NET3 - EuCNC

Wednesday, 17 June 2020, 12:15-16:00 CEST, Non-Live interaction (Chat), ... Our architecture aims at crossdomain security & trust ... NET3HomeNET3NET3–Software-basedandSelf-drivingNetworksWednesday,17June2020,12:15-14:30CEST,Recommendedre-viewing,https://www.youtube.com/playlist?list=PLjQu6nB1DfNBOrnpJ9fJmXRM0acK6q0QTWednesday,17June2020,12:15-16:00CEST,Non-Liveinteraction(Chat), linksentonlytoRegistered people LearningSDNTrafficFlowAccurateModelstoEnableQueueBandwidthDynamicOptimizationEnricoReticcioli,GiovanniDomenicoDiGirolamo,FrancescoSmarra,AlessioCarmeniniandAlessandroD’Innocenzo(UniversityofL’Aquila,Italy);FabioGraziosi(Universityofl’Aquila,Italy)SoftwareDefinedNetwork(SDN)architecturesdecouplecontrolandforwardingfunctionalitiesbyenablingthenetworkdevicestoberemotelyconfigurable/programmableruntimebyacontroller.Asadirectconsequenceidentifyinganaccuratemodelofanetworkandforwardingdevicesiscrucialinordertoapplyadvancedcontroltechniquestooptimizethenetworkperformance.AnenablingfactorinthisdirectionisgivenbyrecentresultsthatappropriatelycombineSystemIdentificationandMachineLearningtechniquestoobtainpredictivemodelsusinghistoricaldataretrievedfromanetwork.Inthispaperweproposeanovelmethodologytolearn,startingfromhistoricaldataandappropriatelycombiningARXidentificationwithRegressionTreesandRandomForests,anaccuratemodelofthedynamicalinput-outputbehaviorofanetworkdevicethatcanbedirectlyandefficientlyusedtooptimallyanddynamicallycontrolthebandwidthofthequeuesofswitchports,withintheSDNparadigm.WecompareourpredictivemodelwithNeuralNetworkpredictorsanddemonstratethebenefitsintermsofPacketLossesreductionandBandwidthsavingsintheMininetnetworkemulatorenvironment. AReal-timeQoS-Demand-AwareComputationalResourceSharingApproachinC-RANMojganBarahmanandLuisM.Correia(INESC-ID/INOV/IST,UniversityofLisbon);LúcioStuderFerreira(ISTEC/ULHTCOPELABS/INESC-ID,Lisbon)Thispaperpresentsadynamicresourcesharingapproachaimingatoptimizingcomputationalresourceperformanceofabasebandunit(BBU)poolinacloudradioaccessnetwork.Basedonthebargainingconceptingametheory,resourcesharingisformulatedasanoptimizationproblemconsideringqualityofservice,real-timedemandandtheminimumresourcesthatarerequiredtopreventBBUcrashes.TheperformanceoftheproposedmodelisevaluatedintermsofBBUfulfilmentlevel,resourceusageandefficiencyovertime.Simulationresults,forheterogeneousservicesinatidaltrafficenvironment,demonstratethattheproposedmodelallocatescomputationalresourcesinproportiontotheinstantaneousdemandofBBUsandthepriorityoftheongoingservices.Resultsalsoshowaminimum97%enhancementintheefficiencyofresourceallocationinoff-peakhours,comparedtofixedallocationstrategiesbasedonpeak-hourtrafficdemand. PredictingBandwidthUtilizationonNetworkLinksUsingMachineLearningMaximeLabonne(CEALIST&InstitutPolytechniquedeParis,France);CharalamposChatzinakis(CommunicatingSystemsLaboratoryCEA,France);AlexisOlivereau(CEA,LIST,France)Predictingthebandwidthutilizationonnetworklinkscanbeextremelyusefulfordetectingcongestioninordertocorrectthembeforetheyoccur.Inthispaper,wepresentasolutiontopredictthebandwidthutilizationbetweendifferentnetworklinkswithaveryhighaccuracy.Asimulatednetworkiscreatedtocollectdatarelatedtotheperformanceofthenetworklinksoneveryinterface.Thesedataareprocessedandexpandedwithfeatureengineeringinordertocreateatrainingset.Weevaluateandcomparethreetypesofmachinelearningalgorithms,namelyARIMA(AutoRegressiveIntegratedMovingAverage),MLP(MultiLayerPerceptron)andLSTM(LongShort-TermMemory),inordertopredictthefuturebandwidthconsumption.TheLSTMoutperformsARIMAandMLPwithveryaccuratepredictions,rarelyexceedinga3\%error(40\%forARIMAand20\%fortheMLP).WethenshowthattheproposedsolutioncanbeusedinrealtimewithareactionmanagedbyaSoftware-DefinedNetworking(SDN)platform. FairShareofLatencyinInter-Data-CenterBackboneNetworksNitinVaryaniandZhi-LiZhang(UniversityofMinnesota,USA)Theinter-data-centerbackbonenetworksinitiallycarriedbandwidth-intensivetrafficwhichdoesnothavestringentlatencyservice-level-objectives(SLOs).Fairallocationpolicieswereusedinsuchnetworkstoachieveequitabledistributionofbandwidthtotheflows.However,thesenetworkshavestartedcarryingtrafficthatissignificantlytiedtotheend-userexperienceandthushavestringentlatencySLOs.But,theliteraturelacksroutingalgorithmsforinter-data-centerbackbonenetworkswhichimposelatencySLOsonitstrafficinadditiontoachievingfairallocationofbandwidth.We,therefore,introduceaconceptcalled“fairshareoflatency”thatinvolvesroutingtrafficfordifferentflowssuchthattheviolationoflatencySLOsisminimum.Weproposealinear-programmingbasedroutingalgorithmforinter-data-centerbackbonenetworksthatincorporatesboth“fairshareoflatency”andfairallocationofbandwidth.Wealsointroducelatencyutilitycurvesthatdepicttheperceivedworthofdifferentlatenciestoanapplication.Simulationresultsonthetopologiesofinter-data-centernetworksofGoogle,Microsoft,Amazon,andIBMrevealthatourroutingalgorithmachievessignificantimprovementinmeetingthelatencySLOsofdifferenttrafficclasseswithaslightreductioninthefairnessofbandwidthallocation.AI-drivenZero-touchOperations,SecurityandTrustinMulti-operator5GNetworks:aConceptualArchitectureGinoCarrozzo(Nextworks,Italy);MuhammadShuaibSiddiqui(Fundaciói2CAT,InternetiInnovacióDigitalaCatalunya,Spain);AugustBetzler(i2CATFoundation,Spain);JoseBonnet(AlticeLabs,Portugal);GregorioMartinezPerez(UniversityofMurcia,Spain);AuroraRamos(Atos,Spain);TejasSubramanya(UniversityofTrento&FBKCREATE-NET,Italy)The5Gnetworksolutionscurrentlystandardisedanddeployeddonotyetenablethefullpotentialofpervasivenetworkingandcomputingenvisionedin5Ginitialvisions:networkservicesandsliceswithdifferentQoSprofilesdonotspanmultipleoperators;security,trustandautomationislimited.Theevolutionof5Gtowardsatrulyproduction-levelstageneedstoheavilyrelyonautomatedend-to-endnetworkoperations,useofdistributedArtificialIntelligence(AI)forcognitivenetworkorchestrationandmanagementandminimalmanualinterventions(zero-touchautomation).Alltheseelementsarekeytoimplementhighlypervasivenetworkinfrastructures.Moreover,DistributedLedgerTechnologies(DLT)canbeadoptedtoimplementdistributedsecurityandtrustthroughSmartContractsamongmultiplenon-trustedparties.Inthispaper,weproposeaninitialconceptofazero-touchsecurityandtrustarchitectureforubiquitouscomputingandconnectivityin5Gnetworks.Ourarchitectureaimsatcrossdomainsecurity&trustorchestrationmechanismsbycouplingDLTswithAI-drivenoperationsandservicelifecycleautomationinmulti-tenantandmulti-stakeholderenvironments.Threerepresentativeusecasesareidentifiedthroughwhichwewillvalidatetheworkwhichwillbevalidatedinthetestfacilitiesat5GBarcelonaand5TONIC/Madrid.