t h e m a x i m u m n u m b e r Ν of G C R A ( T r ) s t r e a m s t h a t t h e n e t w o r k c a n accept. Tb a n s w e r t h e question, w e observe t h a t t h e m a x i m u m backlog m u s t b e at m o s t D χ C cells. Indeed, a cell t h a t faces a backlog o f D x C e x p e r i e n c e s a delay e q u a l to D. C o n s e q u e n t l y , t h e m a x i m u m n u m b e r Ν is o b t a i n e d b y solving ;

B(T, τ, C,

N)=DxC,

t h a t is, N x M - ( M - l ) x C ^ N

τ + T - l

Τ-1

Cr

=

DxC.

By solving this e q u a t i o n w e find £>(Τ-1) + τ

N^C—

T + T -

1

.

Recall t h a t t h e s e derivations a s s u m e t h a t C > N/T. A n e q u i v a l e n t w a y to look at t h e above e q u a t i o n is to write t h e c o n s t r a i n t o n Ν as Ν χ a (T, G

r , D ) < C,

(8.5)

Control of Networks

402 where a (T, G

Τ + τ - 1

ί

τ, D) := m a x {

1 ,- [ .

(8.6)

We c a n i n t e r p r e t t h e i n e q u a l i t y (8.5) as-stating t h a t e a c h GCRA(T, τ ) c o n n e c t i o n w i t h m a x i m u m delay D r e q u i r e s a b a n d w i d t h αο(Τ,τ, D) given b y (8.6). We c a n call αο(Τ,τ, D) t h e effective bandwidth of a G C R A ( T r ) c o n n e c t i o n w i t h m a x i m u m delay D. If w e recall t h e formula (8.4) for t h e m a x i m u m n u m b e r Μ of back-to-back cells, w e find t h a t /

Μ

ί

a (T, G

τ, D) : = m a x \

11

,-

.

(8.7)

T h u s , if D is small c o m p a r e d w i t h M, t h e effective b a n d w i d t h is close to 1. T h a t is, t h e switch m u s t treat s u c h a source as a c o n s t a n t bit rate source w i t h a rate e q u a l to t h e line rate R. At t h e o t h e r extreme, if D is m u c h larger t h a n M, t h e n t h e effective b a n d w i d t h is close to M/D a n d a b o u t D/M c a n b e a c c o m m o d a t e d (so long as Τ > D/M). For instance, o n e finds t h a t a (3, G

40, 30) = 0.42, a (3, G

40, 20) = 0.52, a ( 3 , 40, 10) = 0.70. G

Using t h e ATM F o r u m p a r a m e t e r s for a VBR c o n n e c t i o n , w e see t h a t a (R/SCR, G

BT + CDVT, D) = m a x { - j ^ , λ 1 , [β + D J 1

(8.8)

where we introduce the parameters λ :=

SCR R

_| BT + CDVT and θ := . R/SCR - 1 H

(8.9)

This effective b a n d w i d t h is a m e a s u r e of t h e cost of c a r r y i n g s u c h a VBR c o n n e c t i o n . Note t h a t t h e cost i n c r e a s e s as t h e acceptable d e l a y d e c r e a s e s . T h e cost also i n c r e a s e s with t h e b u r s t i n e s s m e a s u r e d b y β. T h e delay constraint b e c o m e s b i n d i n g if D is small e n o u g h or if β is large e n o u g h . Otherwise, t h e b a n d w i d t h is d e t e r m i n e d b y t h e average r a t e λ. S u m m a r i z i n g this section, w e h a v e explored t h e implications of t h e GCRA control m e c h a n i s m r e c o m m e n d e d b y t h e ATM F o r u m . We h a v e a n a l y z e d t h e m a x i m u m n u m b e r of c o n n e c t i o n s t h a t c a n go t h r o u g h a buffer subject to a m a x i m u m delay constraint. T h e analysis is b a s e d o n t h e worst-case b e h a v i o r of t h e c o n n e c t i o n s a n d ignores t h e likelihood of s u c h behavior. T h e result c a n

8.4 •••

ATM Networks M' • • mnm-mmmmm

•»»MO

• m, •

•• π τιττιο• ^mmmmmmmmmmmmmmammmm

b e v i e w e d as r e q u i r i n g a n effective b a n d w i d t h p e r c o n n e c t i o n . T h e effective b a n d w i d t h i n c r e a s e s w i t h t h e b u r s t i n e s s of t h e c o n n e c t i o n a n d d e c r e a s e s w i t h t h e acceptable delay.

Pricing Calls We explore t h e pricing i m p l i c a t i o n s of d e t e r m i n i s t i c a p p r o a c h e s in a s i m p l e m o d e l t h a t highlights s o m e features of t h e p r o b l e m . Consider calls of two t y p e s : 1 a n d 2. E a c h call r e q u i r e s r e s o u r c e s of two classes, A (e.g., buffer space) a n d Β (e.g., b a n d w i d t h ) , i n o r d e r to b e carried b y t h e n e t w o r k . A s s u m e t h a t a call of t y p e i r e q u i r e s αχ u n i t s of r e s o u r c e s of class A a n d b{ u n i t s of class Β (for i = 1, 2). T h e total a m o u n t of r e s o u r c e s of classes A a n d Β available are a a n d β, respectively. D e n o t e b y ni t h e n u m b e r of calls of t y p e i carried b y t h e n e t w o r k for i = 1, 2. T h e set of admissible pairs (ni, n ) is 2

S := {(m, n ) € {0, 1 , 2 , . . . } | α ι " ι + a n 2

2

2

2

< a a n d b\n\ + b n 2

2

< β}.

This set is illustrated in Figure 8.22. A s s u m e for t h e t i m e b e i n g t h a t t h e p a r a m e t e r s of t h e p r o b l e m are s u c h t h a t t h e t w o oblique b o u n d a r y l i n e s cross e a c h o t h e r at s o m e p o i n t X, as s h o w n in t h e figure. We consider t h a t t h e n e t w o r k will charge pi for a call of t y p e i (i = 1, 2) p e r u n i t of time, a n d w e e x a m i n e t h e u n i t p r i c e s (pi,p ) t h a t t h e n e t w o r k c a n charge in a c o m p e t i t i v e e n v i r o n m e n t . Tb simplify t h e p r o b l e m , w e a s s u m e t h a t 2

8.22 FIGURE

Set S of acceptable n u m b e r s of calls of two types that require resources of two classes. We show that w h e n the constraint lines intersect as here at a point X, a competitive network should operate at the intersection point.

403

Control of Networks

404

t h e d e m a n d is larger t h a n t h e supply, so t h a t w e c a n attract as m a n y c u s t o m e r s as w e c a n carry, if o u r price for t h e service is competitive. Let u s call t h e n e t w o r k N. We define t h e competitive e n v i r o n m e n t as follows. A s s u m e t h a t Ν offers t h e prices (pi, pi) a n d t h a t s o m e o t h e r n e t w o r k N ' offers t h e prices (p[, p' ). (Ν' h a s t h e s a m e resources, a a n d β.) If t h e r e v e n u e s p e r u n i t of t i m e R of Ν a n d R of Ν ' are s u c h t h a t R > R, t h e n it m u s t b e t h a t p\ < p\ if Ν carries calls of t y p e i, for i = 1, 2. T h e i n t e r p r e t a t i o n of this condition is t h a t if Ν ' is as profitable as N, to stay in b u s i n e s s Ν c a n n o t charge m o r e t h a n Ν'; otherwise, Ν w o u l d lose its c u s t o m e r s . Network Ν w a n t s to find t h e prices it c a n charge to collect r e v e n u e s at rate R. (R m a y equal t h e cost of t h e r e s o u r c e s plus r e t u r n o n i n v e s t m e n t . ) O u r first observation is t h a t if t h e n e t w o r k chooses to c a r r y o n l y calls of t y p e 1, t h e n it c a n obtain t h e rate of r e v e n u e s R b y charging p\ = Ra\/a since t h e n e t w o r k c a n carry a/a\ calls of t y p e 1 (see figure). C o n s e q u e n t l y , it m u s t be that 2

f

f

Pi < Rdi/cx. Indeed, if Ν charges (p\,P2) w i t h ^ i > Ra\/a, t h e n it c a n n o t c o m p e t e w i t h a n e t w o r k Ν ' t h a t charges (Ra\/a, oo) a n d carries o n l y c u s t o m e r s of t y p e 1. Similarly, w e find t h a t p 0} is a c o n t i n u o u s - t i m e Markov c h a i n o n s o m e finite state space Ζ w i t h rate m a t r i x Q = {q(i,j), i,j € Z} a n d r : Ζ —• [0, oo) is a given function, (r m a y b e m e a s u r e d in bits or cells p e r second.) T h e i n t e r p r e t a t i o n of t h e definition is t h a t r(z ) is t h e rate at t i m e t of s o m e bit s t r e a m . T h e rate fluctuates r a n d o m l y . T h i s m o d e l is m o t i v a t e d b y observing t h e bit s t r e a m t h a t a variable bit rate video c o m p r e s s i o n a l g o r i t h m g e n e r a t e s . A s s u m e t h a t t h e MMF r(z ) goes t h r o u g h a buffer e q u i p p e d w i t h a t r a n s m i t ­ ter w i t h rate c. We w a n t to a n a l y z e t h e evolution in t i m e of t h e buffer o c c u p a n c y process x . T h e analysis e n a b l e s u s to d e t e r m i n e h o w large t h e buffer capacity should b e a n d w h a t delays t h e MMF faces t h r o u g h t h e buffer. t

t

t

t

Control of Networks

406

We s h o w in section 9.3.4 t h a t t h e buffer o c c u p a n c y process h a s a n i n v a r i a n t distribution of t h e form

In this expression, {β , ζ e Ζ} are t h e e i g e n v a l u e s of t h e m a t r i x A = [A(i,j), i,j e Z], w h e r e ζ

A(i,j)=q(i j)/(r(j)-c). f

Tb d e t e r m i n e t h e coefficients a(z), o n e m u s t calculate all t h e eigenvalues a n d eigenvectors. Tb simplify t h e analysis, o n e u s u a l l y a p p r o x i m a t e s this s u m of e x p o n e n t i a l s b y a single exponential, P(x >x)& t

Κβ- , βχ

for χ » 1,

(8.10)

w h e r e β is t h e smallest of t h e e i g e n v a l u e s {β , ζ € Ζ} a n d Κ is t h e c o r r e s p o n d i n g coefficient. T h e calculation of Κ r e q u i r e s c o m p u t i n g all t h e e i g e n v a l u e s a n d eigenvectors. A useful e x a m p l e is w h e n r(z(t)) is t h e s u m of t h e rates of Ν i n d e p e n d e n t a n d identically distributed sources, e a c h m o d e l e d b y on-off Markov fluid. T h a t is, e a c h source is m o d e l e d b y a Markov c h a i n o n {0,1} w i t h rate m a t r i x s u c h t h a t q(0 1) = λ > 0 a n d q(l, 0) = μ > 0. W h e n t h e source is in state 1, it p r o d u c e s fluid at rate 1. W h e n it is in state 0, t h e source does n o t p r o d u c e fluid. A s s u m e t h a t t h e s u p e r p o s i t i o n of t h e s e Ν s o u r c e s goes t h r o u g h a buffer w i t h t r a n s m i s s i o n rate C. T h e n o n e finds ζ

f

1+ λ 1 -

Νλ/C

(8.11)

C/N

and (8.12) w h e r e |_cj is t h e i n t e g e r p a r t of C. Tb e s t i m a t e Κ o n e m u s t solve for all t h e e i g e n v a l u e s β w h i c h c a n b e d o n e n u m e r i c a l l y easily as long as Ν is n o t too large. This a p p r o a c h h a s t h e m e r i t of yielding c o m p l e t e results for s i m p l e m o d e l s . However, t h e analysis of n e t w o r k s w i t h M a r k o v - m o d u l a t e d fluid s o u r c e s u s i n g this a p p r o a c h a p p e a r s b e y o n d r e a c h . In t h e following section w e discuss a different a p p r o a c h . Ζ)

8.4

ATM Networks

407

More General Model T h e M a r k o v - m o d u l a t e d traffic m o d e l describes t h e r a n d o m fluctuations of t h e rate of a bit s t r e a m . A m o r e g e n e r a l traffic m o d e l is a r a n d o m p r o c e s s A : = {A(f), t > 0} t h a t specifies t h e n u m b e r A(r) of bits c a r r i e d b y t h e traffic d u r i n g t h e interval [0, t] for t > 0. W i t h o u t further a s s u m p t i o n s , this m o d e l is too g e n e r a l for u s to b e able to analyze h o w t h e n e t w o r k c a n t r a n s p o r t s u c h traffic. S o m e h o w w e m u s t specify t h e average rate of t h e traffic a n d s o m e m e a s u r e of burstiness. T h e average rate is [A(r + T) - A(t)]/T for large T. U n d e r suitable a s s u m p ­ tions, this average rate is a well-defined q u a n t i t y λ. T h a t is, [A(t + T) — A(t)]/T a p p r o a c h e s λ as Τ increases, a n d t h a t value λ does n o t d e p e n d o n t h e realiza­ t i o n of t h e stochastic p r o c e s s A n o r o n r. T h e p r o c e s s A h a s t h e s e p r o p e r t i e s if it is stationary a n d ergodic.

Averaging Rate Fluctuations T h e rate of a traffic s t r e a m fluctuates. For instance, t h e p e a k rate of a video s t r e a m m a y b e 10 t i m e s larger t h a n its average rate. Tb p r e v e n t losses, a n e t w o r k n o d e could allocate to e a c h s t r e a m a b a n d w i d t h e q u a l to t h e p e a k rate of t h a t s t r e a m . However, s u c h a n allocation is overly conservative, a n d it is s o m e t i m e s possible for t h e t r a n s m i t t e r to allocate a b a n d w i d t h m u c h closer to t h e average rate t h a n to t h e p e a k rate. T h e r e are two f u n d a m e n t a l l y different m e t h o d s t h a t a n e t w o r k c a n u s e to r e d u c e t h e b a n d w i d t h it m u s t allocate to e a c h c o n n e c t i o n . T h e first m e t h o d is m u l t i p l e x i n g m a n y sources. T h e o t h e r m e t h o d is buffering. T h e multiplexing m e t h o d exploits t h e fact t h a t different s o u r c e s fluctuate i n d e p e n d e n t l y so t h a t w h e n t h e rate of a source is larger t h a n average, t h e rate of a n o t h e r m a y b e s m a l l e r t h a n average. C o n s e q u e n t l y , t h e rate of t h e s u p e r p o s i t i o n of m a n y s o u r c e s t e n d s to b e close to its average value. T h i s observation is similar to t h e fact t h a t if o n e t h r o w s 1,000 fair coins, a b o u t 500 of t h e m l a n d o n h e a d s . T h e probability t h a t m o r e t h a n 600 coins will l a n d o n h e a d s is v e r y small, a b o u t 1 0 ~ . T h u s , if e a c h of 1,000 s o u r c e s is off 50% of t h e t i m e a n d t r a n s m i t s at rate a t h e o t h e r 50% of t h e time, t h e n t h e total r a t e of t h e sources rarely exceeds 600 χ a. T h e analysis of t h e m u l t i p l e x i n g m e t h o d d e t e r m i n e s h o w m a n y s o u r c e s m u s t b e m u l t i p l e x e d a n d t h e t r a n s m i s s i o n rate p e r source r e q u i r e d so t h a t t h e probability t h a t t h e total rate of t h e s o u r c e s exceeds t h e t r a n s m i t t e r rate is s m a l l e r t h a n s o m e specified value, say 1 0 ~ . T h e buffering m e t h o d u s e s t h e fact t h a t over a long d u r a t i o n [r, i + T ] a s t r e a m A w i t h rate λ p r o d u c e s a total n u m b e r of cells close to AT. C o n s e q u e n t l y , if c> λ, t h e n A(r + T) — A(f) < cT w i t h a large probability. Now, if it w e r e t r u e t h a t A(t + T) - A(t) < cT for all t > 0, t h e n a n o d e t h a t t r a n s m i t s t h e s t r e a m 10

10

Control of Networks

408

w i t h rate c w o u l d d e l a y it b y at m o s t T. Tb see this, n o t e t h a t t h e buffer c a n n o t r e m a i n n o n e m p t y for Τ consecutive t i m e units. I n d e e d , if t h e buffer is e m p t y at t i m e t a n d n o n e m p t y d u r i n g [t, t + T], t h e n d u r i n g t h a t interval it h a s o u t p u t cT bits a n d it m u s t b e t h a t m o r e t h a n cT bits e n t e r e d t h e buffer, t h a t is, A(t + T) - A(t) > cT, a contradiction. If t h e buffer c a n n o t r e m a i n n o n e m p t y for Τ consecutive t i m e units, t h e n e v e r y bit t h a t e n t e r s m u s t leave before Τ t i m e units. This a r g u m e n t s h o w s that, b y u s i n g a buffer, a n o d e c a n t r a n s m i t at a rate c only slightly larger t h a n t h e average rate λ. T h e n o d e delays t h e bit s t r e a m b y at m o s t Τ if A(t + T) - A(t) < cT for all t > 0. In t h e statistical approach, w e do n o t insist t h a t this i n e q u a l i t y h o l d all t h e time, b u t o n l y t h a t it hold with a large probability. We t h e n c o n c l u d e t h a t t h e n o d e delays t h e bits b y at m o s t Τ w i t h a large probability. T h e analysis of this m e t h o d e s t i m a t e s t h e t r a n s m i t t e r rate n e e d e d so t h a t t h e n o d e delays t h e bit s t r e a m b y m o r e t h a n Τ s e c o n d s w i t h s o m e specified small probability, say 1 0 ~ . 10

H o w do t h e s e m e t h o d s c o m p a r e ? Multiplexing is effective for real-time traffic a n d buffering is effective for non-real-time traffic. T h e intuitive justifi­ cation for this s t a t e m e n t is t h a t b u r s t y real-time traffic c a n n o t b e buffered long e n o u g h to average o u t its rate fluctuations. For instance, a video c o n n e c t i o n h a s a rate close to t h e p e a k rate for as long as a few seconds, say 10 s. T h e s e long periods of h i g h bit rate c o r r e s p o n d to active s c e n e s in t h e video, say a car chase in a n action movie. T h e m o t i o n c o m p e n s a t i o n video c o m p r e s s i o n algorithm is n o t effective d u r i n g fast-changing s c e n e s and, c o n s e q u e n t l y , pro­ d u c e s a large bit rate to reflect t h e rapid modifications of t h e successive frames. If t h e n o d e buffers t h e video bit s t r e a m for less t h a n a few seconds, t h e n it still m u s t t r a n s m i t t h e s t r e a m at a rate close to t h e p e a k rate. O n e m i g h t t h i n k t h a t b y s u p e r p o s i n g a large n u m b e r , say 100, of s u c h video bit s t r e a m s , buffering w o u l d b e c o m e m u c h m o r e effective. However, analysis a n d s i m u l a t i o n s s h o w t h a t n o t to b e t h e case. We s h o w b e l o w t h a t buffering is ineffective for real-time s t r e a m s s u c h as video bit s t r e a m s . However, multiplexing c a n b e effective for s u c h s t r e a m s . In contrast to t h e real-time traffic case, buffering is effective for s t r e a m s t h a t c a n b e d e l a y e d b y a few s e c o n d s in t h e n e t w o r k , s u c h as s t r e a m s p r o d u c e d b y interactive applications. Obviously, buffering is t h e w a y to h a n d l e best-effort traffic (UBR or ABR).

Multiplexing We n o w p r e s e n t t h e m a i n results o n t h e multiplexing of m a n y sources. R e m e m ­ b e r t h a t t h e objective of multiplexing is to u s e a t r a n s m i s s i o n rate p e r source close to t h e average rate of e a c h source i n s t e a d of r e q u i r i n g a rate close to t h e

8.4

ATM Networks

409

p e a k rate. N e t w o r k e n g i n e e r s call this possibility of r e d u c i n g t h e r e q u i r e d r a t e p e r s o u r c e t h e multiplexing

gain.

C o n s i d e r Ν sources. For η = 1 , . . . , Ν , let Y b e t h e r a t e at s o m e t i m e t of n

s o u r c e n u m b e r n. We a s s u m e t h a t t h e s o u r c e s are stationary, i n d e p e n d e n t , a n d identically distributed. T h a t is, t h e r a t e s { Y \ , . . . , Υχ} are i n d e p e n d e n t r a n d o m variables t h a t h a v e a c o m m o n distribution t h a t d o e s n o t d e p e n d o n t. We w a n t to find t h e rate c s u c h t h a t P{Yi + · · · + Y

> cN] < 1 0 " . 9

N

I n words, if a n o d e t r a n s m i t s t h e s u p e r p o s i t i o n of t h e Ν s o u r c e s w i t h t h a t r a t e c, t h e n it d r o p s at m o s t a fraction 10~~ of t h e bits. 9

As w e m a y expect, t h e r a t e c is slightly larger t h a n t h e average rate, say λ, of e a c h source. T h e p r e c i s e value of c d e p e n d s o n Ν a n d o n t h e distribution of t h e r a n d o m variables Yj. Using t h e Bahadur-Rao t h e o r e m (see section 9.4.5), w e find P(Yi + · · · + Y

N

> Nc) ^

— e~

m c )

.

(8.13)

Λ/2πσθ Λ/Ν 0

I n t h e above expression, 6 a c h i e v e s t h e m a x i m u m in C

1(c) = sup[(0)=log£[exp(0Yi)] and σ

2

= φ"(θα) =

-φ(θ ). 0

( T h e t e r m φ(θ) is called t h e logarithmic moment generating function.') I n t h e case of on-off s o u r c e s w i t h P(on) =p a n d p e a k r a t e a, t h e coefficients of (8.13) h a v e t h e following form: 0 =C

1 a

log

(c(\-p)\ — \p(a-c)J

c (c(\-p)\ , /(c) = - log — a \p(a-c)J

- log

(a(\-p)\ \(a-c)J

\ ,a

l

2

, = c(a-

c).

As a n u m e r i c a l example, w e u s e t h e following v a l u e s for t h e h o m o g e n e o u s on-off sources: λ = 1 / 2 0 s , μ = 1 / 5 s , a = 18 Mbps, c = 8 Mbps. T h e s e p a r a m e t e r s c o r r e s p o n d to P(on) = λ / ( λ + μ) = 0.2 =p a n d therefore to a m e a n rate e q u a l top χ a = 3.6 Mbps.

Control of Networks

410

We find t h a t 0 = 0.0646,1(c) = 0.1523, σ = 80. We c a n t h e n calculate t h e value of Ν n e e d e d so t h a t P{ Y\ + · · · Υχ > cN] < 1 0 ~ . T h e c o m p u t e r finds t h a t Ν > 118 is t h e r e q u i r e d condition. Note t h a t t h e n u m b e r of sources t h a t are in t h e o n state h a s a b i n o m i a l distribution. T h u s t h e probability t h a t t h e aggregate i n p u t rate exceeds t h e o u t p u t rate c a n b e r e p r e s e n t e d exactly as 2

C

9

We c a n evaluate this expression directly a n d find P{Y\ Η YN > cN] < 10~~ for Ν > 114. You will n o t e t h e r e m a r k a b l e a c c u r a c y of t h e Bahadur-Rao approximation. T h e Bahadur-Rao a p p r o x i m a t i o n c a n b e u s e d for c o m p l e x distri­ b u t i o n s of t h e r a n d o m variables Υ& w h e r e a direct calculation is v e r y complex. Note t h a t for s u c h distributions, t h e evaluation of t h e p a r a m e t e r s t h a t e n t e r (8.13) m u s t b e p e r f o r m e d n u m e r i c a l l y . T h e Bahadur-Rao t h e o r e m e n a b l e s u s to analyze t h e case of m u l t i r a t e sources. T h a t t h e o r e m c a n also b e u s e d to analyze t h e overflows of m i x t u r e s of different t y p e s of sources. Tb do this, say t h a t w e h a v e 50% of s o u r c e s of t y p e Y a n d 50% of sources of t y p e Z. We c a n t h e n c o n s t r u c t a h y b r i d source t h a t is of t y p e Υ + Z. T h e analysis of s u c h a situation reveals t h a t t h e value of c r e q u i r e d to achieve a small loss probability c a n n o t b e w r i t t e n as t h e s u m of t h e n e c e s s a r y rates for t h e Y s o u r c e s a n d t h e Ζ sources. T h u s , u n f o r t u n a t e l y , for small buffers t h e r e is n o additive result similar to t h e effective b a n d w i d t h . 9

On-Line Estimation T h e discussion above s h o w s t h a t t h e n e t w o r k n e e d s detailed i n f o r m a t i o n a b o u t t h e statistics of t h e s o u r c e s in o r d e r to d e t e r m i n e t h e capacity t h a t it s h o u l d allocate to c o n n e c t i o n s . In practice it m a y n o t b e realistic to expect t h e u s e r s to k n o w t h a t information w h e n t h e y set u p t h e c o n n e c t i o n . T h e s e contradicting aspects s e e m to m a k e statistical p r o c e d u r e s impractical. We b e l i e v e t h a t this conclusion is n o t correct. T h e n e t w o r k could g u a r a n t e e a quality of service b y b e i n g conservative in its initial a d m i s s i o n control a n d m e a s u r e t h e traffic to d e t e r m i n e t h e actual r e s o u r c e s t h a t t h e traffic r e q u i r e s . For instance, t h e a d m i s s i o n control could b e b a s e d o n t h e p e a k rate of t h e traffic. O n c e t h e c o n n e c t i o n was in progress, t h e n e t w o r k could m o n i t o r its actual r e q u i r e m e n t s . ( T h e n e t w o r k also could calculate t h e price from t h e actual r e s o u r c e s t h a t t h e c o n n e c t i o n utilized.) Such a p r o c e d u r e p r e s e n t s a risk t h a t all t h e ongoing

8.4

ATM Networks

411

• • « ^ ^ c o n n e c t i o n s might, after a while, s u d d e n l y b e c o m e m u c h m o r e active a n d r e q u i r e m o r e b a n d w i d t h . However, s u c h a n e v e n t is v e r y unlikely. According to t h e above description, w e p r o p o s e four m e t h o d s for e s t i m a t i n g t h e actual b a n d w i d t h t h a t c o n n e c t i o n s r e q u i r e . T h e m e t h o d s differ in t h e i r n u m e r i c a l complexity a n d in t h e i r efficiency. A s s u m e t h a t identical a n d i n d e p e n d e n t s o u r c e s w i t h k n o w n m e a n rate m a n d p e a k rate a w a n t to b e t r a n s m i t t e d b y a t r a n s m i t t e r w i t h rate C. We w a n t to design a n on-line a d m i s s i o n p r o c e d u r e t h a t a c c e p t s t h e m a x i m u m n u m b e r of sources subject to a loss probability of 1 0 ~ . We a s s u m e t h a t n o t h i n g is k n o w n a b o u t t h e sources o t h e r t h a n t h e i r m e a n a n d p e a k rates. We first a s s u m e t h a t t h e s o u r c e s are on-off w i t h P(on) = m/a. On-off sources are easily s e e n to b e t h e m o s t b u r s t y s o u r c e s w i t h given m e a n a n d p e a k rates in t h a t fewer of t h e m c a n b e a c c e p t e d for a given loss rate. We u s e t h e BahadurRao formula to d e t e r m i n e t h e m a x i m u m n u m b e r No of s u c h s o u r c e s t h a t c a n b e accepted, as w e did in t h e p r e v i o u s section. We t h e n a c c e p t t h e No sources, a n d w e m e a s u r e t h e i r statistics. T h e four m e t h o d s w e p r o p o s e differ in h o w t h e y infer t h e actual n u m b e r of sources t h a t c a n b e carried b y t h e transmitter. I n m e t h o d s 1 a n d 2, w e calculate t h e b a n d w i d t h n e e d e d to c a r r y t h e No s o u r c e s w i t h a loss rate of 10~~ . In m e t h o d s 3 a n d 4, w e e s t i m a t e t h e p a r a m e t e r s of t h e Bahadur-Rao formula, a n d w e calculate t h e m a x i m u m value Ν t h a t c a n b e carried. 9

9

Method 1 We divide t h e interval b e t w e e n t h e m e a n rate m χ No of t h e No sources a n d C into a n u m b e r of e q u a l parts. T h a t is, w e calculate C = m χ N , C C , . . . , C = C so t h a t C\ - C = C - Ci = · · · = C C -\. We t h e n m o n i t o r t h e i n s t a n t a n e o u s rate of t h e No o n g o i n g c o n n e c t i o n s a n d d e t e r m i n e t h e loss rate t h a t t h e s e c o n n e c t i o n s w o u l d face if t h e service rate w e r e i n s t e a d of C, for k = 0, 1, . . . , K. T h i s m o n i t o r i n g is p e r f o r m e d in parallel, b y a device t h a t does n o t p e r t u r b t h e c o n n e c t i o n s . We t h e n find C(N ) = rnm{C \L B)




—

\ = 10" . ΰ

In addition, w e w a n t N(c + 0.5) < 155. T h e justification is t h a t t h e rate of source η is m o d e l e d b y 8.5Y + 0.5. We t h e n m a x i m i z e N(c) over c. T h e solution to t h e exercise is Ν = 34. T h u s , it is plausible t h a t a statistical m e t h o d t h a t a c c e p t s a s m a l l level of risk (a few b a d s e c o n d s e v e r y few days or w e e k s ) c a n d o u b l e t h e n u m b e r of c o n n e c t i o n s carried b y t h e switch. n

Pros and Cons of Deterministic Approaches T h e d e t e r m i n i s t i c a p p r o a c h e s b a s e d o n l e a k y b u c k e t s h a v e t h e following ad­ vantages: •

t h e source c a n easily e n s u r e t h a t its traffic m e e t s t h e specifications,

•

t h e n e t w o r k c a n easily verify t h a t t h e traffic m e e t s t h e specifications,

•

t h e n e t w o r k c a n g u a r a n t e e h a r d b o u n d s o n delays a n d avoid all losses b e c a u s e of buffer overflows, a n d

• b e c a u s e t h e quality of service is specified in t e r m s of h a r d b o u n d s , t h e u s e r s c a n verify t h a t t h e n e t w o r k p r o v i d e s t h e r e q u e s t e d quality of service. T h e m a i n disadvantage of d e t e r m i n i s t i c a p p r o a c h e s is t h a t t h e y are b a s e d o n worst cases: worst Τ s e c o n d s of a s t r e a m . T h e m e t h o d a s s u m e s t h a t all t h e c o n n e c t i o n s t h a t go t h r o u g h a n y o n e n o d e will exhibit t h e i r worst-case b e h a v i o r d u r i n g t h e same T-second period.

Difficulties with Statistical Approaches T h e e n f o r c e m e n t a n d policing of statistical traffic d e s c r i p t i o n s p o s e s u n r e ­ solved issues. T h e n e t w o r k u s e s s o m e q u a n t i t i e s for a d m i s s i o n a n d r o u t i n g of real-time traffic a n d o t h e r s for interactive traffic. For real-time traffic, t h e

Control of Networks

n e t w o r k n e e d s to k n o w t h e p a r a m e t e r s of t h e Bahadur-Rao b o u n d (8.13). T h e n e t w o r k also n e e d s to k n o w t h e effective b a n d w i d t h a n d t h e d e c o u p l i n g b a n d ­ w i d t h of t h e traffic. For interactive traffic, t h e n e t w o r k n e e d s to k n o w t h e effective a n d d e c o u p l i n g b a n d w i d t h s of t h e traffic. T h e s e q u a n t i t i e s are n o t readily available. Although w e c a n develop p r o c e d u r e s to m e a s u r e t h e n e c e s s a r y statistics, it is n o t clear t h a t w e s h o u l d ask t h e u s e r s to p e r f o r m t h e s e m e a s u r e m e n t s , a l t h o u g h protocols for doing so are conceivable. It is likely t h a t r e p e a t e d videoc o n f e r e n c e s w i t h t h e s a m e e q u i p m e n t will h a v e similar statistics. Also, m o v i e s t h a t are distributed for v i d e o - o n - d e m a n d applications could r e m e m b e r t h e i r statistics. However, all this b o o k k e e p i n g a p p e a r s r a t h e r c u m b e r s o m e , a n d it is t e m p t i n g to look for solutions t h a t m a k e it u n n e c e s s a r y .

A Proposal We p r o p o s e a n a p p r o a c h t h a t h a s t h e a d v a n t a g e s of b o t h t h e d e t e r m i n i s t i c a n d statistical a p p r o a c h e s . This a p p r o a c h m a k e s s i m p l e e n f o r c e m e n t a n d policing possible, a n d it also p e r m i t s efficient r e s o u r c e utilization. We p r o p o s e t h a t u s e r s specify d e t e r m i n i s t i c b o u n d s , s u c h as PCR, SCR, BT, a n d CDVT, for t h e i r traffic a n d statistical d e s c r i p t i o n s of t h e quality of service. T h e n e t w o r k a d m i t s a n d r o u t e s a n e w c o n n e c t i o n b y a s s u m i n g worst-case behavior. T h e n e t w o r k t h e n m e a s u r e s t h e statistics of t h e traffic: effective a n d d e c o u p l i n g b a n d w i d t h s a n d t h e Bahadur-Rao p a r a m e t e r s if it is a real­ t i m e c o n n e c t i o n . T h e n e t w o r k t h e n u s e s t h e s e statistics to k e e p track of t h e r e s o u r c e s t h a t are n o w o c c u p i e d b y t h e ongoing c o n n e c t i o n s . Tb p r e v e n t t h e u s e r s from always b e h a v i n g in t h e worst possible way, w e p r o p o s e t h a t t h e billing b e b a s e d o n t h e actual r e s o u r c e s t h e y use. In this way, t h e r e s o u r c e utilization is c h a r g e d fairly to t h e different users. T h e r e is o n e q u e s t i o n t h a t this proposal does n o t a n s w e r : h o w do u s e r s verify t h a t t h e y are getting t h e statistical quality of service t h a t t h e y r e q u e s t e d ?

8.5

SUMMARY Virtual circuit n e t w o r k s like ATM s e e k to c o m b i n e t h e gains from statistical multiplexing t h a t packet switching creates w i t h t h e g u a r a n t e e d p e r f o r m a n c e t h a t circuit switching offers. W h e n t h e y s u c c e e d in this goal, virtual circuit n e t w o r k s will b e able to r e a p t h e benefits of e c o n o m i e s of scale, service integration, a n d n e t w o r k externalities. In o r d e r to succeed, however, n e t w o r k

8.6

Notes

425 e n g i n e e r s m u s t solve a n u m b e r of control p r o b l e m s c o n c e r n i n g admission, routing, flow a n d congestion, a n d r e s o u r c e allocation. P r o b l e m s of r o u t i n g a n d a d m i s s i o n are similar to t h o s e t h a t o c c u r in circuitswitched n e t w o r k s ; flow- a n d congestion-control p r o b l e m s arise in packetswitched n e t w o r k s . A n d w e r e v i e w e d t h e a p p r o a c h e s d e v e l o p e d in t h e context of circuit a n d p a c k e t switching. P r o b l e m s of r e s o u r c e allocation are p e c u l i a r to virtual circuit n e t w o r k s . (In circuit-switched n e t w o r k s e a c h c o n n e c t i o n gets a fixed resource, w h e r e a s in d a t a g r a m n e t w o r k s n o r e s o u r c e s are allocated to a c o n n e c t i o n . ) T h e major difficulty in designing resource-allocation m e c h a n i s m s is t h a t t h e y d e p e n d o n (1) characteristics of t h e u s e r traffic, (2) t h e available r e s o u r c e s , a n d (3) t h e quality of service g u a r a n t e e d to t h e user. Of t h e s e t h r e e items, t h e first is t h e m o s t difficult to characterize, a n d w e p r e s e n t e d two a p p r o a c h e s . T h e d e t e r m i n i s t i c a p p r o a c h is b e i n g p u r s u e d b y t h e ATM F o r u m . It is easy to i m p l e m e n t , b u t its worst-case a s s u m p t i o n s will give v e r y conservative a n s w e r s , leading to significant u n d e r u t i l i t z a t i o n of t h e n e t w o r k . T h e statistical a p p r o a c h at t h e p r e s e n t t i m e is still b e i n g d e v e l o p e d in r e s e a r c h laboratories. It is far from standardization, b u t progress is rapid. We h a v e d e s c r i b e d t h e m o s t i m p o r t a n t a c c o m p l i s h m e n t s of b o t h a p p r o a c h e s . T h e discussion of t h e statistical a p p r o a c h does m a k e u s e of t h e c o n c e p t s a n d framework of probability t h e o r y . Without a n i n t r o d u c t i o n to probability t h e o r y , t h e r e a d e r will find it difficult to follow t h e discussion. However, w e h a v e tried to m i n i m i z e t e c h n i c a l c o n c e p t s so t h a t t h e m a t e r i a l is accessible. T h e full r a n g e of t e c h n i c a l detail is c o n c e n t r a t e d in C h a p t e r 9. T h a t c h a p t e r is m e a n t o n l y for t h e r e a d e r w i t h a strong b a c k g r o u n d in probability.

8.6

NOTES For a m o r e detailed discussion of r o u t i n g in circuit-switched n e t w o r k s , see [K94]. T h e gradient projection a l g o r i t h m of section 8.3.3 is s t u d i e d in [BG92]. For a n analysis of t h e w i n d o w congestion-control s c h e m e d e s c r i b e d in sec­ tion 8.3.4, see [J88, FJ91, J96, MW98]. T h e receiver-driven rate-based conges­ tion control for t h e I n t e r n e t is p r o p o s e d in [GCMW99]. For a s t u d y of congestion control for multicast, see [GS99]. RED w a s p r e s e n t e d in [FJ93]. For details of t h e ATM F o r u m ' s r e c o m m e n d a t i o n s s u m m a r i z e d in sec­ tion 8.4.2, see [A93, A96d]. A m o r e detailed analysis of t h e pricing m o d e l of section 8.4.2 c a n b e found in [CWW96]. T h e formulas (8.11) a n d (8.12) are d e r i v e d in [AMS82].

Control of Networks

426

T h e Bahadur-Rao t h e o r e m a p p e a r s in [BR60]. O u r discussion is b a s e d o n [H95]. T h e r e is b y n o w a significant literature u s i n g large deviations t h e o r y to s t u d y buffer overflow probabilities a n d to calculate effective b a n d w i d t h ; see [W86, H u 8 8 , B90, K91, GH91, CW96, DZ93, DV93, KWC93]. T h e small buffer analysis is d e v e l o p e d in [SW95]. T h e o r e m 8.4.1 is obtained i n d e p e n d e n t l y in [CW96] a n d [BD94]. Admission control p r o c e d u r e s b a s e d o n effective b a n d w i d t h a n d decou­ pling b a n d w i d t h are p r e s e n t e d in [HW94]. A n i n t r o d u c t i o n w i t h a n extensive bibliography to this m a t e r i a l is given in [W95].

8.7

PROBLEMS 1. What are r e a s o n a b l e delays for interactive database q u e r i e s , for r e m o t e con­ trol of a video server, for videoconferences, for t e l e p h o n e conversations, a n d for transferring X rays? 2. Consider t h e t r a n s m i s s i o n of a video p r o g r a m w i t h a rate of 1.5 Mbps. A s s u m e t h a t t h e cell loss rate is equal to 1 0 . W h a t is t h e average t i m e b e t w e e n losses? W h a t is t h e probability t h a t t h e t r a n s m i s s i o n of a o n e - h o u r video will n o t h a v e a n y error? - 8

3. W h a t is t h e t r a n s m i s s i o n rate r e q u i r e d to t r a n s m i t in 1 s a 4-inch b y 6-inch p h o t o g r a p h w i t h a resolution of 1,200 dots p e r i n c h a n d 8 bits p e r pixel? 4. A s s u m e a typical Web surfer m a k e s 1-MB r e q u e s t s at r a n d o m t i m e s at a n average rate of 20 r e q u e s t s p e r day. A s s u m e also t h a t t h e r e are 10 million Web surfers a n d t h a t t h e y access 10,000 Web servers. Tb simplify, w e m a k e t h e gross a s s u m p t i o n s t h a t t h e s e servers are equally likely to b e c o n s u l t e d . W h a t is t h e average rate w i t h w h i c h o n e server serves 1-MB r e q u e s t s ? W h a t is t h e m i n i m u m c o n n e c t i o n b a n d w i d t h a n d t h r o u g h p u t of a s e r v e r t h a t is 100 t i m e s m o r e p o p u l a r t h a n average? H o w w o u l d caching t h e i n f o r m a t i o n in distributed servers h e l p r u n this application? 5. Consider t h e exchange of i m p o r t a n t data over a n e t w o r k w i t h a typical endto-end delay of 5 s a n d a n average t r a n s m i s s i o n t i m e of 30 Kbps. Discuss t h e effects of a cell e r r o r rate of 1 0 o n s u c h a n application. W h a t is t h e probability t h a t a 1-MB file will b e c o r r u p t e d b y t r a n s m i s s i o n errors? A s s u m e t h a t errors are c h e c k e d for e a c h block of 64 KB in this file. H o w m a n y blocks are likely to b e c o r r u p t e d ? W h a t is t h e average t i m e u n t i l t h e file is successfully received? H o w w o u l d this t i m e c h a n g e if t h e e r r o r c h e c k i n g w e r e p e r f o r m e d only for t h e c o m p l e t e file? - 4

8.7

Problems

II

1

1

427

•••MWWWMMflM^

Hint: Tb solve this p r o b l e m , y o u n e e d to k n o w t h a t if a coin h a s p r o b a b i l i t y ^ of l a n d i n g o n h e a d s in a n y o n e flip, t h e n it takes o n t h e average 1 /p coin flips u n t i l t h e coin first l a n d s o n h e a d s . 6. C o n s i d e r a large office b u i l d i n g w i t h 1,000 t e l e p h o n e sets. A s s u m e t h a t a n e m p l o y e e u s e s a t e l e p h o n e a b o u t 30 m i n u t e s d u r i n g a typical 8-hour b u s i n e s s day. W h a t is t h e average n u m b e r of t e l e p h o n e calls o n g o i n g at a typical t i m e d u r i n g t h e b u s i n e s s day? A s s u m e t h a t 15% of t h e s e calls are outside calls. Give a r o u g h e s t i m a t e of t h e n u m b e r of outside l i n e s t h a t are required. 7. Consider t h e t r a n s m i s s i o n of m e s s a g e s over a t r a n s m i s s i o n l i n e e q u i p p e d w i t h a buffer. T h e m e s s a g e s h a v e l e n g t h s t h a t are e x p o n e n t i a l l y distributed w i t h m e a n L bits. T h e t r a n s m i t t e r h a s rate C b p s . T h e m e s s a g e s arrive as a Poisson p r o c e s s w i t h rate λ m e s s a g e s p e r second. (a) Using formula (8.1), find t h e average d e l a y p e r m e s s a g e . For λ = 10/s a n d L = 8 χ 10 , find t h e m i n i m u m value of C so t h a t t h e average delay does n o t exceed 0.1 s. (b) H o w does t h e average d e l a y p e r p a c k e t c h a n g e if C a n d L are m u l t i p l i e d b y the same constant? (c) Let u s p r e t e n d t h a t o u r q u e u e m o d e l s a Web s e r v e r t h a t a n s w e r s r e q u e s t s . We decide to divide u p t h e files in t h e s e r v e r into subfiles t h a t are Κ t i m e s smaller. As a result, t h e rate of r e q u e s t s for t h e subfiles b e c o m e s Κλ. W h a t is t h e n e w average d e l a y p e r subfile? 6

8. Consider t h e n e t w o r k of Figure 8.11. A s s u m e t h a t w e c a n choose t h e p a r a m e t e r s (ρ, μ ι , μ ) subject to μ ι + 4μ < 2λ. Find t h e v a l u e s of t h e p a r a m e t e r s t h a t m i n i m i z e t h e average delay p e r packet. 2

2

9. As in t h e p r e v i o u s p r o b l e m , c o n s i d e r t h e n e t w o r k of Figure 8.11. Describe a possible i m p l e m e n t a t i o n of t h e distributed-gradient r o u t i n g a l g o r i t h m for this n e t w o r k . Explain t h e e s t i m a t e s t h a t t h e n o d e s m u s t p e r f o r m a n d t h e i n f o r m a t i o n t h e y m u s t exchange. 10. W i n d o w congestion control is c l a i m e d to b e inefficient for c o n n e c t i o n s w i t h a large b a n d w i d t h χ d e l a y p r o d u c t . We discuss a s i m p l e m o d e l t h a t traces t h e inefficiency to t h e long feedback delay. I m a g i n e a n M / M / l q u e u e fed b y two s t r e a m s w i t h r e s p e c t i v e r a t e s λ a n d a . T h e service rate is μ. T h e average d e l a y t h r o u g h t h e q u e u e is 1/r, w h e r e r = μ — λ — α. We a s s u m e t h a t λ is controlled after a feedback of Τ t i m e u n i t s . We also a s s u m e t h a t t h e delay t h r o u g h t h e q u e u e stabilizes after o n e u n i t of t i m e . T h e feedback a l g o r i t h m is as follows. At t i m e r, t h e m e a s u r e d d e l a y is r(r) = μ — X(t — 1) — ce(t — 1). T h e target d e l a y is 1/r, w h i c h c o r r e s p o n d s

428

Control of Networks

to a target value of μ — λ — a. If r(t) φ r, t h e n t h e feedback informs t h e source to d e c r e a s e λ b y r — r(t). T h i s m e s s a g e r e a c h e s t h e source at t i m e t + T. If t h e source is a w a r e of t h e delay, it r e p l a c e s λ b y λ(ί) — r + r(t). Thus, k(t + T) = max{0, X(t) - r + μ - X(t - 1) - a(t - 1)}. Try this algorithm w i t h r = 1, Τ = 10, μ = 3, λ(ί) = a(t) = 1 for t < 20, a(21) = 2, a n d a(t) = 1 for f > 22. Propose m e t h o d s to stabilize t h e algorithm. 11. Consider t h e following analogy to a call a d m i s s i o n p r o b l e m . A n elevator c a n c a r r y 2,000 kg. We w a n t to decide a n a d m i s s i o n m e c h a n i s m for p e o p l e to get into t h e elevator. T h e rule of t h e g a m e is t h a t o n c e s o m e b o d y is in w e c a n n o t ask h i m or h e r to leave. A s s u m e t h a t e v e r y b o d y w e i g h s less t h a n 125 kg. T h e first p r o c e d u r e is to a d m i t 16 people, a s s u m i n g t h e worst case. T h e s e c o n d p r o c e d u r e is b a s e d o n statistics t h a t give u s t h e distribution of t h e weight of a n arbitrary p e r s o n . For simplicity, a s s u m e t h a t t h e weights are Gaussian w i t h m e a n 70 a n d s t a n d a r d deviation 15. (a) Find t h e n u m b e r of p e r s o n s t h a t c a n b e a d m i t t e d so t h a t t h e probability t h a t t h e elevator is overloaded does n o t exceed 10~~ . (b) T h e third m e t h o d consists in m e a s u r i n g t h e total load of t h e elevator as p e o p l e get in a n d closing t h e doors w h e n t h a t load exceeds 1,875 kg. Estimate t h e e x p e c t e d n u m b e r of p e o p l e t h a t c a n get into t h e elevator using t h e s a m e a s s u m p t i o n s as in p a r t (a). 5

12. Write a p r o g r a m to s i m u l a t e t h e GCRA(T, τ ) a l g o r i t h m a n d verify t h e g r a p h s in Figures 8.20 a n d 8.21. 13. Please refer to t h e figure o n t h e n e x t page. Let's a s s u m e t h a t w e h a v e a time-varying traffic source S. T h i s source g e n e r a t e s a "flow" (instead of cells or packets; w e do this for simplicity) w i t h a n i n t e n s i t y according to s = 2 + sin(i) b p s . T h e flow travels t h r o u g h a m e a s u r e m e n t device M. This device calculates t h e leaky-bucket state (i.e., t h e w a t e r level). T h e l e a k y b u c k e t h a s p a r a m e t e r s a (credit arrival rate in b p s ) a n d b (bucket size in bits). T h e m e a s u r e m e n t device j u s t m e a s u r e s t h e traffic as it p a s s e s t h r o u g h (i.e., it does n o t buffer, delay, or discard "nonconforming" traffic). T h e traffic t h e n e n t e r s a buffer B. T h i s buffer h a s capacity χ (bits) a n d is e m p t i e d at t h e c o n s t a n t rate y (bps). We will call t h e leaky-bucket state w(t) a n d t h e buffer o c c u p a n c y o(t). T h e initial conditions for t h e s e states are w(0) = b a n d o(0) = 0 (i.e., t h e b u c k e t is "full" a n d t h e buffer is e m p t y . )

8.7

Problems wmm ι». m wmtwmmmmmmmmmmmMMmmmmmmmm

429

(a) Let a=y = 2 a n d b = χ = 2. Plot s(f), wit), a n d o(t). H o w is t h e leakyb u c k e t state r e l a t e d to t h e buffer o c c u p a n c y ? (b) Let a = y = 2 a n d b = χ = 1/2. Plot s(r), w(t), a n d o(r). (Note: h e r e w e h a v e t h e s a m e i n p u t p r o c e s s a n d o u t p u t l i n k r a t e as in p a r t (a) above y e t w e h a v e less buffer space, therefore w e s h o u l d expect buffer overflow.) W h a t value of y m u s t w e c h o o s e to avoid buffer overflow? (c) Express t h e p a r a m e t e r y as a function of χ t h a t is r e q u i r e d to avoid buffer overflow. (d) Suppose w e h a d n\ traffic s o u r c e s of t y p e 1 a n d n traffic s o u r c e s of t y p e 2. Each traffic source is r a n d o m . T h e t y p e 1 traffic s o u r c e s e a c h "conform" to a leaky-bucket traffic m o d e l w i t h p a r a m e t e r s a\ a n d b\. Similarly, t h e t y p e 2 traffic s o u r c e s h a v e p a r a m e t e r s a a n d b . If all t h e s e traffic s o u r c e s are to s h a r e a single buffer a n d link, w h a t size buffer a n d link data r a t e are r e q u i r e d to avoid buffer overflow? 2

2

s

Μ

2

Β

y — Η χ

t

14. Consider a square-wave traffic s o u r c e s(t) = 2 if 0 < t < 1/2 or 1 < t < 3/2 or 2 < t < 5/2, a n d so on, a n d s(f) = 0 o t h e r w i s e . T h e traffic from this source p a s s e s b y a leaky-bucket m e a s u r e m e n t device (with p a r a m e t e r s b bits a n d a b p s ) a n d t h e n into a b u f f e r / l i n k (with p a r a m e t e r s χ bits a n d y b p s ) . (a) W h a t r a n g e of link s p e e d s y (or a) result in a large b u t finite buffer n o t overflowing? (b) Over t h a t r a n g e of l i n k s p e e d s , w r i t e a n expression for χ (or b~) for t h e m i n i m u m buffer r e q u i r e d to avoid buffer overflow. (c) Let y — 1.5 a n d χ = 0.25, s k e t c h s(t), w(t), a n d o(t) (i.e., t h e source rate, t h e leaky-bucket state, a n d t h e buffer state). (d) Sketch t h e o u t p u t of t h e buffer for t h e above p a r a m e t e r s . (e) Calculate t h e average delay of s(t) as it p a s s e s t h r o u g h t h e buffer. You s h o u l d leave χ a n d / o r y as p a r a m e t e r s .

9

Control of

CHAPTER

Networks: Mathematical Background

X n this c h a p t e r y o u will l e a r n t h e m a t h e m a t i c a l a n a l y s e s t h a t u n d e r l i e t h e control t e c h n i q u e s u s e d a n d t h e r e s u l t i n g n e t w o r k p e r f o r m a n c e m e a s u r e s de­ scribed in C h a p t e r 8. S o m e sections of this c h a p t e r d e m a n d from t h e r e a d e r a sophisticated b a c k g r o u n d in stochastic processes. A b a s i c k n o w l e d g e of m u l ­ tivariate r a n d o m variables a n d Markov c h a i n s is essential for t h e s e sections. However, y o u c a n skip t h e s e sections a n d still find accessible t h e discussion o n d e t e r m i n i s t i c m o d e l s . If y o u are able to follow t h e m o r e m a t h e m a t i c a l material, y o u will b e able to participate in t h e m a t h e m a t i c a l discussions o n n e t w o r k i n g p u b l i s h e d i n t h e r e s e a r c h j o u r n a l s . But e v e n if y o u are u n a b l e to follow t h e a r g u m e n t , C h a p t e r 8 m a k e s t h e c o n c l u s i o n s of t h o s e discussions accessible. We start b y r e v i e w i n g s o m e k e y results o n M a r k o v c h a i n s in section 9.1. We a p p l y t h e s e results to t h e s t u d y of circuit-switched n e t w o r k s in section 9.2 a n d of d a t a g r a m n e t w o r k s in section 9.3. Section 9.4 explains t h e analysis of virtual circuit n e t w o r k s .

9.1

MARKOV CHAINS In this section, w e r e v i e w Markov c h a i n s a n d discuss s o m e k e y results.

9.1.1

Overview A Markov c h a i n is a m o d e l of t h e r a n d o m m o t i o n of a n object i n a discrete set of possible locations. Ttoo v e r s i o n s of this m o d e l a r e of i n t e r e s t to u s : discrete

Control of Networks: Mathematical Background

432

t i m e a n d c o n t i n u o u s t i m e . In discrete time, t h e position of t h e object—called t h e state of t h e Markov chain—is r e c o r d e d e v e r y u n i t of time, t h a t is, at t i m e s 0, 1, 2, a n d so on. In c o n t i n u o u s time, t h e state is o b s e r v e d at all t i m e s t > 0. O n e c a n t h i n k of t h e c o n t i n u o u s - t i m e m o d e l as b e i n g a discrete t i m e m o d e l w h e r e t h e t i m e u n i t is infinitesimally small. T h e state of t h e Markov c h a i n c h a n g e s r a n d o m l y . In discrete time, t h e r e is a die at e v e r y location. Every t i m e unit, t h e Markov c h a i n tosses t h e die at its c u r r e n t location to decide w h e r e to j u m p next. In t h a t way, t h e law of t h e future m o t i o n of t h e state d e p e n d s only o n t h e p r e s e n t location a n d n o t o n p r e v i o u s locations. T h i s k e y p r o p e r t y t h a t t h e Markov c h a i n h a s of "forgetting" its past locations greatly simplifies t h e analysis. E n g i n e e r s u s e Markov c h a i n s to m o d e l t h e progression of t h e calls t h a t a t e l e p h o n e n e t w o r k carries a n d of p a c k e t s t h a t a d a t a g r a m or virtual circuit n e t w o r k t r a n s p o r t s . T h e r a n d o m n e s s in t h e s e m o d e l s reflects t h e u n c e r t a i n t y about w h e n u s e r s place calls or s e n d p a c k e t s a n d a b o u t t h e l e n g t h of p a c k e t s a n d t h e i r destination. T h e r a n d o m n e s s also c a p t u r e s t h e t r a n s m i s s i o n e r r o r s a n d failures of devices. T h e t h e o r y of Markov c h a i n s tells u s h o w to calculate t h e fraction of t i m e t h a t t h e state of t h e Markov c h a i n s p e n d s in t h e different locations. N e t w o r k e n g i n e e r s u s e t h a t t h e o r y to e s t i m a t e t h e delays a n d losses of p a c k e t s in n e t w o r k s or t h e fraction of t i m e t h a t t e l e p h o n e calls are b l o c k e d b e c a u s e all t h e circuits are b u s y . T h e e n g i n e e r s t h e n u s e t h e s e e s t i m a t e s to design a n d control n e t w o r k s , as w e explained in C h a p t e r 8. In this section, w e r e v i e w t h e m a i n results of t h e t h e o r y of Markov chains, a n d w e illustrate t h e s e results w i t h examples.

9.1.2

Discrete Time O n e is given a set X, called t h e state space. We call t h e e l e m e n t s of X states. T h e set X is countable. T h a t is, X is e i t h e r a finite set X = . . . , IN} (for s o m e finite n u m b e r N ) , or X is infinite b u t w e c a n e n u m e r a t e its e l e m e n t s exhaustively as X = 1 2 , 1 3 , . . . } . For instance, t h e set of n o n n e g a t i v e i n t e g e r s Z+ = {0, 1, 2, 3 , . . . } is countable, b u t t h e set of real n u m b e r s b e t w e e n 0 a n d 1, t h a t is, t h e interval [0, 1], is n o t countable. For o u r p u r p o s e , t h e r e l e v a n c e of a set b e i n g countable is t h a t if t h e e l e m e n t s in a collection A of positive real n u m b e r s add u p to 1, t h e n t h e collection m u s t b e countable. We d e n o t e typical e l e m e n t s of X as i,;, k. For i e X w e are given a list of n o n n e g a t i v e n u m b e r s {P(i,j),j e X} t h a t a d d u p to 1. T h a t is, 0 < P(i,j) < 1 for all ij € X a n d ^

P(i,j) = 1 for all i e X.

9.1

Markov Chains

433

We t h i n k of Ρ = {P(i,j) i,j £ X} as a m a t r i x . T h i s m a t r i x is Ν b y Ν if X is finite a n d h a s Ν e l e m e n t s . If X is infinite, t h e n t h e m a t r i x Ρ is infinite. T h e m a t r i x Ρ is called a transition probability matrix. We are also given a probability distribution πο o n X, t h a t is, a collection of n o n n e g a t i v e n u m b e r s {no(i) i € X} t h a t add u p to 1. T h a t is, t

t

0 < 7To(i) < 1 for all i e X a n d

^

π (ι) 0

=

1.

We n o w define a r a n d o m s e q u e n c e χ = {xo, x\, x ,. X as follows: 2

P(*o = io, *i = h, . · · , *n = i ) = no(io)P(io, h)P(h, n

·.} t h a t takes v a l u e s in

h) x · · · x Ρ(ΐη-\,

in)

(9.1)

for all η > 0 a n d all ίο, h,..., z € X. T h e r a n d o m s e q u e n c e χ is called a Markov c h a i n w i t h t r a n s i t i o n probability m a t r i x Ρ a n d initial distribution no. T h e definition of χ specifies t h a t xo is selected in X according to t h e probability distribution πο a n d t h a t if x = i, t h e n x + i = ; w i t h probability P(i,j), i n d e p e n d e n t l y of t h e v a l u e s x for m 0. T h e state t r a n s i t i o n n

n

n

m

n

n

a

b 9.1 FIGURE

State transition diagrams of three discrete-time Markov chains.

Control of Networks: Mathematical Background

434

diagram is a directed graph. A path in s u c h a g r a p h is a succession of arrows s u c h t h a t t h e e n d of o n e a r r o w is t h e start of t h e next arrow. A p a t h c o r r e s p o n d s to a possible trajectory of t h e Markov c h a i n in t h e state space. We c a n calculate t h e probability distribution of x as follows. For η = 1 w e have n

7Ti(i) := P(*i = 0

= Σ *° P(

= j

' *i = 0

=Σ

;'EX

*o(7)P(7, 0.

(9.2)

;"€X

Define 7τ (ι) := P ( x = ι) for i e X a n d let π b e t h e r o w vector w i t h e l e m e n t s {π„(0, i e X}· We c a n rewrite (9.2) as η

n

η

7TI = 7T()P

w h e r e t h e right-hand side d e n o t e s t h e p r o d u c t of t h e r o w vector no b y t h e m a t r i x P. W h e n X is infinite, t h e m a t r i x multiplication involves infinite s u m s . By r e p e a t i n g t h e a r g u m e n t t h a t led u s to (9.2) y o u c a n verify t h a t 7Γ +ι = π Ρ , η

η

η > 0.

(9.3)

F r o m this identity w e c o n c l u d e t h a t π = πΡ

(9.4)

η

η

0

w h e r e P is t h e n t h p o w e r of t h e m a t r i x Ρ defined as t h e p r o d u c t of Ρ b y itself η times. For instance, w i t h t h e transition m a t r i x of t h e leftmost d i a g r a m of Fig­ u r e 9.1, n

w e find 1

pn

—

~ L n b

#n

1 - bn J '

where a + b(l - a - h) — a+b

n

a = n

Λ Ί

and b = n

b+

a(l-a-b) — . a+b

We say t h a t a probability distribution π is invariant m a t r i x Ρ if π=πΡ.

n

for t h e transition probability

(9.6)

9.1

Markov Chains

435

T h e e q u a t i o n s (9.6) are t h e balance equations for t h e t r a n s i t i o n probability m a t r i x P . Note t h a t if π is i n v a r i a n t for Ρ a n d if πο = π, t h e n (9.4) i m p l i e s t h a t π — π for all η > 0. For t h e t r a n s i t i o n probability m a t r i x Ρ given in (9.5) w e find t h a t t h e o n l y solution 7Γ = [7T(0), 7T(1)] of t h e b a l a n c e e q u a t i o n s (9.6) s u c h t h a t 7 r ( 0 ) 4- ττ(1) = 1 is η

(9.7) Before describing t h e m a i n results a b o u t discrete-time Markov chains, w e m u s t i n t r o d u c e a few definitions. Definition 9.1.1 Let Ρ b e a t r a n s i t i o n probability m a t r i x o n t h e state space X. T h e t r a n s i t i o n m a t r i x Ρ is irreducible if it is possible for a Markov c h a i n w i t h t r a n s i t i o n m a t r i x Ρ to m o v e from a n y state i to a n y o t h e r state ; in finite t i m e . In o t h e r words, Ρ is irreducible if t h e r e is a p a t h from e v e r y i to e v e r y o t h e r ; in t h e state transition d i a g r a m t h a t c o r r e s p o n d s to P . A Markov c h a i n w i t h t r a n s i t i o n probability m a t r i x Ρ is said to b e irreducible if Ρ is irreducible. For instance, looking at t h e state t r a n s i t i o n d i a g r a m s of Figure 9.1, w e find t h a t t h e two leftmost Markov c h a i n s are irreducible, w h e r e a s t h e r i g h t m o s t is not. T h e following t h e o r e m tells u s a b o u t t h e possible i n v a r i a n t distributions of a n irreducible Markov c h a i n . Theorem 9.1.1 A n irreducible Markov c h a i n h a s at m o s t o n e i n v a r i a n t distri­ b u t i o n . It certainly h a s o n e if it is finite. T h e Markov c h a i n is said to b e positive r e c u r r e n t if it h a s o n e i n v a r i a n t distribution. We n o t e d earlier t h a t t h e leftmost M a r k o v c h a i n of Figure 9.1 h a s a u n i q u e in­ v a r i a n t distribution t h a t is given b y (9.7). T h e t h e o r e m tells u s t h a t t h e Markov c h a i n in t h e c e n t e r of Figure 9.1 also h a s a u n i q u e i n v a r i a n t distribution. T h e i n v a r i a n t distribution, w h e n it exists, m e a s u r e s t h e fraction of t i m e t h a t t h e Markov c h a i n s p e n d s in t h e various states. T h i s r e l a t i o n s h i p is e x p r e s s e d in t h e following t h e o r e m . Theorem 9.1.2

Let χ = {x , η > 0} b e a n irreducible Markov c h a i n o n X. T h e n n

N-l

(9.8) n=0

Control of Networks: Mathematical Background

436

w h e r e π is t h e u n i q u e i n v a r i a n t distribution of t h e Markov c h a i n if it exists a n d π (ι) : = 0 for ι e X if t h e Markov c h a i n h a s n o i n v a r i a n t distribution. In t h e identity (9.8), t h e n o t a t i o n 1 {x = i} h a s t h e value 1 if x = i and t h e value 0 otherwise. T h u s , t h e q u a n t i t y n

n

is t h e fraction of t i m e t h a t t h e Markov c h a i n s p e n d s in state i d u r i n g t h e first Ν u n i t s of t i m e . T h i s fraction of t i m e is r a n d o m b e c a u s e it d e p e n d s o n t h e particular realization of t h e s e q u e n c e χ t h a t h a p p e n s to occur. T h e t h e o r e m says that, in t h e long t e r m , this fraction of t i m e a p p r o a c h e s a n o n r a n d o m q u a n ­ tity π(ι). T h u s , if t h e Markov c h a i n h a s a n i n v a r i a n t distribution π , t h e n π(ϊ) is t h e long-term fraction of t i m e t h a t t h e Markov c h a i n s p e n d s in state i, for i e X. If t h e Markov c h a i n does n o t h a v e a n i n v a r i a n t distribution, t h e n t h e fraction of t i m e t h a t t h e Markov c h a i n s p e n d s in a n y o n e state is negligible. W h a t h a p p e n s in t h a t case is e i t h e r t h a t t h e Markov c h a i n is w a n d e r i n g off to infinity, if o n e e n u m e r a t e s t h e states in a n y arbitrary way, or t h a t it visits all t h e states infinitely often b u t m a k e s s u c h large excursions in t h e state space t h a t it s p e n d s a negligible fraction of t i m e in a n y finite set of states. T h e following t h e o r e m gives a n o t h e r i m p o r t a n t i n t e r p r e t a t i o n of t h e in­ variant distribution. However, w e n e e d o n e m o r e definition to state t h a t result. Let Ρ b e t h e transition probability m a t r i x of a n irreducible Markov chain. Define (9.9)

d(i) = gcd{n > l | P ( i , i) > 0}, i e X. n

In this definition, gcd A d e n o t e s t h e greatest c o m m o n divisor of t h e e l e m e n t s of t h e set A. For instance, gcd{6, 9 , 1 2 , 1 5 , . . . } = 3 a n d g c d f l , 4, 6, 8 , 1 0 , 1 2 , . . .} = 1. T h e set in (9.9) is t h e collection of n u m b e r s of steps η s u c h t h a t t h e Markov c h a i n c a n go from t h e state i b a c k to itself in η steps. For instance, consider t h e leftmost Markov c h a i n of Figure 9.1 a n d a s s u m e t h a t 0 < α < 1 a n d 0 < b < 1. O n e finds t h a t {n > l | P ( 0 , 0) > 0} = {1, 2, 3, 4 , . . . } so t h a t d(0) = gcd{l, 2, 3, 4, . . . } = 1. Similarly, o n e c a n verify t h a t = 1. Now a s s u m e t h a t α = b = 1. T h e n o n e finds d(0) = gcd{2, 4, 6, 8, . . . } = 2 a n d d ( l ) = gcd{2, 4, 6, 8 , . . . } = 2. It c a n b e p r o v e d that, for a n y irreducible Markov chain, d(i) = d for all ι e X. This leads u s to t h e following definition. n

9.1

Markov Chains

437

Definition 9.1.2 Define

Let Ρ b e a n irreducible probability t r a n s i t i o n m a t r i x o n X.

d = gcd{n > l\P (i n

t

i) > 0}, i e X.

If d > 1, t h e n Ρ is said to b e periodic with period d.lfd=l, t h e n Ρ is said to b e aperiodic. A Markov c h a i n w i t h t r a n s i t i o n probability m a t r i x Ρ is also said to b e aperiodic or periodic w i t h p e r i o d d. T h u s , t h e leftmost Markov c h a i n in Figure 9.1 is aperiodic if 0 < a < 1 a n d 0 < b < 1; it is periodic w i t h p e r i o d 2 if a = 1 = b. Note t h a t in t h e latter case, t h e values x a l t e r n a t e b e t w e e n 0 a n d 1 for η > 0. T h u s , in t h a t case, if πο(0) = α w i t h 0 < a < 1, t h e n 7T2 (0) = a for η > 0 a n d 7Τ2 +ι (0) = 1 - a for η > 0. T h i s e x a m p l e s h o w s t h a t t h e distribution at t i m e η, n a l t e r n a t e s b e t w e e n t w o v a l u e s a n d therefore does n o t converge. T h e periodicity 2 of t h e M a r k o v c h a i n is reflected in t h e periodicity of its distribution as a function of t i m e . T h e Markov c h a i n in t h e c e n t e r of Figure 9.1 is aperiodic. I n d e e d , {n > l\P (0, 0) > 0} = {2, 3, 4, 5 , . . . } so t h a t d = 1. O n e m a y h o p e t h a t if t h e Markov c h a i n is aperiodic, t h e distributions n m a y converge. This is i n d e e d t h e case, as t h e following t h e o r e m m a k e s precise. n

n

η

n

n

n

Theorem 9.1.3 Let χ b e a n irreducible a n d aperiodic M a r k o v c h a i n w i t h i n v a r i a n t distribution n. T h e n , for a n y initial distribution no, π>ι(0 - * n(i) as n - > oo, for all i e X. T h e above t h e o r e m tells u s t h a t if w e start a Markov c h a i n χ t h a t is irreducible a n d aperiodic a n d h a s a n i n v a r i a n t distribution η a n d if w e wait l o n g e n o u g h , t h e n t h e probability of finding t h e Markov c h a i n in state i is close to n(i). T h e i n t e r p r e t a t i o n is t h a t t h e Markov c h a i n approaches steady state. It is often comforting to b e able to s h o w t h a t a Markov c h a i n is positive r e c u r r e n t e v e n if o n e is n o t able to calculate its i n v a r i a n t distribution. For instance, if t h e Markov c h a i n is t h e m o d e l of a buffer o c c u p a n c y , t h e n s h o w i n g t h a t it is positive r e c u r r e n t tells u s t h a t t h e buffer e m p t i e s relatively f r e q u e n t l y a n d gives u s a s e n s e of t h e s y s t e m ' s stability. T h e following result is o n e of a n u m b e r of useful sufficient c o n d i t i o n s for positive r e c u r r e n c e . T h i s r e s u l t h a s t h e a d v a n t a g e of b e i n g intuitive a n d of b e i n g applicable to m a n y q u e u i n g s y s t e m s .

Control of Networks: Mathematical Background

438

Theorem 9.1.4 Let χ b e a n irreducible Markov c h a i n o n X a n d V : X -> [0, oo) s o m e function. Define t h e drift A(x) o f / ( . ) at χ b y Δ(χ) : = E [ V ( x i ) - V(x )\x n +

n

n

= x], for χ e X.

A s s u m e t h a t t h e r e is s o m e finite subset S of X a n d s o m e c o n s t a n t s D > 0 a n d A < oo s u c h t h a t A(x)0if Ρ(

T h e r a n d o m variable τ is exponentially

>t) = e~ , xt

τ

distributed

for t > 0.

with

rate

(9.12)

For c o n v e n i e n c e , w h e n λ = 0, w e define τ to b e a n infinite r a n d o m variable (i.e., τ = + o o w i t h probability 1). T h e following p r o p e r t i e s follow from this definition. Theorem 9.1.5 Let τ b e e x p o n e n t i a l l y distributed w i t h r a t e λ. T h e n (a) T h e m e a n value Ε ( τ ) of τ is given b y

9.1

Markov Chains

439

(b) T h e r a n d o m variable is memoryless. Ρ[τ > t

+ S|T

T h a t is,

> s] = Ρ ( τ > r), for all s, r > 0.

T h e m e m o r y l e s s p r o p e r t y c a n b e i n t e r p r e t e d as follows. A s s u m e t h a t a light b u l b h a s a n e x p o n e n t i a l l y d i s t r i b u t e d lifetime. T h e n k n o w i n g h o w old t h e b u l b is d o e s n o t h e l p p r e d i c t h o w l o n g it will still live. I n o t h e r words, a n old b u l b is exactly as good as a n e w o n e : t h e b u l b d o e s n o t age. ( T h a t is, u n t i l it s u d d e n l y dies.) Using e x p o n e n t i a l distribution w e c a n c o n s t r u c t a c o n t i n u o u s - t i m e Markov chain. We first define a rate m a t r i x Q . Definition 9.1.4 d = {q(i,j)

Let X b e a c o u n t a b l e set. A rate matrix Q o n X is a collection

i,j € X} of real n u m b e r s s u c h t h a t

t

0 < q(i, f) < oo, for all χφ] e X, a n d —q(i, i) = q(i) : = ^

q(i,j) < oo, for all i e X.

We are n o w r e a d y to define a c o n t i n u o u s - t i m e M a r k o v c h a i n . Definition 9.1.5 Let X b e a c o u n t a b l e set a n d Q a r a t e m a t r i x o n X. Let also π b e a probability distribution o n X. We define a c o n t i n u o u s - t i m e M a r k o v c h a i n x : = {*t, t > 0} o n X w i t h rate m a t r i x Q. a n d initial distribution π as follows. First o n e c h o o s e s XQ w i t h distribution π i n X. T h a t is, P(*o = 0 = TT(0 for ieX. Second, if xo = i, o n e selects a r a n d o m t i m e τ t h a t is e x p o n e n t i a l l y dis­ t r i b u t e d w i t h rate q(i). T h e p r o c e s s χ is defined so t h a t x = i for 0 < t < τ. t

Third, at t i m e t = τ, t h e p r o c e s s χ m a k e s a j u m p from its initial v a l u e i to a n e w value ; t h a t is s e l e c t e d i n d e p e n d e n t l y of τ a n d so t h a t Ρ[χ =j\ τ

Xo

= i,r] = r(i,j)

:=

φ i.

T h e c o n s t r u c t i o n t h e n r e s u m e s from x =j at t i m e τ, i n d e p e n d e n t l y of t h e p r o c e s s before t i m e τ. x

Figure 9.2 s h o w s a typical realization of t h e p r o c e s s x. T h u s , t h e p r o c e s s starts i n s o m e state £o a n d k e e p s t h a t v a l u e for ro, t h e n visits a s e q u e n c e of states

Control of Networks: Mathematical Background

440

ξο

4 0. T h a t is, w e h a v e t h e following p r o p e r t y . t

Theorem 9.1.6 Let χ b e a c o n t i n u o u s - t i m e Markov c h a i n w i t h rate m a t r i x Q o n X. T h e n , for a n y set A of trajectories in X, P[(x , s>t)e s

A\x = i; x t

Ut

u 0) € A\x s

0

= i\

T h e set A in t h e t h e o r e m is a n y set of trajectories of χ i n X of t h e form A = {x|^€S*forfc=l,...,lC}, w h e r e Κ < oo, 0 < t\ < t < · · · < t , a n d Sfc is a s u b s e t of X for k = 1 , . . . , K. 2

K

9.1

Markov Chains

•

•

• •

441

m

T h i s result says t h a t t h e o n l y i n f o r m a t i o n a b o u t t h e trajectory of χ u p to t i m e t t h a t is useful for p r e d i c t i n g t h e trajectory after t i m e r is t h e c u r r e n t value x . T h e intuitive justification of this t h e o r e m is t h a t t h e p a s t v a l u e s of χ are i r r e l e v a n t b e c a u s e of t h e w a y successive v a l u e s are selected. Moreover, t h e p a s t h o l d i n g t i m e s are i r r e l e v a n t b e c a u s e t h e future o n e s d e p e n d o n l y o n t h e future states. Finally, t h e e l a p s e d value of t h e c u r r e n t h o l d i n g t i m e is i r r e l e v a n t for p r e d i c t i n g w h e n t h e n e x t j u m p will o c c u r b e c a u s e t h a t h o l d i n g t i m e is m e m o r y l e s s . t

Next w e s t u d y t h e i n v a r i a n t distribution of a c o n t i n u o u s - t i m e Markov chain. We first n e e d to define irreducibility. Definition 9.1.6 A rate m a t r i x Q o n X is irreducible if q(i) > 0 for all i e X a n d if t h e t r a n s i t i o n probability m a t r i x Γ defined b y

Γ(ί,;) =

i=jeX

is irreducible. A c o n t i n u o u s - t i m e Markov c h a i n w i t h rate m a t r i x Q. is said to b e if its rate m a t r i x Q is irreducible.

irreducible

T h u s , a c o n t i n u o u s - t i m e Markov c h a i n is irreducible if it c a n go from a n y state to a n y o t h e r state in finite t i m e . With this definition w e c a n state t h e m a i n result a b o u t c o n t i n u o u s - t i m e M a r k o v c h a i n s . Theorem 9.1.7 Let χ b e a n irreducible M a r k o v c h a i n o n X w i t h r a t e m a t r i x Q. a n d initial distribution π. (a) T h e distribution π is invariant, t h a t is, P(x = i) = 7 τ ( ϊ ) , for all i e X a n d t > 0, t

if a n d o n l y if π solves t h e following balance

equations:

^2n(i)q(i j) = 0, for all; € X. f

ieX

(b) T h e Markov c h a i n is s t a t i o n a r y if a n d o n l y if its initial distribution is invariant. (c) T h e Markov c h a i n h a s e i t h e r n o or o n e i n v a r i a n t distribution. It c e r t a i n l y h a s o n e if X is finite.

Control of Networks: Mathematical Background

442

(d) If t h e Markov c h a i n h a s o n e i n v a r i a n t distribution π , t h e n l i m P(x = ι) = π(ι), for all i e X, a n d t

t-»oo

r

T

ι

lim — /

l{x = i}ds = π(ϊ), for all i e X. s

(e) If t h e Markov c h a i n h a s n o i n v a r i a n t distribution, t h e n l i m P(x = 0 = 0, for all i e X, a n d t

t-»oo

r

ι

T

lim - / l{x = i}ds = 0, f o r a l l i e X . Τ Jo It is often useful to consider a Markov c h a i n r e v e r s e d i n t i m e . T h e following t h e o r e m s u m m a r i z e s t h e k e y features of t h a t process. s

T^oc

Theorem 9.1.8 (a) Let χ b e a stationary c o n t i n u o u s - t i m e Markov c h a i n o n X w i t h rate m a t r i x Q a n d i n v a r i a n t distribution π. T h e n χ : = { x _ , 0 < t < T} T

r

is a stationary c o n t i n u o u s - t i m e Markov c h a i n o n X w i t h i n v a r i a n t distribution π a n d w i t h rate m a t r i x Q given b y ·\

q(hJ) =

*(j)q(j,i)

. ·

v

——,i,;eX. 7T(0

T h e process χ is said to b e χ r e v e r s e d i n t i m e . (b) Let χ b e a c o n t i n u o u s - t i m e Markov c h a i n o n X w i t h rate m a t r i x Q . If π is a distribution o n X a n d Q/ is a rate m a t r i x o n X s u c h t h a t niOqihJ) = x(j)q'(j, 0 , for all*, 7 € X, t h e n 7Γ is i n v a r i a n t for χ a n d Q.' is t h e rate m a t r i x of χ r e v e r s e d i n t i m e . T h e expression for Q. given i n p a r t (a) of t h e t h e o r e m c a n b e u n d e r s t o o d as follows. A s s u m e t h a t χ is stationary w i t h i n v a r i a n t distribution π. T h e n P{x = i, x, =j) = n(i)q(i,j)t Q

+ o(t), for all

i^jeX.

9.2

Circuit-Switched Networks

443

But if χ is χ r e v e r s e d in time, t h e n P(x = i, x =j) = n(i)q(i,j)t 0

+ o(t)

t

= P(x = i, x =;) = P(xo =j, *t = 0 t

0

= *(j)q(j, i)t + o(t)

t

w h i c h s h o w s t h a t n(i)q(i,j)

= n(j)q(j,

i).

With this rapid r e v i e w of t h e m a i n results o n Markov chains, w e are r e a d y to analyze m o d e l s of c o m m u n i c a t i o n n e t w o r k s . We start w i t h circuit-switched n e t w o r k s before m o v i n g o n to packet-switched n e t w o r k s .

9 2

CIRCUIT-SWITCHED NETWORKS In this section w e explain t h e b a s i c t h e o r y of circuit-switched n e t w o r k s . We start b y discussing t h e case of a single switch in section 9.2.1. In section 9.2.2 w e e x a m i n e t h e case of a n e t w o r k .

9.2.1

Single Switch Consider t h e situation s h o w n in Figure 9.3. P h o n e calls arrive at a single switch as a Poisson process w i t h rate p . T h a t is, t h e t i m e s b e t w e e n successive calls are i n d e p e n d e n t a n d e x p o n e n t i a l l y distributed w i t h rate p . T h e r e are Ν circuits to c a r r y t h e calls. If a call arrives w h e n all t h e l i n e s are b u s y , t h e n t h e call is rejected (blocked). T h e d u r a t i o n s of t h e calls a r e i n d e p e n d e n t a n d e x p o n e n t i a l l y distributed w i t h rate 1.

Switch Ρ Arriving calls

Ν Circuits Markov chain model of the number of busy circuits

9.3 FIGURE

Model of a circuit switch. Calls arrive at rate ρ and are carried b y Ν circuits.

Control of Networks: Mathematical Background

444

Markov Chain Model We w a n t to calculate t h e fraction of calls t h a t are blocked. Tb p e r f o r m this calculation, w e observe t h a t t h e n u m b e r x of calls i n progress at t i m e t > 0 is a Markov c h a i n w i t h t h e transition d i a g r a m s h o w n in Figure 9 . 3 . I n this diagram, a n a r r o w from state i to state ; is l a b e l e d w i t h t h e value q(i,j) of t h e e n t r y of t h e rate m a t r i x t h a t c o r r e s p o n d s to t h a t pair of states. t

Tb explain t h e t r a n s i t i o n diagram, a s s u m e t h a t t h e r e are η calls in p r o g r e s s at t i m e r, so t h a t x — n. T h e next t r a n s i t i o n of χ will o c c u r w h e n e i t h e r a n e w call is placed or w h e n o n e of t h e η ongoing calls is c o m p l e t e d . T h e t i m e σ until t h e next call arrives is e x p o n e n t i a l l y d i s t r i b u t e d w i t h r a t e p . T h e t i m e τ until t h e first call in progress t e r m i n a t e s is t h e m i n i m u m of η i n d e p e n d e n t exponentially distributed r a n d o m variables τι, . . . , τ t h a t are t h e residual values of t h e d u r a t i o n s of t h e calls i n progress at t i m e f. T h e s e residual v a l u e s are e x p o n e n t i a l l y distributed w i t h rate 1 b e c a u s e t h e original call d u r a t i o n s are e x p o n e n t i a l l y distributed a n d are therefore m e m o r y l e s s , so t h e residual d u r a t i o n s are distributed exactly as t h e original d u r a t i o n s . Now, t h e m i n i m u m τ of t h e r a n d o m variables τι, . . . , τ is e x p o n e n t i a l l y distributed w i t h r a t e η because t

η

η

Ρ ( τ > 0 = Ρ(τι > r, τ > r , . . . , τ„ > r) 2

= Ρ(τι > Γ ) Ρ ( τ > 0 · · · Ρ ( τ > Γ) = e ~ e f

2

η

_ i

· · · e~* =

e~ . nt

T h e next transition of χ after t i m e t o c c u r s after t h e m i n i m u m of τ a n d σ. T h i s t i m e is e x p o n e n t i a l l y distributed w i t h r a t e ρ + η b e c a u s e Ρ(ηιίη{τ, σ} > t) = Ρ ( τ > r, σ > t) = Ρ ( τ > t)P(a >t) = e~ ent

pt

= eT

( n + p ) i

.

Now, w h e n this t r a n s i t i o n occurs, t h e probability t h a t it is d u e to a n e w call i n s t e a d of a call t e r m i n a t i o n is given b y Ρ\σ < τ\ m i n { a , τ} e (s, s + 6)1 = L

J

Ρ ( τ > s, σ e (s, s + e)) •— — (n + p)e e x p { - ( n + p)s} [exp{—ns}] χ [p€ exp{—ps}] (n + p)e e x p { - ( n + p)s}

=

Ρ n + p'

T h u s , t h e Markov c h a i n χ is s u c h t h a t ρ q(n) = p + n a n d Γ(η, η + 1) =

. η + ρ

9.2

Circuit-Switched Networks

Since Γ(η, η + 1) = q(n, η + l)/q(n),

w e c o n c l u d e t h a t g(n, η + 1) = ρ a n d con­

s e q u e n t l y t h a t q(n, η — 1) = η . O n e c a n u n d e r s t a n d t h e d i a g r a m of Figure 9.3 m o r e directly t h a n b y doing t h e above calculations. T h e d i a g r a m s h o w s t h a t w h e n x = n, t h a t is, w h e n t h e r e t

are η calls in progress, a n e w call a r r i v e s w i t h r a t e p , so t h a t χ j u m p s from η to η + 1 w i t h r a t e p . Also, a call t e r m i n a t e s , a n d χ j u m p s from η to η — 1 w i t h rate n. T h i s r a t e η is t h e s u m of t h e η r a t e s of c o m p l e t i o n of t h e calls in progress.

Invariant Distribution T h e d i a g r a m s h o w s t h a t χ is irreducible. By T h e o r e m 9.1.3, w e calculate t h e i n v a r i a n t distribution of χ b y solving t h e b a l a n c e e q u a t i o n s

^ 7 r ( i ) g ( i , ; ' ) = 0, for a l l ; e X. ieX

By u s i n g t h e definition q(i) = Σ^ι t i o n s as

q(hj) = -q(h

0, w e c a n r e w r i t e t h e s e e q u a ­

(9.13)

T h e i n t e r p r e t a t i o n of t h e e q u a t i o n (9.13) for a p a r t i c u l a r v a l u e of i is t h a t t h e rate of t r a n s i t i o n s out of state i s h o u l d e q u a l t h e r a t e of t r a n s i t i o n s into state i. T h u s , for t h e M a r k o v c h a i n of Figure 9.3, t h e b a l a n c e e q u a t i o n s are as follows: 7Τ(0)ρ = 7Γ(1)

ττ(1)(ρ + 1) = π ( 0 ) ρ + π ( 2 ) 2 π(Ν)Ν

= π(Ν -

l)p.

R e m e m b e r i n g t h a t t h e distribution π m u s t s u m to 1, w e find t h a t t h e solution of t h e s e b a l a n c e e q u a t i o n s is given b y p /n\ n

π(η)

=

, for n = 0, 1, . . . , N .

In particular, w e find t h a t

Control of Networks: Mathematical Background

446 Erlang Loss Formula

T h e formula E(p, N) is called t h e Erlang loss formula. It r e p r e s e n t s t h e fraction of t i m e t h a t t h e Ν circuits are b u s y . This fraction of t i m e is also t h e fraction of t h e calls t h a t arrive w h e n all t h e circuits are b u s y a n d are therefore blocked. Tb see w h y this is t h e case, w e calculate t h e probability a(n) t h a t a call t h a t arrives finds η circuits b u s y : We find a(n) = P[x = n | a call arrives in(t, t + β)] t

P(x = n)P[a call arrives in(t, t + e)\x = n] t

t

P ( a call arrives in(t, t + €)) =

π(η)ρ€

=ττ(η).

Ρ* In t h e last equation, w e u s e d t h e fact t h a t a n arrival occurs in t h e n e x t e t i m e u n i t s w i t h probability pe, i n d e p e n d e n t l y of x b y definition of t h e arrival process. This calculation s h o w s t h a t t h e fraction of calls t h a t find η circuits b u s y is e q u a l to t h e fraction of t i m e t h a t η circuits are b u s y . In particular, for n = N, t h e fraction of calls t h a t are b l o c k e d is t h e fraction of t i m e t h a t all t h e circuits are b u s y . C o n s e q u e n t l y , t h e blocking probability is given b y t h e Erlang loss formula. tt

Insensitivity We h a v e calculated t h e blocking probability at a switch b y a s s u m i n g t h a t t h e call d u r a t i o n s are e x p o n e n t i a l l y distributed. It t u r n s o u t t h a t t h e b l o c k i n g probability does n o t c h a n g e if t h e call d u r a t i o n s h a v e a n o t h e r distribution w i t h t h e s a m e m e a n value. In o t h e r words, t h e blocking probability is insensitive to t h e actual distri­ b u t i o n of t h e call durations. We explain a proof of t h a t i m p o r t a n t result in section 9.3.5.

9.2.2

Network Consider t h e n e t w o r k of Figure 9.4. T h e n e t w o r k h a s Κ switches t h a t are c o n n e c t e d b y L links. Link i h a s ni circuits. A r o u t e is a n acyclic set of links. W h e n a call is r o u t e d along a route, it u s e s o n e circuit in e a c h link along t h e route.

9.2

Circuit-Switched Networks

447

We a s s u m e t h a t calls are p l a c e d along r o u t e r as a Poisson p r o c e s s w i t h r a t e λ for r = 1 , . . . , R. T h e call d u r a t i o n s are i n d e p e n d e n t a n d e x p o n e n t i a l l y distributed w i t h r a t e 1. We w a n t to a n a l y z e t h e b l o c k i n g probability of calls. Γ

D e n o t e b y x\ t h e n u m b e r of calls i n progress a l o n g r o u t e r for r = 1 , . . . , R a n d for t > 0. Let x = (#[, r = l,...,R). Because of t h e m e m o r y l e s s p r o p e r t y of t h e e x p o n e n t i a l distribution, t h e p r o c e s s χ = {x , t > 0} is a M a r k o v c h a i n . T h e state s p a c e X of t h a t Markov c h a i n is t

t

R

X = {χ € Ζ* I Σ

x (> rA r

0 ^ n

for i = 1 , . . . , L } .

u

I n this expression, A(r; i) t a k e s t h e v a l u e 1 if r o u t e r goes t h r o u g h l i n k i a n d t h e value 0 o t h e r w i s e . T h u s , £ f xrA(>;0 is t h e n u m b e r of calls t h a t go t h r o u g h l i n k i. T h a t n u m b e r m u s t b e at m o s t Πχ since l i n k i h a s Πχ circuits a n d c a n c a r r y at m o s t πχ calls. = 1

If t h e links h a d a n infinite n u m b e r of circuits, t h e n t h e n u m b e r s of calls in progress along different r o u t e s w o u l d b e i n d e p e n d e n t b e c a u s e t h e y w o u l d n o t interfere w i t h o n e another. I n t h a t case, t h e n u m b e r of calls i n p r o g r e s s along r o u t e r w o u l d b e m o d e l e d b y a M a r k o v c h a i n o n {0, 1, 2, . . . } w i t h a r a t e m a t r i x Q. s u c h t h a t q(n, η + 1) = λ a n d q(n, η - 1) = η . By solving t h e b a l a n c e e q u a t i o n s of this Markov c h a i n w e w o u l d find t h a t its i n v a r i a n t distribution is given b y Γ

k 7 r ( n ) = — e~ , η! n

kr

r

for η > 0.

Control of Networks: Mathematical Background

448

T h u s , if t h e links h a d infinitely m a n y circuits, t h e i n v a r i a n t distribution π of χ would be, b y i n d e p e n d e n c e ,

However, since t h e links h a v e a finite n u m b e r of circuits, calls along different r o u t e s interfere w i t h o n e another. Remarkably, t h e i n v a r i a n t distribution πχ in this case r e m a i n s p r o p o r t i o n a l to π. T h a t is, t h e i n v a r i a n t distribution is given by

**) = * \ ν * * , π(

*x(*

1

}

for (x

x )eX,

1

γ

R

7T(X)

w h e r e π(Χ) : = Σχεχ ( )· t h a t t h e d e n o m i n a t o r n o r m a l i z e s t h e distribu­ tion so t h a t it adds u p to 1 over X. Tb prove this result w e d e n o t e b y Q t h e rate m a t r i x of t h e Markov c h a i n χ o n X, a n d w e observe t h a t π χ

*x(x)q(x,

N

o

y) = xx(y)q(y,

t

e

x), for all x,yeX.

(9.14)

Tb see this, let χ e X a n d y = x + e , w h e r e e is t h e u n i t vector in direction r. T h a t is, y = x + 1 {i = r } . T h e n q(x, y) = k since a transition from χ toy c o r r e s p o n d s to a n arrival of a call along r o u t e r a n d q(y, x) = x + 1 s i n c e t h e t r a n s i t i o n from y to χ is t h e c o m p l e t i o n of a call along r o u t e r w h e n x + 1 calls are in progress along t h a t route. T h u s , e q u a t i o n (9.14) r e a d s r

l

r

l

r

Y

r

-^—π (χ ) ι

λ

· · · n (x ) r

r

7T(X)

· · · n (x )k R

R

= -^—π (χ )

· · · n (x

ι

r

+ 1) · . · n (x )(x

r

λ

R

r

r

R

+ 1),

7T(X)

a n d w e see t h a t this e q u a t i o n is satisfied b e c a u s e of t h e form of n . O t h e r i n s t a n c e s of pairs of states χ a n d y c a n b e c h e c k e d along similar lines. We c a n u s e t h e i n v a r i a n t distribution πχ to calculate t h e fraction of calls t h a t are blocked. Tb do this, w e consider a call t h a t is p l a c e d along r o u t e r. Such a call is b l o c k e d if it is placed w h e n t h e n e t w o r k is in a state χ e X s u c h t h a t χ + e is n o l o n g e r in X. Let u s d e n o t e t h e set of s u c h states b y X . T h e fraction of t i m e t h a t t h e n e t w o r k state is in X is 7Γχ(Χ ). Arguing as w e did for a single switch, w e c a n s h o w t h a t t h e fraction of calls placed along r o u t e r t h a t find t h e n e t w o r k state in X is e q u a l to t h e fraction of t i m e 7Γχ(Χ ) t h a t t h e n e t w o r k state is in t h a t set. T h u s , t h e blocking probability B for a call p l a c e d along r o u t e r is given b y r

r

r

r

Γ

r

Γ

r

(9.15)

9.2

Circuit-Switched Networks

449

Complexity Considerations In principle, t h e n e t w o r k e n g i n e e r s c a n u s e t h e s e formulas to calculate block­ ing probabilities a n d r a t e s of r e v e n u e s of t h e n e t w o r k , as w e discussed in C h a p t e r 8. However, t h e calculation of t h e n u m e r a t o r a n d d e n o m i n a t o r in (9.15) is v e r y t i m e - c o n s u m i n g b e c a u s e of t h e large n u m b e r of e l e m e n t s in t h e sets X a n d X . T h a t n u m b e r of e l e m e n t s is of t h e o r d e r of r

Tb r e d u c e t h e complexity of t h e calculations, it is possible to d e v e l o p r e c u r s i o n s in t h e n*. It is also possible to e v a l u a t e t h e n u m e r a t o r a n d d e n o m i n a t o r of (9.15) b y M o n t e Carlo s i m u l a t i o n s . T h e idea is to g e n e r a t e at r a n d o m a vector χ according to t h e distribution π. T h i s g e n e r a t i o n is m a d e particularly s i m p l e b y t h e p r o d u c t form of π . By r e p e a t i n g t h e e x p e r i m e n t w e t h e n e s t i m a t e π(Χ) as t h e fraction of t h e s a m p l e s χ t h a t h a p p e n to fall in X a n d similarly for n(X ). A s i m p l e calculation c a n b e u s e d to e s t i m a t e t h e n u m b e r of r a n d o m s a m p l e s t h a t s h o u l d b e g e n e r a t e d in o r d e r to e s t i m a t e t h e b l o c k i n g probability w i t h i n a few p e r c e n t w i t h a h i g h degree of confidence (say 95%). T h i s s i m u l a t i o n c a n also b e s p e e d e d u p b y exploiting i m p o r t a n c e s a m p l i n g m e t h o d s . r

Erlang Fixed Point Tb c i r c u m v e n t t h e n u m e r i c a l c o m p l e x i t y of e v a l u a t i n g (9.15), r e s e a r c h e r s h a v e d e v e l o p e d a n a p p r o x i m a t i o n called t h e Erlang fixed-point approximation. This a p p r o x i m a t i o n is b a s e d o n a s s u m i n g t h a t a call is b l o c k e d i n d e p e n d e n t l y b y t h e different links along its route. With this a s s u m p t i o n , a call along r o u t e r will n o t b e b l o c k e d a n d will a p p e a r o n link i along r o u t e r w i t h probability

n - (l - B j ) , ¥;

€r

w h e r e Bj is t h e probability t h a t l i n k ; b l o c k s t h e call a n d w h e r e t h e p r o d u c t is over all t h e links ;' o t h e r t h a n i along r o u t e r. T h u s , t h e rate k of calls along r o u t e r c o n t r i b u t e s a rate r

k x IT r

l#;er

(l - Bj)

of calls o n link i since this rate is t h e rate of calls t h a t are n o t b l o c k e d b y o t h e r links along r o u t e r. T h u s , b y s u m m i n g over t h e different r o u t e s r w e find t h a t t h e rate pi of calls t h a t are p l a c e d o n link i is given b y R

r=l

Control of Networks: Mathematical Background

450

Now, t h e blocking probability Bj at link ; is a function of t h e rate pj of calls o n t h a t link. In fact, Bj = E(pj, nj) w h e r e E(p, n) is t h e Erlang loss f o r m u l a for η circuits faced b y a Poisson p r o c e s s of calls w i t h rate p. T h u s , w e find t h a t R

ρ = Σ λ τ

τ

χ

Π · ( 1 - E(pj, n )), for ι = 1, . . . . L. ΐ # ;

€ Γ

;

(9.16)

r=l

T h e s e e q u a t i o n s form a set of fixed-point e q u a t i o n s t h a t define t h e r a t e s Pi in t e r m s of t h e m s e l v e s . T h e s e e q u a t i o n s are t h e Erlang fixed-point equations. It h a s b e e n s h o w n t h a t t h e s e e q u a t i o n s provide a good a p p r o x i m a t i o n of t h e blocking probabilities w h e n t h e n e t w o r k is large.

9.3

DATAGRAM NETWORKS In this section w e a p p l y t h e t h e o r y of c o n t i n u o u s - t i m e M a r k o v c h a i n s to t h e analysis of d a t a g r a m n e t w o r k s . I n t h e process, w e d e r i v e t h e k e y r e s u l t s o n product-form q u e u i n g n e t w o r k s .

9.3.1

M/M/1 Queue A n M/M/1

queue is a waiting r o o m e q u i p p e d w i t h a service facility w h e r e

c u s t o m e r s arrive as a Poisson p r o c e s s w i t h rate λ; t h e c u s t o m e r s are s e r v e d b y a single server, a n d t h e i r service t i m e s are i n d e p e n d e n t a n d e x p o n e n t i a l l y distributed w i t h rate μ > λ. T h e first Μ in t h e n o t a t i o n M / M / 1 m e a n s t h a t t h e arrival p r o c e s s is m e m o r y l e s s (Poisson); t h e s e c o n d Μ m e a n s t h a t t h e service t i m e s are m e m o r y l e s s ( e x p o n e n t i a l l y distributed). T h e 1 i n d i c a t e s t h a t t h e r e is a single server. Because of t h e m e m o r y l e s s p r o p e r t y of t h e e x p o n e n t i a l distribution, t h e n u m b e r x of c u s t o m e r s in t h e q u e u e at t i m e t, i n c l u d i n g t h e o n e i n service, is t

a M a r k o v c h a i n w i t h t h e t r a n s i t i o n d i a g r a m s h o w n in Figure 9.5. T h e b a l a n c e e q u a t i o n s for t h e Markov c h a i n are as follows: π ( 0 ) λ = 7Γ(1)μ

π ( η ) ( λ + μ) = π(η - 1)λ + π(η + 1)μ for η > 1 T h e solution of t h e s e e q u a t i o n s is easily verified to b e 7T(n) = (1 - ρ)ρ , η

η > 0, w i t h ρ — - .

(9.17)

9.3

Datagram Networks

9.5

451

λ

λ

λ

λ

U

U

U

μ.

Markov chain model of an M / M / l q u e u e with arrival rate λ and service rate μ. The state of the Markov chain is the n u m b e r of customers in the q u e u e or in service.

In particular, it follows t h a t this i n v a r i a n t distribution is s u c h t h a t n(i)q(ij)

= n(j)q(j, i), for all i,j > 0.

For instance, n(n)q(n, η + 1) = (1 — ρ ) ρ λ = (1 - ρ ) ρ μ = π(η + \)q(n + 1, η). It follows from T h e o r e m 9.1.8 t h a t t h e rate m a t r i x Q. is also t h e rate m a t r i x of t h e Markov c h a i n χ r e v e r s e d in t i m e . T h u s , t h e Markov c h a i n h a s t h e s a m e rate m a t r i x w h e n it is r e v e r s e d in time, a n d it is therefore statistically t h e s a m e after t i m e reversal. We say t h a t t h e Markov c h a i n is time reversible. Note t h a t t h e d e p a r t u r e s from t h e M / M / l q u e u e before t i m e t b e c o m e t h e arrivals after t i m e t w h e n t h e t i m e is reversed. Since t h e q u e u e r e m a i n s a n M / M / l q u e u e after t i m e reversal, t h e arrivals after t i m e t are a Poisson process w i t h rate λ t h a t is i n d e p e n d e n t of t h e q u e u e l e n g t h x at t i m e t. We t h e n r e a c h t h e following conclusion. η

η + 1

t

Theorem 9.3.1 Consider a stationary M / M / l q u e u e . T h e d e p a r t u r e s before t i m e t from t h a t q u e u e are a Poisson p r o c e s s w i t h rate λ t h a t is i n d e p e n d e n t of t h e state of t h e q u e u e at t i m e t. A q u e u e w i t h t h a t p r o p e r t y is said to b e quasi reversible. F r o m t h e i n v a r i a n t distribution of t h e q u e u e l e n g t h , w e c a n derive t h e average q u e u e l e n g t h a n d t h e average d e l a y t h r o u g h t h e q u e u e . For t h e average q u e u e l e n g t h w e find

For t h e average delay t h r o u g h t h e q u e u e , w e first p e r f o r m a direct calculation. T h e probability t h a t a c u s t o m e r arrives w h e n t h e r e are η c u s t o m e r s in t h e q u e u e is π(η). Tb see this, n o t e t h a t P[x = n | a c u s t o m e r arrives in(t, t + e)] t

P[a c u s t o m e r arrives in(t, t + e)\x = n]P(x = η) t

P ( a c u s t o m e r arrives in(t, t + €))

t

λβπ(ή) ke

=

π(η),

Control of Networks: Mathematical Background

452

as claimed. Now, t h e average delay of a c u s t o m e r w h o arrives w h e n t h e r e are η c u s t o m e r s in t h e q u e u e is t h e average of η + 1 service times, t h a t is, (η + 1 ) / μ . Indeed, s u c h a c u s t o m e r m u s t wait for t h e η c u s t o m e r s w h o w e r e a h e a d of h i m to b e served a n d t h e n for his o w n service t i m e before h e c a n leave t h e q u e u e . C o n s e q u e n t l y , t h e average d e l a y Τ of a c u s t o m e r in t h e M / M / 1 q u e u e is given b y

Note t h a t t h e average q u e u e l e n g t h L := E(x ) related b y t h e relationship t

a n d t h e average delay Τ are

L = XT. This relationship, called Little's result, h o l d s for v e r y g e n e r a l q u e u i n g s y s t e m s . O n e intuitive justification for Little's result is as follows. A s s u m e that, o n average, λ c u s t o m e r s go t h r o u g h s o m e service s y s t e m p e r u n i t of time, e a c h c u s t o m e r s p e n d s Τ u n i t s of t i m e in t h e system, a n d L c u s t o m e r s are in t h e s y s t e m at a n y t i m e . We w a n t to argue t h a t L = λ Τ . If e a c h c u s t o m e r p a y s t h e s y s t e m at a u n i t rate while in t h e s y s t e m , t h e s y s t e m gets paid at a n average rate equal to t h e average n u m b e r L of c u s t o m e r s in t h e s y s t e m . O n t h e o t h e r h a n d , e a c h c u s t o m e r p a y s t h e s y s t e m a n average a m o u n t Τ e q u a l to t h e average t i m e s p e n t in t h e s y s t e m . Since λ c u s t o m e r s go t h r o u g h t h e s y s t e m p e r u n i t of t i m e a n d e a c h p a y s a n average of T, w e c o n c l u d e t h a t t h e s y s t e m gets paid at a n average rate e q u a l to λ Τ . H e n c e , L = XT, as w e claimed. T h e M / M / 1 q u e u e m o d e l s a t r a n s m i t t e r w i t h rate c b p s w h e r e p a c k e t s arrive as a Poisson process w i t h rate λ a n d h a v e i n d e p e n d e n t , identically distributed (iid) l e n g t h s e x p o n e n t i a l l y distributed w i t h rate [ic bits. I n d e e d , t h e successive t r a n s m i s s i o n t i m e s of t h e p a c k e t s are t h e n e x p o n e n t i a l l y distributed w i t h rate μ. T h e results of this section e n a b l e u s to calculate t h e average delay of t h e p a c k e t s going t h r o u g h t h e transmitter. Note t h a t if t h e utilization ρ of t h e t r a n s m i t t e r d o e s n o t exceed 80%, t h e n t h e average d e l a y Τ p e r packet does n o t exceed 5 / μ . T h a t is, t h e average delay does n o t exceed five p a c k e t t r a n s m i s s i o n t i m e s . ( T h e average p a c k e t t r a n s m i s s i o n t i m e is e q u a l to μ" ·) In m a n y c o m m u n i c a t i o n s y s t e m s , t h e interarrival t i m e s of p a c k e t s a n d t h e service t i m e s are n o t e x p o n e n t i a l l y distributed. We i n t r o d u c e a m o d e l for d e t e r m i n i s t i c service t i m e s ( s u c h as in ATM n e t w o r k s ) w i t h b a t c h arrivals. 1

93

Datagram Networks

93.2

453

Discrete-Time Queue We m o d e l a n ATM t r a n s m i t t e r w i t h a q u e u e t h a t h a s c o n s t a n t service t i m e s (equal to 1) a n d w h e r e A c u s t o m e r s arrive at t i m e n, for η > 0. We a s s u m e t h a t t h e r a n d o m variables {A , η > 0} are iid, w i t h m e a n λ a n d v a r i a n c e σ . I b analyze t h e q u e u e w e c a n observe t h e n u m b e r of c u s t o m e r s in t h e s y s t e m at t h e successive t i m e s . Let Y b e t h e n u m b e r of c u s t o m e r s i n t h e q u e u e at t i m e n. We a s s u m e t h a t t h e q u e u e o p e r a t e s as follows. At t h e b e g i n n i n g of t h e n t h t i m e epoch, t h e r e are Y c u s t o m e r s i n t h e q u e u e . T h e s e r v e r t h e n s e r v e s o n e of t h e s e c u s t o m e r s (if Y > 0), so t h a t t h e r e are (Y — 1 ) c u s t o m e r s j u s t after t h e service c o m p l e t i o n . T h e n e x t b a t c h of A c u s t o m e r s t h e n e n t e r s t h e q u e u e . Consequently, n

2

n

n

n

+

n

n

n

Y

= A

n + 1

+ (Y

n

- 1)+ = A

n

+ Y ~ l{Y

n

n

n

> 0}.

Because t h e A are iid, t h e p r o c e s s {Y , η > 1} is a discrete-time M a r k o v c h a i n . If w e a s s u m e t h a t Y a n d Y +\ h a v e t h e s a m e distribution, t h a t is, t h a t t h e q u e u e is in s t e a d y state, t h e n w e c a n u s e t h e above e q u a t i o n to calculate t h e m e a n q u e u e l e n g t h as follows. We first calculate t h e m e a n v a l u e of b o t h sides of t h e i d e n t i t y above. We find n

n

n

£Y„+i = EA

n

+ EY

n

- P(Y

n

> 0).

n

Since EY +\ = EY , w e c o n c l u d e t h a t P(Y > 0) = EA = λ. Next w e take t h e m e a n value of t h e s q u a r e s of b o t h sides of t h e identity. W h e n w e u s e t h e i n d e p e n d e n c e of A a n d Y a n d t h e i d e n t i t y Y l(Y > 0) = Y , w e find n

n

n

n

E(Y if

n

= E(A

n+

n

+ Yn~ UYn > 0})

n

= E(A )

+ E(Y )

2

+

2

n

n

+ 2EA EY n

n

n

n

n

2

P(Yn>0)

- 2EA P(Y n

> 0) -

n

2EY

n

= σ + λ +£(Υ„) + λ 2

2

2

+ 2λΕΥ

- 2λ 2

η

I n s t e a d y state, E(Y +i) solve for E Y . We find

= E(Y ) ,

2

2

n

n

2EY . n

so t h a t w e c a n simplify t h e last e q u a l i t y a n d

n

σ -λ 2

EY

n

= L =

2

++ λ

2(1-λ)

2(1-λ)

2*

(9.18)

T h e m a x i m u m d e l a y e x p e r i e n c e d b y a c u s t o m e r is e q u a l to t h e m a x i m u m backlog of t h e q u e u e . Let u s c o n s i d e r a b o u n d D o n t h e average delay. Since

Control of Networks: Mathematical Background

454

λ < 1, t h e above equality s h o w s t h a t t h e c o n s t r a i n t L < D is satisfied if a

2

0, w h e r e for e a c h k = 1, . . . , Κ t h e r a n d o m variables {Α*;, η > 0} are iid w i t h m e a n λ& a n d v a r i a n c e σ^. T h e n w e find t h a t λ = λ Η h λχ and or = σ Λ h σ | . C o n s e q u e n t l y , t h e i n e q u a l i t y (9.19) b e c o m e s n

2

2

< 1.

2D-1

We c a n write this e q u a t i o n as

2

Κ

Σ

σ

t]

η

93

Datagram Networks

455 Ut)

9.6

The arrival and service times of customers at an M / G I / o o queue. The points in the infinite triangle L(t) correspond to customers in the q u e u e at time t. The points in the infinite parallelogram D(s, f) correspond to customers that leave the queue during (s, t].

c o r r e s p o n d to c u s t o m e r s t h a t are i n t h e q u e u e at t i m e t. I n d e e d , if τ < t, t h e n c u s t o m e r η arrived before t i m e t. Also, if τ + σ > t, t h e n t h e c u s t o m e r l e a v e s at t i m e τ + σ > t. η

η

η

η

η

T h e p o i n t s x in t h e infinite p a r a l l e l o g r a m D(s, i] defined as n

D(s, t) := {χ = (τ, a)\s < τ + a < t] c o r r e s p o n d to c u s t o m e r s t h a t leave t h e q u e u e d u r i n g (s, i]. I n d e e d , r + σ is t h e d e p a r t u r e t i m e of c u s t o m e r n. We claim t h a t t h e r a n d o m set of p o i n t s X defines a Poisson m e a s u r e in t h e p l a n e . T h a t m e a n s t h a t t h e n u m b e r s N(A\),..., N ( A M ) of p o i n t s of X i n disjoint sets A i , . . . , A M are i n d e p e n d e n t r a n d o m variables t h a t are Poisson distributed w i t h m e a n s λι, . . . , AM, w h e r e n

X := m

λ I

I f(a)dxdo,

JA

M

η

m = 1,..., Μ

J

a n d w h e r e / ( . ) is t h e probability d e n s i t y of t h e service t i m e s . Tb verify t h e claim, c h o o s e e 1 a n d c o n s i d e r L = (me, (m + l)e] x[0, oo) for meZ. T h e r a n d o m variables N ( L ) for m e Ζ are i n d e p e n d e n t r a n d o m variables t h a t are Poisson w i t h m e a n ke b e c a u s e t h e y are t h e i n c r e m e n t s of t h e Poisson arrival p r o c e s s w i t h r a t e λ over i n t e r v a l s of d u r a t i o n e. N o w = (me (m + l ) e ] χ [ne, (n + l)e] for meZ a n d η > 0. c o n s i d e r t h e sets S for η > 0 are o b t a i n e d b y s a m p l i n g t h e Poisson T h e r a n d o m variables N(S ) T h a t is, e a c h of t h e r a n d o m variable N ( L ) w i t h probabilities p :=f(ne)e. N(L ) p o i n t s of X in t h e set L is a p o i n t of S w i t h p r o b a b i l i t y ^ . I n d e e d , e a c h M

M

m n

}

mn

M

m

n

M

m n

Control of Networks: Mathematical Background

456

c u s t o m e r w h o arrives d u r i n g [me, (m + l ) e ) h a s a service t i m e in t h e interval [ne, (η + 1)β) w i t h probability p . It is a s i m p l e exercise to verify (see section 9.7, p r o b l e m 10) t h a t s u c h a s a m p l i n g of a Poisson r a n d o m variable p r o d u c e s i n d e p e n d e n t Poisson r a n d o m variables w i t h m e a n values λ€ρ , respectively. Also, b y adding i n d e p e n d e n t Poisson r a n d o m variables, o n e obtains a Poisson r a n d o m variable. T h u s , b y a p p r o x i m a t i n g t h e sets A b y u n i o n s of s q u a r e s S and b y using the indepen­ d e n c e a n d Poisson distribution of t h e r a n d o m variables J V ( S ) , o n e c o n c l u d e s t h a t t h e r a n d o m variables N(A ) are i n d e e d i n d e p e n d e n t Poisson r a n d o m vari­ ables w i t h t h e m e a n values i n d i c a t e d in t h e claim. n

η

m

mn

mn

m

In particular, if w e choose a collection of t i m e s t\ < r < · · · < t < t, w e see b y c o n s i d e r i n g t h e disjoint sets D(fi, r ), . . . , D ( r , f), a n d L(r) t h a t t h e depar­ t u r e s from t h e q u e u e d u r i n g t h e intervals [t\, r ) , . . . , [t , t) are i n d e p e n d e n t Poisson r a n d o m variables w i t h m e a n v a l u e s A(r - f i ) , . . . , λ(ί — r ) , respec­ tively, a n d t h a t t h e y are i n d e p e n d e n t of t h e n u m b e r N(L(f)) of c u s t o m e r s in t h e q u e u e at t i m e r. T h e m e a n value of t h e q u e u e l e n g t h N(L(f)) is e q u a l to 2

2

m

m

2

m

2

m

= λ We h a v e p r o v e d t h e following result. Theorem 9 3 . 2 T h e M / G I / o o q u e u e is quasi reversible. T h a t is, t h e d e p a r t u r e process from t h e q u e u e before t i m e t is a Poisson p r o c e s s t h a t is i n d e p e n d e n t of t h e q u e u e l e n g t h at t i m e r. Moreover, t h e q u e u e l e n g t h is a Poisson r a n d o m variable w i t h m e a n ρ := λΕσ\.

933

Jackson Network We n o w e x a m i n e a n e t w o r k of / M / M / 1 q u e u e s as s h o w n in Figure 9.7. C u s t o m e r s arrive from outside as i n d e p e n d e n t Poisson p r o c e s s e s w i t h rate γι into q u e u e i. W h e n a c u s t o m e r leaves q u e u e i, h e j o i n s q u e u e ;' w i t h probability r(i,j), i n d e p e n d e n t l y of t h e past evolution of t h e n e t w o r k . O n l e a v i n g q u e u e i, a c u s t o m e r leaves t h e n e t w o r k w i t h probability

r(i, 0 ) : = 1 - J r y .

9.3

Datagram Networks

457

Average rate through

9.7 FIGURE

Average service

Jackson network. This is a network of single-server queues. The service times in the various queues are i n d e p e n d e n t a n d are exponentially distributed with rate μι in queue i. Customers arrive from outside as i n d e p e n d e n t Poisson processes with rate γι into q u e u e i. T h e routing is Markov.

T h e service t i m e s of t h e c u s t o m e r s are i n d e p e n d e n t , a n d t h e y are e x p o n e n ­ tially distributed w i t h rate μ\ i n q u e u e i. We a s s u m e t h a t t h e probabilities r(i,j) are s u c h t h a t e v e r y c u s t o m e r in t h e n e t w o r k e v e n t u a l l y l e a v e s it. U n d e r this a s s u m p t i o n , t h e r e is a u n i q u e solution to t h e following flow-conservation e q u a t i o n s 7

ki = γι + Σ

;=i

i J>

0, for i = 1, . . . , / .

X r(

(9.21)

For t > 0 define x = (x], t

· · · ,4),

w h e r e d is t h e l e n g t h of q u e u e ; at t i m e t. Because of t h e m e m o r y l e s s p r o p e r t y of t h e e x p o n e n t i a l distribution a n d of t h e w a y t h e r o u t i n g decisions are t a k e n , t h e p r o c e s s χ = {x , t > 0} is a M a r k o v chain. T h e following t h e o r e m gives t h e i n v a r i a n t distribution of t h a t M a r k o v chain. t

t

Theorem 9.3.3 A s s u m e t h a t t h e s o l u t i o n (λι, . . . , λ;) of (9.21) is s u c h t h a t λί < μι for i = 1 , . . . , / . T h e n t h e M a r k o v c h a i n χ a d m i t s t h e following i n v a r i a n t distribution: π(χ\

. . . ,χΙ) = π (χ ) 1

1

· · · 7T (x ), 7

7

(9.22)

Control of Networks: Mathematical Background

458 w h e r e , f o r ; = 1 , . . . , J, ττ ·(η) = (1 ;

pj)pf

t

for η > 0 w h e r e p, : = — .

(9.23)

We give a s i m p l e proof of this t h e o r e m t h a t u s e s t i m e reversal. Define a n e w n e t w o r k w i t h t h e s a m e M / M / 1 q u e u e s b u t w i t h different arrival r a t e s γ- a n d different r o u t i n g probabilities r'(i,j). λζΓ(ι,;) = λ / 0 · ι),

T h e s e v a l u e s a r e s e l e c t e d so t h a t

fori,;€{l,...,/}

and γ[ = λ τ{ί ί

0).

1

T h e s e r a t e s are calculated b y reversing the arrow i n Figure 9.7. D e n o t e b y Q/ t h e rate m a t r i x of this n e w n e t w o r k . A direct verification s h o w s t h a t n(x)q(x,

y) = n(y)q'(y,

x), for all χ a n d y.

(9.24)

For instance, let y = χ + β — t\ so t h a t a t r a n s i t i o n from χ to y o c c u r s w h e n a c u s t o m e r m o v e s from q u e u e ι to q u e u e ; . You t h e n find t h a t ;

q(x, y) = frr(i j) a n d q'(y, χ) = μ / ( / , i). t

E q u a t i o n (9.24) t h e n r e a d s n(x)viir{i j) t

= π(χ + η - ei)^r {j f

t

i),

and this e q u a t i o n is satisfied in v i e w of t h e form of π a n d of t h e definition of r'(j, i). T h e o r e m 9.1.8 t h e n i m p l i e s t h a t π is i n d e e d t h e i n v a r i a n t distribution for t h e Markov c h a i n χ as w e w a n t e d to prove. It follows from t h e above r e s u l t t h a t t h e distribution of t h e vector of q u e u e l e n g t h s i n t h e n e t w o r k is identically t h e s a m e as it w o u l d b e if t h e arrival p r o c e s s e s at all t h e q u e u e s w e r e i n d e p e n d e n t w i t h r a t e Xi at q u e u e i. (It c a n b e s h o w n t h a t t h e p r o c e s s e s are n o t Poisson in g e n e r a l a n d t h a t t h e y a r e c e r t a i n l y not independent.) I n particular, w e c o n c l u d e t h a t t h e average q u e u e l e n g t h in q u e u e i is t h e s a m e as t h e average q u e u e l e n g t h of a n M / M / 1 q u e u e w i t h arrival r a t e Xi a n d w i t h service r a t e μχ. T h i s average q u e u e l e n g t h is

93

Datagram Networks

459

C o n s e q u e n t l y , t h e average n u m b e r L of c u s t o m e r s in t h e n e t w o r k is given b y 7

λ·

7

ι=1 ^

1=1

1

Using Little's result, w e c o n c l u d e t h a t t h e average d e l a y Τ of a c u s t o m e r in t h e n e t w o r k is given b y

T = - T

where γ = Σ

93.4

—,

(9.25)

L

7=1

Yj * ^ e total rate at w h i c h c u s t o m e r s e n t e r t h e n e t w o r k . s

Buffer Occupancy for an Μ Μ F Source In section 8.4.3 w e described a stochastic source as a M a r k o v - m o d u l a t e d fluid (MMF). We c o n s i d e r e d t h e situation in w h i c h this source feeds into a buffer t h a t is d r a i n e d at a c o n s t a n t rate. We derive t h e s t a t i o n a r y distribution of t h e buffer o c c u p a n c y . T h e state of t h e M a r k o v source is z t h e o c c u p a n c y of t h e buffer is x . W h e n z =j, t h e s o u r c e e m i t s fluid at r a t e r(j). T h e t r a n s m i s s i o n rate is c. W h e n t h e source is in s t a t e ; , t h e buffer fills u p at r a t e r(j) — c, so t h a t t h e derivative w i t h r e s p e c t to t of x is r(j) — c w h e n x > 0. (See Figure 9.8.) tf

t

t

t

t

Control of Networks: Mathematical Background

460

Define t h e steady-state distribution of t h e Markov c h a i n (z , x ) b y t

7τ(ι,

X)

t

= P(z = i a n d x < x). t

t

C o n s i d e r t h e evolution of t h e buffer o c c u p a n c y a n d of t h e s o u r c e b e t w e e n t i m e r a n d t i m e t + dt. Note that, for χ > 0, z +dt = i a n d x +& < χ if a n d o n l y if for s o m e ; o n e h a s z = ; , Xt < χ + (c — r(j))dt, a n d if t h e source t h e n j u m p s from state ; to state i. I n d e e d , u n d e r t h e s e conditions, d u r i n g m o s t of t h e interval [r, r + dt] t h e source p r o d u c e s fluid at r a t e r(j) so t h a t t h e buffer o c c u p a n c y i n c r e a s e s b y (r(j) - c)dt. t

t

t

C o n s e q u e n t l y , for χ > 0, ττ(ί, χ) = Σ

ϋ>

π

x

+ ( - r(J))dt)q(j, c

i)dt + π(ι, χ + (c - r{i))dt){\

+ q(i, i)dt)

= Σ^Ο', *)q(J, Wt] + π(ι, x)(l + q(i, i)dt) + —π(ι, x)(c - r(i))dt. Hence,

—n(i,x) = -(c-r(i))

Y^n(j x)q(j,i).

l

}

D e n o t e b y π (χ) t h e r o w vector π ( χ ) = [ττ(1,

* ) , . . . ,TT(M,

x)]

a n d b y | ^ ( A : ) t h e r o w vector

We c a n t h e n r e w r i t e t h e differential e q u a t i o n s as -— (χ) π

= π (χ) A

w h e r e A is t h e m a t r i x A = [a(i,;), 1 < i, j < M] w i t h q(hj) r(j) - c Tb solve t h e s e e q u a t i o n s w e m u s t take into a c c o u n t t h e b o u n d a r y c o n d i t i o n s t h a t w e derive b y c o n s i d e r i n g t h e case χ = 0. Define t h e set S of possible v a l u e s

9.3

Datagram Networks

461

; of ζ s u c h t h a t r(j) < c. By r e p r o d u c i n g t h e derivation above for χ = 0 w e find t h a t for i e S o n e h a s ^-7T(z, 0){r(z) - c} = Σ

*(J> 0)4θ*> 0 ·

Similarly, w e find t h a t for i n o t in S, 7r(i, 0) = 0. T h e s e b o u n d a r y e q u a t i o n s e n a b l e u s to solve t h e l i n e a r e q u a t i o n s . We c a n write t h e solution as follows: M-1

π ( * ) = ]T

d(k)e^ x>0, x

t

w h e r e t h e β% are t h e e i g e n v a l u e s of A—that w e a s s u m e distinct for simplicity— a n d t h e a(k) are proportional to t h e c o r r e s p o n d i n g eigenvectors. O n e eigen­ value, say βο, is zero a n d c o r r e s p o n d s to t h e eigenvector φ e q u a l to t h e sta­ t i o n a r y distribution of z ; t h e o t h e r e i g e n v a l u e s are negative if t h e s y s t e m is positive r e c u r r e n t . Since π(ι, oo) = φ(ϊ) w e c o n c l u d e t h a t a(0) = φ. T h a t is, t

M-1

TT(X) = φ + Σ

aOOeP**, χ > 0.

Tb clarify t h e above calculations, let u s solve explicitly for n(i, x) w h e n t h e MMF is a n on-off Markov fluid w i t h on-rate e q u a l to 1. In t h a t case, t h e process z takes t h e values 0 a n d 1 a n d its rate m a t r i x is s u c h t h a t g(0, 1) = λ a n d q(l, 0) = μ. We h a v e r(0) = 0 a n d r ( l ) = 1, a n d w e a s s u m e t h a t 0 < c < r ( l ) = 1; o t h e r w i s e t h e r e is n o q u e u i n g possible. Moreover, w e m u s t a s s u m e t h a t t h e average i n p u t rate is less t h a n c, t h a t is, λ / ( λ + μ ) < c, o t h e r w i s e t h e q u e u e fills u p a n d h a s n o invariant distribution. We find t h a t t

r A =

λ c

c

λ _λ

η

1-c

i—c 1 - c -I

T h e eigenvalues of A are 0 a n d β := λ / c — μ / ( 1 — c) < 0 w i t h t h e c o r r e s p o n d i n g eigenvectors φ = [ μ / ( λ + μ ) , λ / ( λ + μ)] a n d ν : = [1 — c, c]. T h u s w e k n o w t h a t π(χ) = φ + ave^ , a n d w e find t h e c o n s t a n t a b y u s i n g t h e b o u n d a r y condition 7i(l, 0) = 0. T h i s gives a = —(X/c)/(X + μ ) . Putting all this together yields x

TT(0, χ) =φ(0)~

0(1)

1 - c c

exp{j8*}

TT(1, *) = (!)-0(1) exp{£x}

Control of Networks: Mathematical Background

462

w i t h 0(0) = μ / ( λ + μ ) = 1 — 0(1) a n d β := λ/c — μ / ( 1 — c). F r o m t h e s e expres­ sions w e c a n calculate t h e steady-state probability t h a t t h e buffer o c c u p a n c y exceeds x. We find

0(1)

P(x ) = l- TT(0, X) - ττ(1, x) = t

Figure 9.9 s h o w s a few examples.

93.5

exp{0x}.

>x

c

Insensitivity of Blocking Probability We m e n t i o n e d at t h e e n d of section 9.2.1 t h a t t h e blocking probability for a loss n e t w o r k does n o t d e p e n d o n t h e distribution of t h e h o l d i n g t i m e s of t e l e p h o n e calls. We n o w prove t h a t result. Consider t h e n e t w o r k in t h e left side of Figure 9.10 (page 464). T h e figure s h o w s a closed n e t w o r k w i t h o n e M / G I / o o q u e u e a n d o n e M / M / 1 q u e u e w i t h service rate λ. A s s u m e t h a t t h e r e are Ν c u s t o m e r s in t h e n e t w o r k . T h e o u t p u t s of t h e M / M / 1 q u e u e are calls b e i n g placed, a n d t h e o u t p u t s of t h e M / G I / o o q u e u e are calls b e i n g t e r m i n a t e d . Calls are p l a c e d w i t h rate λ as l o n g as t h e r e are fewer t h a n Ν calls in progress ( r e p r e s e n t e d b y all Ν c u s t o m e r s in t h e M / G I / o o q u e u e ) . W h e n t h e r e are Ν calls in progress, n e w calls are blocked. T h u s , t h e call blocking probability is t h e probability t h a t t h e r e are Ν c u s t o m e r s in t h e M / G I / o o q u e u e . We s h o w t h a t this probability d e p e n d s o n t h e distribution of t h e call h o l d i n g t i m e s — t h e service t i m e s in t h e M / G I / o o q u e u e — o n l y t h r o u g h t h e i r m e a n value. I b analyze t h e n e t w o r k , w e i n t r o d u c e arbitrarily small delays of d u r a t i o n € 1 as s h o w n in t h e right-hand side of Figure 9.10. A s s u m e t h a t at t i m e 0, t h e states of t h e M / G I / o o q u e u e a n d t h e M / M / 1 q u e u e are i n d e p e n d e n t a n d h a v e t h e i n v a r i a n t distributions t h a t c o r r e s p o n d to Poisson i n p u t s w i t h rate a < λ. Moreover, a s s u m e t h a t t h e states of t h e little d e l a y l i n e s at t i m e 0 are s u c h t h a t t h e i r o u t p u t s d u r i n g [0, e) are i n d e p e n d e n t Poisson p r o c e s s e s w i t h rate a > 0 a n d t h a t t h e s e o u t p u t s are i n d e p e n d e n t of t h e states of t h e q u e u e s . C o n s e q u e n t l y , d u r i n g [0, e) t h e inputs of t h e two q u e u e s are i n d e p e n d e n t Poisson processes w i t h rate a. Moreover, b e c a u s e t h e q u e u e s h a v e t h e i n v a r i a n t distribution for t h o s e in­ p u t processes a n d are quasi reversible, t h e outputs of t h e s e q u e u e s are i n d e p e n ­ d e n t Poisson p r o c e s s e s d u r i n g [0, e). In addition, t h e s e o u t p u t s are i n d e p e n d e n t of t h e states of t h e q u e u e s at t i m e e. T h u s , at t i m e 6, t h e state of t h e n e t w o r k h a s t h e s a m e distribution as at t i m e 0 a n d w e c a n r e p e a t t h e a r g u m e n t d u r i n g

93

Datagram Networks

463

(λ, μ, c) = (20, 30,0.42)

π(0,*) π(1,*)

P(x >x) t

,.Ι i I i J.. ••J i I L· J ι ·ι· 0 0.6 1.2 1.8 2.4 3 3.6 4.2 4.8 5.4 6 6.6 7.2 7.8 8.4 9 9.6

•^•J!^^ 9.9

The invariant distribution of a n on-off Markov fluid that feeds a buffer with a constant service rate. Three sets of parameter values are illustrated.

Control of Networks: Mathematical Background

464 M/GIA*

M/M/1

M/GI/oo

r©-—dh 9.10 FIGURE

M/M/1

Network model of loss system. T h e network in the left-hand side of the figure models a loss system. A customer who enters the M/GI/oo queue represents a call being placed. A customer who leaves that queue represents a call being terminated. Thus, calls are placed with rate λ and have iid holding times. T h e network on the right-hand side is modified b y inserting two small delays to simplify the analysis.

[e, 26), a n d so on. C o n s e q u e n t l y , d u r i n g a n y t i m e interval [ne, (n + l ) e ) t h e o u t p u t s of t h e q u e u e s a r e i n d e p e n d e n t Poisson p r o c e s s e s w i t h rate a. Moreover, at a n y t i m e r t h e q u e u e l e n g t h s X in t h e M / G I / o o q u e u e a n d Y in t h e M / M / 1 q u e u e are i n d e p e n d e n t a n d are Poisson w i t h m e a n αΕσ a n d g e o m e t r i c w i t h p a r a m e t e r α / λ , respectively. T h a t is,

^ ' l-J aE(T

(aEa)

m

P(X = m,Y = n)= φ(πι, n) :=

e

/a\

n

/

a\ i l - - j , m , η > 0.

(9.26)

We c o n c l u d e t h a t t h e above distribution is i n v a r i a n t for t h e n e t w o r k w i t h t h e small delays e. Now consider w h a t h a p p e n s w h e n e | 0. T&ke a n arbitrary realization of t h e initial state of t h e n e t w o r k . For € small e n o u g h , t h e delay l i n e s a r e e m p t y . By decreasing € further, w e modify t h e arrival t i m e s into t h e q u e u e b y e. Eventually, for € small e n o u g h , t h e l e n g t h s of t h e t w o q u e u e s (Xf, Yf) at t i m e t stop c h a n g i n g as € k e e p s o n decreasing. Moreover, t h e s e q u e u e l e n g t h s a r e e q u a l to t h e value t h e y w o u l d h a v e for € = 0 w h e n e v e r t i m e r is n o t a n arrival t i m e into o n e of t h e q u e u e s w h e n e = 0. H e n c e , P((Xf, Yf) = (m, n)) - » P((X °, y °) = (m, n)) as e I 0. t

t

It follows t h a t t h e distribution (9.26) is i n v a r i a n t for t h e n e t w o r k w i t h o u t t h e delay lines. Note t h a t for this invariant distribution t h e total n u m b e r of c u s t o m e r s in t h e n e t w o r k is r a n d o m . If w e k n o w t h a t t h e total n u m b e r is N, t h e n w e c a n derive t h e invariant distribution of t h e n e t w o r k as follows. A s s u m e t h a t w e h a v e s o m e Markov c h a i n a p p r o x i m a t i o n χ = {x t > 0} of t h e state of t h e n e t w o r k . Such a n a p p r o x i m a t i o n c a n b e c o n s t r u c t e d b y a p p r o x i m a t i n g t h e h o l d i n g t i m e s b y first tt

9.4

ATM Networks

465

passage t i m e s of finite Markov chains. D e n o t e b y π t h e i n v a r i a n t distribution of t h e Markov c h a i n χ o n its state s p a c e X. A s s u m e t h a t t h e r e is a s u b s e t Y of X s u c h t h a t t h e Markov c h a i n χ is irreducible o n Y a n d c a n n o t leave or e n t e r Y. T h a t is, t h e rate m a t r i x Q of χ is s u c h t h a t q(

y) = 0 if (x, y) e Y χ Y or (*, y) £ Υ χ Y , c

x>

c

w h e r e Y : = X - Y. T h e claim is t h e n t h a t t h e distribution πγ is i n v a r i a n t for χ where c

π(*)1{*€Υ} TTYM :=

77^ 7T(Y)

·

(9.27)

Tb verify this claim, n o t e t h a t

xeY

xeX

T h e first equality follows from (9.27) a n d t h e s e c o n d from t h e i n v a r i a n c e of π . T h u s , t h e i n v a r i a n t distribution of t h e n e t w o r k of t h e left-hand side of Fig­ u r e 9.10 is t h e distribution (9.26) n o r m a l i z e d o n t h e set of states t h a t c o r r e s p o n d to Ν c u s t o m e r s in t h e n e t w o r k . In particular, w e find t h a t t h e i n v a r i a n t prob­ ability t h a t t h e r e are m c u s t o m e r s in t h e M / G I / o o q u e u e a n d Ν — m in t h e M / M / l q u e u e is given b y φ(γη, Ν — m) P[X = m,Y

= N-m\X+

Y = N] =

Σ* 0(η,Ν-η) = ο

p /m! m

Ση=οΡ Μ

, for m = 0 , . . . , N,

η

w h e r e ρ := λΕσ. In particular, for m = N, this formula gives t h e b l o c k i n g probability t h a t is t h u s s e e n to d e p e n d o n t h e distribution of t h e service t i m e s o n l y t h r o u g h t h e i r m e a n value.

9.4

ATM NETWORKS T h e analysis of ATM n e t w o r k s reflects o n e i m p o r t a n t characteristic of t h e s e n e t w o r k s : t h e i r large b a n d w i d t h - d e l a y product. As w e e x p l a i n e d in C h a p t e r 2, this large b a n d w i d t h - d e l a y p r o d u c t m a k e s feedback control largely ineffective. C o n s e q u e n t l y , ATM n e t w o r k s u s e open-loop control strategies. I n this section

Control of Networks: Mathematical Background

466

w e discuss s o m e basic results t h a t n e t w o r k r e s e a r c h e r s u s e to a n a l y z e t h e p e r f o r m a n c e of ATM n e t w o r k s . In section 9.4.1 w e discuss d e t e r m i n i s t i c m o d e l s of n e t w o r k s . T h e s e m o d e l s e n a b l e r e s e a r c h e r s to gain s o m e insight into t h e b e h a v i o r of buffers a n d l e a k y b u c k e t s . In section 9.4.2 w e explain large deviations of iid r a n d o m variables. We explore a generalization in section 9.4.3. We t h e n a p p l y t h e results to t h e analysis of loss probabilities in a buffer in section 9.4.4.

9.4.1

Deterministic Approaches In this section w e develop p e r f o r m a n c e b o u n d s for buffers a n d traffic rates u s i n g d e t e r m i n i s t i c r a t h e r t h a n stochastic analysis.

Linear Bounds Consider a source t h a t t r a n s m i t s at m o s t Β + Rt bits in a n y interval of t seconds, for any possible value of t. We say t h a t s u c h a source p r o d u c e s a (B, R)-traffic to recall t h e s e constraints. A s s u m e t h a t a (B, .R)-traffic goes t h r o u g h a first-come, first-served buffer (initially e m p t y ) w i t h a capacity to store Β bits a n d e q u i p p e d w i t h a t r a n s m i t t e r w i t h rate C bps, w i t h C > R. We claim t h a t t h e buffer n e v e r loses a n y bit a n d t h a t it delays t h e i n p u t s t r e a m b y at m o s t B/C seconds. Tb verify t h e claim, a s s u m e t h a t t h e buffer loses bits at s o m e t i m e Τ a n d t h a t it w a s e m p t y for t h e last t i m e before Τ at t i m e Τ — S. T h e n , d u r i n g [T — S, T], t h e buffer t r a n s m i t s exactly CS bits a n d at m o s t Β + RS bits e n t e r t h e buffer. Since Β + RS — CS < B, it is n o t possible for t h e buffer o c c u p a n c y at t i m e Τ to exceed B. This contradicts t h e a s s u m p t i o n t h a t t h e buffer loses bits at t i m e T. Consider a n o d e t h a t t r a n s m i t s bits from its buffer at rate C w h e n e v e r t h e buffer is n o n e m p t y . A s s u m e t h a t o n e (B, £)-traffic a n d a n o t h e r (B , K'Hraffic s h a r e t h a t buffer. A s s u m e t h a t C>R + R'. T h e i n p u t s t r e a m is a (B + B , R + R')traffic. If t h e n o d e t r a n s m i t s t h e bits in t h e i r o r d e r of arrival, t h e n t h e n o d e delays t h e (B, £)-traffic b y at m o s t (B + B')/C. However, if w e do n o t a s s u m e t h a t t h e n o d e s e n d s t h e bits in t h e i r o r d e r of arrival b u t o n l y t h a t t h e n o d e k e e p s t r a n s m i t t i n g w h e n e v e r it is n o n e m p t y , t h e n w e find t h a t t h e n o d e delays its i n p u t traffic s t r e a m s b y at m o s t Τ = (Β + B )/(C - R - R'). Tb see this, n o t e t h a t t h e buffer c a n n o t r e m a i n n o n e m p t y for a n interval w i t h a d u r a t i o n l o n g e r t h a n T, since in t h a t interval t h e n o d e t r a n s m i t s CT bits, a n d at m o s t B + B' + (R+ R')T = CT bits e n t e r t h e buffer. Moreover, if t h e n o d e t r a n s m i t s all t h e bits of t h e (B, # ) - s t r e a m in t h e i r o r d e r of arrival, t h e n t h e n o d e delays t h e s e bits b y at m o s t S = (B + B )/(C — R'). I n d e e d , t h e worst !

f

f

f

9.4

ATM Networks

467

backlog facing a bit of t h e (B, K)-traffic is Β + B a n d t h e n o d e r u n s o u t of t h a t backlog a n d of t h e s u b s e q u e n t arrivals of t h e (B , R')-traffic after S s e c o n d s . Let u s s u m m a r i z e t h e above o b s e r v a t i o n s in t h e form of a t h e o r e m . f

f

Theorem 9.4.1 By definition, a (B, K)-traffic carries at m o s t Β + Rt bits in a n y t i m e interval of d u r a t i o n t s e c o n d s . A n e t w o r k n o d e c a n avoid losing bits of a (B, #)-traffic b y r e s e r v i n g a buffer capacity of Β bits a n d a b a n d w i d t h of R b p s for t h a t traffic. Moreover, a n e t w o r k n o d e t h a t t r a n s m i t s at rate C w h e n e v e r it is n o t e m p t y a n d t h a t is s h a r e d b y a (B, K)-traffic a n d a (B' R')-traffic delays its i n p u t s b y at m o s t (B + B )/(C — R— R ) s. If t h e n o d e t r a n s m i t s t h e bits of t h e (B, £)-traffic in t h e i r o r d e r of arrival, t h e n it delays t h o s e bits b y at m o s t (B + B )/(C — R') s. t

f

f

f

This t h e o r e m c a n b e u s e d to derive b o u n d s o n delays in n e t w o r k n o d e s a n d therefore t h e end-to-end n e t w o r k delay.

Leaky Bucket Tb e n s u r e t h a t t h e source satisfies t h e above conditions, t h e u s e r c a n control t h a t source w i t h a (E, K ) - l e a k y b u c k e t . T h i s l e a k y b u c k e t h a s a c o u n t e r t h a t a c c u m u l a t e s credits (tokens) at t h e c o n t i n u o u s rate R a n d c a n a c c u m u l a t e Β u n i t s of credit. Tb t r a n s m i t a bit, t h e c o u n t e r m u s t c o n t a i n at least o n e u n i t of credit. Let u s verify t h a t t h e o u t p u t of t h e (B, K ) - l e a k y - b u c k e t controller is a (B, £)-traffic. Tb do this, w e c o n s i d e r a n interval of t i m e [ 0). We m u s t s h o w t h a t t h e o u t p u t of t h e leaky-bucket controller p r o d u c e s at m o s t Β + Rt bits in t h a t t i m e interval. Let A b e t h e n u m b e r of t o k e n s in t h e c o u n t e r at t i m e 5. Note t h a t A < B. D u r i n g t h e t i m e interval [5, 5 + t] t h e c o u n t e r a c c u m u l a t e s Rt t o k e n s . T h u s , A+Rt t o k e n s are available to t h e source to t r a n s m i t bits. C o n s e q u e n t l y , t h e o u t p u t p r o d u c e s at m o s t A+Rt 0}. We a s s u m e t h a t t h e function m is integrable a n d t h a t / °° m(t)dt = M. T h e i n t e r p r e t a t i o n is t h a t m(r) is t h e bit rate of t h e m e s s a g e at t i m e t a n d t h a t Μ is t h e total n u m b e r of bits of t h e m e s s a g e . 0

9.4

ATM Networks

469

A s s u m e t h a t m goes t h r o u g h a buffer w i t h service rate c b p s . We d e n o t e t h e m a x i m u m n u m b e r of bits t h a t t h e buffer m u s t store b y b (c). Tb calculate b (c), w e n o t e t h a t t h e buffer o c c u p a n c y x(t) satisfies, w i t h x(0) = 0, m

m

m(t)-c, (m(i) - c ) , +

if*(f)>0 if*(t) = 0.

T h e s e e q u a t i o n s express t h a t t h e buffer o c c u p a n c y grows at a rate e q u a l to t h e i n p u t rate m(t) m i n u s t h e service rate c a n d t h a t t h e buffer o c c u p a n c y c a n n o t b e c o m e negative. By solving t h e e q u a t i o n s , w e c a n find t h e m a x i m u m value b (c) ofx(t) for t > 0. We c a n define a (B, R)-leaky-bucket controller as a device t h a t a c c u m u l a t e s a t o k e n fluid at a c o n s t a n t rate R in a t o k e n buffer t h a t c a n store u p to Β u n i t s of t o k e n fluid. Tb t r a n s m i t e u n i t s of (traffic) fluid, t h e t r a n s m i t t e r m u s t r e m o v e 6 u n i t s of t o k e n fluid. Consider s o m e m e s s a g e m . A s s u m e t h a t this m e s s a g e goes t h r o u g h a (B, R) - l e a k y - b u c k e t controller. T h e o u t p u t is a n e w m e s s a g e n. T h e claim is t h a t b (c) < b (c) for all c > 0. T h a t is, t h e o u t p u t of t h e l e a k y b u c k e t r e q u i r e s less buffering at a t r a n s m i t t e r of a n y fixed rate c. Note t h a t this buffering at t h e t r a n s m i t t e r does n o t i n c l u d e t h e buffering in t h e leaky-bucket controller. We say t h a t t h e m e s s a g e η is less bursty t h a n m e s s a g e m. m

n

m

Tb verify t h e claim, c o n s i d e r t h e situation w h e r e t h e m e s s a g e η is s e r v e d b y a buffer w i t h a t r a n s m i t t e r w i t h rate c. A s s u m e t h a t t h e buffer a c c u m u l a t e s b u n i t s of fluid. We will s h o w t h a t t h e buffer also a c c u m u l a t e s at least b u n i t s of fluid w h e n its i n p u t is m. Tb s h o w this, w e d e n o t e b y Τ t h e first t i m e t h a t t h e buffer o c c u p a n c y r e a c h e s t h e value b w i t h t h e i n p u t η a n d b y S t h e last t i m e before Τ t h a t t h e buffer w a s e m p t y . T h u s , d u r i n g [S, T] t h e i n p u t η m u s t c a r r y b + c(T — S) u n i t s of fluid. Let Κ b e t h e a m o u n t of fluid a c c u m u l a t e d in t h e l e a k y b u c k e t controller at t i m e S. If Κ = 0, t h e n m carries at least b + c(T — S) u n i t s in [S, T] a n d t h e claim is proved. If Κ > 0, let U b e t h e last t i m e before 5 t h a t t h e l e a k y b u c k e t controller is e m p t y . T h e claim is t h a t d u r i n g [U, T], m carries at least b + c(T — U) u n i t s of fluid. Tb see this, n o t e t h a t to a c c u m u l a t e Κ u n i t s d u r i n g [17, S], m carries at least Κ + R(S —U)>K + c(S—U) u n i t s of fluid d u r i n g t h a t t i m e interval. Moreover, d u r i n g [S, T], m m u s t c a r r y at least b + c(T — S) — Κ units, o t h e r w i s e η could n o t c a r r y b + c(T - S) u n i t s d u r i n g [5, T]. T h i s c o m p l e t e s t h e proof. Some r e s e a r c h e r s h a v e p r o p o s e d t h e following variation o n t h e l i n e a r b o u n d s . I n s t e a d of defining a (B, K)-traffic, t h e y define a {(B\, R\),..., (Βχ, RK)}traffic as a s t r e a m t h a t carries at m o s t B^ + Rtf cells for all k = 1 , . . . , Κ a n d for all t > 0. Tb enforce t h a t condition, t h e source traffic s h o u l d go t h r o u g h e a c h of t h e (Bk, ^ - r e g u l a t o r s for = 1, . . . , !C. For instance, w e m a y recall t h a t t h e VBR specification calls for two l e a k y b u c k e t s ; see section 8.4.2.

Control of Networks: Mathematical Background

470

9.4.2

Large Deviations of iid Random Variables I n this section w e explore statistical m e t h o d s for call a d m i s s i o n . We w a n t to justify t h e effective b a n d w i d t h results. We first explain t h e case of iid r a n d o m variables. In t h e next section w e e x t e n d t h e results to n o n i n d e p e n d e n t r a n d o m variables.

Cramer's Theorem Consider a collection {X , η > 1} of i n d e p e n d e n t a n d identically distributed r a n d o m variables w i t h c o m m o n distribution F(.) a n d w i t h finite m e a n value m. Define partial s u m S as n

n

η

We k n o w from t h e strong l a w of large n u m b e r s t h a t — -> m as n—• oo w i t h probability 1. T h u s , t h e probability t h a t S /n is a w a y from m goes to 0 as η i n c r e a s e s . It c a n b e s h o w n t h a t this c o n v e r g e n c e to 0 occurs e x p o n e n t i a l l y fast in n. More precisely, for a>m, n

l i m -logP(S >na)

= -A*(a),

n

(9.28)

η->οο γι

where A*(a) = sup[0a - Λ(0)] w i t h Λ(0) = log E(e ). θ

(9.29)

m

For this result to b e valid w e n e e d A(0) to b e differentiable a n d finite in a n e i g h b o r h o o d of 0. Algebra s h o w s t h a t A*(m) = 0. Roughly, this result states t h a t Ρ(5 ^ηα)^β η

_ η Λ

*

( α )

.

This result is called C r a m e r ' s T h e o r e m . T h a t is, it is unlikely t h a t S /n is a w a y from m, a n d this is e x p o n e n t i a l l y unlikely in n. T h e value of Λ*(α) indicates h o w difficult it is for S /n to b e close to a. If Λ*(α) is large, t h e n it is v e r y difficult for S /n to b e close to a. n

n

n

9.4

ATM Networks

•-•

•

471

»•

For further reference, w e n o t e t h a t t h e function A(0) c a n b e o b t a i n e d from t h e function Λ* (a) b y A( 2Λ*(α),

b y strict convexity of t h e function A*(.). C o n s e q u e n t l y , t h e probability of t h a t e v e n t is m u c h s m a l l e r t h a n e . It c a n b e s h o w n formally that, given t h a t S ^ na, t h e m o s t likely realization of t h a t e v e n t is if S*. « ak for k < n. T h u s , t h e r a r e e v e n t {S ^ na} o c c u r s w h e n t h e partial s u m s X\ Η h Xfc are a p p r o x i m a t e l y e q u a l to ak for k < n. We will r e m e m b e r t h a t fact b y saying t h a t the large deviation occurs in a straight line, as illustrated in Figure 9.11. T h u s , t h e m o s t likely w a y t h a t t h e arrival s e q u e n c e h a s a n e m p i r i c a l r a t e α over s o m e l o n g t i m e interval is for t h e arrivals to o c c u r w i t h a n a p p r o x i m a t e l y c o n s t a n t rate α d u r i n g t h a t interval. 2

n

n

Control of Networks: Mathematical Background

472 x + x +... + x x

2

k

ak - Large deviation slope mk = Mean slope k η

Large deviations of sums of iid random variables occur in straight lines.

9.11 FIGURE

Proof of Cramer's Theorem T h e proof of this t h e o r e m illustrates a g e n e r a l a p p r o a c h u s e d in t h e t h e o r y of large deviations. Tb s h o w a limit, w e prove a n u p p e r b o u n d a n d a lower b o u n d . T h a t is, w e s h o w •

for all η > 1, P(S

n

•

> na) < e x p { - n A * ( a ) } .

(9.31)

for all €, 8 > 0 t h e r e is s o m e Ν s u c h that, for all P(S

n

>na-

n>N,

ne) > (1 - 8) e x p { - n A * ( a ) } .

(9.32)

Let u s first s h o w t h e u p p e r b o u n d (9.31). We u s e Markov's inequality. For all 0 > 0 one has P(S

n

>na) 0. (Recall t h e definition (9.29).) T h e lower b o u n d (9.32) is slightly m o r e difficult to prove. T h e trick is to c h a n g e t h e distribution of t h e X 's to m a k e it easy for t h e m to h a v e a s a m p l e m e a n close to a. Tb do this, w e define r a n d o m variables Y t h a t h a v e a distribution s u c h t h a t n

n

P(Y„ e (κ, χ + dx)) = exp{0x}P(X e(x x n

t

+ dx))E e x p { - 0 X }

= e x p { - A ( 0 ) + θχ}Ρ(Χ

η

e (χ, χ + dx)).

(9.33) (9.34)

T h e last t e r m in t h e right-hand side of (9.33) is t h e r e to n o r m a l i z e t h e distribu­ tion of Y . T h e r a n d o m variables Y for η > 1 are iid a n d w e c h o o s e θ so t h a t E(Y ) = a. Note t h a t n

n

n

9.4

ATM Networks

PI

473 /γ

1

... + γ — e

+

\ [a — €, a + €] I -> 1, as η

00.

b y t h e w e a k l a w of large n u m b e r s ( b e c a u s e E(Y )

= a ) . N o w w e claim t h a t for

n

0 1} (i = 1, 2). T h e additivity of t h e effective b a n d w i d t h also h o l d s for i n d e p e n d e n t pro­ cesses of class S. T h i s result is v e r y a p p e a l i n g b e c a u s e it suggests t h a t w e c a n t r e a t t h e various s t r e a m s as if t h e y h a d a fixed rate e q u a l to t h e i r effective b a n d w i d t h . However, w e s h o u l d k e e p in m i n d t h e c r u d e a p p r o x i m a t i o n s m a d e along t h e way. We will s h a r p e n t h e s e results later. S u m m a r i z i n g , w e h a v e m o t i v a t e d (not really p r o v e d ) t h e following result. η

Theorem 9.4.2 Consider a discrete-time q u e u e w i t h X arrivals at t i m e η (η > 0) t h a t serves c c u s t o m e r s p e r u n i t of t i m e in t h e q u e u e . If t h e s e q u e n c e of arrivals is of class 5(Λ*), t h e n t h e i n v a r i a n t probability P(W > B) t h a t t h e q u e u e l e n g t h W exceeds Β is s u c h that, for large B, n

χ Λ*(α) -Β log P(W > B) « - m m ——. a>c α — c 1

n

Control of Networks: Mathematical Background

478

Moreover, t h e m i n i m u m value of c s u c h t h a t t h e r i g h t - h a n d side of t h e above i n e q u a l i t y is at m o s t — 8 < 0 is called t h e effective bandwidth

of t h e arrival

s e q u e n c e a n d is e q u a l to a(8), w h e r e

w i t h A(5) defined b y (9.41). I n particular, t h e effective b a n d w i d t h of t h e s u m of two i n d e p e n d e n t arrival s e q u e n c e s is t h e s u m of t h e i r effective b a n d w i d t h s . Loosely speaking, w e c a n t h i n k of t h e r a n d o m arrival s t r e a m s as h a v i n g con­ stant arrival r a t e s given b y t h e i r effective b a n d w i d t h s . T h i s effective b a n d w i d t h h a s a value b e t w e e n t h e m e a n a n d t h e p e a k rate of t h e s t r e a m , a n d it m e a s u r e s in a precise w a y t h e b u r s t i n e s s of t h a t s t r e a m .

Network Large Deviations Consider t h e n e t w o r k s h o w n i n t h e left-hand p a r t of Figure 9.12. T h e two q u e u e s are first c o m e , first served. T h e arrival p r o c e s s e s A, B, a n d F are i n d e p e n d e n t a n d h a v e straight-line large deviations w i t h effective b a n d w i d t h &Α(8), a (8), a n d (XF(8), respectively. We want, for * » 1, t h e probability of overflow in t h e two q u e u e s to b e d o m i n a t e d b y exp{—8x], w h e r e χ is t h e capacity of e a c h of t h e two q u e u e s . We expect this b o u n d to b e valid if t h e t r a n s m i s s i o n rates C\ a n d C2 of t h e two q u e u e s are large e n o u g h . B

F r o m t h e result for t h e first q u e u e , w e k n o w t h a t if C\ > a (8) + (XB(8), t h e n t h e loss probability i n t h e first q u e u e is b o u n d e d b y exp{-

»

m

M

M

M

—

4

7

I n d e e d , t h e p r o c e s s D (see figure) is a c o m p l e x c o m b i n a t i o n of A a n d Β t h a t d e p e n d s o n t h e service rate C\. T h e large deviations of D d e p e n d o n t h o s e of B. For instance, if Β h a s a large b u r s t of arrivals, t h e n t h e s e arrivals c r e a t e a backlog of arrivals from A in t h e first q u e u e . Eventually, w h e n t h e arrivals of Β are s e r v e d b y t h e first q u e u e , t h e backlog p r o d u c e s a b u r s t of D. C o n s e q u e n t l y , if D h a s a n effective b a n d w i d t h , it a l m o s t certainly d e p e n d s o n t h e statistics of B. T h i s discussion s h o w s t h a t w e s h o u l d expect t h e c o n d i t i o n o n C2 to m a k e t h e losses in q u e u e 2 rare e n o u g h to d e p e n d i n a c o m p l i c a t e d w a y o n t h e statistics of A, B, a n d F . However, w e c a n derive a s i m p l e sufficient condition, w h i c h w e state below. Theorem 9 . 4 3 If C\ > a (8) + a (8) a n d C > a (8) + a (8), t h e n t h e loss probability in e a c h of t h e t w o q u e u e s is n o t larger t h a n exp{—8x\ for κ ^> 1. A

B

2

A

F

T h e intuitive m e a n i n g of this t h e o r e m is that, a l t h o u g h t h e fluctuations of t h e s e c o n d q u e u e d e p e n d o n t h o s e of q u e u e 1, t h e a s s u m p t i o n s m a k e t h e fluctuations of b o t h q u e u e s unlikely. T h e proof of t h e t h e o r e m is a b i t involved, b u t t h e m a i n idea is r a t h e r straightforward a n d is as follows. A s s u m e t h a t t h e arrival p r o c e s s e s c o n s p i r e to m a k e t h e s e c o n d q u e u e r e a c h a value κ at t i m e T. (See left-hand p a r t of t h e figure.) D e n o t e b y S t h e last t i m e before Τ t h a t q u e u e 2 is e m p t y (with S = 0 if it n e v e r is). Say t h a t t h e p r o c e s s e s m a k e t h e first q u e u e r e a c h s o m e value y at t i m e S. T h e claim is t h a t if A a n d F w e r e s e n t directly to t h e s e c o n d q u e u e d u r i n g [5, T], t h e n t h e y w o u l d m a k e it go from 0 to at least κ — y. If this claim is true, t h e n t h e probability of t h e b e h a v i o r d u r i n g [0, S] is at m o s t exp{—8y] a n d t h a t of t h e b e h a v i o r d u r i n g [S, T] is at m o s t exp{- χ. Now, c o n s i d e r w h a t h a p p e n s if t h e p r o c e s s e s A a n d F arrive directly into a n e m p t y s e c o n d q u e u e at t i m e S. (See t h e r i g h t - h a n d p a r t of t h e figure.) In t h a t case, t h e n u m b e r of arrivals in t h e s e c o n d q u e u e d u r i n g [5, T] is t h e s a m e as in t h e case of t h e c e n t e r p a r t of t h e figure. Also, in t h a t case, t h e q u e u e c a n n o t serve m o r e t h a n C ( T — S) d u r i n g [S, T]. C o n s e q u e n t l y , at t i m e 2

T, t h e s e c o n d q u e u e o c c u p a n c y ν will b e s u c h t h a t v>z—y.

H e n c e , v>x

— y,

as claimed. T h i s a r g u m e n t e x t e n d s to feed-forward n e t w o r k s , b y i n d u c t i o n o n t h e n u m b e r of q u e u e s . We let t h e r e a d e r provide t h e details.

Decoupling Bandwidth Consider o n c e again t h e n e t w o r k of Figure 9.12 b u t a s s u m e t h a t F = 0. In t h e above discussion, w e a r g u e t h a t if C\ a n d C exceed t h e s u m of t h e effective 2

b a n d w i d t h s of t h e arrival p r o c e s s e s t h a t go t h r o u g h t h e s e q u e u e s , t h e n t h e loss probability is a s y m p t o t i c a l l y small e n o u g h . T h i s a r g u m e n t p r o v i d e s a sufficient condition for t h e s e c o n d q u e u e . I n t u i t i o n suggests t h a t t h e c o n d i t i o n m a y b e overly conservative. I n d e e d , it is possible t h a t t h e first q u e u e s m o o t h e s o u t t h e p r o c e s s A a n d t h a t C does n o t n e e d to b e as large as 2

a (8). A

However, it c a n b e s h o w n t h a t if C\ is large e n o u g h , t h e n C m u s t b e at least 2

a (S) A

for t h e loss probability in t h e s e c o n d q u e u e to d e c a y as fast as exp{-ai(i)+a (0), B

t h e n w e n e e d C > a (S) 2

A

+ CXF( 1} b e iid r a n d o m variables w i t h A(0) : = log M(0), w h e r e M(0) : = E [ e x p ( 0 Y i ) ] . T h e n n

P(Yi 4- · · · + YN > Nc) «

— exp{-NA*(c)}.

(9.45)

V2na e VN c

c

In t h e above expression, 0 a c h i e v e s t h e m a x i m u m in C

A*(c) = s u p [ 0 c - A ( 0 ) ] θ

where a

2

= A"(0 ). c

Proof

Define t h e iid r a n d o m variables Z so t h a t n

exp{0 x}P(Y e (χ, χ + d*)) c

P ( Z € (χ, χ + A ) ) =

n

n

M(0 ) C

Note t h a t EZ„ = c a n d v a r ( Z ) = A"(0 ) = σ . With Z 2

n

N

C

= Zi + · · · + Z , N

Y

N

Yi + · · · + YN, Y = (Yi, · · · , YN), a n d Ζ = ( Z . . . , Z ) , w e find l f

P(Y

N

> Nc) = E [ 1 { Y

N

Y

> Nc}] = E

N

>

^1{Z

N

Z

NcJ^j

= M ( 0 ) E [ e x p { - 0 Z } l { Z > Nc}] = exp{-NA*(c)}E[exp{-0 W }l{W N

N

c

N

c

N

> 0}]

N

c

w h e r e W =Z -c n

P(Y

N

n

and W

N

=Z

N

- Nc. Let V = W /(a y/N). N

c

Then

> Nc) = e x p { - N A * ( c ) } E [ e x p { - 0 a N V } l { V > 0}]. /

c

cV

By t h e central limit t h e o r e m , w e k n o w t h a t V is a p p r o x i m a t e l y G a u s s i a n w i t h zero m e a n a n d u n i t variance. We t h e n w r i t e

Control of Networks: Mathematical Background

482 1 P(Y

N

> Nc) π exp{-NA*(c)}—=

f°°

/ ν 2 π Jo

= e x p { - N A * ( c ) } - ^ = exp

exp{-0 a >/N*}

exp{-x /2}dx

{^(0 a VN) J jf

exp J - ^ C ^ v ^ + xf

c

2

c

2

c

c

x

J dx

41

= e x p { - N A * ( c ) } - L exp j\(e a VN) j /

exp

2

c

c

V27T

J Je a VN c

c

^ e x p { - N A * ( c ) } - ^ L exp { ^ σ ^ Ν ) } 2

= -t=L

dx

e x

P

{"^(^cVWJ

exp{-NA*(c)},

as we w a n t e d to prove. In t h e next to last equality w e u s e d t h e a p p r o x i m a t i o n

Γ 9.5

exp

^

2

2

1 αχ ^ - exp u

_^ 2

, for u > 1.

SUMMARY T h e p e r f o r m a n c e evaluation of circuit-switched n e t w o r k s r e q u i r e s c o m p u t a ­ tion of t h e blocking probabilities; t h e evaluation of n e t w o r k - c o n g e s t i o n a n d flow-control strategies in packet-switched n e t w o r k s r e q u i r e s c o m p u t i n g q u e u e l e n g t h distributions; a n d t h e evaluation of (small) buffer overflow probabili­ ties a n d multiplexing gain in virtual circuit n e t w o r k s r e q u i r e s c o m p u t i n g "tail" probabilities. This c h a p t e r p r o v i d e d a rapid b u t r e a s o n a b l y c o m p l e t e s u m m a r y of t h e m a t h e m a t i c a l m o d e l s a n d results u n d e r l y i n g all t h e s e c o m p u t a t i o n s . T h e results o n blocking probabilities first a p p e a r e d 70 y e a r s ago; t h o s e o n q u e u e l e n g t h distributions started in t h e 1960s; a n d t h e first results o n tail l e n g t h distributions w e r e p u b l i s h e d in t h e late 1980s. T h e s e last results are available only in j o u r n a l s a n d c o n f e r e n c e proceedings. T h e s e results are difficult to c o m p r e h e n d fully w i t h o u t a strong b a c k g r o u n d in probability t h e o r y . However, it is w o r t h t h e effort to u n d e r s t a n d t h e results, b e c a u s e t h e y provide w a y s of analyzing r e s o u r c e allocation a n d a d m i s s i o n control for h i g h - p e r f o r m a n c e virtual circuit n e t w o r k s . A n indirect m e t h o d of control is t h r o u g h pricing of n e t w o r k services. T h a t a p p r o a c h is c o n s i d e r e d in C h a p t e r 10.

9.7

Problems

483

9.6

NOTES For g e n e r a l discussions of q u e u i n g t h e o r y see [K75, K79, W88]. Also see t h e r e f e r e n c e s cited in section 8.6. More extensive t r e a t m e n t s of t h e d e t e r m i n i s t i c a p p r o a c h e s of section 9.4.1 are given in [C91, LV95]. For elaboration of t h e intuition p r e s e n t e d at t h e e n d of section 9.4.2, c o n s u l t [DZ93]. For a discussion of t h e Bahadur-Rao t h e o r e m see T h e o r e m 1.3 of [D91].

9.7

PROBLEMS 1. Let χ = {x , η > 0} b e a discrete-time Markov c h a i n w i t h t r a n s i t i o n probabil­ ity m a t r i x Ρ a n d i n v a r i a n t distribution π. Show t h a t {z = (x , x +\), η > 0} is a Markov chain. Find its i n v a r i a n t distribution. n

n

n

n

2. Consider a c o n t i n u o u s - t i m e Markov c h a i n o n {1, 2, 3} w i t h q(l) = q(l, 2) = a > 0, q(2) = q(2, 3) = b > 0, q(3) = q(3, 1) = c> 0. F i n d t h e rate m a t r i x of t h e Markov c h a i n r e v e r s e d in t i m e . 3. Let A = {A t > 0} a n d Β = {B , t > 0} b e t w o i n d e p e n d e n t Poisson p r o c e s s e s w i t h rate λ. Define x = xo + A — B w h e r e XQ is a r a n d o m variable i n d e p e n ­ d e n t of A a n d B. Show t h a t χ = {x , t > 0} is a Markov chain. Is it irreducible? Does it h a v e a n i n v a r i a n t distribution? tt

t

t

t

t

t

4. Consider a c o n t i n u o u s - t i m e M a r k o v c h a i n χ = {x , t > 0} w i t h rate m a t r i x Q = {q(i,j), i,j e X}. A s s u m e t h a t ν is t h e i n v a r i a n t distribution of {ξ , η > 0}, t h e Markov c h a i n χ o b s e r v e d at its j u m p t i m e s . Relate ν to π , t h e i n v a r i a n t distribution of χ = {x , t > 0}. t

η

t

5. Consider a s i m p l e t e l e p h o n e n e t w o r k w h e r e calls t h a t arrive as a Poisson process w i t h rate λ are s e n t w i t h probability ρ to a g r o u p of Μ l i n e s a n d w i t h probability 1 — ρ to a n o t h e r g r o u p of Ν lines. Find t h e value of ρ t h a t m i n i m i z e s t h e b l o c k i n g probability of a call. C o m p a r e t h e resulting blocking probability to t h a t c o r r e s p o n d i n g to t h e calls b e i n g s e r v e d b y Μ + Ν lines. 6. Consider Κ t e l e p h o n e sets t h a t s h a r e Ν < Κ outgoing l i n e s of a PBX. A s s u m e t h a t e a c h t e l e p h o n e g e n e r a t e s a n e w call, w h e n it is n o t b u s y , as a Poisson process w i t h rate p . Calls h a v e i n d e p e n d e n t h o l d i n g t i m e s t h a t are e x p o n e n t i a l l y distributed w i t h rate 1. Model t h e n u m b e r of calls in progress b y a Markov c h a i n a n d calculate t h e probability t h a t t h e Ν l i n e s are b u s y . Relate t h a t probability to t h e probability t h a t a call is blocked. Are t h e s e probabilities t h e s a m e ?

Control of Networks: Mathematical Background

484

7. Show t h a t t h e blocking probability in p r o b l e m 6 is insensitive w i t h r e s p e c t to t h e distribution of t h e call h o l d i n g t i m e s . Hint: Consider a closed n e t w o r k w i t h a n M / G I / o o a n d a n M / M / o o queue. 8. Consider Κ t e l e p h o n e sets t h a t are i n d e p e n d e n t l y b u s y or idle w i t h proba­ bilities a n d 1 — _p, respectively. Use t h e c e n t r a l limit t h e o r e m to e s t i m a t e a n u m b e r Ν s u c h t h a t t h e probability t h a t m o r e t h a n Ν p h o n e s are b u s y is about 1%. Relate this result to t h a t of p r o b l e m 6. 9. Consider t h e following m o d e l of a fixed wireless n e t w o r k . T h e r e are Ν radio t r a n s m i t t e r s t h a t s h a r e Μ radio c h a n n e l s . Each radio station c a n serve Κ t e l e p h o n e sets. Each t e l e p h o n e set g e n e r a t e s calls w i t h rate ρ w h e n it is n o t b u s y . Find a Markov description of t h e n e t w o r k w h e n t h e call h o l d i n g t i m e s are i n d e p e n d e n t a n d e x p o n e n t i a l l y distributed w i t h rate μ. W h a t is t h e blocking probability of this n e t w o r k ? Design a M o n t e Carlo p r o c e d u r e for e s t i m a t i n g t h e call blocking probability. 10. Let {A , t > 0} b e a Poisson process w i t h rate λ. Define t h e p r o c e s s e s {B , t > 0} a n d {Q, t > 0} as follows. At e a c h arrival t i m e of {A , t > 0}, o n e flips a coin. With probability p, t h e o u t c o m e is h e a d s a n d t h e arrival t i m e is defined to b e {B , t > 0}. Otherwise, t h a t arrival t i m e is defined to b e {Q, t > 0}. Show t h a t t h e p r o c e s s e s {B , t > 0} a n d {C , t > 0} are i n d e p e n d e n t Poisson p r o c e s s e s w i t h rate λρ a n d λ(1 — ρ), respectively. t

t

t

t

t

t

11. A s s u m e t h e t i m e s b e t w e e n b u s arrivals at a b u s stop are e q u a l to 10 m i n u t e s w i t h probability 0.5 a n d to 30 m i n u t e s w i t h probability 0.5. (a) H o w long is t h e average wait for a b u s for s o m e o n e w h o s h o w s u p at a r a n d o m t i m e at t h e b u s stop? (b) A s s u m e t h a t p a s s e n g e r s s h o w u p at t h e b u s stop according to a Poisson process w i t h rate 0.4 p a s s e n g e r s p e r m i n u t e . W h a t is t h e average n u m b e r of p a s s e n g e r s in a b u s ? 12. If t h e inter-arrival t i m e s of b u s e s are equally likely to b e 10 a n d 40 m i n u t e s , t h e n t h e average waiting t i m e of a p a s s e n g e r is (a) 25 m i n u t e s . (b) 12.5 m i n u t e s . (c) 17 m i n u t e s . 13. Consider a n e t w o r k of q u e u e s w i t h Poisson e x t e r n a l arrivals a n d Markov routing as in Figure 9.7. A s s u m e t h a t s o m e q u e u e s are M / M / 1 a n d o t h e r s are M / G I / o o . Find t h e i n v a r i a n t distribution of t h e n e t w o r k .

9.7 Problems • » _ ι

485

^

14. Consider a r o u t e r w i t h o u t p u t rate 155 Mbps. Because of t h e TCP protocol, p a c k e t s t e n d to arrive in b a t c h e s , l b a n a l y z e t h e average d e l a y p e r packet, a s s u m e t h a t all t h e p a c k e t s h a v e t h e s a m e size e q u a l to 1 KB a n d t h a t p a c k e t s arrive in g r o u p s of 1, 10, a n d 20 w i t h e q u a l probabilities. T h e arrivals of t h e b a t c h e s form a Poisson process. T h e arrival r a t e at t h e r o u t e r (in bits p e r s e c o n d ) is 80% of t h e o u t p u t rate. Calculate t h e average d e l a y p e r packet. Hint: First t r e a t a b a t c h as a single "big" p a c k e t a n d calculate t h e average delay p e r big packet. Next, derive t h e average d e l a y p e r regular packet. 15. C o n s i d e r a r o u t e r w i t h a n o u t p u t l i n e rate of 1.5Mbps. Two s t r e a m s of p a c k e t s arrive as i n d e p e n d e n t Poisson processes. T h e s t r e a m s h a v e r a t e s LI = 500 p a c k e t s / s a n d L2 = 100 p a c k e t s / s , respectively. T h e p a c k e t s of t h e first s t r e a m h a v e 200 b y t e s . Packets of t h e s e c o n d s t r e a m h a v e i n d e p e n d e n t sizes a n d are 100-byte long w i t h probability 0.5 a n d 1,000-bytes l o n g w i t h probability 0.5. (a) Calculate t h e average delay p e r p a c k e t w h e n t h e r o u t e r serves t h e p a c k e t s in t h e first c o m e , first s e r v e d order. (b) Calculate t h e average d e l a y p e r p a c k e t of t y p e 1 a n d of t y p e 2 w h e n t h e r o u t e r serves t h e p a c k e t s of t y p e 1 w i t h h i g h priority a n d p a c k e t s of t y p e 2 w i t h low priority. 16. Packets arrive in b a t c h e s of 100 at a link. T h e average l e n g t h p e r p a c k e t is 500 b y t e s . T h e link t r a n s m i s s i o n rate is 1 Mbps. T h e average d e l a y p e r p a c k e t is (a) (b) (c) (d) (e)

at least 0.2 s. at least 0.4 s. at least 2 m s . at m o s t 0.4 s. n o n e of t h e above.

17. In a q u e u e w i t h high a n d low priorities ( n o n - p r e e m p t i v e ) , as t h e arrival rate of low priority p a c k e t s increases, t h e average d e l a y of high-priority packets (a) (b) (c) (d)

increases. does n o t c h a n g e . decreases. m a y i n c r e a s e or d e c r e a s e .

18. A s t r e a m of p a c k e t s arrives at p o i n t Q a s a Poisson p r o c e s s w i t h rate 2,000 p a c k e t s / s . T h e p a c k e t s h a v e 400 bits w i t h probability 0.5 a n d 4,000 bits w i t h probability 0.5. With probability ρ = 0.8, e a c h p a c k e t is, i n d e p e n d e n t l y of

Control of Networks: Mathematical Background

486

t h e o t h e r packets, s e n t to n o d e R l ; it is s e n t to n o d e R2 o t h e r w i s e . T h e o u t p u t rate of Rl is 4 Mbps, a n d t h a t of R2 is 1 Mbps. ,——

,

4 Mbps

Ρ

2,000 packets/s 1 Mbps

1-p

Calculate t h e average delay p e r p a c k e t as a function of p. Hint: T h e p a c k e t s arrive as Poisson p r o c e s s e s at Rl a n d R2. Recall t h e average d e l a y formula for a n M / G / l q u e u e w i t h Poisson arrival rate A a n d i n d e p e n d e n t service t i m e s distributed as t h e r a n d o m variable S:

Averagedelay

Ax = 2 { i

E(S ) 2

_

A x E ( s ) } +

E(S).

19. Consider a n M / M / 1 q u e u e w h o s e buffer size is fixed a n d finite. We call this a n M / M / l / J c q u e u e . (Note t h a t k is t h e total n u m b e r of c u s t o m e r s in t h e s y s t e m , w h i c h i n c l u d e s t h e o n e b e i n g s e r v e d in t h e server. W h e n it is omitted, s u c h as in t h e case of M / M / 1 , it i m p l i e s t h a t t h e s y s t e m c a n h o l d a n y n u m b e r of customers.) Suppose t h e q u e u e h a s Poisson arrivals w i t h m e a n rate λ, a n d e x p o n e n t i a l service t i m e s w i t h m e a n rate μ. (a) Identify t h e states of this s y s t e m a n d d r a w t h e state-transition-rate diagram. (b) Derive t h e e q u a t i o n s n e e d e d to solve for t h e stationary probabilities of t h e states. Verify t h a t t h e stationary probabilities are _ P d - P) n

where ρ = λ/μ. (c) We c a n see t h a t t h e buffer m a y overflow, c a u s i n g loss of c u s t o m e r s in this q u e u e . W h a t is t h e probability t h a t a c u s t o m e r c a n n o t get into t h e q u e u e a n d is lost? (d) If w e w a n t o u r loss probability to b e small, say 10~ , h o w big s h o u l d o u r buffer b e ? H e r e let's a s s u m e t h a t ρ = 0.5. 20. ATM cells arrive at a t r a n s m i t t e r e q u i p p e d w i t h a buffer according to a Poisson process w i t h rate λ (in cells p e r cell t r a n s m i s s i o n t i m e s ) . W h a t is t h e m a x i m u m value of λ if t h e average delay m u s t b e less t h a n five cell transmission times? 3

9.7

Problems •

i"« »i» «ι

•

ι

υ

— —

—

21. A n ATM source p r o d u c e s A cells d u r i n g t h e n t h cell t r a n s m i s s i o n t i m e . A s s u m e t h a t P(A = m/p) =p = 1 — P(A = 0). Calculate h o w m a n y s u c h sources c a n go t h r o u g h a t r a n s m i t t e r e q u i p p e d w i t h a buffer if t h e average d e l a y p e r cell m u s t b e less t h a n five cell t r a n s m i s s i o n t i m e s . n

n

n

22. Let χ = {x , t > 0} b e a Markov c h a i n o n a finite-state space X = {1, 2 , . . . , m, m + 1} w i t h rate m a t r i x Q . A s s u m e t h a t P(xo = ι) = π(ι), i e X. Define τ = inf{t > 0|*t = m + 1}. Calculate E(exp{—sr}). t

Hint: Let a (i, s) = E[exp{-sr}|*o = i]. Argue t h a t if *o = i, t h e n τ is e q u a l to t h e h o l d i n g t i m e of state ι p l u s t h e t i m e to go from t h e n e x t state to m + 1. s T h u s , a(i, s) = [1/(^(0 + s)] J2j ) 9(hj) for ΐ ^ m + 1. Derive a set of e q u a t i o n s a n d solve t h e m to obtain E(exp{—sr}). Explain h o w to u s e t h e s e ideas to c o n s t r u c t a Markov c h a i n m o d e l of t h e M / G I / o o q u e u e . x

2 3 . A s i m p l e n e t w o r k consists of t w o n o d e s in t a n d e m w i t h r e s p e c t i v e service r a t e s C\ a n d C . A (B\, Ki)-traffic (see section 9.4.1) e n t e r s n o d e 1 a n d c o n t i n u e s o n t h r o u g h n o d e 2. A (B , K )-traffic also e n t e r s n o d e 2. Find b o u n d s o n t h e delay of s t r e a m 1 t h r o u g h t h e n e t w o r k . 2

2

2

24. Let {Χ , η > 1} b e a s e q u e n c e of iid Poisson r a n d o m variables w i t h m e a n λ. Calculate t h e effective b a n d w i d t h of t h a t s e q u e n c e . C o m p a r e t h a t effective b a n d w i d t h to t h a t of a n iid s e q u e n c e of {0, A} r a n d o m variables w i t h t h e s a m e m e a n rate. η

25. Let {X , η > 1} b e iid r a n d o m variables w i t h v a l u e s in {0, 1}. T h i n k of t h e s e as b e i n g on-off sources. F i n d h o w m a n y s o u r c e s c a n go t h r o u g h a t r a n s m i t t e r w i t h rate C u s i n g t h e Bahadur-Rao a p p r o x i m a t i o n . C o m p a r e y o u r a n s w e r w i t h t h e result t h a t y o u w o u l d obtain if t h e s o u r c e s w e r e Gaussian w i t h t h e s a m e m e a n a n d variance. A s s u m e C = 20, ρ = 0.2, a n d Pe = Ι Ο " . n

8

487

10

Network

CHAPTER

Economics

JLn this c h a p t e r y o u will b e i n t r o d u c e d to t h e b a s i c c o n c e p t s a n d m o d e l s t h a t h e l p to explain s o m e i m p o r t a n t features of t h e e c o n o m y of c o m m u n i c a t i o n n e t w o r k s . T h a t e c o n o m y is diversifying a n d growing rapidly. It i n c l u d e s t h e m a r k e t s for c o m m u n i c a t i o n services, on-line a l t e r n a t i v e s to traditional m a r ­ kets, m a r k e t s for information goods t h a t did n o t exist before, a n d t h e m a r k e t for n e t w o r k i n g e q u i p m e n t . T h i s c h a p t e r is l i m i t e d to t h e m a r k e t for c o m m u ­ n i c a t i o n services. Like t h e s t u d y of o t h e r e c o n o m i c c o m m o d i t i e s , t h e s t u d y of t h e c o m m u n i ­ cation services m a r k e t s focuses o n characteristics of s u p p l y a n d d e m a n d a n d o n t h e i r i n t e r a c t i o n in t h e m a r k e t . T h e m o s t i m p o r t a n t s u p p l y factors are t h e t e c h n o l o g y of n e t w o r k e l e m e n t s ( c o m m u n i c a t i o n links a n d switches) a n d t h e m a n a g e m e n t r u l e s t h a t c a n b e u s e d to control t h e s e e l e m e n t s so t h a t t h e n e t w o r k supplies t h e c o m m u n i c a ­ tion services t h a t u s e r s w a n t . T h e t e c h n o l o g y of n e t w o r k e l e m e n t s is d e s c r i b e d in C h a p t e r s 11 a n d 12, while C h a p t e r s 8 a n d 9 are d e v o t e d to a s t u d y of net­ w o r k control. A m o r e c o m p l e t e s t u d y of s u p p l y s h o u l d i n c l u d e a n a s s e s s m e n t of costs: t h e capital costs of n e t w o r k e l e m e n t s ( h a r d w a r e a n d software) a n d op­ erating costs ( m a i n t e n a n c e , depreciation, a d m i n i s t r a t i o n ) . We c a n n o t discuss costs quantitatively b e c a u s e t h e y are n o t p u b l i s h e d in t h e o p e n literature. As t h e costs of c o m m u n i c a t i o n a n d c o m p u t i n g h a r d w a r e decline e a c h year, o p e r a t i n g costs b e c o m e relatively m o r e i m p o r t a n t . T h e s e costs m a y lock c u s t o m e r s to a p a r t i c u l a r h a r d w a r e or software m a n u f a c t u r e r or service provider, b e c a u s e c u s t o m e r s c a n face h i g h a d m i n i s t r a t i v e a n d t r a i n i n g costs s h o u l d t h e y switch to a different s u p p l i e r or t e c h n o l o g y . T h i s lock-in effect c a n

490

Network Economics

often explain t h e expensive efforts of suppliers of e q u i p m e n t a n d services to i n c r e a s e t h e i r m a r k e t s h a r e as quickly as possible. T h e lock-in effect also explains t h e e n t r e n c h e d n a t u r e of TCP/IP. Suppose y o u h a v e d e v e l o p e d a t r a n s p o r t protocol, SmartTP, t h a t is clearly b e t t e r t h a n TCP for Web traffic. Tb deploy y o u r protocol, however, r e q u i r e s c h a n g e s in client a n d server software. T h e m o r e servers d e p l o y SmartTP, t h e g r e a t e r t h e i n c e n t i v e for clients to do so, a n d vice versa. T h i s is a n e x a m p l e of n e t w o r k externality. T h e externality creates a positive feedback w i t h a critical size: o n c e t h e S m a r t T P installed b a s e exceeds this size, t h e d e m a n d for S m a r t T P will grow rapidly, b u t b e l o w this size, t h e r e is insufficient i n c e n t i v e to d e p l o y SmartTP. As t h e installed b a s e of t h e i n c u m b e n t technology, T C P / I P in this example, grows, t h e critical size t h a t S m a r t T P m u s t o v e r c o m e increases, too. O n t h e o t h e r h a n d , a software l a y e r o n top of T C P / I P that, for example, o p e n s m u l t i p l e c o n n e c t i o n s or m a i n t a i n s a p e r s i s t e n t c o n n e c t i o n t h a t i m p r o v e s Web access faces a lower critical size, e v e n t h o u g h t h e p e r f o r m a n c e gain is m u c h lower t h a n SmartTP. D e m a n d factors d e t e r m i n e t h e trade-off u s e r s m a k e b e t w e e n t h e services t h e y want, t h e i r quality, a n d t h e i r price. T h e d e m a n d for c o m m u n i c a t i o n services in large p a r t is derived demand. T h a t is, u s e r s d e m a n d c o m m u n i c a t i o n services not for t h e m s e l v e s , b u t b e c a u s e those services provide a m e a n s to a n e n d t h a t u s e r s really w a n t . For example, u s e r s m a y access a n on-line database in order to obtain n e w s ; or t h e y m a y subscribe to a v i d e o - o n - d e m a n d service to w a t c h a movie; or t h e y m a y t e l e c o m m u t e . In e a c h case, u s e r s h a v e alternative m e a n s to achieve t h e i r e n d s : n e w s m a y b e obtained from a n e w s p a p e r , radio, or TV; videos m a y b e r e n t e d from a store; a n d o n e m a y travel to work. T h e price a n d c o n v e n i e n c e of u s e of t h e s e alternatives will affect t h e d e r i v e d d e m a n d . We b e g i n in section 10.1 w i t h a discussion of t h e n a t u r e of derived d e m a n d . I n section 10.2 w e analyze t h e m a r k e t for I n t e r n e t access. We argue t h a t flat rate pricing is inefficient a n d retards t h e i n t r o d u c t i o n of quality-differentiated services. In section 10.3 we i n t r o d u c e a r e s o u r c e m o d e l of n e t w o r k access to suggest four t y p e s of charges for c o m m u n i c a t i o n services. I m p l e m e n t i n g t h e s e charges r e q u i r e s a t e c h n o l o g y capable of service provisioning, a c c o u n t i n g a n d billing, a n d u s e r control t h a t does n o t y e t exist. In section 10.4 w e p r e s e n t results from a trial at t h e University of California, Berkeley, called INDEX ( I n t e r n e t D e m a n d E x p e r i m e n t ) . INDEX is b o t h a t e c h n o l o g y a n d a m a r k e t trial. It d e m o n s t r a t e s a t e c h n o l o g y w i t h t h e capabilities m e n t i o n e d above a n d collects data a b o u t h o w u s e r s value different qualities of service. In sections 10.5 a n d 10.6 w e i n t r o d u c e two m o d e l s to s t u d y pricing of c o m m u n i c a t i o n services. T h e first m o d e l is useful for a data n e t w o r k like t h e

10.1

Derived Demand for Network Services

I n t e r n e t , w i t h a single quality of service a n d n o a d m i s s i o n control, m a k i n g t h e n e t w o r k susceptible to congestion. T h e s e c o n d m o d e l is for n e t w o r k s w h e r e service quality g u a r a n t e e s are possible a n d n e w u s e r r e q u e s t s c a n b e rejected if n e t w o r k r e s o u r c e s to serve t h o s e r e q u e s t s are n o t available. We u s e t h e s e c o n d m o d e l to calculate h o w m a n y r e q u e s t s of different t y p e s of service c a n s i m u l t a n e o u s l y b e served. We also d e v e l o p a v a r i a n t of this m o d e l to suggest t h a t a good w a y to charge for t h o s e services is in t e r m s of t h e r e s o u r c e s t h e y consume. Section 10.7 provides a s u m m a r y .

10.1

DERIVED DEMAND FOR NETWORK SERVICES C o m m u n i c a t i o n n e t w o r k s , like road n e t w o r k s , alter t h e absolute a n d relative cost of access to locations of activities t h a t p e o p l e value. Unlike physical locations t h a t are r e a c h e d b y a r o a d n e t w o r k , t h e locations r e a c h e d b y a c o m m u n i c a t i o n n e t w o r k are h o s t s c o n n e c t e d to t h e n e t w o r k . T h e s e h o s t s are virtual locations. Physical a n d virtual locations provide t h e "space" for a variety of activities. Some physical locations are w o r k p l a c e s , o t h e r s are s h o p s or p l a c e s t h a t provide e n t e r t a i n m e n t , a n d o t h e r s are h o m e s w h e r e p e o p l e live. Virtual locations (Web sites, w o r k p l a c e hosts, e-mail servers, etc.) provide i n f o r m a t i o n or t h e m e a n s to b u y a n d sell goods a n d services on-line. W h e n p e o p l e e m b a r k o n a n activity at a r e m o t e physical or virtual location, t h e y i n c u r a c o m m u n i c a t i o n cost a n d a s e a r c h cost. T h e s e costs are s u b t r a c t e d from t h e value of t h e activity itself. If y o u p u r c h a s e a b o o k from a physical bookstore, t h e c o m m u n i c a t i o n cost is t h e cost of t r a n s p o r t a t i o n to o n e or m o r e bookstores, a n d t h e s e a r c h cost is t h e t i m e a n d effort y o u s p e n d in finding t h e b o o k y o u w a n t . If y o u go to w o r k or to visit a friend, t h e c o m m u n i c a t i o n cost is t h e effort in traveling to y o u r destination, a n d t h e s e a r c h cost is negligible. If y o u access a n on-line library catalog to r e s e a r c h s o m e topic, t h e c o m m u n i ­ cation cost m a y b e t h e price y o u p a y for t h e c o m m u n i c a t i o n services, a n d t h e s e a r c h cost is d e t e r m i n e d b y h o w well t h e on-line s e a r c h facility m a t c h e s y o u r need. T h e s u m of t h e c o m m u n i c a t i o n a n d s e a r c h costs is t h e transaction cost as­ sociated w i t h a n activity s u c h as going to work, s h o p p i n g for a book, or Web b r o w s i n g for s o m e information. T h e t r a n s a c t i o n cost h a s a m o n e t a r y c o m p o ­ n e n t , a n d a subjective c o m p o n e n t , w h i c h is a function of h o w t h e t r a n s a c t i o n is

491

Network Economics

492

e x p e r i e n c e d . If t h e t r a n s a c t i o n u s e s n e t w o r k services, t h e e x p e r i e n c e d e p e n d s o n t h e t i m e it takes to c o m p l e t e t h e t r a n s a c t i o n s u c h as d o w n l o a d i n g a Web page. T h a t t i m e is d e t e r m i n e d b y t h e s p e e d of t h e service. O t h e r d i m e n s i o n s of service quality (latency, jitter) also affect t h e e x p e r i e n c e . E c o n o m i s t s s u p p o s e t h a t a n individual assigns a m o n e t a r y value to this subjective e l e m e n t of t h e t r a n s a c t i o n cost. W h e n w e refer to t h e t r a n s a c t i o n cost faced b y a n individual, w e m e a n t h e s u m of t h e m o n e t a r y cost a n d this assigned value. T h e subjective cost of t h e s a m e t r a n s a c t i o n is different for different in­ dividuals a n d for t h e s a m e individual at different t i m e s . For example, w h e n individuals place a h i g h value o n t h e i r time, t h e y will choose a m o r e expensive, high-speed service b e c a u s e of t h e t i m e t h e y save. Individuals m a y h a v e a choice a m o n g physical a n d virtual locations, a n d a m o n g c o m m u n i c a t i o n services of different qualities a n d price. T h e y c o m p a r e t h e total cost of t h e different alternatives a n d choose t h e o n e w i t h t h e least cost. Consider t h e choice b e t w e e n physical a n d virtual locations. For s o m e activities t h e t r a n s a c t i o n cost at a virtual location is m u c h l o w e r t h a n at a physical location. So p e o p l e are increasingly likely to visit virtual locations for t h o s e activities. T h i s explains t h e g r o w t h of on-line c o m m e r c e , especially for goods a n d services t h a t are r e a s o n a b l y s t a n d a r d ( p u r c h a s i n g airline tickets, books, or stocks). T h e shift to on-line c o m m e r c e is altering m a n y b u s i n e s s e s . Airline travel a g e n t s h a v e s e e n t h e i r r e v e n u e s decline precipitously. Soon t h e r e will b e few of t h e m left as travelers save m o n e y a n d t i m e t h r o u g h on-line p u r c h a s e s . T h e growth of on-line trading in stocks b e c a u s e of its c o n v e n i e n c e a n d m u c h lower t r a n s a c t i o n cost is r e d u c i n g t h e i n c o m e s of b r o k e r s . On-line "aggregators" are attracting c u s t o m e r s b y providing i n f o r m a t i o n a b o u t price a n d quality of p r o d u c t s from c o m p e t i n g m a n u f a c t u r e r s . In 1998 1% of total retail t r a d e in t h e U n i t e d States w a s c o n d u c t e d on-line, a n d s o m e e s t i m a t e t h a t this will i n c r e a s e to 6% b y 2002. On-line sales a m o n g b u s i n e s s e s will i n c r e a s e e v e n faster j u d g i n g b y t h e e x a m p l e s of Cisco, Dell Computer, a n d Boeing.

10.1.1

Information Goods T h e g r o w t h in on-line c o m m e r c e is attributable to t h e r e d u c e d t r a n s a c t i o n cost. I b t a l cost r e d u c t i o n s are e v e n larger for on-line sales of information goods. T h e s e goods are t h e m s e l v e s in digital form (e.g., software p a c k a g e s a n d database access). This cost r e d u c t i o n will s t i m u l a t e t h e p r o d u c t i o n of m a n y goods s u c h as CDs, videos, a n d b o o k s in digital form.

10.1

Derived Demand for Network Services

We p o i n t to s o m e distinguishing features of t h e m a r k e t for i n f o r m a t i o n goods. T h e costs of producting, distributing, a n d c o n s u m i n g i n f o r m a t i o n goods difer from t h o s e of o t h e r goods a n d services. While it c a n b e costly to design a n d p r o d u c e t h e first (or m a s t e r ) item, t h e cost to p r o d u c e additional i t e m s or distribute t h e m on-line is negligible. T h e c o n s u m e r of a n i n f o r m a t i o n good incurs, in addition to t h e cost of t h e good itself, t h e t r a i n i n g a n d a d m i n i s t r a t i v e cost of "absorbing" t h e good. For example, t h e cost of u s i n g a n e w software package—one t h a t is different from t h e o n e y o u are c u r r e n t l y using—can b e h i g h a n d serve as a d e t e r r e n t from acquiring t h e n e w package. In s o m e cases, t h e cost of switching to a n e w package is high b e c a u s e it r e q u i r e s o t h e r s to switch as well, as is t h e case w i t h d o c u m e n t packages. T h e h i g h initial cost, t h e low m a r g i n a l cost of p r o d u c t i o n a n d distribution, a n d t h e h i g h absorption cost affect t h e n a t u r e of t h e c o m p e t i t i o n in t h e m a r k e t s for information goods a n d t h e strategies t h a t p r o d u c e r s a n d c o n s u m e r s of t h o s e goods are likely to follow. We n o t e s o m e of t h e s e i m p a c t s . Because t h e m a r g i n a l cost of p r o d u c t i o n is low, price c o m p e t i t i o n c a n l e a d to pricing b e l o w t h e average cost. T h i s b e n e f i t s c o n s u m e r s b u t it c a n r u i n pro­ d u c e r s . In this situation, a p r o d u c e r m a y e n g a g e in price c o m p e t i t i o n to b u i l d u p m a r k e t share. O n c e a n a d e q u a t e s h a r e h a s b e e n acquired, p r i c e s of t h e s a m e or c o m p l e m e n t a r y p r o d u c t s could b e raised w i t h o u t losing c u s t o m e r s w h o w o u l d b e r e l u c t a n t to switch to a c o m p e t i n g p r o d u c t b e c a u s e of its absorp­ tion cost. T h e i n c e n t i v e to i n c r e a s e m a r k e t s h a r e explains w h y Web b r o w s e r s are given a w a y free. T h e a i m is to sell c o m p l e m e n t a r y p r o d u c t s s u c h as servers. Lastly, since t h e m a r g i n a l cost of p r o d u c t i o n is low, b u t c o n s u m e r s face high absorption costs, p r o d u c e r s will t r y price d i s c r i m i n a t i o n . T h e y w o u l d like to charge a high price to c u r r e n t c u s t o m e r s a n d a low price to attract c u s t o m e r s of competitive products. T h i s price d i s c r i m i n a t i o n c a n take a v a r i e t y of forms. O n e form is p r o d u c t differentiation. For example, t h e New York Times gives free on-line access to c u r r e n t articles, b u t y o u h a v e to p a y for a r c h i v e d m a t e r i a l . Often a n i n t r o d u c t o r y version of a software p r o d u c t is given a w a y free, b u t a d v a n c e d versions are sold at a h i g h price.

10.1.2

Site Rents T h e road n e t w o r k a n d t h e c o m m u n i c a t i o n n e t w o r k alter t h e distribution of t r a n s a c t i o n costs over t h e g e o g r a p h y of physical or virtual locations. Suppose location A is r e a c h e d at a l o w e r cost t h a n B. T h e n m o r e p e o p l e will visit A t h a n B, a n d so m e r c h a n t s at A will g e n e r a t e h i g h e r sales t h a n at B. As a result, r e n t s at A will b e h i g h e r t h a n at B. T h e difference in t h e r e n t s is called location or

493

Network Economics

494

site rent. Landlords of c o m m e r c i a l space t r y to i n c r e a s e site r e n t s b y attracting m o r e p e o p l e to t h e i r location. Site r e n t s also a c c r u e to o w n e r s of virtual locations s u c h as Web sites. A c o m m o n form of site r e n t is a charge for advertising at a Web site. T h i s is also h o w TV, radio, a n d n e w s p a p e r s collect site r e n t s . Tb attract visitors Web site o w n e r s subsidize various services (e-mail, b u l l e t i n boards, c h a t groups), j u s t as TV b r o a d c a s t s are free a n d n e w s p a p e r s are sold b e l o w cost. Because t h e y c o m p e t e for t h e s a m e v i e w e r s from t h e g e n e r a l public, Web sites imitate o n e another. So t h e r e is a t e n d e n c y for Web sites to look alike, j u s t as TV p r o g r a m s , n e w s p a p e r s , a n d politicians d u r i n g elections look alike. A n alternative strategy to i n c r e a s e site r e n t s is to specialize a n d target a n a r r o w set of viewers. T h i s occurs in physical locations as well, so r e s t a u r a n t s t e n d to cluster together, as do w a r e h o u s e s a n d i n v e s t m e n t b a n k s . Similarly, s o m e Web sites specialize in attracting viewers s e e k i n g financial information, o t h e r s attract chemists, a n d o t h e r s specialize in i n f o r m a t i o n a b o u t a u t o m o b i l e s or real estate. T h e r e also are s o m e i n t e r e s t i n g differences b e t w e e n t h e site r e n t s at p h y s ­ ical a n d virtual locations. For example, since it is possible to d e t e r m i n e if a n i t e m w a s p u r c h a s e d from a particular Web site, t h e site r e n t m a y take t h e form of a c o m m i s s i o n o n t h e sales. This is n o t g e n e r a l l y possible w i t h n e w s p a p e r or TV advertising (although it is t h e case w i t h on-line TV s h o p p i n g c h a n n e l s ) . Some Web sites obtain d e m o g r a p h i c i n f o r m a t i o n a b o u t t h e i r v i e w e r s (usually in exchange for a free service like e-mail). T h e y c a n t h e n target advertising m o r e selectively a n d charge a h i g h e r rent. T h e attraction of o n e physical location relative to a n o t h e r m a y b e b e c a u s e t h e former is m o r e easily accessible b y road. T h e attraction t h e n persists for a long t i m e as t h e r o a d n e t w o r k c h a n g e s v e r y slowly. Virtual locations, b y contrast, are likely to b e equally accessible a n d it is easy to b u i l d c o m p e t i n g locations, so t h e i r relative attraction will b e d e t e r m i n e d b y h o w m u c h t h e y re­ d u c e t h e s e a r c h cost c o m p o n e n t of t r a n s a c t i o n costs, h o w m u c h t h e y subsidize services t h a t viewers desire, a n d h o w m u c h b r a n d loyalty t h e y c a n create.

10.2

INTERNET SERVICE PROVIDERS In this section w e s t u d y t h e m a r k e t for I n t e r n e t access. We first investigate a theoretical m o d e l a n d t h e n p r e s e n t s o m e e m p i r i c a l e v i d e n c e t h a t s u p p o r t s t h e model.

10.2

Internet Service Providers

10.1 • Η Η Μ Ι -

An ISP aggregates subscriber traffic and forwards their datagrams to the backbone network through a network access point or NAP.

In t h e U n i t e d States, individuals a n d b u s i n e s s e s p u r c h a s e c o m m u n i c a t i o n services from a n I n t e r n e t service p r o v i d e r (ISP), as d e p i c t e d in Figure 10.1. Users c o n n e c t via a d e d i c a t e d or s h a r e d link to t h e ISP's p o i n t of p r e s e n c e (PoP). T h e ISP aggregates t h e traffic at t h e PoP, w h e r e its r o u t e r s forward u s e r d a t a g r a m s to t h e I n t e r n e t b a c k b o n e n e t w o r k t h r o u g h a n e t w o r k access p r o v i d e r (NAP). T h e ISP's subscribers m a y also c o n n e c t to ISP s e r v e r s t h a t provide e-mail, news, a n d Web-caching. Routing of p a c k e t s a m o n g b a c k b o n e NAP n e t w o r k s d e p e n d s o n reciprocal p e e r i n g a r r a n g e m e n t s . NAPs do n o t bill o n e a n o t h e r for this traffic. But a NAP m a y u s e policy-based r o u t i n g to provide b e t t e r t r e a t m e n t of traffic from certain p e e r s d e p e n d i n g o n t h e reciprocal a r r a n g e m e n t . T h i s is different from t h e t e l e p h o n e n e t w o r k in w h i c h t h e charge for a call is divided a m o n g t h e t e l e p h o n e c o m p a n i e s t h a t c a r r y this call according to a s y s t e m of s e t t l e m e n t charges. T h e r e are m o r e t h a n 3,000 ISPs in t h e U n i t e d States. Most of t h e m are local w i t h o n e or a few PoPs. S o m e h a v e PoPs t h a t s p a n a state or region. A few are national. In s o m e c o u n t r i e s t h e n a t i o n a l t e l e p h o n e c o m p a n y is t h e sole ISP, a l t h o u g h this is b e i n g r e p l a c e d b y a s y s t e m of c o m p e t i n g ISPs. A subscriber p a y s two charges to access t h e I n t e r n e t . T h e first is a charge for t h e link to t h e ISP's PoP. N i n e t y p e r c e n t of subscribers p a y a fixed m o n t h l y charge to t h e i r t e l e p h o n e c o m p a n y for this link. (In m o s t countries, b u t n o t in t h e U.S., t h e r e is a n additional t e l e p h o n e c o n n e c t - t i m e charge t h a t h a s a large

495

496

Network Economics

i m p a c t o n d e m a n d . ) In m o r e t h a n 90% of cases this link is a n analog voice line u s e d w i t h a dial-up m o d e m at s p e e d s of 28 or 56 Kbps. A growing n u m b e r of subscribers h a v e a d e d i c a t e d 128 Kbps ISDN line or a DSL line t h a t provides access at d o w n s t r e a m / u p s t r e a m s p e e d s of 3 8 4 / 1 2 8 or 1,500/384 Kbps. Large organizations, w i t h t h o u s a n d s of h o s t c o m p u t e r s , m a y lease l i n e s at s p e e d s of 1.5 to 45 Mbps. A r e c e n t alternative is offered b y CATV operators. In this case u p to 500 subscribers s h a r e a 10/3.8 Mbps link (this s p e e d will i n c r e a s e in t h e future). T h e r e are o t h e r alternatives as well, including wireless a n d satellite links. T h e fixed m o n t h l y charge for t h e link is $15 to $20 for a n analog voice line, $40 to $200 for a n ISDN or DSL line d e p e n d i n g o n speed, a n d $30 for CATV (in addition to t h e $30 m o n t h l y charge for t h e TV p r o g r a m s ) . T h e subscriber p a y s for t h e m o d e m . (In 1999, a b o u t 600,000 subscribers in t h e U.S. h a d CATV a n d DSL access.) T h e subscriber also p a y s a charge to t h e ISP t h a t d e p e n d s o n t h e access speed. Most ISPs in t h e U n i t e d States h a v e m o v e d from a c o n n e c t i o n t i m e b a s e d charge to a flat-rate m o n t h l y charge. D e p e n d i n g o n t h e access speed, this charge r a n g e s from $20 for 28 Kbps a n d CATV to $200 for 1.5 Mbps. T h e ISP i n c u r s two costs. T h e first is t h e cost of aggregation. If it provides access via dial-up m o d e m s , t h e ISP m u s t p u r c h a s e a pool of m o d e m s to t e r m i ­ n a t e subscriber calls a n d p a y t h e t e l e p h o n e c o m p a n y for t r u n k s to t h e m o d e m pool. T h e a m o r t i z e d cost for o n e m o d e m a n d t r u n k line is a b o u t $25 p e r m o n t h . With a c o n c e n t r a t i o n of 10 subscribers p e r m o d e m , this a m o u n t s to a cost to t h e ISP of $2.50 p e r m o n t h p e r subscriber. In t h e case of CATV access, u s e r traffic is already aggregated. DSL traffic is aggregated at t h e DSL multiplexer w h e r e t h e DSL loops t e r m i n a t e . T h e ISP m u s t also p a y t h e NAP for t r a n s p o r t of its traffic over t h e b a c k b o n e . NAPs offer sophisticated service p l a n s r a n g i n g from a fixed m o n t h l y charge d e p e n d i n g o n s p e e d to a charge t h a t d e p e n d s o n t h e average traffic rate p l u s a b u r s t charge. T h e y also offer o t h e r services s u c h as virtual private n e t w o r k s , reliability, t u n n e l i n g , a n d security. T h e NAP charge typically is less t h a n 1 c e n t p e r MB of b a c k b o n e traffic. T h e subscriber's cost of data transfer is high. In t h e U.S. t h e average a m o u n t of data transferred via 28 Kbps m o d e m s is about 60 MB p e r m o n t h , so at a m o n t h l y cost of $20, t h e subscriber is p a y i n g 33 c e n t s / M B . If y o u i n c l u d e t h e m o n t h l y $15 for a t e l e p h o n e line, this b e c o m e s 58 c e n t s / M B . T h e light user, w h o is transferring one-fifth of t h e average traffic, is p a y i n g a n exorbitant $1.65 to $2.91 p e r MB, a n d t h e h e a v y user, w h o transfers five t i m e s t h e average, p a y s 6.6 to 11.6 c e n t s / M B . Data from o n e large ISP indicates t h a t 76% of t h e traffic u s e s t h e H T T P protocol; 12% u s e s N N T P (Network News Transfer Protocol, a protocol for t h e

10.2

Internet Service Providers

497

distribution, retrieval, a n d posting of n e w s articles); 5% is e-mail (POP3 a n d SMTP); 3 % u s e s FTP; a n d 2% u s e s a s e c u r e H T T P S ( s e c u r e H T T P ) .

A Subscriber Demand

102.1

Model We d e v e l o p a m o d e l t h a t w e will u s e to c o m p a r e t h e b e h a v i o r of ISPs a n d t h e i r c u s t o m e r s u n d e r two pricing s c h e m e s : a m o n t h l y flat-rate c h a r g e a n d a usageb a s e d charge. Suppose a n ISP sells I n t e r n e t access at a price of ρ p e r u n i t of usage, m e a ­ s u r e d in m i n u t e s of c o n n e c t t i m e or in MB of data transferred. T h e average u s e r ' s d e m a n d for access is m o d e l e d as a function χ — Dip). T h i s is t h e de­ creasing demand curve in Figure 10.2. T h e user, faced w i t h a u n i t price p, does n o t c o n s u m e m o r e t h a n Dip) b e c a u s e t h e value or b e n e f i t s h e gains from a n y additional usage is w o r t h less to h e r t h a n h e r cost p. She d o e s n o t c o n s u m e less, b e c a u s e h e r benefit from e a c h u n i t of usage u p to Dip) exceeds p. T h a t is t h e m e a n i n g of a d e m a n d curve. (If t h e usage is m e a s u r e d in c o n n e c t time, a n d t h e subscriber m u s t also p a y a p e r - m i n u t e c o n n e c t - t i m e c h a r g e τ to t h e t e l e p h o n e c o m p a n y , t h e d e m a n d c u r v e is D ( r +p).)

P

χ D(p)

10.2 FIGURE

DiO)

At a unit price p, the average user consumes Dip) units; a light user consumes aDip) with a < 1; a heavy user consumes aDip) with a > 1.

Network Economics

498

If access is sold at ρ p e r unit, t h e u s e r p u r c h a s e s D(p). T h e value to h e r of c o n s u m i n g D(p) units, m i n u s h e r cost pD(p), e q u a l s t h e a r e a of t h e triangle b e l o w t h e d e m a n d c u r v e a n d above t h e horizontal line p. T h i s area is called t h e consumer surplus. (If t h e ISP's u n i t cost of providing access is p, t h e n t h e c o n s u m e r ' s cost pD(p) e q u a l s t h e ISP's cost, a n d t h e c o n s u m e r s u r p l u s e q u a l s t h e gain in social welfare.) Different subscribers h a v e different d e m a n d c u r v e s Di. For simplicity, w e s u p p o s e t h a t t h e s e d e m a n d s differ b y a scale factor, so i's d e m a n d is Di(p) = atD{p). For a h e a v y user, cti > 1, a n d for a light user, a; < 1.

Flat-Rate Charge Suppose t h e ISP switches to a flat-rate charge of S p e r m o n t h . T h i s flat-rate s c h e m e is likely to i n d u c e four c h a n g e s . First, after p a y i n g S t h e subscriber faces n o additional charge, so t h e i n c r e m e n t a l price is 0. As a result, i's c o n s u m p t i o n will i n c r e a s e to aiD{U). T h i s is wasteful since e a c h of t h e ai[D(0) - D(p)] additional u n i t s of c o n s u m p t i o n is w o r t h less to h e r t h a n t h e cost ρ of p r o d u c i n g it. Note t h a t h e a v y u s e r s (with oii > 1) waste m o r e t h a n light users. Also n o t e t h a t t h e m o r e price-sensitive t h e u s e r is, t h e larger Dip) — D(0) will be, a n d t h e g r e a t e r t h e waste. I b explain t h e s e c o n d effect, s u p p o s e t h a t t h e ISP sets t h e fixed charge S to cover t h e cost. Since t h e average d e m a n d curve u n d e r flat-rate charge is D(0) a n d ρ is t h e u n i t cost, S=pD(0). T h i s is t h e r e c t a n g u l a r area i n d i c a t e d in Figure 10.2. So i's bill n o w is S c o m p a r e d w i t h h e r p r e v i o u s bill of potiD(p). If at
l

>J

k

G

( > 16

3 2

>

6 4

>

9 6

>

1 2 8

)>

(10.1)

Network Economics

504

w h e r e x^ is t h e n u m b e r of m i n u t e s of c o n n e c t t i m e of s p e e d k Kbps p u r c h a s e d b y a c o n s u m e r d u r i n g a w e e k w h e n facing a price of pj c e n t s p e r m i n u t e for s p e e d ; . With this "log-log" m o d e l , t h e coefficient a£ is t h e own-price elasticity, t h a t is, a\ is t h e p e r c e n t c h a n g e in d e m a n d x^ for s p e e d k d u e to a 1% c h a n g e in its price pr, a n d d is t h e cross-price elasticity, t h a t is, t h e p e r c e n t a g e c h a n g e k

in t h e d e m a n d x^ d u e to a 1% c h a n g e in t h e price pj of s p e e d ;'. T h e prior expectation is t h a t a£ < 0, d e m a n d for s p e e d k will d r o p if its price increases, a n d a/ > 0, d e m a n d for s p e e d k will i n c r e a s e if t h e price of a substitute s p e e d k

increases. T h e e s t i m a t e d d e m a n d e q u a t i o n is X8 X X X Xie 12

96

64

3 2

= = = = =

4.1 2 . 6 2.7 2.33 0.52

- 1 . 6 5 P i 2 8 +0.44P 6 +1.23Pi28 - 3 . 3 4 P + O . O 8 P 1 2 8 +0.84P +0.48Pi -0.58P +0.42Pi28 - 0 . 2 6 P 9

9 6

9 6

2 8

9 6

9 6

+0.55P +1.17P -1.71P +0.88P +0.18P 6

4

6 4

6 4

6 4

6 4

-0.12P +0.23P +0.47P -1.10P +0.97P

3 2

3 2

3 2

3 2

3 2

+0.00Pi +0.00P +0.55Pi +0.08Pi -1.29Pi

6

1 6

(10.2)

6

6

6

w h e r e X^ = log x^, Pi = log Pj. T h e e s t i m a t e d coefficients in (10.2) w h o s e r-static (that is, e s t i m a t e divided b y its s t a n d a r d error) is larger t h a n 3 are w r i t t e n in bold. Observe t h a t t h e d e m a n d for a p a r t i c u l a r s p e e d is v e r y sensitive to its o w n price a n d to t h e price of t h e next l o w e r speed. T h e own-price elasticity is b e t w e e n - 1 a n d —3, a n d t h e cross-price elasticity is a r o u n d 1. T h e h i g h own-price elasticity i m p l i e s t h a t w h e n usage is c o n n e c t time, w a s t e i n d u c e d b y flat-rate pricing is large. Second, t h e large cross-price elasticity indicates t h a t if u s e r s are offered differentiated quality of service, t h e y w o u l d i n d e e d d e m a n d m o r e t h a n o n e service quality. Direct e v i d e n c e of this in INDEX data also c o m e s from t h e fact t h a t o n average, e a c h c u s t o m e r selects 3.5 out of t h e 5 available p r i c e d s p e e d s (in addition to t h e free 8 Kbps s p e e d ) e a c h w e e k . Since c o n s u m e r s do value m u l t i p l e service qualities, t h e r e is a loss to providers a n d c o n s u m e r s w h e n ISPs only offer t i er ed quality service as t h e y do today. (See section 10.4.2 for m o r e o n this topic.)

10.3

NETWORK CHARGES: THEORY A N D PRACTICE In this section w e i n t r o d u c e t h e e c o n o m i c principles u n d e r l y i n g c h a r g e s for c o m m u n i c a t i o n services. O u r e x a m p l e s will largely b e d r a w n from t h e I n t e r n e t , a l t h o u g h t h e discussion applies to o t h e r c o m m u n i c a t i o n n e t w o r k s .

10.3

Network Charges: Theory and Practice

505

A w o r d o n t e r m i n o l o g y : w e will u s e t h e t e r m s charge ana price i n t e r c h a n g e ­ ably, a l t h o u g h a useful distinction c a n b e m a d e . A price is n o r m a l l y associated w i t h o n e u n i t of service: if y o u b u y η u n i t s of service, y o u p a y η t i m e s t h e u n i t price. A charge is a m o r e g e n e r a l form of price. For example, a charge m a y consist of a fixed c o m p o n e n t p l u s a u n i t price. M a t h e m a t i c a l l y , w e m a y t h i n k of a charge as a n o n l i n e a r price a n d a price as a l i n e a r charge.

103.1

A Resource Model Figure 10.5 is a m o d e l of t h e r e s o u r c e s t h a t a n ISP n e e d s to provide I n t e r n e t access to t h e h o m e . T h e r e are four k i n d s of r e s o u r c e s . Subscribers c o n n e c t t h e i r m o d e m s via circuit-switched c o n n e c t i o n s to t h e ISP m o d e m s . T h i s r e s o u r c e is c h a r a c t e r i z e d b y t h e bit rate A b p s of t h e sub­ scriber m o d e m a n d t h e n u m b e r of access p o r t s t h a t t h e ISP provides. T h e figure s h o w s t h a t Ν e n d h o s t s s h a r e Μ access ports. T h e t h i r d r e s o u r c e is t h e packetswitched link w i t h bit rate Β b p s c o n n e c t i n g t h e ISP to t h e b a c k b o n e n e t w o r k via a NAP. T h e fourth r e s o u r c e is t h e b a c k b o n e n e t w o r k itself d e p i c t e d b y t h e cloud. T h e s e r e s o u r c e s a n d t h e w a y t h e y are m a n a g e d d e t e r m i n e t h e quality of service t h a t a subscriber receives. T h e bit rate A o n a subscriber's d e d i c a t e d l i n k limits t h e p e a k rate. Dial-up m o d e m s over analog loops h a v e s p e e d s of 28 or 56 Kbps. Digital subscriber loops (DSL) p e r m i t s p e e d s u p to 1.5 Mbps. In s o m e m e t r o p o l i t a n areas in t h e U.S., I n t e r n e t access a c c o u n t s for 50% or m o r e of t h e traffic over t h e local t e l e p h o n e n e t w o r k . Since Ν subscribers c o n t e n d for Μ access ports, a subscriber m a y b e blocked. T h e b l o c k i n g probability d e p e n d s ,

Subscriber loops (N)

ISP

CO

ISP link

Ports (M)

10.5 FIGURE

Circuit switched

Packet switched

(A bps per circuit)

(B bps total)

The service quality offered b y an ISP depends on the speed of the link, the blocking probability of the m o d e m pool, the congestion in the shared packet link, and the network state.

Network Economics

506

as w e saw in C h a p t e r 9, o n Ν, Μ a n d o n t h e h o l d i n g t i m e s of a subscriber's c o n n e c t i o n . (ISPs typically adopt a c o n c e n t r a t i o n ratio ( N / M ) of 10 or m o r e . Holding t i m e s d e p e n d o n w h e t h e r u s e r s face a flat-rate or a c o n n e c t - t i m e charge. U n d e r flat-rate charges, t h e average h o l d i n g t i m e is 50 m i n u t e s . ) In t h e case of CATV, subscribers access t h e ISP via a s h a r e d link, as w e saw in C h a p t e r 5. T h e link provides s p e e d s u p to 3 Mbps u p s t r e a m a n d 10 Mbps d o w n s t r e a m . Since t h e link is s h a r e d b y u p to 500 subscribers, it m a y b e congested. T h e subscribers s h a r e t h e packet-switched link w i t h a p e a k rate B, a n d so t h e quality d e p e n d s o n h o w c o n g e s t e d this link is. Lastly, t h e service quality d e p e n d s o n t h e state of t h e rest of t h e n e t w o r k d e p i c t e d in t h e figure b y a cloud. In s u m m a r y , t h e r e are d e d i c a t e d r e s o u r c e s s u c h as a subscriber loop, circuit-switched s h a r e d r e s o u r c e s t h a t m a y b e subject to blocking, packetswitched s h a r e d links t h a t m a y b e congested, a n d t h e b a c k b o n e n e t w o r k itself t h a t m a y b e congested.

103.2

Economic Principles Subscribers p a y a charge for I n t e r n e t access. Ideally, t h e charge covers t h e r e s o u r c e cost. W h e n a r e s o u r c e is d e d i c a t e d to a particular subscriber, it is clear t h a t its charge s h o u l d e q u a l its cost. W h e n r e s o u r c e s are shared, t h e charge also serves to ration access so as to limit congestion. It is difficult to assess t h e p r o p e r congestion charge. Although o n e m a y i m a g i n e m a n y k i n d s of charges, it is i m p o r t a n t to distinguish only four t y p e s : a n access fee, a usage charge, a congestion charge, a n d a service quality charge. A n access fee is a m o n t h l y subscription fee for b e i n g c o n n e c t e d to t h e n e t w o r k . T h e access m a y b e l i m i t e d to a certain p e r i o d of t i m e e a c h day, or it m a y b e u n l i m i t e d . T h e fee is paid i n d e p e n d e n t l y of h o w m a n y c o n n e c t i o n s t h e subscriber actually m a k e s , or h o w m u c h data is transferred, t h a t is, it is i n d e p e n d e n t of t h e subscriber's u s e of t h e n e t w o r k . Ideally, t h e access fee s h o u l d e q u a l t h e cost of c o n n e c t i n g a u s e r to t h e n e t w o r k . T h u s t h e m o n t h l y fee charged b y t h e t e l e p h o n e c o m p a n y should p a y for t h e local loop c o n n e c t i n g t h e subscriber to t h e c e n t r a l office. Similarly, for e a c h subscriber a n ISP i n c u r s a fixed cost for h a r d w a r e ( m o d e m , h o s t m e m o r y , etc.), software, a n d a d m i n i s t r a t i o n (setting u p a n d m a n a g i n g t h e c u s t o m e r a c c o u n t ) t h a t s h o u l d b e r e c o v e r e d t h r o u g h t h e access fee. O n t h e d e m a n d side, a u s e r w o u l d subscribe to t h e n e t w o r k o n l y if his willingness to p a y exceeds t h e benefit of access. Because access confers t h e right to u s e t h e n e t w o r k w h e t h e r or n o t o n e actually u s e s it, u s e r s m a y subscribe to p r e s e r v e t h e i r options. ( C o m m o n e x a m p l e s of access fees are t h e

10.3

Network Charges: Theory and Practice

m e m b e r s h i p d u e s o n e p a y s to j o i n a club or a m u s e u m ; m e m b e r s h i p gives o n e t h e right to u s e t h e club or visit t h e m u s e u m . ) In t e r m s of t h e r e s o u r c e m o d e l of Figure 10.5, t h e access fee s h o u l d cover t h e fixed cost of t h e subscriber loop, t h e ISP's facilities, a n d t h e b a c k b o n e network. A usage charge d e p e n d s o n t h e a m o u n t of use. It is, therefore, b a s e d o n e a c h c o n n e c t i o n or call t h e subscriber m a k e s , or t h e a m o u n t of traffic s h e g e n e r a t e s . (We u s e t h e t e r m s connection a n d call i n t e r c h a n g e a b l y to refer to a TCP c o n n e c t i o n , a n ATM virtual circuit c o n n e c t i o n , or a t e l e p h o n e call.) T h e usage charge c a n b e calculated in m a n y ways. It m a y b e a function of t h e d u r a t i o n of t h e c o n n e c t i o n , t h e a m o u n t of data transferred, or t h e end-to-end distance. T e l e p h o n e usage charges, for example, d e p e n d o n t h e call d u r a t i o n a n d t h e distance b e t w e e n t h e calling parties. Usage c h a r g e s s h o u l d e q u a t e cost a n d benefit. More n e t w o r k r e s o u r c e s ( t r a n s m i s s i o n links, buffers, r o u t e r s or switches, s y s t e m o p e r a t i o n s a n d m a i n t e n a n c e , etc.) are d e v o t e d to subscribers or c o n n e c t i o n s w i t h g r e a t e r usage, a n d so t h e y s h o u l d p a y a g r e a t e r s h a r e of t h e n e t w o r k cost. Ideally, t h e usage charge s h o u l d e q u a l t h e cost of t h e additional r e s o u r c e s t h a t a c o n n e c t i o n or call n e e d s . In t h e case of I n t e r n e t access, t h e usage charge for a c o n n e c t i o n could i n c l u d e a c o m p o n e n t t h a t is a function of t h e b a n d w i d t h a n d p r o p o r t i o n a l to t h e c o n n e c t time, a n d a c o m p o n e n t t h a t is p r o p o r t i o n a l to t h e v o l u m e of traffic. A congestion charge d e p e n d s o n t h e a m o u n t of traffic or load t h a t t h e net­ w o r k is carrying at t h e t i m e of t h e subscriber's c o n n e c t i o n . C o n g e s t i o n c h a r g e s are responsive to t h e state of t h e n e t w o r k : t h e charge is h i g h e r w h e n t h e net­ w o r k is congested, a n d t h e r e is n o c o n g e s t i o n charge w h e n it is u n c o n g e s t e d . T h e rationale for a congestion charge is as follows. Service quality deterio­ rates rapidly as t h e n e t w o r k a p p r o a c h e s congestion. In a n y n e t w o r k t h a t u s e s statistical multiplexing, s u c h as t h e I n t e r n e t or a n ATM n e t w o r k , c o n g e s t i o n in­ creases q u e u i n g delays or p a c k e t loss d u e to buffer overflow. In t h e t e l e p h o n e n e t w o r k , w h i c h u s e s time-division m u l t i p l e x i n g to r e s e r v e a fixed b a n d w i d t h for e a c h call, congestion i n c r e a s e s t h e blocking probability. In o r d e r to p r e v e n t congestion a n d t h e resulting quality degradation, t h e n u m b e r of c o n n e c t i o n s m u s t b e limited. T h e r e are m a n y w a y s to do this, b u t it w o u l d b e socially advisable to p e r m i t c o n n e c t i o n s n e t w o r k u s e r s d e e m m o r e valuable a n d p r e v e n t c o n n e c t i o n s u s e r s t h i n k are less valuable. O n e w a y to do this is b y i m p o s i n g a c o n g e s t i o n charge: u s e r s w o u l d initiate o n l y t h o s e c o n n e c t i o n s t h a t t h e y value m o r e t h a n t h e charge a n d p o s t p o n e m a k i n g lessv a l u e d c o n n e c t i o n s to a t i m e w h e n t h e congestion charge is low or zero. In this w a y congestion charges c a n b e u s e d to discourage less-valued c o n n e c t i o n s . T h e level of congestion charge s h o u l d b e j u s t sufficient to p r e v e n t congestion: too

507

508

Network Economics

low a charge will n o t p r e v e n t congestion, a n d too h i g h a charge will r e d u c e t h e n u m b e r of calls to a level b e l o w w h a t t h e n e t w o r k c a n a c c o m m o d a t e . S o m e t i m e s t h e traffic p a t t e r n varies regularly over t h e day, a n d c o n g e s t i o n predictably occurs d u r i n g certain b u s y h o u r s of t h e day. I n t h e s e cases a congestion charge m a y b e a p p r o x i m a t e d b y a "time-of-use" charge: t h e r e is a h i g h e r usage charge d u r i n g b u s y periods. ( T h e difference b e t w e e n t h e usage charge d u r i n g t h e b u s y a n d n o n b u s y p e r i o d s is a c o n g e s t i o n charge.) T h i s is w h a t t e l e p h o n e c o m p a n i e s do w h e n t h e y i m p o s e a l o w e r usage charge at night and on weekends. In t h e I n t e r n e t , congestion h a s t w o different t i m e scales. T h e r e is a pre­ dictable p a t t e r n of congestion t h a t l e n d s itself to a time-of-use charge. T h e r e also is u n p r e d i c t a b l e congestion for w h i c h a congestion charge c a n n o t b e calcu­ lated i n advance. I n this case it is possible to i m a g i n e a c o n g e s t i o n charge o n a p a c k e t t h a t is c o m p u t e d after it h a s b e e n t r a n s m i t t e d : e a c h q u e u e t r a v e r s e d b y t h e packet calculates a charge p r o p o r t i o n a l to t h e c u m u l a t i v e d e l a y i m p o s e d b y this p a c k e t o n t h e o t h e r packets. If s u c h a charge w e r e instituted, a u s e r could b e charged for h a v i n g h i s p a c k e t s p l a c e d a h e a d of o t h e r p a c k e t s i n a q u e u e . We s h o u l d n o t confuse usage a n d congestion charges. A u s a g e charge re­ flects t h e cost of n e t w o r k r e s o u r c e s t h a t a r e b e i n g used. A c o n g e s t i o n charge is a m e a n s to give n e t w o r k access to m o r e valuable c o n n e c t i o n s a n d to d e n y ac­ cess to less valuable c o n n e c t i o n s . Congestion charges reflect overall n e t w o r k d e m a n d a n d b e a r n o direct relation to n e t w o r k cost. Of course, a large con­ gestion charge is a n indication of large d e m a n d , suggesting a n o p p o r t u n i t y for profit b y investing in i n c r e a s e d n e t w o r k capacity. Finally, a n e t w o r k m a y provide different qualities of service, s o m e of w h i c h r e q u i r e m o r e n e t w o r k r e s o u r c e s t h a n others. T h e quality charge reflects this dif­ ference in r e s o u r c e use. T e l e p h o n e n e t w o r k s a n d data n e t w o r k s today typically provide only o n e service quality, so quality charges a r e u n c o m m o n . However, w i t h t h e g r o w t h of h i g h - b a n d w i d t h applications t h a t r e q u i r e g u a r a n t e e d ser­ vice quality (e.g., g u a r a n t e e d delay b o u n d s ) , services will b e differentiated b y quality, a n d quality charges will b e c o m e c o m m o n p l a c e . ATM n e t w o r k s a r e expected to provide different service quality. ( T h e postal s y s t e m c h a r g e s dif­ ferently for overnight, first-class, a n d bulk-mail delivery. T h i s is a n e x a m p l e of quality charges.) In s u m m a r y , e c o n o m i c t h e o r y suggests t h a t a n e t w o r k u s e r s h o u l d p a y a four-part bill c o m p r i s i n g a fixed charge for t h e fixed n e t w o r k costs, a us­ age charge e q u a l to t h e cost of r e s o u r c e s used, a congestion charge t h a t limits less-valued c o n n e c t i o n s , a n d a quality charge for additional r e s o u r c e s n e e d e d for higher-quality service. Practice, however, does n o t strictly follow theory.

10.3

Network Charges: Theory and Practice

1033

Charges in Practice T e l e p h o n e c o m p a n i e s h a v e t h e m o s t elaborate pricing s c h e m e s , i n c l u d i n g ac­ cess fees, usage charges t h a t d e p e n d o n distance a n d call duration, a n d con­ gestion charges t h a t are a p p r o x i m a t e d b y time-of-use charges. Often t h e r e are q u a n t i t y d i s c o u n t s for b o t h fixed a n d usage charges, reflecting t h e e c o n o m i e s of scale. In t h e U n i t e d States t h e s y s t e m of c h a r g e s is closer to w h a t t h e t h e ­ ory prescribes. However, it is n o t ideal. For example, b e c a u s e t h e y are farther a w a y from t h e switch, subscribers in lightly p o p u l a t e d rural a r e a s p a y m o r e for t h e local loop t h a n those in d e n s e l y p o p u l a t e d u r b a n areas. Nevertheless, t h e access charges are t h e s a m e for rural a n d u r b a n subscribers. Data n e t w o r k charges are u n s o p h i s t i c a t e d , b y contrast. A local area net­ w o r k usually b e l o n g s to a single e n t e r p r i s e t h a t provides free access to its m e m b e r s . Because t h e e n t e r p r i s e b e a r s all of t h e n e t w o r k costs, t h e r e is lit­ tle i n c e n t i v e to charge users, a n d n e t w o r k costs are p a r t of t h e e n t e r p r i s e ' s "overhead" costs. However, as t h e u s e a n d cost of LANs h a v e increased, a fixed, i n t e r n a l charge p e r host or p e r u s e r or d e p a r t m e n t is often i m p o s e d to r e c o v e r t h e cost. This is likely to b e t h e case in e n t e r p r i s e s t h a t h a v e large n e t w o r k costs a n d a distinct d e p a r t m e n t t h a t a c q u i r e s n e t w o r k assets a n d provides net­ w o r k services. A n additional b e n e f i t from instituting s u c h a n i n t e r n a l charge is t h a t it elicits information a b o u t t h e value m e m b e r s place o n n e t w o r k services. T h a t information c a n b e u s e d to m a k e decisions a b o u t n e t w o r k e x p a n s i o n . For a flat rate, millions of subscribers receive I n t e r n e t access a n d additional services s u c h as access to databases, chat r o o m s , a n d directories. As w e h a v e seen, t h e a b s e n c e of a n y usage-based c h a r g e s m a k e s this a n inefficient w a y of providing service. S o m e n e t w o r k services s u c h as SMDS a n d F r a m e Relay i n c l u d e a form of usage charge. In SMDS, for example, u s e r s subscribe to a service p a r a m e t e r i z e d in t e r m s of t h e s u s t a i n e d i n f o r m a t i o n rate, t h e m a x i m u m b u r s t size, a n d t h e m a x i m u m n u m b e r of i n t e r l e a v e d m e s s a g e s (see section 3.7), a n d t h e charge for t h e service d e p e n d s o n t h e s e p a r a m e t e r s . Since t h e s e p a r a m e t e r s reflect t h e n e t w o r k r e s o u r c e s d e v o t e d to providing this service, t h e charge i n c l u d e s a usage c o m p o n e n t . ISDN charges also i n c l u d e a usage c o m p o n e n t . We h a v e already n o t e d t h a t quality charges are u n c o m m o n today. However, t h e I n t e r n e t protocol does provide for distinguishing a m o n g different t y p e s of d a t a g r a m s u s i n g t h e TOS field (section 4.3). T h i s e x a m p l e suggests t h a t o n e w a y of achieving service quality differentiation is t h r o u g h priority of service, w h i c h g u a r a n t e e s preferential t r e a t m e n t (but n o t a g u a r a n t e e d delay or loss b o u n d ) . In principle, ATM will p e r m i t t h e m o s t sophisticated forms of c h a r g e s a n d service definition.

509

Network Economics

510

10.3.4

Vulnerability of the Internet Until 1990, t h e I n t e r n e t w a s h e a v i l y subsidized, a n d m o s t u s e r s b e l o n g e d to universities a n d r e s e a r c h institutions. T h e I n t e r n e t w a s rarely congested. W h e n congestion did occur, h o s t s exercised flow control. I n t e r n e t traffic g r e w b e c a u s e c o m m e r c i a l I n t e r n e t service p r o v i d e r s ex­ tended the user group beyond the academic and research communities and b e c a u s e of t h e p o p u l a r i t y of applications s u c h as t h e WWW. (More t h a n 90% of I n t e r n e t traffic is n o w u n d e r t h e com d o m a i n . ) T h i s g r o w t h h a s m a d e t h e I n t e r n e t v u l n e r a b l e a n d h a s exposed t h e difficulty in serving applications t h a t n e e d g u a r a n t e e d service quality. T h e I n t e r n e t provides a single service of u n c e r t a i n quality—best-effort service. T h e n e t w o r k accepts all c o n n e c t i o n s a n d tries to deliver t h e data pack­ ets. T h e r e is n o a d m i s s i o n control, a n d flow control is left to hosts. W h e n a router's buffers b e c o m e full, all c o n n e c t i o n s t h r o u g h t h e r o u t e r suffer p a c k e t loss or i n c r e a s e d q u e u i n g delay. Hosts d e t e c t this condition b e c a u s e of re­ t r a n s m i s s i o n t i m e o u t s . T h e y are t h e n e x p e c t e d to r e d u c e t h e i r w i n d o w size, a n d h e n c e t h e i r p a c k e t rate. However, t h e r e is n o r e q u i r e m e n t t h a t h o s t s adopt s u c h a responsible policy. Such a r e q u i r e m e n t w o u l d b e v e r y difficult to enforce. A selfish u s e r m a y deliberately decide n o t to exercise flow control. Indeed, if o t h e r s act responsibly a n d r e d u c e t h e i r p a c k e t rate, t h e selfish u s e r will receive a g r e a t e r s h a r e of n e t w o r k b a n d w i d t h a n d buffers. T h i s p e r v e r s e incentive further e n c o u r a g e s abuse. T h e resulting congestion c a u s e s r e t r a n s ­ mission, r e d u c i n g n e t w o r k t h r o u g h p u t a n d inflicting p o o r e r service quality o n all. A n I n t e r n e t service provider faces t h e s a m e p e r v e r s e incentive. T h e typical provider h a s access to t h e I n t e r n e t of a certain capacity (link s p e e d a n d r o u t e r or g a t e w a y capacity). T h e provider gives u s e r s access for a fixed charge. Tb m a x i m i z e profits, t h e provider h a s t h e i n c e n t i v e to give access to as m a n y u s e r s as possible, since t h e r e is n o service quality r e q u i r e m e n t . T h e service received b y e a c h u s e r ( m e a s u r e d b y delay or loss) will of course d e t e r i o r a t e as t h e n u m b e r of u s e r s increases. In a best-effort service n e t w o r k , e v e r y o n e receives t h e s a m e service qual­ ity, m e a s u r e d , say, in t e r m s of delay. T h i s quality d e p e n d s o n t h e total n e t w o r k load. Since t h e r e is n o a d m i s s i o n or flow control, t h e load is u n p r e d i c t a b l e , a n d so t h e r e c a n b e n o g u a r a n t e e d b o u n d o n t h e delay. T h e n e t w o r k t h u s c a n n o t m e e t t h e n e e d s of applications t h a t r e q u i r e s u c h d e l a y b o u n d s . More gener­ ally, t h e n e t w o r k p o o r l y serves c o n n e c t i o n s t h a t n e e d a predictable service quality. In C h a p t e r 4 w e p r e s e n t e d proposals to u p g r a d e t h e I n t e r n e t protocols

10.4

A Billing and Provisioning System for Internet Connections

t h a t seek to provide QpS g u a r a n t e e s . If t h e s e proposals are i m p l e m e n t e d , t h e vulnerability of t h e I n t e r n e t will b e r e d u c e d . Because t h e I n t e r n e t is a d a t a g r a m n e t w o r k , it lacks t h e control functions a n d associated protocols (like ATM n e t w o r k s ) t h a t are n e e d e d in o r d e r to provide different, g u a r a n t e e d levels of service quality. It m a y b e possible, n e v e r t h e l e s s , to provide a n i n d i r e c t form of control b y instituting a p r o p e r pricing system, including b o t h u s a g e a n d c o n g e s t i o n charges. We n o w p r e s e n t a design for s u c h a s y s t e m . We will t h e n assess t h e k i n d s of control t h a t c a n b e exercised w i t h s u c h a pricing s y s t e m .

10.4

A BILLING AND PROVISIONING SYSTEM FOR INTERNET CONNECTIONS A billing s y s t e m for usage c h a r g e s m u s t m e t e r t h e traffic. If t h e r e is a c o n g e s t i o n charge t h a t d e p e n d s o n real-time c h a n g e s in t h e n e t w o r k state, t h e billing s y s t e m m u s t m o n i t o r t h e n e t w o r k state a n d give real-time p r i c e feedback to users. If t h e n e t w o r k is able to provide variable quality service, t h e s y s t e m m u s t also b e able to provision t h a t service. T h u s , a practical billing s y s t e m m u s t h a v e t h e following features: •

No changes to existing Internet protocols and applications. Because t h e in­ stalled b a s e of hosts, routers, a n d u s e r software is h u g e , t h e billing s y s t e m m u s t w o r k w i t h o u t r e q u i r i n g c h a n g e s to existing I n t e r n e t protocols a n d widely u s e d applications s u c h as WWW, FTP, a n d e-mail.

•

User involvement. In o r d e r to bill individual users, t h e s y s t e m m u s t deter­ m i n e t h e u s e r ' s identity, explain t h e charges, a n d obtain approval for t h o s e charges in a s e c u r e m a n n e r .

•

On-line reporting of network usage. In o r d e r to institute a c o n g e s t i o n charge, t h e s y s t e m m u s t collect a n d r e p o r t in real t i m e aggregate n e t w o r k usage data so t h a t t h e a p p r o p r i a t e c o n g e s t i o n charge c a n b e calculated.

•

Sharing of information and resources. Applications like t h e WWW e n c o u r ­ age t h e s h a r i n g of i n f o r m a t i o n a n d r e s o u r c e s a m o n g r e m o t e sites. Billing s y s t e m s s h o u l d b e able to c o o p e r a t e a n d identify u s e r s a n d bill t h e m ac­ cordingly.

511

Network Economics

512

We will describe a billing a n d provisioning s y s t e m d e v e l o p e d for INDEX ( I n t e r n e t D e m a n d E x p e r i m e n t ) . We will also p r e s e n t s o m e early a n a l y s e s of t h e data. INDEX is a m a r k e t trial a n d a t e c h n o l o g y trial. T h e m a r k e t trial involves 70 subjects affiliated w i t h t h e University of California at Berkeley. INDEX c u s t o m e r s are offered a series of service plans, e a c h lasting eight w e e k s . Each service p l a n consists of a m e n u of quality-price choices. A c u s t o m e r c a n i n s t a n t a n e o u s l y select o n e of t h e s e choices. We first describe h o w a c u s t o m e r m a k e s a choice, followed b y s o m e a n a l y s e s of t h e data. We t h e n describe s o m e features of t h e technology.

10A1

User Experience After a c u s t o m e r c o n n e c t s to t h e INDEX n e t w o r k , s h e is p r e s e n t e d w i t h t h e choice m e n u . Figure 10.6 is a s c r e e n shot of t h e m e n u for o n e service plan. U n d e r this plan, t h e c u s t o m e r c a n i n s t a n t a n e o u s l y select o n e of t h e offered

10.6 FIGURE

The m e n u offers a choice of access speeds, each priced per m i n u t e of connect time.

10.4

A Billing and Provisioning System for Internet Connections

speeds, e a c h p r i c e d p e r m i n u t e of c o n n e c t t i m e . T h e prices r a n g e from 0 c e n t s p e r m i n u t e for 8 Kbps service to 5.7 c e n t s p e r m i n u t e for 128 Kbps. ( I b b e t t e r e s t i m a t e c u s t o m e r d e m a n d , different c u s t o m e r s face different r a n d o m l y selected prices for t h e s a m e quality choices.) T h e c u s t o m e r indicates h e r choice b y clicking o n o n e of t h e b u t t o n s . T h e b u t t o n is highlighted, a n d t h e status b a r o n t h e b o t t o m indicates t h e c u r r e n t c o n n e c t i o n . T h e c u s t o m e r toggles t h e m e t e r o n t h e b o t t o m right to find o u t h e r e x p e n d i t u r e for t h e c u r r e n t session, for t h e day, a n d for t h e m o n t h to date. T h i s m e t e r is u p d a t e d e a c h m i n u t e so t h e u s e r r e c e i v e s a n indication in real t i m e of t h e value of t h e n e t w o r k r e s o u r c e s s h e is c o n s u m i n g . At a n y t i m e t h e u s e r c a n obtain on-line a detailed a c c o u n t of h e r e x p e n d i t u r e s . T h e choice t h a t c o n s u m e r s m a k e u n d e r this service p l a n indicates h o w m u c h t h e y value different s p e e d s . O t h e r service p l a n s m e a s u r e different di­ m e n s i o n s of quality a n d pricing s t r u c t u r e . T h e s e i n c l u d e : s e p a r a t e selection of t h e u p s t r e a m a n d d o w n s t r e a m bit rates, pricing b y v o l u m e , c o m b i n a t i o n of p e a k rate p e r m i n u t e pricing a n d v o l u m e pricing, a n d w e e k l y flat-rate charge d e p e n d i n g o n speed. O n e a r g u m e n t against t h e i n t r o d u c t i o n of usage-based pricing is t h a t a c u s t o m e r m i g h t b e r e l u c t a n t to subscribe to s u c h a service b e c a u s e of t h e potential to r u n u p a large bill. T h e INDEX c u s t o m e r c a n limit h e r e x p e n d i t u r e b y u s i n g t h e settings p a n e l s h o w n in Figure 10.7. T h i s allows o n e to set limits o n e x p e n d i t u r e b y m o n t h , day, a n d session. W h e n a n y of t h e s e limits is exceeded, t h e c u s t o m e r ' s service is d i s c o n n e c t e d a n d t h e settings p a n e l highlights t h e exceeded limit. T h e u s e r c a n t h e n raise t h e limit or stop c o n s u m i n g further service.

10.4.2

Demand for Variable Quality As w e explained in section 10.2, flat-rate pricing p r e v e n t s ISPs from offering differentiated quality service. W h a t ISPs offer i n s t e a d is a multi-tiered s c h e m e in w h i c h different access s p e e d s are offered at different flat-rate charges. U n d e r this s c h e m e , a c u s t o m e r m u s t select o n e of t h e access s p e e d s , b u t s h e c a n n o t switch from o n e s p e e d to another. T h e m o n t h l y charge for DSL access m i g h t vary b e t w e e n $40 a n d $200 d e p e n d i n g o n t h e s p e e d . T h i s is inefficient o n several c o u n t s . First, t h e difference in charge is m u c h g r e a t e r t h a n t h e difference in fixed costs b e t w e e n DSL l i n e s of different s p e e d s . (In fact t h e DSL m o d e m s for different s p e e d s are identical.) T h e different fixed charges are a form of price

513

Network Economics

514

10.7

The settings panel allows the user to control h e r expenditure.

FIGURE

discrimination. T h e i r p u r p o s e is to s e g m e n t subscribers so t h a t t h e y c a n b e charged different a m o u n t s . Second, since subscribers face n o u s a g e charge, t h e y will waste c o m m u n i c a t i o n r e s o u r c e s . T h e w a s t e i n c r e a s e s w i t h access speed. Third, subscribers w h o w a n t to u s e h i g h e r - s p e e d service for a l i m i t e d a m o u n t of t i m e m a y b e unwilling to b e locked into a h i g h e r - s p e e d tier. Both providers a n d subscribers lose from multi-tiered flat-rate pricing. Providers lose b e c a u s e t h e y exclude low-usage subscribers from t h e highers p e e d service. Subscribers lose b e c a u s e t h e y c a n n o t obtain t h e l i m i t e d a m o u n t of high-speed service for w h i c h t h e y are willing to pay. T h e variable-speed service p l a n of Figure 10.6 allows INDEX u s e r s to switch b e t w e e n s p e e d s w i t h n o friction. T h u s t h e choice u s e r s m a k e u n d e r this service p l a n c a n b e u s e d to gauge t h e potential loss from multi-tiered flat-rate pricing. INDEX c u s t o m e r s o n average u s e d t h r e e different s p e e d s p e r w e e k . Moreover, t h e i r usage of a particular s p e e d w a s v e r y sensitive b o t h to t h e price of t h a t s p e e d as well as to t h e price of t h e adjacent speed, as s e e n from t h e e s t i m a t e d d e m a n d e q u a t i o n (10.2). This s h o w s t h e v e r y h i g h price sensitivity

10.4

A Billing and Provisioning System for Internet Connections

of c o n s u m e r s a n d suggests t h a t t h e loss from multi-tiered flat-rate c h a r g e s for variable quality to providers a n d c o n s u m e r s is large.

10A3

The INDEX Billing and Provisioning System We describe this s y s t e m b y first explaining h o w IP p a c k e t s are t r a n s f e r r e d b e t w e e n INDEX subscribers a n d t h e rest of t h e I n t e r n e t . We t h e n explain a c c o u n t i n g a n d provision of service quality. Figure 10.8 depicts t h e INDEX physical n e t w o r k . A subscriber's c o m p u t e r is c o n n e c t e d via a n E t h e r n e t interface to a n ISDN r o u t e r w i t h a PPP c o n n e c t i o n to t h e INDEX N e t w o r k Access S y s t e m (NAS). Traffic from m a n y u s e r s is aggre­ gated b y t h e NAS. T h e NAS t u n n e l s t h e PPP c o n n e c t i o n s to t h e billing g a t e w a y (BGW). I n this way, t h e n e t w o r k establishes a distinct logical l i n k b e t w e e n e a c h u s e r a n d t h e BGW. In t h e case of t h e INDEX ISDN e x p e r i m e n t , t h e NAS is a r o u t e r t h a t s u p p o r t s m a n y ISDN interfaces. For access via CATV n e t w o r k , t h e CMTS (cable m o d e m t e r m i n a t i o n s y s t e m ) serves as t h e NAS. For DSL access, t h e DSLM (DSL multiplexer) serves this function. T h e BGW r o u t e s IP p a c k e t s b e t w e e n a u s e r a n d t h e WAN.

10.8 FIGURE

INDEX physical network. A PPP connection over ISDN establishes a logical link between a user's computer and the billing gateway.

515

Network Economics

10.9 FIGURE

Control packets b e t w e e n the user and the network operations center provide instructions for accounting and service provision.

Since t h e BGW is in t h e data p a t h b e t w e e n a n INDEX u s e r a n d t h e WAN, it collects statistics s u c h as b y t e c o u n t s a n d c o n n e c t i o n r e c o r d s t h a t are to b e u s e d for accounting. In addition to t h e data p a c k e t s b e t w e e n t h e u s e r a n d t h e WAN, t h e r e also are control or signaling p a c k e t s b e t w e e n t h e u s e r a n d t h e INDEX n e t w o r k o p e r a t i n g c e n t e r (NOC). T h e relations b e t w e e n t h e u s e r a n d t h e NOC are d e p i c t e d in Figure 10.9. T h e left side of Figure 10.9 s h o w s t h e u s e r interface (UI), o n e p e r user. T h e u s e r indicates his choice b y clicking t h e a p p r o p r i a t e b u t t o n o n t h e service p l a n w i n d o w (see Figure 10.6). T h e Service Plan Executive observes t h e choice a n d i n t e r p r e t s it in t e r m s of (1) i n s t r u c t i o n s to t h e BGW for service provision as explained b e l o w a n d (2) a formula t h a t calculates t h e u s e r ' s e x p e n d i t u r e e a c h m i n u t e from t h e statistics collected b y t h e BGW. (For example, in case of a v o l u m e charge, t h e price is multiplied b y t h e b y t e c o u n t r e c e i v e d from t h e BGW.) This is r e c o r d e d in t h e database a n d is u s e d to u p d a t e t h e e x p e n d i t u r e m e t e r s o n t h e u s e r interface. T h e BGW e m u l a t e s a n e t w o r k of p a r a m e t e r i z e d l e a k y b u c k e t s , o n e p e r user. T h e i n s t r u c t i o n from t h e SP Executive to t h e BGW specifies t h e p a r a m e t e r s to m a t c h t h e service choice m a d e b y t h e user. Each l e a k y b u c k e t is specified b y t h r e e p a r a m e t e r s : t h e s u s t a i n e d rate, t h e size of t h e bucket, a n d a small buffer for t h e difference b e t w e e n t h e 128 Kbps ISDN line rate a n d t h e s u s t a i n e d rate. For example, if t h e u s e r selects 64 Kbps, t h e l e a k y b u c k e t limits t h e s u s t a i n e d rate to 64 Kbps.

10.4

A Billing and Provisioning System for Internet Connections

Service over DSL access m a y b e p r o v i s i o n e d in t h e s a m e w a y u s i n g l e a k y b u c k e t s . In CATV access, t h e service choice m a y b e p r o v i s i o n e d b y limiting t h e n u m b e r of r e s e r v a t i o n slots given to a u s e r for u p s t r e a m data, w h e r e a s downstream rates m a y be limited by using leaky buckets.

10A4

Flexibility of INDEX Pricing and Provisioning T h e INDEX pricing s y s t e m is flexible: it p e r m i t s pricing b y c o m b i n a t i o n s of flat rate, c o n n e c t time, v o l u m e , p e a k capacity, a n d t i m e of day. A c e r t a i n a m o u n t of congestion pricing is possible as well. If n e t w o r k state c a n b e e s t i m a t e d , t h e prices p r e s e n t e d to t h e u s e r c a n b e a function of t h a t state. INDEX provisions service b y limiting access. If n e c e s s a r y , t h e s y s t e m c a n exercise a d m i s s i o n control: a u s e r ' s choice m a y b e d e n i e d , or t h e service choices p r e s e n t e d to t h e u s e r m a y d e p e n d o n t h e n e t w o r k state. T h e major limitation of INDEX is t h a t it c a n n o t g u a r a n t e e end-to-end service quality, b e c a u s e o n l y access is provisioned. A distributed v e r s i o n of INDEX could b e e n v i s i o n e d for t h e p u r p o s e of end-to-end quality. T h e idea is d e p i c t e d in Figure 10.10. T h e WAN is divided into a set of billing d o m a i n s , e a c h w i t h a n access link controlled b y a billing gateway. W h e n a u s e r r e q u e s t s a p a r t i c u l a r end-to-end service, t h e billing g a t e w a y a t t e m p t s to provision a r o u t e (using a protocol s u c h as RSVP). A n alternative a p p r o a c h is possible if policy-based r o u t i n g is a d o p t e d in t h e b a c k b o n e n e t w o r k . A quality choice is t h e n m a p p e d into a flow t h a t

10.10 FIGURE

The Internet is modeled as a set of billing domains with access links controlled b y a billing gateway.

517

Network Economics

518

receives t h e a p p r o p r i a t e b a n d w i d t h or priority along its route. N e i t h e r of t h e s e a p p r o a c h e s c a n assure full g u a r a n t e e . W h a t e i t h e r of t h e m c a n a c c o m p l i s h is to provide feedback to t h e u s e r in t h e form of charges t h a t reflect t h e value of t h e r e s o u r c e s in t h e n e t w o r k c o n s u m e d b y t h e user. T h i s w o u l d i m p r o v e t h e total benefit to u s e r s of c o m m u n i c a t i o n services. In t h e n e x t section w e s t u d y this in a simple context.

10.5

au

PRICING A SINGLE RESOURCE T h e link b e t w e e n a billing d o m a i n or o n e large organization a n d t h e I n t e r n e t b a c k b o n e is a n expensive resource. Its cost i n c r e a s e s as t h e link capacity is increased, b u t t h e service to u s e r s will improve, too. T h u s t h e r e is a n e e d to b a l a n c e service benefits against capacity cost. Figure 10.11 s h o w s a trace of WAN traffic from t h e Berkeley c a m p u s ob­ t a i n e d d u r i n g a n e x p e r i m e n t referred to earlier. In t h e trace, SMTP, NNTP, a n d FTP a c c o u n t for 11%, 22%, a n d 47% of all b y t e s transferred, respectively. If w e take t h e 3 3 % of e-mail a n d bulletin-board traffic a n d s p r e a d it o u t over t h e 24 h o u r s of t h e trace duration, t h e total a m o u n t of traffic t h a t lies above t h e aver­ age rate is r e d u c e d b y a b o u t 30%, as s h o w n in t h e left p a n e l of Figure 10.11. T h e p a n e l o n t h e right s h o w s t h a t b y buffering traffic for 1 s, t h e 100-ms p e a k traffic rate is r e d u c e d b y 40% ( c o m p a r e top a n d m i d d l e graphs), a n d b y buffering for 20 m i n , t h e p e a k is r e d u c e d b y 60% ( c o m p a r e m i d d l e a n d lowest graphs). A s s u m e for n o w t h a t t h e capacity of t h e link b e t w e e n t h e Berkeley c a m p u s a n d t h e WAN is fixed. H o w c a n o n e m a x i m i z e u s e r benefits? T h e data s h o w s t h a t if w e divide t h e traffic b e t w e e n delay-insensitive (e-mail a n d bulletinb o a r d services) a n d delay-sensitive (the rest) traffic, t h e r e are two w a y s to r e d u c e p e a k capacity a n d a c c o m m o d a t e m o r e traffic. First, b y shifting delayinsensitive traffic, p e r h a p s b y several h o u r s , to low utilization p e r i o d s d u r i n g t h e day, t h e p e a k rate c a n b e d e c r e a s e d b y 20%. Second, b y buffering delaysensitive traffic for a v e r y short t i m e o n t h e o r d e r of a second, o n e c a n r e d u c e t h e p e a k rate b y 40%. T h e shifting over t i m e of t h e s e two t y p e s of traffic c a n b e a c c o m p l i s h e d b y administrative p r o c e d u r e s , or b y a s y s t e m of t i m e of u s e a n d c o n g e s t i o n charges. T h e administrative p r o c e d u r e s w o u l d classify s o m e traffic t y p e s as delayinsensitive a n d restrict t h e i r t r a n s m i s s i o n to low utilization p e r i o d s of t h e day. T h e r e m a i n i n g , delay-sensitive traffic w o u l d b e buffered for short p e r i o d s of t i m e (on t h e o r d e r of 1 s). T h i s a p p r o a c h is easy to i m p l e m e n t , b u t it h a s

10.5

Pricing a Single Resource

519

— Peak 100 ms — Peak 1 s — Average

20

— All traffic Interactive — Water level 01/19/95

10.11 FIGURE

200 100 01/17/95 01/18/95 01/19/95 01/20/95

Reduction in peak traffic rate by spreading e-mail and bulletin-board traffic uniformly over the day (left); peak traffic rate measured over 0.1-s and 1-s intervals and 20-minute average (right).

t w o defects. It m a k e s t h e d e t e r m i n a t i o n of w h i c h traffic is delay-insensitive a m a t t e r of b u r e a u c r a t i c choice. Such choice is likely to b e c r u d e . For example, a l t h o u g h e-mail typically is delay-insensitive, s o m e e-mail c o n n e c t i o n s m a y b e urgent, b u t a n a d m i n i s t r a t i v e p r o c e d u r e c a n n o t recognize t h e s e u r g e n t m e s s a g e s . Second, t h e a d m i n i s t r a t i v e classification will h a v e to b e e x t e n d e d over t i m e to i n c l u d e n e w applications as t h e y develop. T h i s e x t e n s i o n is b o u n d to b e arbitrary, since t h e r e is n o a c c u r a t e w a y to predict t h e p u r p o s e s for w h i c h t h e n e w applications will b e u s e d . A m u c h b e t t e r w a y to shift traffic over t i m e is to allow u s e r s t h e m s e l v e s to m a k e t h e choice. In a n ideal world, u s e r s w o u l d b e altruistic a n d voluntarily t r a n s m i t t h e i r delay-insensitive traffic d u r i n g p e r i o d s of low d e m a n d . But a l t r u i s m is unreliable. A m o r e reliable m e c h a n i s m is t h r o u g h a s y s t e m of prices. If u s e r s are c h a r g e d a l o w e r (or zero) r a t e at night, t h e n t h e y will h a v e a n i n c e n t i v e to shift t h e i r delay-insensitive traffic to t h o s e periods. For delaysensitive traffic, a congestion charge could b e i m p o s e d to k e e p d o w n t h e p e a k d e m a n d . We will next s t u d y t h e s e two price m e c h a n i s m s . It is useful to k e e p in m i n d t h a t o u r discussion applies to a n y situation in w h i c h a single r e s o u r c e is s h a r e d b y m a n y u s e r s . T h i s r e s o u r c e could b e a n access link, a disk s y s t e m , a pool of c o m p u t e r s , or a h i g h w a y .

Network Economics

520

10.5.1

Usage-Based Prices We develop a n e c o n o m i c m o d e l t h a t will focus o u r discussion o n usage-based pricing. T h e m o d e l is built b y specifying t h r e e e l e m e n t s : t h e u s e r s ' d e m a n d for service; t h e n e t w o r k capacity, t h a t is, t h e a m o u n t of service t h a t t h e n e t w o r k can supply; a n d t h e i n t e r a c t i o n b e t w e e n d e m a n d a n d s u p p l y t h r o u g h prices. We consider a single service (say, best-effort service) t h a t is differentiated b y t i m e of day. We divide t h e day into periods d e n o t e d b y t = 1, . . . , T. For notational simplicity, consider only two periods: t = 1 is t h e "peak" period, a n d t = 2 is t h e "off-peak" period. We consider a p o p u l a t i o n of potential u s e r s indexed b y i = 1, . . . , I. Consider a user's decision regarding o n e c o n n e c t i o n , say, to s e n d e-mail or to b r o w s e t h r o u g h a WWW site. T h e u s e r m u s t choose b o t h t h e p e r i o d (r = 1 or 2) a n d t h e a m o u n t of traffic to t r a n s m i t ( m e a s u r e d in bytes, say). We m o d e l t h e user's p r e f e r e n c e s b y t h e utility function

x>0,

u (x) = u(x) — d x, t

t

r = 1,2.

H e r e χ is t h e a m o u n t of traffic, u(x) is t h e b e n e f i t ( m e a s u r e d in dollars) t h a t t h e u s e r derives from s e n d i n g x, a n d d x is t h e loss or benefit r e d u c t i o n (also m e a s u r e d in dollars) suffered from s e n d i n g χ in period t. Typically, d\ < d , w h i c h is to say, m o s t u s e r s prefer s e n d i n g t h e traffic d u r i n g t h e p e a k period. Suppose t h e price for s e n d i n g 1 b y t e in p e r i o d r is p . If t h e u s e r selects period t, she will decide to t r a n s m i t a m e s s a g e of t h e size t h a t will m a x i m i z e h e r n e t benefit, t h a t is, s h e will solve t h e p r o b l e m t

2

t

m a x u(x) — d x — p x. t

t

T h e o p t i m u m m e s s a g e size, x , is given b y setting to zero t h e derivative w i t h r e s p e c t to x: t

— \u(x) - d x -ptx] dx t

= 0 or u'{x ) =p t

t

+ d.

(10.3)

t

We d e n o t e t h e n e t benefit she will derive from this t r a n s a c t i o n b y V(p

+ dt) = u(x ) - (p + d )x.

t

t

t

(10.4)

t

Here, u (x) := du/dx is t h e d o w n w a r d sloping curve in Figure 10.12; x is t h e size w h e r e this curve intersects t h e line t h r o u g h p + d ; a n d V(p + d ) is t h e s h a d e d area b e t w e e n t h e curve a n d t h e line. We c a n solve (10.3) to obtain h e r d e m a n d as a function of (p + d ). We write this as x = D(p + d ). In t e r m s of l

t

t

t

t

t

t

t

t

t

t

10.5

Pricing a Single Resource

10.12 FIGURE

521

A user's d e m a n d curve D(p + d ) is obtained from the marginal utility u'{x ). ^ sh^ed * user's surplus V(p + d ). t

T

e

a r e a

t

t

s

t

f

Figure 10.12, D(p + ^t) is s i m p l y t h e d o w n w a r d sloping c u r v e u'{x) e x p r e s s e d as a function of t h e ordinate or y coordinate. Note t h a t D d e c r e a s e s as p increases. T h u s , if t h e u s e r decides to t r a n s m i t in period 1, h e r b e n e f i t is V(pi + d\) a n d if s h e decides to t r a n s m i t in p e r i o d 2, h e r benefit is V(p + d ). Clearly, s h e will choose t h e p e r i o d t h a t yields a larger benefit, t h a t is, s h e will choose p e r i o d 1 if p\ + d\ < p + d , a n d s h e will choose p e r i o d 2 o t h e r w i s e . Let u s n o w c o n s i d e r a n arbitrary u s e r i, w h o s e utility function is given b y u\{x) = u\x) — d\x. T h i s u s e r will select p e r i o d 1 \ΐρ\ +d\