英文资料

www.elsevier.com/locate/comnet

PerformanceevaluationofSIP-basedmultimediaservices

inUMTS

DirkPesch

aba,*,MariaIsabelPousa,GerryFoster

bCentreforAdaptiveWirelessSystems,CorkInstituteofTechnology,Bishopstown,Cork,Ireland

SystemsEngineeringAnalysisGroup,MotorolaGlobalSystemsSolutionSector,SwindonSN58YQ,UnitedKingdom

Availableonline4June2005

Abstract

WithaneverincreasingpenetrationofIPtechnologiesandthetremendousgrowthinwirelessdatatraffic,thewire-lessindustryisevolvingthemobilecorenetworkstowardsIPtechnology.ThethirdGenerationPartnershipProject(3GPP)hasspecifiedanIPmultimediasub-system(IMS)inUMTSRelease5/6,whichisadjuncttotheUMTSpacket-switched(PS)GPRScorenetwork.ThisIP-basednetworkprovidesfullpacketcallcontrolcapabilitiesbyusingthetext-basedSessionInitiationProtocol(SIP).Initialindications,astothesignallingdelayassociatedwithSIPmes-sages,haveconcernedmobileoperatorsabouttheviabilityofSIPservicesovertheUMTSairinterface.ThisarticleprovidesaninsightintotheUMTSsystemperformance,focusingonselectedUMTSSIP-basedservices.Typicalserviceswithreal-timerequirementssuchasvoiceaswellasdelay-sensitiveandnon-sensitiveapplications,suchasreal-timechatandinstantmessagingservicesareinvestigated.Furthermore,thepaperdiscussesandanalysestherequirementsandpossiblesolutionsforimprovingefficiencyofSIPusageinawirelessenvironmentthroughsignallingprotocolmessagecompression.ResultsofaperformanceevaluationofSIPsignallingscenariosarepresentedintermsoftimedelayandmessageoverloadinthesystem.ResultsshowthatmessagecompressioncanconsiderablyreduceSIPmessagetransmissiontimeontheradioaccessnetworkwhilecorenetworkdelaycontributionsarefoundtobestillhigh.

Ó2005ElsevierB.V.Allrightsreserved.

Keywords:UMTS;SIPservices;Multimediaservices;Performanceevaluation

1.Introduction

Second-generationwirelesssystems,suchastheglobalsystemformobilecommunications(GSM),wereprimarydesignedtoprovidemobilevoiceservicestotheenduserwithadditionalshort

Correspondingauthor.Tel.:+353214326377;fax:+353214326625.

E-mailaddress:dpesch@cit.ie(D.Pesch).

*13-1286/$-seefrontmatterÓ2005ElsevierB.V.Allrightsreserved.doi:10.1016/j.comnet.2005.05.013

386D.Peschetal./ComputerNetworks49(2005)385–403

messagesandlow-rate(9.6kbps)dataservicesinasecondphase.However,withagrowingdemandformobiledataaccessandtheexplosivegrowthofInternetdataservices,wirelessdataapplicationsareseenasamajornewrevenuestream,whichhasspawnedmuchdevelopmentinitiallytowards2.5GservicessuchasGPRSandnowfocusedonthirdgenerationmobilenetworks,i.e.,theuniversalmobiletelecommunicationsystem(UMTS).

UMTSevolvesthemobilecorenetworkto-wardsanallIPtechnologywithanewradionet-workthatprovideshighercapacityanddataratesrequiredforthesupportofadvancedmulti-mediaservices.UMTSintroducesanewradioac-cessnetworkbasedonWCDMAandevolvesthecorenetworktowardsanovelpacket-switchedIP-basedtransportandserviceplatform,theIPmultimediacorenetworksub-system(IMS),whichisspecifiedinUMTSRelease5/6[1].Throughthisalltheassociatedfeaturesofacircuit-switchedcallbecomeavailabletothepacket-switcheduserintheIPdomain.

TheRelease5/6standardsemploytheIETFde-finedtext-basedSessionInitiationProtocol(SIP)[2]astheircallcontrolprotocol.Text-basedproto-colshavetheadvantagethattheyaregenerallyeasiertodevelopthanbit-wisepresentations,buttheirmessagesarelong.SIPmessageoverheadsareverylargeandthereforepotentiallyinefficientforalimitedbandwidthwirelesssystemsuchasUMTS.ThishasmotivatedtheinvestigationofSIPsignallingperformanceandpossibilitiestoim-proveSIP-basedsignalling[3].TheaimoftheinvestigationpresentedhereisthedevelopmentofaUMTSIMSsub-systemmodeldesignedandimplementedinacomputersimulationenviron-ment,compatiblewiththeproposed3GPPstan-dards.ThedevelopedmodelallowstheevaluationofdifferentSIP-basedservicesandtheircapabilitieswithinUMTS,providesforanunder-standingoftheimplicationofintroducingSIP-basedservicesina3Gwirelessnetworks,andyieldsaninsightintotheUMTSsystemperfor-mance.Previouscontributions[4–6]havepre-sentedSIPsignalling,describedSIPservicesin3Gnetworksandhavepointedatpossibleperfor-mancedifficultiesduetothelargeSIPmessagesizes.Theanalysispresentedheregoesbeyond

priorworkinthatitcontributesanewinsightintotheexpectedperformanceofSIPservicesinUMTSbeyondpreviousarticlesandshowshowSIPpro-tocolmessagecompressionhasthepotentialtosig-nificantlyimprovedelayissuesforSIPsignallinginUMTS.

Inthefollowing,webrieflydiscusstheUMTSnetworkarchitecture,thenpresentabriefoverviewofthesessioninitiationprotocolanddescribetheSIPserviceswemodelled:VoiceoverIPandin-stantmessagingwithpresencecapability.ThenwepresentaphysicalreferenceimplementationoftheIMSUMTSarchitectureandpresentthere-sultsofaperformanceevaluationofselectedSIPmultimediaservicesbasedontheUMTSmodel.

2.UMTSRelease5/6IMS

TheUniversalMobileTelecommunicationsSystem(UMTS)isathirdgenerationmobilecom-municationsystemthatprovidesarangeofbroad-bandservices.UMTSRelease99includestwodistinctcorenetworksforsupportofpacket-switched(PS)andcircuit-switched(CS)servicesasdepictedinFig.1.CallcontrolforCSservicesissimilartocallcontrolinGSM.TheUMTSTer-restrialRadioAccessNetwork(UTRAN)consistoftheNodeBandtheRNC,wheretheNodeBoperatesasthebasestationandtheRNCisthecontrollingunitforanumberofNodeBsandforaccesstotheradionetwork.TheServingandGatewayGPRSSupportNodes(SGSN,GGSN)

Fig.1.UMTSfunctionalarchitecture.

D.Peschetal./ComputerNetworks49(2005)385–403387

constitutethePSCoreNetwork.EachSGSNisconnectedtoanumberofRNCs,theGGSNisthepointofinterconnectionbetweenexternalnet-worksandUMTSandactsasthegatewayforamobileterminalsÕIPconnection.

AsUMTSevolvestoRelease6,thereisamovetowardsall-IPtechnology,abandoningtheCScorenetworkandprovidingallservicesoverIP.Forthis,thePSCNhasbeenextendedwiththeIPmultimediaCNsub-system(IMS)functionalityallowingmobileoperatorstooffertheirsubscribersmultimediaservicesbasedonInternetapplicationsandprotocols.TheIMScomprisesofallcorenet-workelements,showninFig.2,forprovisionofmultimediaservices.Thisincludesthecollectionofsignallingandbearer-relatednetworkelementsasdefinedin[7,8].TheIMSintroducesthreemainlogicalnetworkelementstotheexistinginfrastruc-ture:CallSessionControlFunction(CSCF),MediaGatewayControlFunction(MGCF),andMediaGateway(MGW).TheHomeSubscriberServer(HSS)isalsointroducedprovidinguserprofileinformationsimilartothatoftodayÕsHLR.

TheCallSessionControlFunction(CSCF)isanSIPserverthatprovides/controlsmultimediaser-vicesforpacket-switchedIPterminals,bothmobileandfixed.TheSessionControlFunctioncantakevariousrolesasdefinedin[9].Proxy-CSCFisthefirstcontactpointfortheUserEquipment(UE)withintheIMSofthelocalnetworktheUEresides

Fig.2.IMSsub-systementities.

in.ItsIPaddressisdiscoveredafteroraspartofasuccessfulPDPcontextactivation.TheP-CSCFforwardsSIPmessagesfromUEtothespecificI-CSCFortotheSIPserver(S-CSCF)actingasaSIPproxy.Serving-CSCFisalwaysassignedinasubscriberÕshomenetwork.Itactsasaregistrarmakinginformationavailablethroughthelocationserver(HSS)andsubsequentlyperformstheses-sioncontrolservicesfortheregisteredendpoint.Ithandlesthesessionstatesonthenetworkforsup-portoftheservices.Interrogating-CSCFisthecon-tactpointwithinanoperatorÕsnetworkforallIMSconnectionsdestinedtoahomeorroamingsub-scribercurrentlylocatedwithinthenetwork.TheI-CSCFalsoobtainstheS-CSCFlinkedwiththeuserintheregistrationprocedureandtheS-CSCFoftheterminatingcounterpartinthesessionestab-lishment.BreakoutGatewayControlFunctiondeterminesthenetworkinwhichthebreakoutintothePSTN/CSdomainistooccur.Oncethenet-workisselected,itselectsaMGCFwhichwillberesponsiblefortheinter-working.

TheMediaGatewayControlFunctioncontrolstheMediaGateway(MGW)andperformstransla-tionofcallcontrolsignallingbetweenPSTNandISDNtypenetworksandSIPsignalling.TheMGWterminatesbearerchannelsfromacircuit-switchednetworkandmediastreamsfromapacketnetwork.TheHSSisthemasterdatabaseforUMTSRelease5/6IPusers.Itcontainsthesubscription-relatedinformationtosupportthenetworkentitieshandlingtheIPsession.ThisentityalsointegratestheHLRfunctionalityforbothpacketandcircuitdomain,whichisthereonconsideredasaHSSsubset.

3.SIP-basedapplications

TheSessionInitiationProtocol(SIP)isanapplication-layercontrolprotocoldevelopedbytheIETFanddefinedin[2].Thistext-encodedpro-tocolwasinitiallydesignedtoestablish,modifyandterminatemultimediasessionswithoneormoreparticipantsinanIPenvironment.However,SIPÕsmobilitysupportandextensibilitymakeitidealtosupportotherservicesandapplications,suchasPresenceandInstantMessaging.SIP

388D.Peschetal./ComputerNetworks49(2005)385–403

messagesusetheSessionDescriptionProtocol[10]torequestcertaincharacteristicsofmediastreamsforsessions.

Inthispaper,wehavemodelledandsimulatedseveralSIPcall/sessionsetupscenarios,accordingtothe3GPPspecifications[11]:MobileOriginated(MO)call,destinedtoafixedphone(callee),MobileTerminating(MT)call,whereafixedphone(caller)initiatesthecalltothemobile,andmobile-to-mobile(M2M)callwithtwomobiles,locatedinthesamenetwork.3.1.Presenceoverview

SIPcapabilitieshavebeenextendedtoalsohan-dlePresenceandInstantMessaging(IM)services.Thepresenceservicedefinedin[12]byIETFiscurrentlybeingstandardisedin3GPPRelease6[13,14]forsupportinUMTS.APresenceserviceallowsausertosubscribetootherusersÕstateandbenotifiedofchangesinusersÕstate(e.g.,usergoingoff-line,changingcontactdetails,etc.).Pres-enceanditscombinationwithamessagingserviceandotherservicessuchaspush-to-talkprovidesasimpleandfastwayofcommunicationbetweenonlineusers.Presenceconveystheabilityandwill-ingnessofausertocommunicateacrossasetofdevices(presentity).Fig.3showsthepresencemodelarchitectureasdefinedinIETFRFCanddrafts[12,15].

3.2.Instantmessagingserviceoverview

Theexchangeofcontentbetweencommunicat-ingpartiesinnearrealtimeisrealisedwithinstantmessaging.Typically,thecontentofinstantmes-sagesisshorttextanditstransferistypicallyfastenoughtomaintainsomeformofinteractive

Fig.3.Presenceservice.conversation(chat).EachmessagecanbesentindependentlyusingtheSIPMESSAGEmethod,ormessagescanbeassociatedintosessionsthatareinitiatedusingSIPINVITE.Thefirstapproachisoftenreferredtoaspager-mode,duetoitssimi-laritytothebehaviouroftwo-waypagerdevices,andisusedwhenshortinstantmessagesaresenttoasingleorsmallnumberofrecipients.Incon-trast,thesecondapproach,calledsession-mode,isrequiredforextendedconversations,joiningchatgroups,etc.Bothapproaches,definedbytheIETFSIMPLEgroup,areconsideredinourmodel(seeFig.4).

MessageSessionModel.InthismodeltheIMtrafficisviewedasamediastream,whichispartofanormalSIPsession.Asthedataisalwayssentoverareliablelink,themessagesizeisnotre-stricted.Thismodeloffersadvantageswhenthenumberofmessagesprocessedincreases.OncetheinitialINVITErequesthasbeenprocessed,subsequentSIPmessagessentwithintheestab-lishedsession,bypassanyinterveningSIPproxy.Thereforethemessageloaddecreasesonthoseproxies.Themodelisusedintextconferenceandchatapplicationswhereitisusefulandmoreeffi-cienttohavemessagesassociatedwithsessions.PagingModel.Herenosetuporsessionestab-lishmentisrequiredbeforesendingamessage.ThereforeeachmessageissentindependentlyusingtheMESSAGEmethod.Itmimicstheoper-ationofSMSintodayÕswirelessnetwork.Thismethodhaslimitationsonthemessagesize(<1300bytes)duetonetworkcongestionconcerns.

Fig.4.IMtransportmodels.

D.Peschetal./ComputerNetworks49(2005)385–4033

4.SIPinUMTS

TheSessionInitiationProtocolhasbeenstan-dardisedinUMTSRelease5/6astheapplicationlayersignallingprotocolforpacket-switched(PS)sessionsbetweentheUEandtheIMS.ThenewnetworkelementsintroducedintheIMSactasSIPproxies/serversroutingandhandlingSIPmes-sagesthroughthenetwork.Thisallowsthewire-lessnetworktoprovideasimplevoicetelephonecalltoaPSuseraswellasawiderangeofIP-basedservicessuchasinstantmessagingwithpresence.SIPmessagesinUMTSaretransmittedusingUDP/TCPoverIPovertheappropriateUMTSBearerService.ThisrequiresPDPcontextactiva-tionforIMSsignallingtraffic(SIPsignalling).OnceaPDPcontextisestablishedtheuserper-formsaserviceregistrationwiththeIMS.Afterserviceregistration,theUEcanestablishsessionsusingtheserviceÕsrequiredSIPsignalling.ThenextsectionsprovideabriefoverviewofSIPsig-nallingprocedures.Thepresentedsignallingproce-duresandmessageflowshavebeenimplementedinasimulationenvironmentandtheirperformanceevaluationisdetailedinsubsequentsections.4.1.UMTSsignallingprocedures

Atthestartofapacket-switcheduserapplica-tion,aBearerServiceconnection(PDPcontextwithspecificRadioAccessBearerandRadioBearer)needstobeestablishedtoenabledatatransfer.However,beforeanRNCcancontrol

anyrequestedbearer,itneedstocreateasignallingconnectionbetweentheUEandtheCN.TheRNCusestheRadioResourceControl(RRC)connec-tionservicestocreateoneormoreSignallingRadioBearers(SBRs)betweenUEandtheUTRAN.Thesignallingconnectionsetupproce-dureisshowninFig.5.ItisalwaysinitiatedbytheUE.

4.2.UMTSbearerservice:PDPcontextactivationThePDPContextActivationproceduremaybeinitiatedbytheUEorbythenetworkdependingonthedirectionofthesession.APDPcontextestablishesanassociationbetweentheUEandtheCNforagivenQoSacrossaUMTSBearerService.ItcontainsroutinginformationthatisusedtotransferthePDPPDUsbetweentheUEandtheGGSN.OnceaprimaryPDPcontexthasbeenestablishedforagivenPDPaddress,asecondaryPDPcontextcanbeactivatedwiththealreadyactivePDPcontext,butwithadifferentQoSprofile.Fig.6showsthesignallingflowofthePDPContextActivationprocedure.BesidesacquiringaPDPcontext,theUEalsodiscoverstheaddressoftheProxyCSCFinthenetworkitresidesin.

4.3.IMSsignallingprocedures

OnceaPDPcontexthasbeenestablished,theUEcaninitiateSIPsignallingprocedures.Inthe

Fig.5.RRCconnection.

390D.Peschetal./ComputerNetworks49(2005)385–403

Fig.6.PDPcontextactivation.

followingthesignallingflowsforapplicationregistrationandsessionsetupproceduresarebrieflyintroduced.Thosemethodsaregenerallyre-quiredforanyIMSapplication.ThesessionsetupsignallingflowsforvoiceoverIPorinstantmes-sagingsessionsarethenintroduced.Inadditiontothat,theSIPextensionsinSIMPLEfortheIMandPresenceservicearediscussed.

4.3.1.ApplicationlevelSIPregistration

InordertorequestservicesoftheIMSdomain,ausermustperformanapplicationlevelregistration.ThisisdoneonceauserhasanactivatePDPcon-textwithappropriateQoSparametersfortransferofIMS-relatedSIPsignalling.TheQoSparametersspecifiedinPDPcontextactivationmustbeappro-priateforIMSsignalling.Fig.7showstheflowofmessagesforregistrationoftheUEwithitsServingCSCF,assumingtheUEwasnotpreviouslyregis-tered.Asshown,theS-CSCFauthenticatesthemo-bilebeforeregistrationissuccessful.

4.3.2.Subscription

OncethemobileisconnectedtotheIMS,sub-scriptiontothepresenceproviderserversisre-quiredforthoseusersusingpresencecapabilities.Firstthewatcherentitysubscribestoitspresencelistserver(PLS).ThePLSwillthenforwardthesubscriptionrequesttothedesiredpresentityser-ver.Assoonasthemessagearrivesatthepresen-tityserver,allwatcherPLSsareprovidedwithpresentityÕsdetailedinformationandarerequestedtonotifythewatcherentitywiththelatestinfor-mation.Fig.8showsthemessageflowforpresencesubscriptionandwatchernotification.

4.3.3.Sessionsetup

AfterregistrationwiththeIMSandsubscrip-tiontothepresenceservice(whenpresenceisused),theuserisreadytoaccessthemultimediaservicesusingSIPsessioncontrolprocedures.Thesessiontypessupportedinthisanalysisin-cludesessionsetupbetweenmobileandfixed

D.Peschetal./ComputerNetworks49(2005)385–403391

Fig.7.SIPregistermethod.

Fig.8.Presencesubscriptionmethod.

392D.Peschetal./ComputerNetworks49(2005)385–403

Fig.9.SessionsetupmobiletoPSTN(MO).

phoneforvoicesessionsandmobiletomobilesessionsforinstantmessaging.

Figs.9and10showtheflowofmessagesre-quiredforthesessionsetupbetweenamobileandafixedphone.Fig.9showsthesignallingmes-sageflowforthecasewhenthemobileinitiatesthesession,whereasFig.10showstheflowwhenthefixedphoneinitiatesthesession.Aftermediachar-acteristicsforthesessionhavebeendetermined,re-sourcereservationisrequired,whichentailscreatingasecondaryPDPcontextfortransportoftherequiredmedia,andsettingupthecorre-spondingradioaccessbearers.Fig.11showsthesameSIPmessageexchangetosetupasessionbetweentwomobiles.

5.Simulationmodelandenvironment

InordertoevaluatetheperformanceofSIP-basedmultimediaservicesinUMTSRelease5/6

wehavedevelopedaneventdrivensimulationenvironmentthatallowstheimplementationandsimulationofSIPmessageflowswithinamodelofaUMTSnetwork.Inordertofacilitatetheeffi-cientsimulationofafullUMTSnetworkincludingalargenumberofradioaccessnetworkandcorenetworkelementsaswellasIMSnetworkele-ments,thesimulatorimplementsnetworkelementsasconsistingofprocessorsandfiniteFIFObuffers.Eachprocessorhasaparticularmessageprocess-ingcapabilitywhichisbasedonmeasurementsta-kenfromrealnetworkelements.ThisapproachallowsthesimulationofalargescaleUMTSnet-workmodelsasdetailedinTable1onasinglepro-cessorworkstationwithinareasonabletimeframe.TypicalprocessingtimesforUMTSandSIPsignallingmessagesarebasedonmeasurementsta-kenwithrealnetworkingequipmentincludingNo-deBs,RNCs,SGSN,andGGSN,aswellasatypicalserverplatformsuchasSUNNetraservers.Themessageprocessing(turn-around)timesfor

D.Peschetal./ComputerNetworks49(2005)385–403393

Fig.10.SessionsetupPSTNtomobile(MT).

UMTSnetworkelementsandeachelementassoci-atedFIFObufferaresummarisedinTable2(seealsoTables3and4).Networkelementsareinter-connectedusingtypicaltransmissionlinksbasedonE1,STM-1,andEthernet.Weassumethat10%oftheavailabledatarateoneachtransmis-sionlinkisreservedforcontrolsignalling.

Thedynamicsignallingsimulator,SigSim,isdesignedtoestimatetheend-to-endsignallingloadintermsofnumberofmessageshandledpernet-workelementandproceduraldelays.SigSimin-cludesmodelsofusermobilitywithinaparticularenvironmentaswellasuserbehaviourintermsofaccessingdifferentservices.TheeffectsofradiotransmissionbasedonWCDMAarecap-turedinlook-uptablesgeneratedbyaphysicalradiointerfacesimulator.Eventhoughonly

signallingtrafficissimulated,trafficmodelsareimplementedandaccountedfortheperiodoftimeauserisusingaparticularservice.5.1.Networkmodel

Intheprevioussectionwepresentedthefunc-tionalarchitectureoftheIMSasdefinedintheUMTSRelease5/6.However,inordertoanalyseitsperformanceintermsofsignallingfloweffi-ciencyweproposeamorepracticalrealisationhere.Firstthemodelreferenceforthebasicsignal-lingservicesaccordingtoUMTSRelease5specifi-cationsispresented.Thenweadaptthemodeltoincludeenhancedservicesandapplicationcapabil-itiesintroducedinUMTSRelease6includingthePresenceandInstantMessagingservices.

394D.Peschetal./ComputerNetworks49(2005)385–403

Fig.11.Mobiletomobilesessionestablishment.

Table1

NetworkconfigurationdataParameter

Numberofcells:(rows·columns)NumberofRNCNumberofSGSNNumberofGGSNNumberofCSCFNumberofMGCFNumberofI-CSCFNumberofHSSURAsizeRAsizeCellradiusDenseurbanUrbanruralSuburbanRural

Value(28,28)

42111112425050010004000

UnitsCells

Table2

NetworkelementprocessingtimesNetworkelementNodeBRNCSGSNGGSNP-CSCFS-CSCFI-CSCFMGCF

Turn-aroundtime(ms)1818

SeeTable3SeeTable425252550

Buffer

size(Mbits)3232323216163232

CellsCellsMeters

Table3

SGSNturn-aroundtimes

5.1.1.BasicUMTSnetworkmodel

Accordingto3GPPspecifications,basicses-sionsbetweenmobileusersalwaysinvolvetwoS-CSCFs(oneforeachuser)andanI-CSCFtoselectthem.Ontheotherhand,asessionbetweenauserandaPSTNendpointinvolvesanS-CSCFfortheUE,aBGCFtoselectthePSTNgatewayanda

SignallingflowtypeMOPDPcontextactivationPDPcontextdeactivationMOservicerequestNIservicerequestRoutingareaupdateSIPsignalling

Turn-aroundtime(ms)394046391950

D.Peschetal./ComputerNetworks49(2005)385–403

395

Table4

GGSNturn-aroundtimesSignallingflowtypeTurn-aroundtime(ms)MOPDPcontextactivation75NIPDPcontextactivation59PDPcontextdeactivation75SIPsignalling

50

MGCFforthePSTN.Therefore,SIPmessagesareroutedthroughfourSIPproxiesinthemobiletofixedscenario,i.e.,P-CSCF,S-CSCF,BGCFandMGCF.Thisisworseforthemobile-to-mobilecasewhereaP-CSCFandanS-CSCFarerequiredforbothentitiesinadditiontotheI-CSCF,addinguptoatotaloffiveSIPproxiesorservers.ThemoreSIPproxiesthemessagehastotraversethegreaterthetransmissiondelay.Consequently,inordertoreducethetransmissiondelay,weproposetoco-locateIMSlogicalnetworkelementswithsimilarfunctionalityintothreephysicalnodesasillustratedinFig.12.

SIPServerNode.IntegratestheP-CSCFandtheS-CSCFinacommonnodewithinaparticularoperatorÕsnetwork.EverymobilecontactstheIMSthroughaProxy-CSCF.AfterregistrationtheP-CSCFroutestheSIPmessagestotheS-CSCFSIPcontrolelement.TheP-CSCFresidesinthenetworkwherethemobileresides,visitedorhomenetwork,whereastheS-CSCFalwaysresidesinthehomenetwork.Thescenarios

Fig.12.Basicmodelreference.

consideredhereassumedthatallmobilesareintheirhomenetworks,thereforebyco-locatingthosetwoentitiesthenumberofmessagestrans-mittedthroughthenetworkisdecreasedby34%,thusalsoreducingtransmissiondelay.

IMSGatewayNode.Whenthesessionisestab-lishedbetweenamobileuserandaPSTNendpointsuchasafixedtelephoneuser,theBGCFandMGCFhandletheSIPsignallingforthePSTNendpoint.TheBGCF,atthestartofthesessionsetup,selectsthePSTNnetworkwithwhichthein-ter-workingistooccurandforwardsthemessagetothecorrespondingMGCF.AlthoughtheBGCFdoesnothaveconsiderableimpactonthesessionsetup,asitisnotincludedintheSIPmessagepath,thecollocationwiththeMGCFcontributestominimisingmessagetransactiontime.

DatabaseNode.HoststheI-CSCFandtheHSS,whichisalargedatabasewithextendedHLRcapabilities.TheI-CSCFfunctionalityforanon-roaminguserisreducedtocontactingtheHSSforinformation.ItqueriestheHSStoassigntheServing-CSCFattheregistrationpointandalsoobtainstheS-CSCFaddressoftheterminatingcounterpartduringsessionsetup.Thereforeitseemsreasonabletocollocateboth.

5.1.2.EnhancedUMTSnetworkmodel

Thepresenceservice,whichresidesintheIPMultimediaSub-system(IMS),isbeingstandar-disedinRelease6ofthe3GPPstandards[13,14].Thepresenceservermanagesthepresenceinfor-mationofauser(presentity),whichisuploadedbydifferentagents(networkelements,terminalsorexternalelements)andcombinesitintoasinglepresencedocumentinastandardisedXML-basedformat.Furthermore,theserverallowsotherusers(watchers)tosubscribetoitforreceivingpresenceinformation.Forsimplicity,weconsiderthatbothwatcherandpresentityentitiesresideinthesamenetwork,thehomenetwork.Assuch,theycom-municatethroughthehomenetworkÕsCSCFsproxiesandnoexternalagentsareinvolved.Basedonthissimplifiedarchitecture,apracticalrealisa-tionoftheUMTSpresenceservicemodelisproposedhere,wheredifferentelementsarecollo-catedinordertoreducethemessagetransmissiondelay.

396D.Peschetal./ComputerNetworks49(2005)385–403

5.3.Trafficmodel

Dependingonthetypeofuserdatacarriedwithinasessionaparticularusertrafficmodelischosen.Intheanalysispresentedhereweconsiderthreetrafficmodels,aconversationalpacketvoiceservice,aninteractive—PagingInstantMessag-ing—andastreaming—MessageInstantMessaging—instantmessagingservice.

ThevoicetrafficmodelisbasedonatwonodeMarkovchainmodelbasedonBradyÕsmodel[16],withtalkspurtsandsilenceperiodsmodelledbyatruncatedexponentialdistribution(seeTable5);themessagingmodelisbasedonahierarchicalUMTSdatamodelbasedon[17]andisshowninFig.15.

ThePagingInstantMessagingserviceusesthesessionandpacketlevelparametersasshowninTable6.Theparametervaluesarebasedonasur-veycarriedoutamongCorkInstituteofTechnol-ogystudents,whereweidentifiedthefrequencyandamountofGSMShortMessagingService(SMS)messagesstudentssendperweek.AsSMSisaparticularlypopularserviceamongthe

Fig.13.Enhancedmodelreference.

Thepresenceserveriscollocatedwiththeregis-terserver,i.e.,theS-CSCF.Furthermore,thewatcherandthepresentityentitiesresideontheUserEquipmentandcommunicatewiththeserveracrosstheSIPproxies,P-CSCF,S-CSCFandI-CSCF.Fig.13showstheconsideredreferenceUMTSarchitectureincludingthetwointroducedapproaches.5.2.Sessionmodel

SessionarrivalsaremodelledbyaPoissonpro-cesswithameansessionholdingtimeof168minwithsessioninter-arrivaltimesdependentonthenextservicerequest(seeFig.14).Eachsessionmayholdoneormoreservices(usersessions)andiftheirQoSsdifferamodificationinthesessionisundertaken.Although3GPPspecificationsallowmorethanonePDPcontextperuserwehaveonlysimulatedsinglePDPcontexts.

Table5

VoicetrafficmodelparametersLevelSession

ParameterArrivalrate

DurationTalk-SpurtSilent

UnitsSession/hsss

DistributionTruncatedexponentialTruncatedexponentialTruncatedexponentialTruncatedexponential

Mean1.612011.3

Periods

Fig.14.Sessionmodel.

D.Peschetal./ComputerNetworks49(2005)385–403397

Fig.15.Packetservicesessionmodel.

Table6

Paginginstantmessagingparameters,1/l=mean,T=trunca-tionvalueLevelParameterUnits

DistributionUL/DL

values

Session

ArrivalSession/dayTruncated

l=2.9rate

exponential

T=0.36Duration

sTruncated

1/l=5exponential

T=250Packet

SizeBytesTruncated

l=160pareto

T=1000Inter-arrivalsTruncated

1/l=0.032exponential

T=1000

15–24-yearagegroupinIreland,wefeltthatsuchasurveywouldyieldareasonableexpectationforthesessionsgeneratedinafutureUMTS-basedmobileinstantmessagingservice.

TheMessageInstantMessagingserviceparame-tersforsession,file,andpacketsizedistributionareshowninTable7,andarebasedonfigurespublishedin[18].

6.Performancecharacteristics

InordertoobtainperformancecharacteristicsforSIP-basedmultimediasessionsinUMTSRe-lease5/6wehavesimulatedpacket-voicecallsetupdelaysformobileoriginatedandterminatedcallinitiationandforthetwoinstantmessagingschemesusingthemodelspresentedabove.First

Table7

Messageinstantmessagingparameters,1/l=mean,T=trun-cationvalueLevelParameterUnitsDistributionUL/DLvaluesSession

ArrivalrateSession/dayTruncatedl=1.7exponentialT=0.36Duration

MinTruncated1/l=7.5exponentialT=15FileSize

UnitsGeometricn=188Inter-arrivalUnitsGeometricn=5Packet

SizeBytesTruncatedl=80paretoT=1000Inter-arrival

Sec.

Truncatedl=0.024exponential

T=1000

wepresenttypicalSIPmessagesizesonwhichthesignallingflowsformultimediaservicesarebasedandthenwepresentresultsforsessionsetupdelaysforvoiceandinstantmessagingsessions.6.1.SIPmessagesizes

InordertoanalyseSIP-basedmultimediases-sionsetupandmessagetransmissiondelays,weas-sumedSIPmessagesizesshowninTables8–12.AcriticalissuewithSIPmessagesisthattheyaresubstantialinsizecomparedtothebinaryencodedcallcontrolmessagesusedinGSMorUMTSRe-lease99.Asshownintheperformanceanalysis,SIPmessagesizescontributetosignificantdelayontheradioaccessnetworkofUMTS.Oneap-proachtoreducethedelayistocompressSIPmes-sages.TheIETFROHCgrouphasproposedaSIPmessagecompressionframeworkcalledSigComp[19]togetherwithanumberofcompressionmeth-ods.Intheperformanceevaluationpresentedhere,weusetheText-basedCompressionusingtheCacheandBlank(TCCB)approach[20,21],whichachievesveryhighcompressionratiosandcontri-butestoasignificantdelayreduction.

Tables8and9showtypicalSIPmessagesizesofsignallingflowsformobileoriginatedandmo-bileterminatedpacket-voicecallsetup.Itisobvi-ousthatmessagesizesaresubstantialbutusingTCCBcompression,messagescanbecompressedandreducedbyupto92%inthecaseofa180Ringingmessagesforexample.

398D.Peschetal./ComputerNetworks49(2005)385–403

Table8

MobileoriginatedsessionSIPsignallingmessagesizesUPLINKINVITEHeaderSDPTotalUPDATEHeaderSDPTotal

PRACK(180)Total(SDP=0)ACK

Total(SDP=0)DOWNLINK183SessionProgressHeaderSDPTotal180Ringing

Total(SDP=0)200OKHeaderSDPTotal

UE

Notcompressed507229736352194546370314P-CSCFNotcompressed614233847361363194557

Compressed2027427629257095

676867929383

Compressed310190500982111910691

Compressionratio(%)39173272787171

Table10showsSIPmessagesizesfortheRegis-termethodincludingsignallingforapplicationlevelauthenticationwithandwithoutSIPmessagecompressionbasedonTCCB.Herewedonotachievethesameimpressivecompressionratiosasachievedinotherflowssuchassessionsetup.ThereasonisthattheTCCBapproachcachespre-viousSIPmessagecontentandblanksitoutforsubsequentSIPmessages.AstheRegistermethodistypicallythefirstSIPflow,cachecontentislim-itedornotavailableatthatstageandthecompres-sionratioislow[19].

Tables11and12showthemessagesandtheirrespectivesizeswithandwithoutmessagecom-pressionfortheSIPsubscribemethodusedforsubscribingtoanotheruserÕsbuddylistandthemobile-to-mobileSIPsessionsetupdelayfortheMessageInstantMessagingflow.Substantialreductionsinmessagesizesareachievedwithcom-pressionandtheperformanceevaluationshowstheimprovementinsignallingdelayreduction.

6.2.Voicecallsetupdelayanalysis

WehavesimulatedthesignallingflowshowninFig.9inthenetworkmodeldescribedabove.Theresultswereobtainedbasedonthebasicnetworkmodeldescribedabove,thenetworkconfigurationdetailedinTable1andatotalof210,000usersroamingtheUMTSnetwork.InordertoallowacomparisonwithUMTSRelease99CallControl,weconsideredthatSIPsignallingmessageswouldbecarriedina3.4kb/sDTCHasallUMTSsignal-lingistransmittedona3.4kb/sradiobearerwithMACblocksizeof136[22].Table13showstheUMTSRelease99circuit-switchedcallsetupde-laysasobtainedwiththeSigSimsimulationenvi-ronment.WecancomparethistoaUMTSRelease5/6packetvoicecallsetupdelayusingSIPsignallingasshowninTable14.WecanseesignificantsessionsetupdelaysusingSIPsignallingontheradiointerfaceandintheIMScomparedtoUMTSRelease99callsetupdelays.Duetoa

D.Peschetal./ComputerNetworks49(2005)385–403

Table9

MobileterminatedsessionSIPsignallingmessagesizesDOWNLINKINVITEHeaderSDPTotal

PRACK(183)HeaderSDPTotal

PRACK(180)Total(SDP=0)ACK

Total(SDP=0)UPLINK

183SessionprogressHeaderSDPTotal180Ringing

Total(SDP=0)200OKHeaderSDPTotal

P-CSCFNotcompressed6342268603751965713315P-CSCFNotcompressed5181967143613194558

Compressed219943139610170171

585256737269

Compressed47618766332573043

2517239171849286

399

Compressionratio(%)

Table10

SIPregistermethodmessagesizesUPLINK

REGISTER

Total(Body=0)REGISTERAuth.Total(Body=0)DOWNLINK401unauthorisedTotal(Body=0)200OK

Total(Body=0)

UE

Notcompressed534639P-CSCFNotcompressed363426

Compressed203218

4449

Compressed429405

2037

Compressionratio(%)

significantdelayontheradiointerface,weproposetocompressSIPmessagesusingSIPmessagecom-pressiontechniquessuchastheText-basedCom-pressionusingCacheandBlank(TCCB)orothermethodsasproposedbytheIETFROHCgroup.EmployingSIPmessagecompressionreducestheRANdelayby66%(seeTable14)withanoverallreductioninsessionsetupdelayof46%.EventhoughSIPmessagecompressionreducesdelay,othermeanshavetobefoundtobringthe

400

Table11

SIPsubscribemethodmessagesizesUPLINK

SUBSCRIBE

Total(Body=0)200OK

Total(Body=0)DOWNLINK

D.Peschetal./ComputerNetworks49(2005)385–403

UE

Notcompressed472215P-CSCFNotcompressed

Compressed102276322

FunctionCompressed296112

Compressionratio(%)

3748

200OK

Total(Body=0)NOTIFY

Total(XML=0)NOTIFY(state)HeaderXML

302420458

Function

663430

Function

Table12

Mobile-to-mobileSIPinstantmessagingsessionsetupmessagesizesUPLINKINVITEHeaderSDPTotalUPDATEHeaderSDPTotal

PRACK(180)Total(SDP=0)ACK

Total(SDP=0)DOWNLINK183SessionProgressHeaderSDPTotal180Ringing

Total(SDP=0)200OK

Total(SDP=0)

UE

Notcompressed424184608300184484288263P-CSCFNotcompressed500184684293263

Compressed302184486103102

400296561

Compressed2051843142184326147137

52036530334948

Compressionratio(%)

SIP-basedcallsetupintosimilardelayregionsasUMTSRelease99.Oneapproachistoco-locatenetworkelementsintheIMStoreducemessagetransmissionacrossexternalinterfacesortobaseIMSnetworkelementsonhigh-performanceserv-erswithhighermessagethroughputthancurrent

D.Peschetal./ComputerNetworks49(2005)385–403

401

Table13

UMTSRelease99circuit-switchedcallsetupdelayFlowID

Meandelay(s)RANCoreTotalMOUMTSCScallsetup1.350.551.90MTUMTSCScallsetup

1.63

0.66

2.28

Table14

MobileoriginatedSIPcallsetupdelayFlowID

SIPcompr.Meandelay(s)RANCoreTotalPrimaryPDPcontextN/A1.570.772.34MOSIPsessionsetupOff4.381.936.31MOSIPsessionsetupOn1.491.933.42SecondaryPDPContext

N/A

1.18

0.76

1.94

referenceequipmentascorenetworkdelaystillcontributestoover50%delaywhenSIPmessagecompressionisenabled.

CallsetupdelaysforPSTNoriginatedandmo-bileterminatedcallinitiationusingSIPsignallingisshowninTable15.Herewecanobservesimilarresultsasinthemobileoriginatedcase.

Inordertoassestheimpactofthelowbitrate3.4kb/ssignallingbearerusedforcallsetupdelaysimulation,wesimulatedthesamemobileorigi-natedandmobileterminatedSIPcallsetupusingakb/sdatabearerwithaMACblocksizeof1280.Theuseofkb/sbearersinsteadofa3.4kb/scontrolsignallingbearerandusingSIPmessagecompressionbringsmobileoriginatedcallsetupalmostinlinewithUMTSRelease99delaysandmobileterminatedcallsetupdelaysarecom-parableifnotbetterthanUMTSRelease99callsetupdelaysasshowninTables16and17.Inboth

Table15

MobileterminatedSIPcallsetupdelayFlowID

SIPcompr.Meandelay(s)RANCoreTotalNIPDPcontextN/A2.1.674.31activation

MTSIPsessionsetupOff4.911.566.47MTSIPsessionsetupOn1.971.3.61SecondaryPDPcontext

N/A

1.18

0.76

1.94

Table16

MobileoriginatedSIPcallsetupdelay,kb/sbearerFlowID

SIPcompr.

Meandelay(s)RANCoreTotalMOSIPsessionsetupOff0.701.902.6MOSIPsessionsetupOn0.251.902.15SecondaryPDPcontext

N/A

1.18

0.76

1.94

Table17

MobileterminatedSIPcallsetupdelay,kb/sbearerFlowID

TCCBDelay(s)RANCoreTotalMTSIPsessionsetupOff0.751.632.38MTSIPsessionsetupOn0.341.631.97SecondaryPDPcontext

N/A

1.18

0.76

1.94

casesthemaincontributingfactortodelayisthecorenetwork.ThisrequiresacarefulchoiceforcorenetworkSIPservercomputingplatformsinordertoachieveservicequalityintermsoflowcallsetupdelaysforfutureUMTSIMS-basedmulti-mediaservices.

6.3.Instantmessagingserviceanalysis

Resultsobtainedfortheinstantmessagingser-viceanalysisarebasedon5hofsimulatedreal-timeand10,000IMusersintheUMTSnetwork.TheenhancedUMTSnetworkmodeldescribedabovewasimplementedintheSigSimsimulator.Akb/sdatabearerisusedontheUMTSradioaccessnetworktotransmitSIPmessages.Table18

Table18

Mobile-to-mobilemessageinstantmessaging(streaming),with-outSIPmessagecompressionFlowID

Meandelay95%DelayRANCoreTotaltotalPSsessionsetup1.460.772.233.36SIPregistration0.410.881.281.48SIPsubscriptionLS0.240.530.760.SIPsubscription0.160.350.510.56presentity

MMOSIPsessionsetup

0.82

1.25

2.07

2.39

402D.Peschetal./ComputerNetworks49(2005)385–403

Table19

Mobile-to-mobilemessageinstantmessaging(streaming),withSIPmessagecompressionFlowID

SIPregistrationSIPsubscriptionLS

SubscriptionpresentityMOMMOSIPsessionsetupSubscriptionpresentityMTMMTSIPsessionsetupSIPnextsubscriptionNextMMOinvite

MeanDelayRANCoreTotal0.310.170.140.5800.710.160.41

0.880.530.351.250.051.030.451.25

1.190.700.491.830.051.740.611.66

95%Totaldelay1.330.780.512.010.051.740.701.87

showsthemeansignallingdelaysinregisteringwithanSIPserver,subscribingtoapresentity,andthesessionsetupfortransmittinginstantmes-sages.Thetablealsoshowsthe95%percentileofmessagetransmissiondelays.

Table19showsthesameresultsbutwithSIPmessagecompressionenabled.Herewecanob-servethatcompressingmessagesimprovesignal-lingdelaysontheRANsignificantly.However,delaysinthecorenetworkstaythesameasdecom-pressionofmessagestakesplaceintheP-CSCF.Table20showsdelaysforSIPsignallingflowsforthePagingInstantmessagingservice.ResultsareshownwithandwithoutSIPmessagecompres-sionandformeanvaluesand95%percentileoftotaldelays.

7.Conclusions

WehavepresentedanoverviewofSIP-basedmultimediaservicesproposedforimplementationinUMTSRelease5/6usingthenewIP-based

MultimediaCoreNetworkSubsystem.PartoftheoverviewwastoshowSIPmessagesflowsasso-ciatedwiththenewservicepropositions.Inordertoefficientlyimplementtheseservices,anSIP-basedpacketvoiceservice(voiceoverIP)andapresenceandinstantmessagingservice,wealsoproposedphysicalreferenceimplementationsoftheIMSinordertolimitexchangeofalargeamountofSIPmessagesbetweennetworkelementsoverexternalinterfaces.

Acomputersimulation-basedperformanceevaluationrevealedinterestinginsightintotheper-formanceofSIPmessageflowsfortheservicesweinvestigated.Aswastobeexpected,thelargeSIPmessagesizesandalsothenumberofmessagesre-quiredtosetupavoicesessioncontributetosignif-icantcallsetupdelayscomparedtocallcontrolsignallinginUMTSRelease99,whichisstillthereference.SIPmessagecompressionachievedareductioninsignallingdelayovertheradioaccessnetworks,however,aconsiderableamountofde-laywasincurredonthecorenetworkduetothelargenumberofmessagesandthenumberofnet-workelementsinvolvedinthesessionsetup.CorenetworkdelaywillneedtobeaddressedinfutureimplementationsofUMTSRelease5/6throughselectionofappropriatecomputingplatformsforSIPservers.Co-locationisalsoadesignoptiontoreducedelay.

PerformanceofSIP-basedinstantmessaging(IM)andpresencealsoimprovedwhenSIPmes-sagecompressionwasusedeventhoughdelayhereisnotascritical.ThemessagebodyofIMmes-sageswillrequirecompressioniflargeamountsofmultimediadatasuchasimages,audio,etc.,areincludedinthemessagetoreducetransmissiondelays.

Table20

Mobile-to-mobilepaginginstantmessaging(interactive)FlowID

Meandelay

(withoutcompression)RAN

PSsessionsetupSIPregistrationSIPsubscriptionLS

SIPsubscriptionpresentity

1.480.420.250.16

Core0.770.880.530.35

Total2.251.300.770.51

2.831.420.850.5395%Totaldelay

Meandelay

(withcompression)RAN1.480.320.180.14

Core0.770.880.530.35

Total2.251.200.710.49

2.831.290.750.5195%Totaldelay

D.Peschetal./ComputerNetworks49(2005)385–403403

References

[1]H.Kaaranen,A.Ahtiainen,L.Laitinen,S.Naghian,V.Niemi,UMTSNetworks,JohnWileyandSons,Chichester,2001.

[2]J.Rosenbergetal.,SIP:SessionInitiationProtocol,IETFRFC3261,June2002.

[3]M.I.Pous,G.Foster,A.Sesmun,V.Kenneally,D.Pesch,PerformanceevaluationofUMTSpacketvoicecallcon-trol,in:Proceedingsofthe5thEuropeanPersonalMobileCommunicationsConference,Glasgow,Scotland,May2003.

[4]I.D.D.Curcio,M.Lundan,SIPcallsetupdelayin3G,in:ProceedingsoftheIEEEISCC2002,Taormina,Italy,July2002.

[5]C.Urrutia-Valde

´s,A.Mukhopadhyay,M.El-Sayed,Z.Sayeed,D.Sun,SIPin3Gwirelessnetworks:Servicemodels,architecture,andnetworkdesign,BellLabsTechnicalJournal3(2004)101–126.

[6]M.Cortes,J.R.Ensor,J.O.Esteban,OnSIPperformance,BellLabsTechnicalJournal3(2004)155–172.[7]3GPPTS23.002,NetworkArchitecture,2002.

[8]3GPPTR23.922,ArchitectureforanAllIPnetwork,2002.

[9]3GPPTS23.228,IPMultimediaSubsystem(IMS);Stage2,2002.

[10]M.Handley,V.Jacobson,SDP:SessionDescription

Protocol,IETFRFC2327,April1998.

[11]3GPPTS24.228,SignallingflowsfortheIPmultimedia

callcontrolbasedonSIPandSDP;Stage3,2003.

[12]M.Day,J.Rosenberg,H.Sugano,Amodelforpresence

andinstantmessaging,IETFRFC2778,February2000.[13]3GPPTS22.141,Presenceservice;Stage1,2003.

[14]3GPPTS23.141,Presenceservice;Architectureand

functionaldescription;Stage2,2003.

[15]M.Dayetal.,Instantmessaging/presenceprotocol

requirements,IETFRFC2779,February2000.

[16]P.T.Brady,Amodelforgeneratingon-offspeechpatterns

intwo-wayconversation,BellSystemTechnicalJournal48(1969)2445–2472.

[17]A.Klemm,C.Lindemann,M.Lohmann,Trafficmodelling

andcharacterizationforUMTSnetworks,in:ProceedingsoftheIEEEGLOBECOM,November2001.

[18]Telecompetition,Inc.,3GWirelessdata—trafficcharacter-istics,UMTSForum,June2001.

[19]R.Priceetal.,Signalingcompression(SigComp),IETF

RFC3320,January2003.

[20]V.Kenneally,D.Pesch,I.Majumdar,EvaluationofSIP

compressionforIP-basedwirelessmultimediacommuni-cations,in:ProceedingsoftheIT&TConference,Water-ford,Ireland,October2002.

[21]I.Majumdar,V.Kenneally,D.Pesch,ImprovingSIPcall

controlperformancethroughmessagecompression—the

TCCBalgorithm,in:PresentationatSIP2003,Paris,France,January2003.

[22]3GPPTS34.108,Commontestenvironmentforuser

equipment(UE)conformancetesting,2002.DirkPeschwasborninKrefeld,Ger-manyin1966.HereceivedaDipl.IngdegreefromAachenUniversityofTechnologyin1993andaPh.D.degreefromtheUniversityofStrath-clyde,Glasgow,UK,in1999,bothinElectricalandElectronicEngineering.HewaswithNokiaMobilePhonesfrom1993to1995.From1996to1998hewasaresearchfellowwiththeMobileCommunicationsGroupinthe

DepartmentofElectricalandElectronicEngineeringatUni-versityofStrathclyde.Since1999hehasbeenalecturerintheDepartmentofElectronicEngineeringatCorkInstituteofTechnology.Heisco-founderandresearchdirectoroftheCentreforAdaptiveWirelessSystemsatCorkInstituteofTechnology.Heisaregularreviewerforanumberofinter-nationaljournalsandhasbeenontheprogrammecommitteeofanumberofinternationalconferenceincludingIEEEGlobe-com,IEEEISWCS,IFIPMWCN,EuropeanWireless,andothers.Hehaspublishedover45papersinpeer-reviewedjournalsandconferences.Hisresearchinterestsareintheareaofdesignandevaluationofalgorithmsandprotocolsformobileandwirelessnetworks,sensorandad-hocnetworks,andperformanceevaluationofmultimediaservicesinmobilenetworks.

MariaIsabelPousreceivedanengineeringdegreefromtheUniversityofValencia,Spainin2001andaMasterofEngi-neeringdegreefromCorkInstituteofTechnologyin2003,bothinElectronicEngineering.Sincelate2003shehasbeenwithaconsultantwithAccentureinMadrid.Herresearchinterestsareintheareaofperformanceevaluationofmultimediaprotocolsinmobilenetworks.

GerryFostergraduatedincommunicationsengineeringatPlymouthPolytechnicin1986.HeworkedatBritishAero-spaceuntil19designingcomponentsformicrowaveandmillimetrewavecommunicationssystems.HethenmovedtoSTCfortwoyears,wherehedevelopedmicrowavecompo-nentsfortheSquarialsatelliteflat-plateantennaandworkedonthecommunicationslinkforareconnaissanceaircraft.Next,hemovedintodatacomforBritishGas,whereheworkedonPMRmodemdesign,scanningtelemetry,under-water,andpipe-bornecommunicationssystems.In1997hemovedtoLucenttoworkonGPRSGSNs,andforthelastfiveyearshehasbeenatMotorolaworkingonUMTSsystemsanalysis.