【研报】新计算面使能移动算网融合.docx

5G-A6GNewComputingP1.aneforMobi1.eNetworkandComputingConvergence(2024Edition)©卷膏1中超智库jnteIgPW浪潮ZTE中兴NO<IA1.enovoOAsia1.nfo:伉中信科移动亚恒科技/1/CDC1.CTMobi1.eH3CForewordNewservicesof6Gandfuturenetworksrequireu1.timateexperience,whichca1.1.sfortheconvergenceofmobi1.enetworkandcomputing.Thiswhitepaperintroducesmotivation,featuresandrequirementsofmobi1.enetworkandcomputingconvergence,proposes6Garchicciureandkeytechno1.ogies,andprovidestypica1.usecasesandpotentia1.so1.utions.Itisexpectedhatthiswhitepapercou1.dhe1.pthefo11nationofindustria1.consensus,Thecopyrightofthiswhitepaperbe1.ongstoa1.1.thecontributorcompanies.Withoutauthorization,anyentitiesorindividua1.sarenota1.1.owedtorertx1.uccorcopyIhecon1.en1.softhiswhi1.epaperinpartorinwho1.e.1. Introduction1.1. MotivationOfmobi1.enetworkandcomputingconvergenceWiththedcvc1.opnwntofcommunicationtechno1.ogyandthe1.arge-sca1.ecommercia1.usageofedgecomputing.InOreandmoredatawi1.1.beprocessed1.oca1.1.yattheedgeofmobi1.enetworks11.AccordingtoGartnefsforecast,morethan75%ofdatawi1.1.beprocessedattheedgeby2025.Incurrent5Gsystems,thenetworkandcomputing(c.g.cdge/c1.oudcomputing)arcseparate,whichmakesitcha1.1.engingtomeettheexperiencerequirementsofsomenewSerViCeS.AccordingtoIM-2030perspective,newservicessuchasimmersecommunicationandho1.ographiccommunicationwi1.1.bewide1.yusedin6G(2.Forexamp1.e,XRservicerequiresupto100Mbpsofdown1.inkbitrate.5msof1.atencyandover99%framere1.iabi1.ity.AIc1.oudgamingca1.1.sforsignificantGPUresourcestoimp1.ementrendering.Industria1.roboticssystemneedsare1.iabi1.ityofover99.9999%.A1.1.thescenariosabovehaverequirementsformassivein-networkcomputing,whichcannotbefu1.fi1.1.edviacurrentcommunicationnetwork.Thus,thenextgenerationsystemwi1.1.enab1.eubiquityofnetworkaccessandcomputingcapabi1.ity.wherecomputingandnetworkneedtobedeep1.yconvergedanddeve1.opedtogethertomaximizeresourceefficiencyandoptimizetheusers,experience.Figure1-1Newcapabi1.ityrequirementsofIMT-2030T1.ieconvergenceofnetworkandcomputingprovidesmu1.tifacetedbenefitsooptimizingtheprocessingandtransmissionofdata,whichiscrucia1.Ioeniergingtechno1.ogiesandapp1.ications,inc1.udingXR.AIc1.oudgaming,industria1.roboticscontro1.,andotherA1.usages.Aconvergedarchitectureofcomputingandnetworkingsupportshigh-qua1.itybut1.ow-costservicede1.iver)'throughcoordinatedperformancemonitoringandjointQoScontro1.InscenariosCfIime-Varyingservicedemandsandnetworktrafficstates,theconvergenceframeworkdynamica1.1.yande1.astica1.1.yop1.imizcswhereandwhenthedatacanbeprocessed,toreduce1.atency,maximizere1.iabi1.ity,andenab1.erea1.-timeprocessing.Furthermore,theconvergence1.everagesresourceuti1.izationefficiencybyjoint1.ya1.1.ocatingresourcesandba1.ancingwork1.oads,therebypreventingIxxi1.enecksfromeitherIhecommunicationorcomputationcomponent.1.2. Featuresofmobi1.enetworkandcomputingconvergenceMobi1.enetworkandcomputingconvergence(MNCC)isanewcapabi1.itywhichmobi1.enetworkandcomputingaredeep1.yintegrated.Withit,goa1.sOfautomateddep1.oyment,optima1.routing,and1.oadba1.ancingofservicesarcachievedthroughubiquitous1.ydep1.oyedcomputingnodes.Thenewnetworkinfrastructureiscapab1.eofperceivingcomputingresourcesbothinsideandoutsidethenetwork.Thisinfrastructureenab1.esthenetwork(operfb11ncomputingresourceschedu1.ingondemandandinrea1.-time,therebyimprovingcomputingresourcesuti1.izationefficiency,andenhancinguserexperience.MNCCisbasedonthe11M)etworkcapabi1.itiesandthecomputinginfrastructure3).TheobjectiveistoprovideintegratedcommunicationandcomputingserviceswiththerequiredE2EQoS(rate,1.atency,re1.iabi1.ity).MNCChasthefo1.1.owingfeatures:1) MNCCprovidesondemandandintegratedcommunicationandcomputingservices,hrea1.izes(heoptima1.connectionamonguserapp1.icationsandavai1.ab1.eCOmPU1.ing/networkingresourcesbyse1.ectingtheoptima1.accesspathfromusertoapp1.icationforbestuserexperience.Ita1.sofeaturesubiquitousmobi1.eaccess,te11nina1.-neworkcoordination,contro1.1.ab1.eexperience,andhigh-stabi1.itycommunicationandcomuiing.Therefore,byincorporatingcomputingcapabi1.ity,Ihenetworkcanprovidehigh-qua1.itymobi1.eaccesscapabi1.itiesforbothCommuniea1.ionandcomputingservices.2) MNCCisbui1.tonmobi1.enetworkinfrastructure.Inadditiontomaintainingtraditiona1.corecapabi1.itiesofmobi1.enetworks,itintroducestwokeyinnovativearchitectura1.capabi1.ities:theintegrationofnetworkandcomputing,andunifiedschedu1.ingofcommunicationandcomputingresources.Itencompassescomputingresourcesspanningfromtc11nina1.in-nc1.work.toedgeandc1.oud,1.everagingtheuniquedistributednetwork1.ocationadvantage.MNCCa1.soenhancesQoSinwideareabyjointM?hedu1.ingOfcoinmunicationandcomputingresources.Consequent1.y.MNCCprovidesadeterministicandoptimizedend-to-endserviceexperience,wherethenetworksupportsintegratednc1.working-p1.us-c11putingresourcesco-schet1.u1.ing.2. KeyrequirementsTorea1.izethefeaturesofMNCCmentionedabove,thefo1.1.owingkeyrequirementsmustbesatisfied:wide-areacontinuouscomputingcapabi1.itycoverage,computingresourceawarenessanddiscovery.E2Eexperienceguarantee,te11nina1.-in-nework-edge-c1.oudcomputingcoordination,andpathse1.ectionandoptimisation.2.1. Wide-areacontinuouscomputingservicecoverageTheMNCCrequiresthecapabi1.ityofwide-areacontinuouscomputingservicecoverage,necessitatingthejointp1.anningofnetworkcoverage,userp1.aneandedgecomputingnodes.Inwidc-arcamobi1.ityscenario.MNCCdeterminestheconsistencyinodeforserviceexperienceaccordingtotherequirementsofcomputingtasks,e.g.thecomputingtaskisde1.ay-sensitiveorcontinuity-sensitiveorboth.Forde1.ay-sensitivetasks,MNCCse1.ects,adjustsandSChedUIeSputingnode(CgthCcomputingnodec1.osesttothetermina1.)forthecomputingtask,basedonrea1.-timetermina1.1.ocationtoensurethe1.owestde1.ay.Forcon1.iniii1.y-scnsitivctasks,aftercomputingnodese1.ection,theMNCCshou1.dguaranteethecomputingtaskisanchoredtothese1.ectedcomputingnodeandadjuststheforwardingpathtoensuredataconsistency.2.2. Computingresourceawarenessandse1.ectionComputingresourceisthefoundationofMNCC.andthereforetheawarenessofcomputingresourceisthekeycapabi1.ityopossess.Incomputingandnetworkconvergence,differentdataprocessingscenarioshavedifferentcomputingrequirements.Thekeytasksforbui1.dingtheconvergence,dep1.oyingservices,andaccurate1.yschedu1.ingcomputingresourcesarcasfo1.1.ows:howtomode1.andmeasurecomputingcapabi1.itywhi1.eshie1.dingheterogeneityofthecomputingresources;howtoatomica1.1.yabstractthecomputingservicestoprovideunifiedcomputingcapabi1.itiesforapp1.ications:andhowtoestab1.ishaunifiedmode1.fordifferentwork1.oaddistributionmodes:Networkorc1.oudsidecomputing.Datasentbythetermina1.needscomputingatnetworkorc1.oudside.MNCCrequiresthecapabi1.ityofse1.ectingcomputingresourcebasedonthetaskrequirementandassigningcomputingtasksoappropriatecomputingnodes.TCnninaI-nciwork-c1.oudco1.1.aboration.Whenaserviceisinitiatedbyatermina1.,thenetworkneedstodeterminewhetherthetermina1.possessesenoughcomputingresourcetoexecutethecomputation1.oca1.1.y,ortermina1.-edgeco1.1.aborationcomputingisneeded.Ifinhe1.attercase,MNCCshou1.dhavethecapabi1.itytodiscovertherequiredcomputingresourcesandschedu1.ethemforthetask.Forbothmodes,these1.ectionofcomputingresourcesrequirespo1.iciesbasedontermina1.1.ocation.E2EQoSrequirements,networkcoverageandcapacity,workQoSinformation,computingnodedep1.oyment(c.g.1.ocation,resourcetype,resourcequantity,bandwidth)andusersubscriptions.2.3. E2EexperienceguaranteeTheE2Eexperienceguaranteeofanapp1.icationforauserstemsfrombothcommunicationandcomputingsegments.,Vhe1.egacymobi1.ecommunicationandedge/c1.oudcomputingareintwoseparatesystemsandeschedu1.ingisdoneindcpcndcn1.1.yatbothsides.Therefore,thereisnodeterministicmechanismtoguaranteetheE2Eexperience.Withcommunicationandcomputingconvergedinonesystem,theschedu1.ingOfcoinmunicaiionandcomputingtaskscan1.>ejoint1.ycontro1.1.edioensureIheE2Eexperience.Byknowingthecapabi1.ities,statusandperformanceofbothcommunicationandcomputingsegments,schedu1.ingofIhccommunicationandcomputingresourcescanbedynamica1.1.yadjustedtonwettheE2Eexperiencerequirementswithoutover-provisioning.2.4. Termina1.-in-network-edge-c1.oudcomputingnetworkcoordinationMNCCshou1.dbestuti1.izethecomputingresourcesacrossmu1.tip1.edomains,inc1.udingcentra1.c1.oud,edgec1.oud,in-networkcomputing,andtermina1.s.Specifica1.1.y,MNCCprovidesubiquitouscomputingresourcesphysica1.1.yinterconnectedamongthesedomains,andsynergistica1.1.yorchestratescomputingtasksamongthem.Forexamp1.e,ifanobjectrecognitionanddetectiontaskbeforeXRrenderingisimp1.c11wntcdonthec1.oudside,itwi1.1.1.eadtoa1.argeamountofdatatransmissionandcorrespondinghighnetworkbandwidthrequirement.However,ifthecomputationisperformedon1.yonthetermina1.side,itmayriska1.argerendering/processingde1.aydue1.othe1.imitationOfcomputingresources,(hereby1.oweringuserexperience.Therefore,thecomputingresourcesofthetermina1.endorin-networkoredgeneedtobejoint1.yuti1.ized,whereeachentityhand1.esaportionofthecomputingtasktominimizethesericede1.ayandteminasenergyconsumption.Asanexamp1.e,edgecomputingSUPPOrIScomp1.exinference(e.g.1.argeAI/M1.mode1.),whi1.ethetermina1.performsasma1.1.amountofdatapreprocessing.Moreover,putingresources,andbyCo1.1.abonitivcschedu1.ingofcomputingandnetworkresourcesinaccordancewiththescnriccrequirements.MNCCcanfurtherimprovetheexperienceofusersatsca1.e.2.5. Pathse1.ectionandoptimizationTOaccesscomputingnodesforservices,pathse1.ectionandoptimizationshou1.dbeperformedinthenetwork.T1.ieexistingpathse1.ectionnchanisnwhereIhenetworkse1.ectseUPF(UserP1.aneFunction)bySMF(SessionManagementFunction),isbasedontheavai1.abi1.ityand1.oadofexistingUPFresourceswithintheSMFsservicearea.However,itneitherfu1.1.yconsiderstheba1.anceofresourcesonag1.oba1.sca1.enorthea1.ignmentbetweenuserQoSrequirementsanduserp1.anepaths.Therefore,itcannotse1.ectIheoptima1.userp1.anepathdynamica1.1.yandefficient1.ytoprovideu1.timatecomputingexperienceIotheuser.Forcontro1.p1.anepathse1.ection,whenauseraccessesthemobi1.ecommunicationnetwork,itisnecessarytoconsidertherea1.-timestatusmanagementandcontro1.capabi1.itiesofthemobi1.enetwork,workresourcesandcomputingnodes'work1.oads,workfunctionfortheuser.Asaresu1.t,MNCCneedstouti1.izethenetworktodynamica1.1.ybui1.dintcrvonncc1.ionbetweendistributedcomputingresourcesandstorageresources.Throughmu1.ti-dimensiona1.coordinateda1.1.ocationofnetwork,storage,andcomputingresources,theapp1.icationcanbeschedu1.edamongwide1.ydistributedcomputingresourcesintheMNCC.Byjoini1.yconsideringandcontro1.1.ingthecommunicationde1.ay(inc1.udingpropagationde1.ayandtransmissionde1.ay),g1.oba1.optimizationofcomputingresourceusagecanbeachievedwithguaranteedservicequa1.ityanduserexperience.2.6. IdentificationofservicerequestsandassociatedQoS/resourcerequirementsInsomete11nina1.-in-network-c1.oudco1.1.aborationusecases,suchasfederated1.earningandUEcomputa1.inoff1.oading,app1.icationmodu1.esorcomputingtaskswi1.1.bedynamica1.1.ydep1.oyedoroff1.oadedtothecomputingnodesinthenetworkbasedonUE'srequestsandavai1.ab1.eresources.TheUE,srequestcanbeover-the-topapp1.icationrequestsorspecificcomputingservicerequeststothenetworkentities.Regard1.esshowservicerequestsareexpressedandtransported,thenetworkneeds(oidentifytherequestsandreso1.vetherequiredresourceandQoS11xuircncnts,inc1.udingthenumberofapp1.icationmodu1.esorcomputingtaskstobedep1.oyed,theamountofcomputingresourcesrequired,theexpecied1.atency,thebandwidthacrossse1.ectedpath,andothernecessaryrequirements.Theserequirementsneedtoberef1.ectedexp1.icit1.yorimp1.icit1.yintheUE'srequest.3. Architectureconsiderations3.1. Evo1.utionof5(JandEdgesystemarchitecturesIn5Gera,the5GCandEdgeDataNetworkaretwoseparatenetworksthatareoperated,contro1.1.edandmanagedindependent1.y.TheEdgeDataNetworkcou1.dconsumesome5GCnetworkcapabi1.itiesthroughNEF(forexamp1.e.EDGE_7.EDGE_2andEDGE_8interfaces(see3GPPTS23.5584)tofu1.fi1.therequirementoftheedgeserviceoperations,suchasAF(App1.icationFunction)trafficinf1.uencefunctiona1.ity.UE1.ocationinformation,estab1.ishmentofanAFsessionwithQoS.andQoSmonitoringcapabi1.ity.However,thesecapabi1.itiesarcnotsufficienttoensureandoptimizetheE2EQoSthatstemsfrombothcommunicationandcomputingsegments.Withthis1.imitedcoordination,the5GcorenetworkisnotawareoftheQoSprovidedbytheEdgeDataNetwork,andtheEdgeDataNetworkhasnoknow1.edgeaboutthenetworkconditionsoftheAN(accessnetwork)ortermina1.,thereforethereisnowaytocoordinateresourcemanagementandtrafficschedu1.ingatbothsidestoensureandoptimizetheE2EQoS.Toenab1.eMNCCandcontro1.theE2EQoSbetweenthe5GsystemandEdgeDataNetwork,thecoordinationbetweenthetwosystemsneedstobeextendedtosupx>rt:5Gsystemtogetthein1.br11aiionandstatusoftheEdgeserver,suchas1.ocation,capacity,usage,bandwidth,1.atency,throughput,cic.;and5Gsystemaru1./orEdgeDataNetworkogetthein(b11nationofAccessNetworkandtermina1.,suchas1.atency,throughput,capacity,coveragearea,numberofusers,re1.iabi1.ity,etc.TheextendedCoOrdinaIiOncou1.dbeachievedthroughenhancingtheinterfaces(EDGE_7.EDGE_2andEDGE_8)between5GCandtheEdgeDataNetwork,theinterfaces(e.g.EDGE_M)betweentheManagementandOrchestrationSystemsofbothsides,asi1.1.ustratedintheFigure3-1.resources,communication1.atency,andcomputing1.atency,which,togetherwithadditiona1.informationsuchasUE1.ocation,arcnecessarytosupportcomputingtaskschedu1.ingandcomputingresourcea1.1.ocation.Thenetworkestab1.ishessessionsbasedonbothcommunicationandcomputingQoSparameters.Thecomputingandnetworkingsessionmanagementfunctionscanse1.ectdifferentforwardingpathsandcomputingnodestoprocesstherequestedcomputingtasks.Thetypeandsizeofacomputingtaskaffectsthese1.ectionofcomputingnodeswithrespecttoconditionsofnetworkandcomputingsegments,whi1.ethecommunication1.atencydeterminesthese1.ectionofuserp1.anepaths.Forthesereasons,these1.ectionOfcomputingresourceandcommunication>athshou1.dfo1.1.owunifiedpo1.icytoguanintccE2EQoS.4.3. DynamicUEwork1.oadoff1.oadingandschedu1.ingWithpervasivecomputingresourcedisiribuionwithinthenetwork,(heefficiencyofwork1.oadoff1.oadingandschedu1.ingarcheavi1.yimpactedbynetworkconditions.Thissectionexp1.oreshowwork1.oadscanbedynamica1.1.yoff1.oadedandschedu1.edacrossdifferentremoteexecutionenvironmentsanddiscussestheimpactofcommunication1.ink1.atencyandcongestiononoff1.oadingdecisions,wherethecomputingcapabi1.itiesinremoteexecutionenvironmentsinck1.cHardwareasaService(HaaS).P1.atformasaScnicc(PaaS),andSoftwareasaService(SaaS).Effectiveoptimizationstrategiesofwihrespecttothesefactorsareessentia1.forenhancingE2EQoSperformanceandOpiimizingresourceUti1.iziiiionandconsistsofthefo1.1.owingaspects:1) Work1.oadmode1.ing.Mode1.ingof(asksinvo1.vesidentifyingthecomputationa1.requirements,datadependencies,bandwidthrc<uircnntsandexecutiontime1.inesoftasks.Accuratemode1.she1.pinpredictingtheperformanceimpactsofoff1.oadingandinmakinginformeddecisionsa