ATM and Internet Protocol A convergence of technologies Mark Bentall Cogent Defence Systems Newport, Gwent UK Chris Hobbs Nortel Enterprise Networks Ottawa Canada Brian Turton Division 'ofElectronic Engineering Cardiff School of Engineering University of Wales Cardiff UK ELSEVIER A memberoftheHodderHeadlineGroup LONDON·SYDNEY·AUCKLAND Copublished in North,Centraland South America by John Wiley& Sons Inc., New York•Toronto FirstpublishedinGreatBritain in1998by Arnold, amember oftheHodder Headline Group, 338Euston Road, LondonNWI 3BH http://www.arnoldpublishers.com ElsevierLtd. LinacreHouse,JordanHill,OxfordOX28DP 200WheelerRoad,Burlington,MA01803 Transferredtodigitalprinting2007 © 1998M.Bentall, C.Hobbs andB.Turton Allrightsreserved. Nopartofthispublicationmaybereproducedor transmittedinanyformorbyanymeans,electronicallyormechanically, includingphotocopying,recordingoranyinformationstorageorretrieval system, withouteither priorpermissioninwritingfromthepublisherora licence permittingrestrictedcopying. IntheUnitedKingdom suchlicences areissuedbytheCopyrightLicensingAgency: 90TottenhamCourtRoad, LondonW1P9HE. . Whilsttheadvice andinformationinthisbook isbelievedtobetrueand accurateatthedateofgoingtopress,neithertheauthornorthepublisher canacceptanylegalresponsibilityorliability foranyerrorsoromissions thatmaybemade. BritishLibraryCataloguinginPublicationData Acataloguerecord forthisbookisavailable fromtheBritishLibrary LibraryofCongress Cataloging-in-PublicationData °Acatalogrecord forthisbookisavailable fromtheLibraryofCongress ISBN 0340719214 ISBN 0471 314277(Wiley) CommissioningEditor: SianJones ProductionEditor: JamesRabson ProductionController:RoseJames Coverdesigner: TerryGriffiths This bookis dedicatedto the memoryof IsambardKingdom Brunel, an engineer PREFACE The world'stelecommunicationsnetworks were created tocarry voice trafficbut are now being called upon tocarry various types ofdata, typicallybetweencom puters. 1996 is believed to have been the first year when data traffic exceeded voice traffic. Over the last hundred years, many large telecommunications companies have beenestablishedtotransfer voice fromone'customer toanother. Suchcompanies are seeing their traditional markets being eroded and are, ingeneral, responding as quickly as their size allows. Broadband ISDN (B-ISDN) is a key element in their strategies. There are also large companieswhich produce the devices required totransfer data. Now that data traffic predominates, and since voice can be considered nothing more than a particular kind ofdata, business that was traditionally the sole preserve of the telecommunications companies is now being targeted by their new competitors whose weapon is the Internet Protocol - IP. As these two forces square up to each other, the interplay ofB-ISDN and IP has become a crucial factor on the battlefield. Thisbook attempts toexplain B-ISDNtoareader entering themodemtelecom municationsfieldeither from thedataarenaorfrom themore traditionaltelecom munications disciplines (X.25 or Frame Relay). It then explores the inter relationship ofIP and ATM. Little more than a basic understanding oftelecom munications is assumed from the reader. As with most areas of modem telecommunications, B-ISDN is complex, tightly coupled to other techniques and encompasses a forest ofabbreviations, jargon and acronyms. The best advice that the authors can give a newcomer to this field is to have been born in about 1950. This would bring them into the telecommunications industry in the mid 1970s, a time when the installed systems were still relatively simple and the techniques which now dominate the field were still in their infancy. Anyone statisfying this requirement would have been able to participate in the definition of the ISO OSI reference model in the late 1970s,produce packet switching devices to the X.25 standards and collaborate with the IEEE on the creation of the LAN standards in the early 1980s, work with the emerging digital voice switching standards in the mid 1980s, design a few ISDN terminal viii Preface adaptors in the later 1980sand see the labyrinth oflater standards - on Frame Relay, ATM and Network Access - emerge from the CCITT and ETSI in a reasonable historical sequence. Readers bornsignificantlylaterthan 1950arecoming intoanareaofenormous complexity without this historical perspective and are likely to find themselves among experts talking apparent gibberish and quoting from thousands ofstan dards documents, each hundreds of pages long and each referring in tum to several dozen others. This is the state of the telecommunications industry today. A few years ago, B-ISDN and ATM, the primary subjects ofthis book, were envisaged as unifying andsimplifying technologies, designed toremove thelayers ofbaroque complexity which had grown in the previous 15 to 20 years. In practice, by combining speech, data and video on one network, B-ISDN has sparked the battle for survival between the large, slow-moving telephone companies, bent on deploying ATM and mopping up the data traffic business, and the small, fast-moving data companies, meeting the challenge by moving into the speech transmission business. The battle is not yet over and it is not yet obvious who will win. This is an exciting time in the telecommunications field. The purpose of this book is to lead readers gently onto the battlefield and, by introducing the techniques ofB-ISDN in relation to IP, to arm them for the fight. Thisbook includes aglossary, startingon page 129,with shorttutorial articles on terms of general telecommunications applicability. Mark Bentall Cogent Defence Systems, Newport, Wales Chris Hobbs Nortel Enterprise Networks, Ottawa, Canada Brian Turton University ofWales, Cardiff, Wales 1 AN INTRODUCTION TO B-ISDN ANDATM Readers with little telecommunications backgroundmayfind the glossary entry for telecommunications onpage 154 useful when reading this chapter. 1.1 Broadband ISDN This chapter gives an outline ofBroadband ISDN and ATM, an outline which is amplified in the following chapters. The road towards ATMbegan in the 1980s with Packet-Switched Networks based ontheX.25 standards (seereference [1]). Inthese networks, datatrafficis X25 broken into variable-length frames andtheframes arerouted individually across the network for re-assembly at their destination. In the slow and unreliable networks for which X.25 was designed, a fairly heavy protocol (i.e. a protocol requiring substantial computing power and carrying a significant overhead with the data) was required to ensure that errors in transmission affecting one frame did not require whole files to be retransmitted. Networks are now much more reliable - the bit error rate probability of a modemfibreopticlinkisapproximately 10-7to 10-12perbit(seereferences [2, 3, 4]) - and with this improvement, it is possible to lighten the protocol and switch variable-length packets very quickly; the new, lighter protocol is known as Frame Relay. Frame Relay Frames, however, have different lengths and even when associated with a light protocol, they are difficult for hardware to handle; they generally require software intervention and this slows down their transit through switches. A logical step, therefore, was to fix the size ofthe frame (and to call it a cell), cell andATM,thetransport mechanism forBroadband ISDN,wasborn, combining a lightweight protocol withfixed-length cells. Itshouldbe said, however, that due totheinflexibilityoffixed-lengthcells,therenowappearstobesomepressure on integrated circuit manufacturers to support variable-length frames in hardware and a resurgence offrame- rather than cell-switching is apossibility in some environments. Broadband ISDN (B-ISDN) is a set of communications protocols designed B-ISDN to transport a wide range ofservices simultaneously, services which previously 2 An introductionto B-ISDNandATM traversed different networks. In general, theseservices require more bandwidth than is offered by a traditional speech telephony circuit and may eventually include residential services such as distributed video (including home shop ping, TV, broadcast distance learning, high-speed Internet connections, video on-demand, catalogue andadvertising services)andinteractive multimedia(such as home doctor, virtual reality, interactive games, individualised distance learn ing and on-line yellow pages). While the economics ofresidential B-ISDN stillrequires agreat deal ofstudy (forexample, itiseasytocalculate that, scaledagainst thecostofaconventional telephone call, watching a2hour video by thismethod should cost inthe region of£1500),the business applications ofB~ISDN make more immediate financial sense. Although mostofthepopular publicity hasconcerned broadband delivery to the home, while some companies are trialling broadband to the home, many more are selling B-ISDN to businesses. Business applications include the interconnection oflocal area networks, mul timediatelephony andteleconferencing, computerload-sharing, working athome (tele-commuting) andthetransmission ofimages to centralised pools ofexperts. Of course, these services have been available for some time, but only by using a plethora of different devices (routers, bridges, etc.) and transmission techniques (leased lines, public packetswitching networks, etc.). The purpose ofB-ISDN is to simplify this for the business customers while reducing their costs - instead ofrenting a permanent leased line between offices a company can now access the B-ISDN services and pay only for usage. MAC Underlying B-ISDN is the low-level Media Access Control Protocol (MAC) for transferring the actual information. This is common to all services and is ATM known as Asynchronous Transfer Mode (ATM). B-ISDN and ATM are com monly used synonymously, but the distinction is quite clear: B-ISDN is the means ofcarrying traffic seen by the end-user, ATM is the low-level protocol which actually does the carrying. Thus, several trends areencouraging thewidespread introductionofATM:the availability ofhigh-speed, low error-rate communication links between switch ing centres, the availability oftechnology to digitise video and speech and the pressure to reduce operating costs by integrating previously separate telephony and data networks. 1.2 Traffic types Speech has been digitised for transmission across the voice network for many years; intercomputerdatatrafficis,byitsnature, digitised; andwiththemore so phisticatedstandards andtechnology forcompressed video (and with consumers lowering their standards ofvideo acceptability), video can now also be digitised within a reasonable bandwidth. Thus, three services which have traditionally been carried across differentnetworks arenow simply strings ofnumbers which Traffic types 3 need to be transmitted from point to point. This is, ofcourse, a very simplistic view; although the information for each of the services is a stream of numbers, the necessary characteristics of their transmission differ greatly: • Speech services must'have areasonably short and constant delay between source anddestination. Longdelays, suchasthosesometimes encountered ontrans-Atlantic satellite links, make two-way conversationawkward and largevariations indelayproduce broken sound effects which make speech difficult to understand. A generally accepted maximum delay is 25 ms and delays longer than this are not only unacceptable to users, they also make echo-suppression equipment very complex and expensive. It is also very important to preserve the order in which audio samples are received; ifthe samples are misordered intransit then they have to be discarded or buffered, causing delay and increased cost at the receiver. However, the situation is not all bad: humans are more tolerant of interference on signals than is a computer. The analogue telephone net work reduces audio quality significantly in any case and subscribers are familiar with this. This tolerance of subscribers to reduction in speech quality is also exploited by carriers who reduce the bandwidth on their lines by compressing the speech and by not transmitting during silent periods. Since over 500/0 of a telephone conversation in each direction is normally silent, this technique, known as Silence Suppression, reduces the amount of information to be sent at the cost ofturning what was a constant bit-rate datastream into a variable bit-rate stream. • Video broadcast services have characteristics similar to speech though the rate at which information is transported is much higher and absolute delayis no longer ofprimary importance. Delay variance, however, and misordering ofreceived packets isstillamajorproblem although alimited amount ofsuch distortion isacceptable tohuman consumers watching the result. Toreduce the network bandwidth, compression techniques (see, for ex ample, reference [5]) are used to remove repetition within and between successive video frames. Usingcompression produces avarying data rate; forexample twosuccessive similarvideo frames would require fewer data to betransmitted than two very dissimilar frames. Within the telecommu nications community, the video of the film Star Wars has become almost a de.facto standard for test purposes because ofits sudden (and violent) changes in colour and background. The use ofcompression with video, as with audio, has the characteristic of turning what is essentially a con stant bit-rate signal into a variable bit-rate stream. Data rates ofbetween 1.5Mb/s and 8 Mb/s are typical. • Inter-computer traffic includes document text, data files, still pictures, audio and video. Generally, this type oftransmission is sensitive neither 4 An introductionto B-ISDNandATM Service Absolute Delay Corruption Misordering delay variance Audio Must be Intolerant Tolerant Intolerant small Video Tolerant Intolerant Tolerant Intolerant Inter- Tolerant Tolerant Intolerant Tolerant computer Table1.1Service characteristics to delay nor to delay variance since nothing is done with the data until they have all arrived and the original file has been re-constructed. The receiving computer is usually also tolerant of mis-ordering of the parts of the message in transmission, being able to hold parts which arrive early and insert them into their correct position later. The computers are generally, however, intolerant ofany corruption ofthe information being transferred: these services rely on an error-free connection. All services can be considered as a combinationofthese three elements: real time audio, real-time video and inter-computerdata. The characteristics ofthese basic elements are summarised in Table 1.1where it can be seen that providing the necessary Quality of Service to each of these simultaneously is likely to prove difficult. One characteristic ofthese services, hinted at above but not included in this table, isthe variability ofthe bit rate. Four different bit rate types are supported by the ATM standards: Constant Bit Rate (CBR)~ Variable Bit Rate (VBR)~ Available Bit Rate (ABR)~ andUnspecified Bit Rate (UBR). These are described CBR inmore detail below but, asthe name suggest, CBR trafficderives from.asource where information ispresentedataconstant rate (for example, telephony speech without silence suppression). VBR VBR traffic derives from a variable source (for example, compressed video or voice with silence suppression). Two types of VBR have been defined: VBRrt VBRrt (VBR real time) and VBRnrt (VBR non-real time) to handle traffic with VBRnrt the respective characteristics. A great deal ofresearch has been carried out in recent years into the observed very 'bursty' nature ofsome data traffic. Much inter-LAN and compressed video traffic has been shown to be self-similar or fractal: that is, over whatever time interval the traffic is observed, the same type of 'burstiness' appears. Thus the bursts themselves appear in bursts. This type oftraffic causes significant difficulties for practical switches because ofthe difficulty in estimating (and supplying) the correct amount ofbuffering. Since Traffictypes 5 the beginning oftelephony, assumptions have been made about the aggregation oftraffic: forexample, assuming thatwhile thetrafficfrom onetelephone source is likely to be somewhat unpredictable, the combined traffic from thousands of telephones is highly predictable and stable. This, so-called, Poisson model breaks down very badly with the self-similar traffic where there is little or no 'averaging' ofthe peaks and troughs from multiple sources. See reference [6] for more details of this very important phenomenon which was observed and recorded as early as 1951, albeit not in ATMnetworks (see reference [7]). ABR is a later addition to the ATM world. Once a carrier has allocated the ABR necessary bandwidth on his links to carry the CBR traffic and the guaranteed minimum for the VBR traffic, ABR is the mechanism to share the remaining bandwidthfairly, between subscribers. TosupportABR,~customer'sequipment must accept signals fromthenetwork to.indicatethat itshould reduce orincrease the bandwidth it isusing (that is, send cells more slowly or more quickly). This type oftrafficrate, while convenientforanapplication where the transfer ofdata isneither delay-nortime-critical, places highdemands onthenetwork butallows the network providers to operate the network highly efficiently. ABR has also been termed Best Effort Delivery and isexpected bysometoreplace theInternet Best Effort Protocol (IP) since it offers higher transmission speeds (see reference [4]). The Delivery mechanisms whereby the network signals congest tothe attached terminals have been another area ofhot debate- between the rate-based schemes, which would appear to work better in wide area networks, and the credit-based schemes, which work better in local area networks. The basic principles of the two schemes are: • In a rate-based scheme, the network provider signals amaximum bit-rate Rate-based to the user's terminal and the terminal is not allowed to transmit at a Flow Control higherrate. When the network congests, it signals a lower rate to each of the sources; as the congestion clears it signals a higher rate. • In a credit-based system, the terminals and switches in the network ex- Credit-based change so-called credits with theirneighbours and only devices inposses- Flow Control sion ofcredits may transmit. Ifthe number offree buffers in a particular switch is becoming dangerously low, that switch will withhold credits from its neighbours. The neighbours will then not be able to send it ABR traffic. For a time it appeared that a compromise between these would prevail, us ing rate-based algorithms across the wide area and credit-based algorithms in local areas, but this was finally abandoned in favour ofrate-based techniques. This scheme relies on Resource Management (RM) cells which are generated RM artificially by the source device and which are inserted into the data stream. These cells are reflected by the destination back onto the originator. As theRM cells return, they are modified by intermediate switches which are experiencing congestion to indicate to the source that it should reduce its flow rate. UBR is the simplest technique ofall to describe: there are no guarantees at UBR