無線傳感器網(wǎng)絡(luò)第五講

無線傳感器網(wǎng)絡(luò)第五講第一頁，共102頁。第五講Transportlayer第二頁，共102頁。TRANSPORTLAYER

(PRIORKNOWLEDGE)ENDTOENDRELIABILITYCONGESTIONCONTROLCongestiondetectionFairnessissuesTCP(TransmissionControlProtocol)forDataTrafficUDP(UserDatagramProtocol)forRealTimeTraffic第三頁，共102頁。TransportLayerInternet,Satellite,etcSinkSinkUserEnd-to-endcommunicationbetweenasensornodeanduserEndtoendreliableeventtransfer第四頁，共102頁。ChallengesEnd-to-EndMeasuresApplication-DependentOperationEnergyConsumptionBiasedImplementationConstrainedRouting/Addressing第五頁，共102頁。End-to-EndMeasuresTCP:End-to-endretransmission-basederrorcontrolWindow-basedadditive-increasemultiplicative-decrease(AIMD)congestioncontrolmechanisms.End-to-endandpoint-to-pointreliabilityandcongestioncontrolsolutionsResourcewastageinWSNs第六頁，共102頁。ThepacketlossesandcongestionmitigationareperformedthroughcommunicationbetweenasourceandadestinationwithoutanyinvolvementfromtheintermediatepartiesThetransportcontrolmechanismsresideonlyonthesourceanddestination.Eachflowisconsideredindependentlytoprovideapoint-to-pointcommunicationsolution.第七頁，共102頁。Collectiveinformationfromagroupofsensorsismuchmoreimportantthantheindividualinformationfromeachsensornode.Conventionalend-to-end,point-to-pointtransportlayertechniquesmayleadtowasteofscarcewirelesssensorresources.LocalmeasuresforreliabilityandcongestioncontrolareusuallyexploitedtoimprovetheenergyefficiencyofthetransportlayerprotocolsReliabilityofthecollectiveinformationfromagroupofsensorsiscontrolledinsteadofthereliabilityofinformationfromeachindividualsensornode.第八頁，共102頁。RelatedWorkWirelessTCPvariantsareNOTsuitableforsensornetworksDifferentnotionofend-to-endreliabilityHugebufferingrequirementsACKingisenergydrainingBOTTOMLINE:Traditionalend-to-endguaranteedreliability(TCPsolutions)cannotbeappliedhere.NewReliabilityNotionisrequired!!!第九頁，共102頁。CongestionControlACKfeedbackClosed-loopcontrolFeedbackshouldbefrequent,butnottoomuchotherwisetherewillbeoscillationsCannotcontrolthebehaviorwithatimegranularitylessthanthefeedbackperiod第十頁，共102頁。TCPcongestioncontrolcwndgrowsexponentially(slowstart),thenlinearly(congestionavoidance)with1moresegmentperRTTIfloss,dividesthresholdby2(multiplicativedecrease)andrestartwithcwnd=1packetTimeSequenceNoCongestionwindowdoubleseveryround-triptimepacketack第十一頁，共102頁。ChallengesEnd-to-EndMeasuresApplication-DependentOperationEnergyConsumptionBiasedImplementationConstrainedRouting/Addressing第十二頁，共102頁。Application-DependentOperationDeployedwithaspecificsensingapplicationobjectiveMonitoringapplication:ReliabilityEventdetectionapplications:TimelinessTransportlayersolutions----tailoredtotheapplication.Theimportanceofthesemetricscanalsovaryaccordingtotheapplicationarea.ThespecificobjectiveofWSNsinfluencesthedesignrequirementsofthetransportlayerprotocols.第十三頁，共102頁。ChallengesEnd-to-EndMeasuresApplication-DependentOperationEnergyConsumptionBiasedImplementationConstrainedRouting/Addressing第十四頁，共102頁。EnergyConsumptionScarceenergyresourcesaffectthedesignoftransportlayersolutions.Energyawaretransportlayerfunctionalities:reliabilityandenergypower---end-to-endmeasuresnotbescalableforWSNs.Tradedoffbetweenreliabilityandenergyconsumption----localreliabilitymeasures第十五頁，共102頁。ChallengesEnd-to-EndMeasuresApplication-DependentOperationEnergyConsumptionBiasedImplementationConstrainedRouting/Addressing第十六頁，共102頁。BiasedImplementationResource-constrainedsensornodesandresource-richsink:MostofthefunctionalitiesareperformedatthesinkMinimumfunctionalitiesarerequiredatthesensornodes.Flowdirectiondependingtraffic:sensors-to-sinkdirectiontrafficrequiretimelydeliverywithloss-tolerantoperationSink-to-sensorsdirectionrequiresahighdeliveryratio.第十七頁，共102頁。ChallengesEnd-to-EndMeasuresApplication-DependentOperationEnergyConsumptionBiasedImplementationConstrainedRouting/Addressing第十八頁，共102頁。ConstrainedRouting/AddressingNotemployend-to-endglobaladdressingAttribute-basednaminganddata-centricrouting第十九頁，共102頁。Transportlayerprotocols第二十頁，共102頁。TransportlayerprotocolsReliableMulti-SegmentTransport(RMST)ProtocolPumpSlowly,FetchQuickly(PSFQ)ProtocolCongestionDetectionandAvoidance(CODA)ProtocolEvent-to-SinkReliableTransport(ESRT)ProtocolGARUDA第二十一頁，共102頁。RMSTRMSTprovidesend-to-enddata-packettransferreliabilityRMSTisatransportlayerprotocolfordirecteddiffusionBedesignedasafilterthatcouldbeattachedtothedirecteddiffusionprotocol.RMSTisaselectiveNACK-basedprotocolthatcanbeconfiguredforin-networkcachingandrepair.TherearetwomodesforRMST:CachingModeNon-CachingMode第二十二頁，共102頁。PacketerrorsdetectionEachpacketofaflowislabeledbyauniquesequencenumber.Thepacketerrorsaredetectedwheneverthereisaholeinthesequencenumbersreceived.Incaseofpacketerrors,nodesrequestretransmissionbysendingaNACKpackettowardthereverseroutefromthesinktothesensor.第二十三頁，共102頁。ReliableMulti-SegmentTransport(RMST)SinkRMSTNodeSourceNode

EachRMSTnodecachesthefragmentsidentifiedbyFragNoofaflowidentifiedbyRmstNo.Watchdogtimersaremaintainedforeachflow.Whenafragmentisnotreceivedbeforethetimerexpires,anegativeacknowledgementissentbackwardinthereinforcedpath.ThefirstRMSTnodethathastherequiredfragmentalongthepathretransmitsthefragment.SinkactsasthelastRMSTnode.Innon-cachingmode,sinkistheonlyRMSTnode.RMSTreliesondirecteddiffusionschemeforrecoveryfromthefailedreinforcedpaths.第二十四頁，共102頁。NON-CACHINGMODESimilartoconventionaltransportlayerprotocolsOnlythesourceanddestinationplayaroleinprovidingreliabilityThepacketlossesaredetectedatthesinkandrequestedfromthesourcenodeinanend-to-endfashionthroughaNACKpacket.Advantage:additionalprocessing,storage,andenergyconsumption–fromtheintermediatenodesinthemulti-hopnetworkarenotrequired.第二十五頁，共102頁。errorSinkdetectpacket4lossNACKRe-transmission第二十六頁，共102頁。CACHINGMODEAnumberofnodesalongareinforcedpath,(pathbeingusedtoconveythedatatothesinkbydirecteddiffusion),areassignedasRMSTnodes.第二十七頁，共102頁。errorCachenodedetectpacket3lossNACKtoupstreamcacheRe-transmission第二十八頁，共102頁。QualitativeEvaluationAdvantages

Thecostassociatedwithend-to-endretransmissionsisminimizedCreatereliablesegmentsbetweentwoconsecutivecachingnodesandtheretransmissionsareperformedinsidethesesegmentsinsteadofthroughtheend-to-endroute.RMSTaimstoprovideguaranteeddeliveryforeachflowintheWSN——helpfulforapplicationswhereindividualnodeinformationisimportant第二十九頁，共102頁。QualitativeEvaluationDisadvantagesAdditionaloverhead:increasetheoverallcomplexityandenergyconsumptionofthenetwork.Treatseachflowseparately:overutilizationofresourcesGuaranteedreliabilityviain-networkcaching:bringsignificantoverheadAlargeamountofinformationmayflowinsidethenetwork:congestionandassociatedpacketdrops第三十頁，共102頁。TransportlayerprotocolsReliableMulti-SegmentTransport(RMST)ProtocolPumpSlowly,FetchQuickly(PSFQ)ProtocolCongestionDetectionandAvoidance(CODA)ProtocolEvent-to-SinkReliableTransport(ESRT)ProtocolGARUDA第三十一頁，共102頁。PumpSlowly,FetchQuickly(PSFQ)pacethedatafromasourcenodeatarelativelylowspeedtoallowintermediatenodestofetchmissingdatasegmentsfromtheirneighbors,Assumption:nocongestion,lossesdueonlytopoorlinkqualityhop-by-hoprecoveryGoalsRecoverfromlosseslocally.EnsuredatadeliverywithminimumsupportfromtransportinfrastructureMinimizesignalingoverheadfordetection/recoveryoperationsOperatecorrectlyinpoorlinkqualityenvironmentsProvideloosedelayboundsfordatadeliverytoallintendedreceiversThreebasicoperations:

pump,fetch,andreportAlternatebetweenmulti-hopforwardingwhenlowerrorratesandstore-and-forwardwhenerrorratesarehigher.第三十二頁，共102頁。PSFQAddressthepathfromsinktosensors.Differenttosensors-to-sinkpathNetworkmanagementtasksRe-taskingofthesensornodes.ConcernreliabilityOne-to-onecommunication:reliablydisseminatecontrolinformationissentbythesinktoeachsensor.第三十三頁，共102頁。C.Y.Wan,A.T.CampbellandL.Krishnamurthy,“PSFQ:AReliableTransportProtocolforWirelessSensorNetworks,”InProc.ACMWSNA’02,September2002,Atlanta,GAPSFQSlowinjectionofpacketsintothenetworkAggressivehop-by-hoprecoveryincaseofpacketlosses“PUMP”performscontrolledfloodingandrequireseachintermediatenodetocreateandmaintainadatacachetobeusedforlocallossrecoveryandin-sequencedatadelivery.Applicableonlytostrictsensor-sensorguaranteeddeliveryAndforcontrolandmanagementend-to-endreliabilityforthedownlinkfromsinktosensorsDoesnotaddresscongestioncontrol第三十四頁，共102頁。PSFQ

PSFQcomprisesthreefunctions:*MessageRelaying(PUMPoperation),*Relayinitiatederrorrecovery(FETCHoperation)and*Selectivestatusreporting(REPORToperation).第三十五頁，共102頁。PSFQEveryintermediatenodemaintainsadatacache.Anodethatreceivesapacketchecksitscontentagainstitslocalcache,anddiscardsanyduplicates.Ifthereceivedpacketisnew,theTTLfieldinthepacketisdecremented.IftheTTLfieldishigherthan0afterbeingdecremented,andthereisnogapinthepacketsequencenumbers,thepacketisscheduledtobeforwarded.ThepacketsaredelayedforarandomperiodbetweenTminandTmax,andthenrelayed.AnodegoestoFETCHmodeonceasequencenumbergapisdetected.ThenodeinFETCHmoderequestsaretransmissionfromneighboringnodes.第三十六頁，共102頁。PumpReliabilityismoreimportantthantimelinessPSFQemploysaslowpumpmechanismSlowinjectionofpacketsintothenetworkEachnodeontheroutetothedestinationswaitsforaspecificamountoftimebeforerelayingthemessages.第三十七頁，共102頁。PumpNodebroadcastsapackettoitsneighborseveryTminuntilallthedatafragmentshavebeensentoutNeighborswhoreceivethepacketcheckagainsttheirlocalcachediscardinganyduplicatesIfitisjustanewmessagethepacketisbufferedandtheTime-To-Live(TTL)fieldintheheaderisdecreasedby1IfTTLisnotzeroandthereisnogapinthesequencenumberthepacketthenisscheduledfortransmissionwithinarandomtimeTtx,whereTherandomdelaybeforeforwardingthemessageallowsadownstreamnodetorecovermissingsegmentsbeforethenextsegmentarrivesfromanupstreamnodeItalsoallowsreducingthenumberofredundantbroadcastsofthesamepacketbyneighbors第三十八頁，共102頁。PSFQ

PUMPOPERATIONIfnotduplicateandin-orderandTTLnot0CacheandScheduleforForwardingattimet(Tmin<t<Tmax)TminTmaxTminTmax111t12第三十九頁，共102頁。FETCHIncaseofpacketerrors,eachnodeperformsaggressivehop-by-hoprecoverytofetchthelostpacketsfromneighbornodes.Ifanodeonthepathfromthesinktosensorsdetectsasequencenumbergapinthissequenceofpackets,thefetchoperationisinitiated.Throughthefetchoperation,anodeaggressivelysendsoutNACKpacketstoquicklyrecoverthelosspacketsfromitsimmediateneighbors.第四十頁，共102頁。PSFQ

FETCHMODE(RecoveryfromErrors)

2lostRecover212321234112233第四十一頁，共102頁。PSFQ

FetchQuickly112lost23TminTmaxTrRecover2Tr2212第四十二頁，共102頁。FETCHLocalerrorrecovery:persistentNACKmessagesbetweentwopackettransmissions.Preventmessageimplosion:limitsNACKmessagetransmissiontotheone-hopneighborhoodThemechanismisnotsuitableforcaseswherethelastpacketofaflowislostorallthepacketsrelatedtoaflowarelost.PSFQemploysaproactivefetchoperation,wherethereceiversfollowatimer-basedfetchoperation.第四十三頁，共102頁。ProactivefetchAnodedoesnotreceiveapacketforTproSendaNACKpackettoneighbors.Tpro

=α(Smax?Slast)Slast：thelasthighestsequencenumberSmax:thelargestsequencenumber

α≥1ThenodeproactivelysendsaNACKpacketearlierifitisclosertotheendofthemessage.InthecasewherebufferlengthislimitedTpro

=αnTmaxnisthebufferlength.第四十四頁，共102頁。ReportReportingfunctionality---createsclosed-loopcommunicationbetweensensorsandsink.AllowthesinktorequestfeedbackfromthesensornodesinascalablemannerThesinkcancollectinformationrelatedtooperationofthenetwork.Initiatebythesinkbysettingareportbitinthepacketheader.Uponreceivingthereportrequest,asensornoderespondsimmediatelybytransmittingareportmessage.第四十五頁，共102頁。ReportEachnodealongthepathtothesinkaddsitsstatusinformationtothisreportbypiggybacking

IfanodeintheupstreampathdoesnotreceivethereportresponseinTreport,itcreatesitsownreportpacketandsendsittothesink.Thereportoperationisalsousedforsinglepacketdelivery.Ifamessagethatthesinkwillsendcanfitintoonepacket,thenthereportoperationisinitiatedbysettingthereportbit.Whenevertheintendedsensornodereceivesthispacket,itrespondswithareportpacket.Anend-to-enderrorcontrolmechanismisintegratedintoPSFQforsinglepacketflows.第四十六頁，共102頁。PSFQ-ConclusionsLightweightandenergyefficientSimplemechanismScalableandrobustNeedtobetestedforhighbandwidthapplicationsCachesizelimitationDoesnotaddresscongestioncontrol第四十七頁，共102頁。TransportlayerprotocolsReliableMulti-SegmentTransport(RMST)ProtocolPumpSlowly,FetchQuickly(PSFQ)ProtocolCongestionDetectionandAvoidance(CODA)ProtocolEvent-to-SinkReliableTransport(ESRT)ProtocolGARUDA第四十八頁，共102頁。CODAAimtodetectandavoidcongestionThreemaincongestionscenariosWirelesschannelcontentionHotspotscongestionPersistenthotspotslocalcongestioncontrolmechanismsend-to-endmechanisms第四十九頁，共102頁。CODAEnergyefficientcongestioncontrolschemeThreemechanismsReceiver-basedcongestiondetectionOpen-loophop-by-hopbackpressuresignalingtoinformthesourceaboutthecongestionClosed-loopmulti-sourceregulationforpersistentandlargerscalecongestionconditions.CODAusesacombinationofbufferoccupancylevelandchannelloadconditionasanindicationofcongestion.第五十頁，共102頁。CongestionDetectionandAvoidance(CODA)

CODAmainlyaimstodetectandavoidsCONGESTIONontheforwardpathviareceiver-basedcongestiondetection,open-loophop-by-hopbackpressuresignalingtoinformthesourceaboutthecongestion,closed-loopmulti-sourceregulationforpersistentandlarger-scalecongestionconditions.SimulationresultsshowthatCODAcanincreasethenetworkperformancebycongestionavoidance.CODAprotocoldoesnotaddressthereliableeventtransportinthesensornetworks.Onthecontrary,ithasbeenobservedintheexperimentsthatthecongestioncontrolperformedatthesensornodeswithoutconsideringthereliabilityimpairstheend-to-endreliability.第五十一頁，共102頁。ReceiverbasedcongestiondetectionmechanismCongestionusuallyoccursinreceivernodes.Themechanismreliesonbothbufferoccupancyandchannelload.Thechannelloadisestimatedbylisteningtothechannelwheneveranodehasapackettosend.CODAdecidesoncongestionifthechannelloadishigherthanafractionofthemaximumchannelutilization.Consequently,congestionisdetectedatthereceivers.第五十二頁，共102頁。congestionSuppressionmessagetransmissionCongestionmitigationand(optional)re-routing

open-loophop-by-hopbackpressure第五十三頁，共102頁。CongestionDetection AccurateandefficientcongestiondetectionisimportantBufferqueuelengthorBufferoccupancy–notagoodmeasureofthecongestion.Channelloading–samplechannelatappropriatetimetodetectcongestion.Reportrate/Fidelitymeasurement–slow,observedoveralongerperiodCODA第五十四頁，共102頁。Open-LoopHop-by-HopBackpressureCODACongestiondetected123452第五十五頁，共102頁。ClosedLoopMulti-SourceRegulation1,2,3ACK4,5,6Congestiondetected7,8RegulatebitissetACK12CODA第五十六頁，共102頁。CODAPerformance–CostMetricsAverageEnergyTax=

TotalPacketsdroppedinsensorNW/TotalPacketsreceivedatSinkAverageFidelityPenalty=MeasuresdifferencebetweenaveragenumberofpacketsdeliveredatasinkusingCODAandusingidealcongestionscheme

第五十七頁，共102頁。QualitativeEvaluationAcompletecongestioncontrolmechanismfordownstreamtrafficefficientcongestiondetection:acombinationofbufferoccupancylevelandchannelloadconditiontoprovideAddressbothlocalandend-to-endcongestionDonotaddressreliabilityOnlyprovidescongestionavoidanceanddoesnotprovideareliableoperation.AdditionallatencyTheclosed-loopmulti-sourceregulationmechanism第五十八頁，共102頁。TransportlayerprotocolsReliableMulti-SegmentTransport(RMST)ProtocolPumpSlowly,FetchQuickly(PSFQ)ProtocolCongestionDetectionandAvoidance(CODA)Protocol

Event-to-SinkReliableTransport(ESRT)ProtocolGARUDA第五十九頁，共102頁。Event-to-SinkReliableTransport(ESRT)InatypicalsensornetworkapplicationthesinknodeisonlyinterestedinthecollectiveinformationofthesensornodeswithintheregionofaneventandnotinanyindividualsensordataWhatisneededisameasureoftheaccuracyoftheinformationreceivedatthesink,i.e.andevent-to-sinkreliability第六十頁，共102頁。ThebasicassumptionisthatthesinkdoesallthereliabilityevaluationusingparametersthatareapplicationdependentOnesuchparameteristhedecisiontimeintervalτAttheendofthedecisionintervalthesinkderivesareliabilityindicatorri

basedonthereportsreceivedfromthesensornodesriisthenumberofpacketsreceivedinthedecisionintervalIfRisthenumberofpacketsrequiredforreliableeventdetectionthen

ri>RisneededforreliableeventdetectionThereisnoneedtoidentifyindividualsensornodesbutinsteadthereistheneedtohaveaneventIDThereportingrate,f,ofasensornodeisthenumberofpacketssentoutperunittimebythatnodeTheESRTprotocolaimstodynamicallyadjustthereportingratetoachievetherequireddetectionreliabilityRatthesinkESRT第六十一頁，共102頁。ESRTEvent-to-sinkreliabilitynotionProvidereliableeventdetectionwithoutanyintermediatecachingrequirements.Aimtoaddressboththereliabilityandcongestionevent-to-sink—collectionof“flows”fromagroupofsensorsassociatedwithasingleevent.Providereliabilityoftheeventinformationflowthroughcongestioncontrolmechanisms.Algorithmsrunonthesink第六十二頁，共102頁。EventRadiusSinkSensornodesESRT

O.B.Akan,I.F.AkyildizandY.Sankarasubramaniam,

toappearinIEEETransactionsonNetworking,Fall2004.

AlsoinProc.ofACMMobiHoc’03,Annapolis,Maryland,June2003.Sensornetworksareevent-drivenMultiplecorrelateddataflowsfromeventtosinkGOAL:Toreliablydetect/estimateeventfeaturesbasedonthecollectivereportsofseveralsensornodesobservingtheevent.Event-to-sinkcollectivereliabilitynotion

第六十三頁，共102頁。Event-to-SinkReliableTransport(ESRT)ESRTisthefirstschemethatfocusesontheend-to-endreliableeventtransfer.Theend-to-endeventtransferreliabilityiscontrolledbasedonthereportingfrequenciesofsensornodes.Sinkabcd第六十四頁，共102頁。End-to-endReliableEventTransferSinkrabcdreventregionsensorcoveragesensorrange第六十五頁，共102頁。Event-to-SinkReliabilitySinkdecidesabouteventfeatureseveryτtimeunits(decisionintervals)DEFINITION1:ObservedEventReliabilityriisthenumberofdatapacketsreceivedindecisionintervali

atsinkDEFINITION2:DesiredEventReliability

Risthenumberofpacketsrequiredforreliableeventdetection(applicationspecificandisknowna-prioriatthesink)

(Ifri>R,thentheeventisreliablydetected.Else,appropriateactionsmustbetakentoachieveR.)DEFINITION3:ReportingRate

fisthefrequencyofpackettransmissionsatasourcenode第六十六頁，共102頁。TRANSPORTPROBLEMINSENSORNETWORKSToconfigurethereportingrate,f,ofsourcenodessoastoachievetherequiredeventdetectionreliability,R,atthesinkwithminimumresourceutilization.第六十七頁，共102頁。rvsfrelationshiprshowsinitiallinearincreasewithfuntilf=fmax

Forf>fmax,rdropsduetocongestionbecausethenetworkisunabletohandletheincreasedinjectionofdatapacketsThisbehaviorisindependentofthenumberofnodesn

fmaxdecreaseswithincreasingn

(congestionoccursatlowerreportingfrequencieswithgreaternumberofsourcenodesn)第六十八頁，共102頁。ESRT:Event-to-SinkReliableTransportOBJECTIVE:Achievereliableeventdetectionwithminimumenergyexpenditureandcongestionresolution.SALIENTFEATURES:Self-configuration–Adaptstorandom,dynamicnetworktopologyCollectiveidentification–DoesnotrequireindividualnodeIDsBiasedimplementation–GracefultransferofcomplexitytothesinkSensornodesneedonlytwoadditionalfunctionsImplementacongestiondetectionmechanismListentosinkbroadcastsforfrequencyupdates第六十九頁，共102頁。ESRT:ProtocolOverview

DeterminereportingfrequencyftoachievedesiredreliabilityRwithminimumresourceutilizationSource(Sensornodes):Senddatawithreportingfrequencyf

MonitorbufferlevelandnotifycongestiontothesinkSink:Measurestheobservedeventreliability

ri

attheendofdecisionintervali

Normalizedreliabilityηi=

ri/R

Performscongestiondecisionbasedonthefeedbackfromthesourcesnodes(todeterminef><fmax).

Updatefbasedonηiandf><fmax

(congestion)toachievedesiredeventreliability

R第七十頁，共102頁。StateDescriptionCondition(NC,LR)(Nocongestion,Lowreliability)

f<fmaxandηi

<1-ε(NC,HR)(Nocongestion,Highreliability)f≤fmaxandηi

>1+ε(C,HR)(Congestion,Highreliability)f<fmaxandηi

>1(C,LR)(Congestion,Lowreliability)f<fmaxandηi≤1OOROptimalOperatingRegionf<fmaxand

ηi

E

[1-ε,1+ε]ESRT:NetworkStates第七十一頁，共102頁。ESRT:

CongestionDetectionMechanismACK/NACKnotsuitableWeuselocalbufferlevelmonitoringinsensornodes

MarkCongestionNotification(CN)fieldinpacketifcongested,i.e.,

bk+Db>B(thenodeinfersthatitwillexperiencecongestioninthenextreportinginterval)bkbk-1DbB

fEventIDCN(1bit)DestinationTimeStampPayloadFECbk:BufferfullnesslevelattheendofreportingintervalkDb:BufferlengthincrementB:Buffersizef:reportingfrequency第七十二頁，共102頁。StateFrequencyUpdateComments(NC,LR)

fi+1=fi/ηiiMultiplicativeincrease,achievedesiredreliabilityasap(NC,HR)fi+1=fi(ηii+1)/2ηiiConservativedecrease,nocompromiseonreliability(C,HR)fi+1=fi/ηiiAggressivedecreasetostate(NC,HR)(C,LR)

fi+1=fiηiiExponentialdecrease,relievecongestionasapOORfi+1=fiUnchangedESRT:FrequencyUpdate第七十三頁，共102頁。S0=(NC,LR)S0=(NC,HR)ESRTPerformance第七十四頁，共102頁。S0=(C,HR)S0=(C,LR)ESRTPerformance第七十五頁，共102頁。ConclusionsSensornetworkparadigmnecessitatesthenotionofevent-to-sinkreliabilityExistingend-to-endguaranteedreliabilitysolutionsleadtoover-utilizationofscarcesensorresourcesESRTisanovelsolutionproposeexclusivelyforreliableeventtransportinsensornetworksTailoredforsensorenvironmentsBiasedimplementationEnergyconservationCollectiveidentification,self-configurationESRTcanalsoaddressconcurrentmultipleevents第七十六頁，共102頁。OpenResearchIssuesExtendESRTtoaddressreliabletransportofconcurrentmultipleeventsinthesensorfield.ExplorepossibleotherreliabilitymetricsTotalexpectedmeansquaredistortionMinimummeansquarederrorestimationDevelopunifiedtransportlayerprotocolsforsink-to-sensorsandbi-directionalreliabletransportinWSNResearchtointegrateWSNdomainintoNGWI(NextGenerationWirelessInternet)AdaptiveTransportProtocolsforWSN-AdHocenvironments第七十七頁，共102頁。TransportlayerprotocolsReliableMulti-SegmentTransport(RMST)ProtocolPumpSlowly,FetchQuickly(PSFQ)ProtocolCongestionDetectionandAvoidance(CODA)ProtocolEvent-to-SinkReliableTransport(ESRT)ProtocolGARUDA

第七十八頁，共102頁。GARUDAGARUDAisalargemythicalbirdorbird-likecreaturethatappearsinbothHinduandBuddhistmythologyTransportreliably第七十九頁，共102頁。GARUDAGARUDAalsosupportsotherreliabilitysemanticsthatmightberequiredforsink-to-sensorscommunicationsuchasreliabledeliverytoallnodeswithinasub-regionofthesensornetwork;reliabledeliverytominimalnumberofsensorsrequiredtocoverentiresensingarea;andreliabledeliverytoaprobabilisticsubsetofthesensornodesinthenetwork.第八十頁，共102頁。GARUDA第八十一頁，共102頁。IntroductionReliabledownstreampoint-to-multipointdatadeliveryTheneedforthereliabilityisdependentonthetypeofapplications.Ex)securityapplicationReliabilityinmultihopwirelessnetworksvsReliabilityinwirelesssensornetworksEnvironment

considerationsLimitedlifetime,bandwidth,energy,sizeofthenetworkMessageconsiderationsInasensornetworks,small-sizedqueriesReliabilityconsiderationsDependentonreliabilitysemantics第八十二頁，共102頁。CharacteristicsofGARUDAAnefficientpulsing-basedsolutionforreliableshortmessagedeliveryAvirtualinfrastructurecalledthecore,whichapproximatesanoptimalassignmentoflocaldesignatedserversAtwo-stagenegativeacknowledgment(NACK)basedrecoveryprocess

andout-of-sequenceforwardingAsimplecandidacybasedsolutiontosupportthedifferentnotionsofreliability第八十三頁，