Chapter3SwitchTechnologyFundamentalsRequiredreading:Garcia1.14.14.47.1&7.33.1CircuitSwitching3.2MessageSwitching3.3PacketSwitching

3.3.1virtualcircuit 3.3.2datagram3.4TheDelayAnalysisinNetworks

TheNetworkCoremeshofinterconnectedroutersthefundamentalquestion:howisdatatransferredthroughnet?circuitswitching:dedicatedcircuitpercall:telephonenetpacket-switching:datasentthrunetindiscrete“chunks”VirtualCircuitDatagram3.1CircuitSwitchingEnd-endresourcesreservedfor“call”●linkbandwidth,switchcapacity●dedicatedresources:nosharing●circuit-like(guaranteed)performance●callsetuprequiredCircuitSwitchingnetworkresources(e.g.,bandwidth)dividedinto“pieces”●piecesallocatedtocalls●resourcepieceidleifnotusedbyowningcall(nosharing)●dividinglinkbandwidthinto“pieces”frequencydivisiontimedivisionCircuitSwitching:FDMandTDMFDMfrequencytimeTDMfrequencytime4usersExample:TiminginCircuitSwitchingDATACircuit

EstablishmentData

Transmission

Circuit

TerminationHost1Host2Node1Node2propagationdelaybetweenHost1andNode1propagationdelaybetweenHost2andhost1processingdelayatNode1CircuitSwitchingAnode(switch)inacircuitswitchingnetworkincominglinksoutgoinglinksNodeCircuitSwitching:Multiplexing/DemultiplexingTimedividedinframesandframesdividedinslotsRelativeslotpositioninsideaframedetermineswhichconversationthedatabelongstoNeedssynchronizationbetweensenderandreceiverIncaseofnon-permanentconversationsneedstodynamicallybindaslottoaconservationhowtodothis?Ifaconversationdoesnotuseitscircuitthecapacityis

lost!3.2MessageSwitchingeachend-enddatastreamasmessagesuserA,Bmessagesharenetworkresourceseachmessageusesfulllinkbandwidthresourcesusedasneeded,Bandwidthdivisioninto“pieces”DedicatedallocationResourcereservationresourcecontention:

●aggregateresourcedemandcanexceedamountavailable●congestion:messagesqueue,waitforlinkuse●storeandforward:messagesmoveonehopatatimetransmitoverlinkwaitturnatnextlinkTiminginMessageSwitchingDATAHost1Host2Node1Node2DATADATA3.3PacketSwitchingeachend-enddatastreamdividedintopacketsuserA,Bpacketssharenetworkresourceseachpacketusesfulllinkbandwidthresourcesusedasneeded,resourcecontention:

●aggregateresourcedemandcanexceedamountavailable●congestion:packetsqueue,waitforlinkuse●storeandforward:packetsmoveonehopatatimetransmitoverlinkwaitturnatnextlinkBandwidthdivisioninto“pieces”DedicatedallocationResourcereservationPacketSwitchingAnodeinapacketswitchingnetworkincominglinksoutgoinglinksNodeMemoryPacket-switching:store-and-forwardCEFABDCsendpacketstoERouterpacket-switchednetworksPacket1Packet2Packet3Packet1Packet2Packet3TimingofDatagramPacketSwitchingPacket1Packet2Packet3processing

delayofPacket1atNode2Host1Host2Node1Node2propagationdelaybetweenHost1andNode2transmissiontimeofPacket1atHost1Packet-switching:store-and-forward●TakesL/Rsecondstotransmit(pushout)packetofLbitsontolinkorRbps●Entirepacketmustarriveatrouterbeforeitcanbetransmittedonnextlink:storeandforward●delay=3L/R(assumingzeropropagationdelay)Example:L=7.5MbitsR=1.5Mbpstransmissiondelay=15secRRRLPacketSwitching:MessageSegmentingNowbreakupthemessageinto5000packetsstoreandforwardbehaviorEachpacket1,500bits,1msectotransmitapacketononelinkpipelining:eachlinkworksinparallelDelayreducedfrom3*5=15secto5.002secPacketSwitching:Multiplexing/DemultiplexingDatafromanyconversationcanbetransmittedatanygiventimeAsingleconversationcanusetheentirelinkcapacityifitisaloneHowtotellthemapart?usemeta-data(header)todescribedataHeaderDataTrailerPacketsStructurePacketSwitching:StatisticalMultiplexingSequenceofA&Bpacketsdoesnothavefixedpattern,bandwidthsharedondemandstatisticalmultiplexing.TDM:eachhostgetssameslotinrevolvingTDMframe.ABC100Mb/sEthernet1.5Mb/sDEstatisticalmultiplexingqueueofpacketswaitingforoutputlink45MbsHowdolossanddelayoccur?packetsqueueinrouterbuffers

●packetsqueue,waitforturn

●packetarrivalratetolinkexceedsoutputlinkcapacityABpacketbeingtransmitted(delay)packetsqueueing(delay)free(available)buffers:arrivingpacketsdropped(loss)ifnofreebuffersDelayinpacket-switchednetworkspacketsexperiencedelayonend-to-endpath●foursourcesofdelayateachhopnodalprocessing:

checkbiterrorsdetermineoutputlinkqueuingtimewaitingatoutputlinkfortransmissiondependsoncongestionlevelofrouterABpropagationtransmissionnodalprocessingqueuingDelayinpacket-switchednetworksTransmissiondelay:R=linkbandwidth(bps)L=packetlength(bits)timetosendbitsintolink=L/RPropagationdelay:d=lengthofphysicallinks=propagationspeedinmedium(~2x108m/sec)propagationdelay=d/sABpropagationtransmissionnodalprocessingqueueingNote:sandRareverydifferentquantities!Nodaldelaytproc=processingdelaytypicallyafewmicrosecsorlesstqueue

=queuingdelaydependsoncongestionttrans=transmissiondelay=L/R,significantforlow-speedlinkstprop

=propagationdelayafewmicrosecstohundredsofmsecsQueueingdelay(revisited)R=linkbandwidth(bps)L=packetlength(bits)a=averagepacketarrivalratetrafficintensity=La/RLa/R~0:averagequeueingdelaysmallLa/R->1:delaysbecomelargeLa/R>1:more“work”arrivingthancanbeserviced,averagedelayinfinite!“Real”InternetdelaysandroutesWhatdo“real”Internetdelay&losslooklike?Tracerouteprogram:providesdelaymeasurementfromsourcetorouteralongend-endInternetpathtowardsdestination.Foralli:sendsthreepacketsthatwillreachrouterionpathtowardsdestinationrouteriwillreturnpacketstosendersendertimesintervalbetweentransmissionandreply.3probes3probes3probesPacketswitchingversuscircuitswitching1Mbpslinkeachuser:100Kbpswhen“active”active10%oftimecircuit-switching:10userspacketswitching:10usersorlessnoproblem>10usersqueuingdelaybutstillpossibletosupportwith35users,probability>10%activelessthan.004Nusers1MbpslinkPacketswitchingversuscircuitswitchingGreatforburstydataresourcesharingsimplercongestion:packetdelayandlossprotocolsneededforreliabledatatransfer,congestioncontrolQ:Howtoprovidecircuit-likebehavior?bandwidthguaranteesneededforaudio/videoappsVC(VirtualCircuit虛電路)所謂虛電路是面向連接的服務(wù)，在數(shù)據(jù)傳輸之前需要先建立一條邏輯鏈路；數(shù)據(jù)交換后，必須釋放這個連接。同一呼叫的數(shù)據(jù)分組只需標識邏輯電路號，并沿著虛電路按序傳輸，提供可靠的報文服務(wù)。它不同于電路交換中的物理連接，而是邏輯連接。DG(Datagram數(shù)據(jù)報)是無連接的服務(wù)，無需要預先建立連接，網(wǎng)絡(luò)隨時接受主機發(fā)送的分組；每個數(shù)據(jù)報需要標識出完整的目標地址；每個分組獨立的選擇路由，因此不能防止報文的丟失、重復或失序，它提供“盡最大努力交付”的服務(wù)，是一種不可靠的服務(wù)。分組交換提供兩種不同形式的服務(wù):虛電路和數(shù)據(jù)報。3.3.1Virtualcircuitscallsetupforeachcallbeforedatacanflow,andteardownaftereachpacketcarriesVCidentifier(notdestinationhostID)everyrouteronsource-destpathmaintains“state”foreachpassingconnectiontransport-layerconnectiononlyinvolvedtwoendsystemslink,routerresources(bandwidth,buffers)maybeallocatedtoVCtogetcircuit-likeperf.“source-to-destpathbehavesmuchliketelephonecircuit”performance-wisenetworkactionsalongsource-to-destpathVirtualcircuits:signalingprotocolsusedtosetup,maintainteardownVCusedinATM,frame-relay,X.25notusedintoday’sInternetapplicationtransportnetworkdatalinkphysicalapplicationtransportnetworkdatalinkphysical1.Initiatecall2.incomingcall3.Acceptcall4.Callconnected5.Dataflowbegins6.Receivedata子網(wǎng)采用虛電路方式，只需在建立連接時做一次路由選擇。VirtualCircuitDelayPacket1TransmissiontimeVirtualCircuit

EstablishmentPacket2Packet1Packet3Packet2Packet1Packet3Packet2Packet3VirtualCircuit

Termination3.3.2Datagramnetworks:theInternetmodelnocallsetupatnetworklayerrouters:nostateaboutend-to-endconnectionsnonetwork-levelconceptof“connection”packetsforwardedusingdestinationhostaddresspacketsbetweensamesource-destpairmaytakedifferentpathsapplicationtransportnetworkdatalinkphysicalapplicationtransportnetworkdatalinkphysical1.Senddata2.Receivedata子網(wǎng)采用數(shù)據(jù)報方式，每個包都要做路由選擇DatagramnetworksH1H5H6H4H2H3H1send

packets

toH5H2sendpacketstoH6RouterABCDEpacket-switchednetworksThereisnoroundtriptimedelaywaitingforconnectionsetup;ahostcansenddataassoonasitisready.Sourcehosthasnowayofknowingifthenetworkiscapableofdeliveringapacketorifthedestinationhostisevenup.Sincepacketsaretreatedindependently,itispossibletoroutearoundlinkandnodefailures.Sinceeverypacketmustcarrythefulladdressofthedestination,theoverheadperpacketishigherthanfortheconnection-orientedmodel.Packet1Packet2Packet3Packet1Packet2Packet3DatagramDelayPacket1Packet2Packet3processing

delayofPacket1atNode2Host1Host2Node1Node2propagationdelaybetweenHost1andNode2transmissiontimeofPacket1atHost1DatagramorVCnetwork:why?Internet*dataexchangeamongcomputers“elastic”service,nostricttimingreq.*“smart”endsystems(computers)canadapt,performcontrol,errorrecoverysimpleinsidenetwork,complexityat“edge”*manylinktypesdifferentcharacteristicsuniformservicedifficultATMevolvedfromtelephonyhumanconversation:stricttiming,reliabilityrequirementsneedforguaranteedservice“dumb”endsystemstelephonescomplexityinsidenetworkVirtualCircuitPK.Datagram(1)分組頭：DG方式的每個分組頭中要包含詳細的目的地址VC方式由于預先已建立邏輯連接，分組頭中只要含有對應(yīng)于所建立的VC的邏輯信道標識選路：VC方式預先有建立過程，但一旦虛電路建立，在端到端之間所選定的路由上的各個交換節(jié)點都具有映象表，存放出入邏輯信道的對應(yīng)關(guān)系，每個分組到來時只要查找映象表，而不需要進行復雜的選路。DG方式則不需要有建立過程，但對每個分組都要獨立地進行選路。分組順序：VC方式中，屬于同一呼叫的各個分組在同一條虛電路上傳送，分組會按原有順序到達終點，不會產(chǎn)生失序現(xiàn)象。DG方式中，各個分組由于是獨立選路，可以從不同的路由轉(zhuǎn)送，會引起失序VirtualCircuitPK.Datagram(2)故障敏感性VC方式對故障較為敏感，當傳輸鏈路或交換節(jié)點發(fā)生故障時可能引起虛電路的中斷，需要重新建立。（有些分組網(wǎng)具有再連接功能，出現(xiàn)故障時可自動建立新的虛電路，并做到不丟失用戶數(shù)據(jù)）DG方式中各個分組可選擇不同路由，對故障的防衛(wèi)能力較強，從而可靠性較高。應(yīng)用VC方式適用于較連續(xù)的數(shù)據(jù)流傳送，其持續(xù)時間應(yīng)顯著地大于呼叫建立的時間，如文件傳送、傳真業(yè)務(wù)等。DG方式則適用于面向事務(wù)的詢問／響應(yīng)型數(shù)據(jù)業(yè)務(wù)。

DelayComparison采用存儲轉(zhuǎn)發(fā)方式的分組交換兼有電路交換和報文交換的優(yōu)點，它與報文交換的不同在于：分組交換將用戶要傳送的信息分割為若干個分組（packet），每個分組中有一個分組頭，含有可供選路的信息和其他控制信息。3.4ThedelayanalysisinNetworksDelay

inCircuitSwitchingAssume:Numberofhops=MPer-hopprocessingdelay=PLinkpropagationdelay=LTransmissionspeed=Wbit/sMessagesize=BbitsTotalDelay=totalpropagation

+totaltransmission

+totalprocessing=4ML+B/W+(M-1)PPLB/WTotalDelayDelayginDatagramPacketSwitchingAssume:Numberofhops=MPer-hopprocessingdelay=PLinkpropagationdelay=LPackettransmissiondelay=TMessagesize=NpacketsTotalDelay=totalpropagation

+totaltransmission

+

totalstore&forward+totalprocessing=ML+NT+(M-1)T+(M-1)PPTLTotalDelayPTDelayVirt.Circ.PacketSwitchingAssume:Numberofhops=MPer-hopprocessingdelay=PLinkpropagationdelay=LPackettransmissiondelay=TMessagesize=NpacketsTotalDelay=totalpropagation

+totaltransmission

+

totalstore&forward+totalprocessing=4ML+NT+(M