OffPiste QoSaware Routing Protocol By Yigal Eliaspur.
-
date post
19-Dec-2015 -
Category
Documents
-
view
220 -
download
0
Transcript of OffPiste QoSaware Routing Protocol By Yigal Eliaspur.
OffPisteQoSawareOffPisteQoSawareRoutingProtocolRoutingProtocol
ByYigalEliaspurByYigalEliaspur
ProblemBackgroundProblemBackground ThemainchallengeinQoSroutingistobeabletoThemainchallengeinQoSroutingistobeableto
respondtorespondtoonlineonlinerequestswithrequestswith ReasonableresponsetimeReasonableresponsetime Minimalnetworkoverhead(Messages,MemoryProcessingMinimalnetworkoverhead(Messages,MemoryProcessing
time).time). Minimalprobabilityofblocking(requestsfailure).Minimalprobabilityofblocking(requestsfailure).
AtypicalrequestistoreserveacertainresourcealongaAtypicalrequestistoreserveacertainresourcealongapathfromatransactionsourcetoatransactionpathfromatransactionsourcetoatransactiondestination.destination.
Theresourcemaybeanadditiveresource(e.g.delay)Theresourcemaybeanadditiveresource(e.g.delay)or/andanonadditiveresource(e.g.bandwidth).or/andanonadditiveresource(e.g.bandwidth).
TherequestmaybeappliedtoUnicastorMulticastTherequestmaybeappliedtoUnicastorMulticasttraffic.traffic.
AvailableSolutionsAvailableSolutions ThesolutionstodaycanbepartitionedtothreebroadThesolutionstodaycanbepartitionedtothreebroad
classes:classes: SourceroutingalgorithmsSourceroutingalgorithms
• Transformsadistributedproblemintoacentralizedone.Transformsadistributedproblemintoacentralizedone.• MaintainingacompleteglobalstateineachnodeofdynamicsMaintainingacompleteglobalstateineachnodeofdynamicsnetworkresources.networkresources.
DistributedroutingalgorithmsDistributedroutingalgorithms• ThepathcomputationisdistributedamongtheintermediatenodesThepathcomputationisdistributedamongtheintermediatenodesbetweenthesourceandthedestination.betweenthesourceandthedestination.
• Singlepathsearch–usuallyassumesglobalstateineachnode.Singlepathsearch–usuallyassumesglobalstateineachnode.• Multipathsearch(flooding)–usesonlylocalstateineachnode.Multipathsearch(flooding)–usesonlylocalstateineachnode.
HierarchicalroutingalgorithmsHierarchicalroutingalgorithms• EachnodemaintainsonlyapartialglobalstateEachnodemaintainsonlyapartialglobalstate• TocopewiththescalabilityproblemofglobalstateinlargeTocopewiththescalabilityproblemofglobalstateinlargeinternetworks.internetworks.
RelatedWorkRelatedWork OurOPsARprotocol(OffPisteQoSawareRouting)canbeclassifiedOurOPsARprotocol(OffPisteQoSawareRouting)canbeclassified
totheMultipath,totheMultipath,distributedroutingalgorithmsfamily.distributedroutingalgorithmsfamily. Othersworksrelatedtothisfamilyare:Othersworksrelatedtothisfamilyare:
Selective flooding Selective flooding –multipathsearchwillbedoneonlyonpre-–multipathsearchwillbedoneonlyonpre-computedroutes.computedroutes.
Ticket based probingTicket based probing-Everyprobing(search)messageissupposed-Everyprobing(search)messageissupposedtocarryatleastoneticket,andthusthetotalnumberofticketslimitthetocarryatleastoneticket,andthusthetotalnumberofticketslimitthemultipathsearch.multipathsearch.
QMRP and S-QMRPQMRP and S-QMRP(Scalable-DistributedQoSMulticastRouting(Scalable-DistributedQoSMulticastRoutingProtocol):Protocol):• Theunicastroutetowardsthedestinationischeckedfirst.Theunicastroutetowardsthedestinationischeckedfirst.• Ifthatfailedselectivescanningmechanismisapplied.Ifthatfailedselectivescanningmechanismisapplied.• ThescanningiscontrolledbyMaximumBranchingDegreeandMaximumThescanningiscontrolledbyMaximumBranchingDegreeandMaximum
BranchingLevelparameters.BranchingLevelparameters. QOSMicQOSMic-bestsuitedtoamulticastenvironment,sinceitlooksfora-bestsuitedtoamulticastenvironment,sinceitlooksfora
pointonamulticasttreeto``hook''onthenewreceiver.pointonamulticasttreeto``hook''onthenewreceiver.
TheOPsARProtocolTheOPsARProtocol OurmainOurmainmotivationmotivationofOPsARistoimprovetradeoffbetweentheofOPsARistoimprovetradeoffbetweenthe
overheadoftheprotocolandthesuccessratioitproduces.overheadoftheprotocolandthesuccessratioitproduces. InOPsAR,anodekeepstrackofrecentQoSmessagestolearnInOPsAR,anodekeepstrackofrecentQoSmessagestolearn
aboutresourceavailabilityaboutresourceavailabilitytoandfromtoandfromvarioustargetpoints.varioustargetpoints. Thelearningisreflectedinthenode'sThelearningisreflectedinthenode's``knowledge state''.``knowledge state''. EfficientpathselectionisbeendonebyleveragingontheEfficientpathselectionisbeendonebyleveragingonthe
knowledgestateatthenodes.knowledgestateatthenodes. TheOPsARprotocolisbuiltof3mainstages:TheOPsARprotocolisbuiltof3mainstages:
Try PhaseTry Phase–inwhichasinglepathsearchandreservationisapplied–inwhichasinglepathsearchandreservationisappliedfromthetransactionsourcetothetransactiondestination.fromthetransactionsourcetothetransactiondestination.
Scan PhaseScan Phase–inwhichamultipathsearchwithoutreservationis–inwhichamultipathsearchwithoutreservationisappliedfromthetransactiondestinationbacktothesource.appliedfromthetransactiondestinationbacktothesource.
Try 2 PhaseTry 2 Phase–inwhichthescanphaseresultsareevaluatedandthe–inwhichthescanphaseresultsareevaluatedandthebestcandidatepathisreservedfromthetransactionsourcetothebestcandidatepathisreservedfromthetransactionsourcetothedestination.destination.
TryPhaseTryPhase
ApathsearchfromthetransactionsourcetowardtheApathsearchfromthetransactionsourcetowardthetransactiontarget.transactiontarget.
TryphasefollowstheshortestpathaslongasithastheTryphasefollowstheshortestpathaslongasithastherequiredresources.requiredresources.
Thedeviationfromtheshortestpathtakesan``offpiste''Thedeviationfromtheshortestpathtakesan``offpiste''routethatleveragesontheroutethatleveragesontheknowledge stateknowledge statetooptimizetooptimizetheroutingprotocol.theroutingprotocol.
Itsdeviationfromtheshortestpathisbounded.Itsdeviationfromtheshortestpathisbounded. Ifresourcescannotbereservedwithinthatboundary,theIfresourcescannotbereservedwithinthatboundary,the
resourceswhichhavealreadybeenreservedareresourceswhichhavealreadybeenreservedarereleased,andarequestissenttothetransactiontargetreleased,andarequestissenttothetransactiontargettobegintheScanphase.tobegintheScanphase.
ScanPhaseScanPhase ThescanprocessisbasedonlimitedBreadthFirstSearch(BFS)ThescanprocessisbasedonlimitedBreadthFirstSearch(BFS)
fromthetransactiontargettowardthetransactionsource.fromthetransactiontargettowardthetransactionsource. WeneitherreserveresourcesintheScanphasenorkeepanystateWeneitherreserveresourcesintheScanphasenorkeepanystate
thatrelatestothespecificscan.thatrelatestothespecificscan. AsintheTryPhasethescanningprocesstakesadvantageoftheAsintheTryPhasethescanningprocesstakesadvantageofthe
knowledgestatetooptimizethesearch.knowledgestatetooptimizethesearch. Thebranchinglimitationisdoneby:Thebranchinglimitationisdoneby:
TicketingTicketingschemetoboundthetotalnumberofpaths.schemetoboundthetotalnumberofpaths. Maximum branching degree (MBD)Maximum branching degree (MBD)ateachnodeinordertoincreaseateachnodeinordertoincrease
thevarietyofpotentialpathstotraverse.thevarietyofpotentialpathstotraverse. Off-piste counterOff-piste countertolimitthedistancefromtheshortestpath,similartotolimitthedistancefromtheshortestpath,similarto
theTryphase.theTryphase. Branchisterminatedduringthescanningprocessif:Branchisterminatedduringthescanningprocessif:
OffpistelimitisreachedandtheunicastroutedoesnothavetheOffpistelimitisreachedandtheunicastroutedoesnothavetheresources.resources.
Orwhennooutgoinglinkhastherequestedresources.Orwhennooutgoinglinkhastherequestedresources.
Try2PhaseTry2Phase
IfthetransactionsourcereceivesseveralsuccessfulscanIfthetransactionsourcereceivesseveralsuccessfulscanmessages,itinitiatestheTry2phase.messages,itinitiatestheTry2phase.
Itchoosesthe``best''routefromthesuccessfulscanItchoosesthe``best''routefromthesuccessfulscanmessagesandaskstoreservetheresourcesalongthatmessagesandaskstoreservetheresourcesalongthatpath.path.
Ifreservationfailurealongtheexplicitrouteisdetected,Ifreservationfailurealongtheexplicitrouteisdetected, TheOPsARtriestoroutethereservationrequestTheOPsARtriestoroutethereservationrequest
messagetothetransactiondestinationusingalternativemessagetothetransactiondestinationusingalternativeroutesthattheoffpistemechanismoffers.routesthattheoffpistemechanismoffers.
IfthatfailsanackmessageisreturnedtothetransactionIfthatfailsanackmessageisreturnedtothetransactionsourceindicatingtheneedtochooseanotherexplicitsourceindicatingtheneedtochooseanotherexplicitroutefromthepreviousscanresult.routefromthepreviousscanresult.
KnowledgeState-definitionKnowledgeState-definition EachnodemaintainsEachnodemaintains
Alocalstateinwhichitholdsitslinks'statusandtheAlocalstateinwhichitholdsitslinks'statusandtheresourceavailabilityonthem.resourceavailabilityonthem.
AboundedlistofrecordsAboundedlistofrecords• <targetnodes,outgoinglink><targetnodes,outgoinglink>
ForeachrecordtheresourcesavailabilityisForeachrecordtheresourcesavailabilityismaintainedwithrespecttothatoutgoinglink:maintainedwithrespecttothatoutgoinglink:• MaxBWtowardthetargetnode.MaxBWtowardthetargetnode.• MaxBWfromthetargetnode.MaxBWfromthetargetnode.
ThisinformationisupdatedoccasionallyandisThisinformationisupdatedoccasionallyandismarkedtoidentifythetimeofitslastupdate.markedtoidentifythetimeofitslastupdate.
Thistimeisusedforagingmechanism.Thistimeisusedforagingmechanism.
KnowledgeState-usageKnowledgeState-usage AnyAnyOPsARprotocolmessagetraversinganodeisusedtoupdatetheknowledgestate(KS).OPsARprotocolmessagetraversinganodeisusedtoupdatetheknowledgestate(KS). EachOPsARprotocolmessageincludesthefollowingrelevantfields:EachOPsARprotocolmessageincludesthefollowingrelevantfields:
MaxBWToOriginMaxBWToOrigin MaxBWFromOriginMaxBWFromOrigin
TherearethreemainoperationtheKSisinvolvewith:TherearethreemainoperationtheKSisinvolvewith: KSrecordcreation/updateKSrecordcreation/update
JJ JJ
OPsARmessagefieldsupdateOPsARmessagefieldsupdate
JJ
RoutingdecisionRoutingdecision• Thechoiceismadeaccordingtotheresourceavailabilityalongthevariouslinkstowardthetarget,andaccordingtohowrecentThechoiceismadeaccordingtotheresourceavailabilityalongthevariouslinkstowardthetarget,andaccordingtohowrecent
thatinformationis.thatinformationis.• ThisisBasedonthreelevelsofoutgoinglinks(neighbors)maintainedpertargetnode:ThisisBasedonthreelevelsofoutgoinglinks(neighbors)maintainedpertargetnode:
FreshFresh StaleStale OldOld
KnowledgeState–RoutingKnowledgeState–Routingdecision(cont.)decision(cont.)
CandidateCandidate
NeighborsgroupNeighborsgroup
CandidateCandidate
NexthopNexthop
ResourceResourceinformationinputinformationinput
FreshFresh unicastunicast End2EndEnd2End
FreshFresh AllAll End2EndEnd2End
~Fresh~Fresh unicastunicast NextHopNextHop
StaleStale AllAll End2EndEnd2End
~Fresh~Fresh AllAll NextHopNextHop
AllAll AllAll NextHopNextHop
SimulationModelSimulationModel NS2simulatorNS2simulator PowerLawnetworktopologyPowerLawnetworktopology
Asthenodedegreeincreases,thenumberofnodeswiththatdegreeAsthenodedegreeincreases,thenumberofnodeswiththatdegreedecreasesexponentially.decreasesexponentially.
UsedthetopologygeneratordescribedinOsnat’swork(OntheUsedthetopologygeneratordescribedinOsnat’swork(Onthetomographyofnetworksandmulticasttrees)tomographyofnetworksandmulticasttrees)
ThegeneratorwasextendedtosupportBWallocation.ThegeneratorwasextendedtosupportBWallocation. ThebandwidthonthelinkswasuniformlydistributedfromThebandwidthonthelinkswasuniformlydistributedfrom
{10,34,45,100}Mb/s.{10,34,45,100}Mb/s. InordertomakesurethatthecongestionwouldfirstoccurinthecoreInordertomakesurethatthecongestionwouldfirstoccurinthecore
networkwereassignedthebandwidthoftheendpointsto1000Mb.networkwereassignedthebandwidthoftheendpointsto1000Mb. WealsoconductedtestswithhierarchicalbandwidthassignmentWealsoconductedtestswithhierarchicalbandwidthassignment
chosenfrom{10Mb,100Mb,1G,10G}bitspersecond.chosenfrom{10Mb,100Mb,1G,10G}bitspersecond.• Thisbackbone/metrotypeofoverprovisioningBWallocationshowedThisbackbone/metrotypeofoverprovisioningBWallocationshowed
almostnocongestionforBWreservationrequests.almostnocongestionforBWreservationrequests.• Therefore,thetopologiessimulatedwereonlylargeedgenetworksandISPTherefore,thetopologiessimulatedwereonlylargeedgenetworksandISP
likenetworks.likenetworks.
SimulationBasisSimulationBasis 600nodeswasused.600nodeswasused. Transactionendpointswerechosenoutof120edgenodes.Transactionendpointswerechosenoutof120edgenodes. Mostofthegraphsaretheresultof10,000transactionsperformedMostofthegraphsaretheresultof10,000transactionsperformed
onsixdifferentgeneratedtopologies.onsixdifferentgeneratedtopologies. Weraneachsimulationon5differentprotocoltypes:Weraneachsimulationon5differentprotocoltypes:
Traditional RSVPTraditional RSVP-AllocatestheQoSrequirementalongtheunicast-AllocatestheQoSrequirementalongtheunicastroutetowardthetransactiondestination;routetowardthetransactiondestination;
S QMRP*S QMRP*-IsS-QMRPadaptationtounicastrouting-IsS-QMRPadaptationtounicastrouting• Basicallyit’sthesameasOPsARbutwithoutKSandOff-PistecounterBasicallyit’sthesameasOPsARbutwithoutKSandOff-Pistecounter
support.support. S QMRP*DS QMRP*D–IsS-QMRP*withOff-Pistecountersupport.–IsS-QMRP*withOff-Pistecountersupport. OPsAROPsAR OPTOPT-ImplementedasaBFSwhichfindstheshortestpaththatfulfills-ImplementedasaBFSwhichfindstheshortestpaththatfulfills
thebandwidthQoSrequirements.thebandwidthQoSrequirements. Messageoverhead;Messageoverhead;
SimulationBasis(cont.)SimulationBasis(cont.)
WeperformedthefollowingsimulationandWeperformedthefollowingsimulationandevaluatetheirrelationshiptothereservationevaluatetheirrelationshiptothereservationsuccessratio:successratio: MemoryusageMemoryusage AmountofconcurrenttransactionsAmountofconcurrenttransactions NumberofedgenodesNumberofedgenodes NumberofdestinationnodesNumberofdestinationnodes TheCostandPerformanceGainofUsingTry&ScanTheCostandPerformanceGainofUsingTry&ScanPhasesPhases
GradualDeploymentwithinRSVPFrameworkGradualDeploymentwithinRSVPFramework
MemoryUsagevs.SuccessRatioMemoryUsagevs.SuccessRatio
MemoryUsagevs.SuccessRatioMemoryUsagevs.SuccessRatio(cont.)(cont.)
Theamountofmemorysufficienttoachieveabout85%Theamountofmemorysufficienttoachieveabout85%ofsuccessratioisveryreasonable.ofsuccessratioisveryreasonable.
ThememoryistheoreticallyboundedbyThememoryistheoreticallyboundedby Theoutdegreeofacorenode(orbytheagedoutthresholdTheoutdegreeofacorenode(orbytheagedoutthreshold
whichis9inourcase)whichis9inourcase) Timesthenumberofpossibletransactiondestinationnodes.Timesthenumberofpossibletransactiondestinationnodes.
InthelargestsimulationdonethistheoreticalnumberInthelargestsimulationdonethistheoreticalnumberwas160KB.was160KB.
Theaveragememoryconsumptionwasabout10%ofTheaveragememoryconsumptionwasabout10%ofthetheoreticalbound.thetheoreticalbound.
Where60KBwaslimit-theactualboundsetintheWhere60KBwaslimit-theactualboundsetinthesimulationcode.simulationcode.
ConcurrentTransactionsvs.ConcurrentTransactionsvs.SuccessRatioSuccessRatio
MessageOverheadvs.SuccessMessageOverheadvs.SuccessRatioRatio
MessageOverheadvs.SuccessMessageOverheadvs.SuccessRatio(cont.)Ratio(cont.)
Westudiedalltheparameter'spossiblecombinationwithinaspecificrangeWestudiedalltheparameter'spossiblecombinationwithinaspecificrange branchingdegreebranchingdegree scanningdeviationscanningdeviation andnumberoftickets.andnumberoftickets.
Eachsimulationresultgeneratedonepointinthegraph.Eachsimulationresultgeneratedonepointinthegraph. OPsARvs.S-QMRP*OPsARvs.S-QMRP*
Forthesameamountofmessageoverhead,theOPsARimprovesthesuccessForthesameamountofmessageoverhead,theOPsARimprovesthesuccessratioupto30%morethanSQMRP*.ratioupto30%morethanSQMRP*.
OPsARvs.RSVPOPsARvs.RSVP Increaseofoverheadbyfivesetimesyieldsaboutthreetimesthesuccessratio.Increaseofoverheadbyfivesetimesyieldsaboutthreetimesthesuccessratio. Anotherpointtoconsideristhattheaveragepathlengthisabout8hopswhenAnotherpointtoconsideristhattheaveragepathlengthisabout8hopswhen
deviationisallowedandthe4hopswhendeviationisforbidden(e.g.RSVP).deviationisallowedandthe4hopswhendeviationisforbidden(e.g.RSVP). OPsARvs.OPTOPsARvs.OPT
Theoverhead/sucessratioofOPTis20.6whiletheoverhead/sucessratiooftheTheoverhead/sucessratioofOPTis20.6whiletheoverhead/sucessratiooftheOPsARin200Kmessages,isin29.8whichisonly30%morethentheOPT.OPsARin200Kmessages,isin29.8whichisonly30%morethentheOPT.
NumberofEdgeNodesvs.NumberofEdgeNodesvs.SuccessRatioSuccessRatio
NumberofDestinationsNodesvs.NumberofDestinationsNodesvs.SuccessRatioSuccessRatio
NumberofDestinationsNodesvs.NumberofDestinationsNodesvs.SuccessRatio(cont.)SuccessRatio(cont.)
WeranthesimulationwithaconstantnumberofWeranthesimulationwithaconstantnumberof25%edgenodes(asopposedtothe20%we25%edgenodes(asopposedtothe20%weusuallyused).usuallyused).
ThenumberofcandidatedestinationnodesThenumberofcandidatedestinationnodesvariedfrom1%uptothewholesetofedgevariedfrom1%uptothewholesetofedgenodes(25%).nodes(25%).
ThecandidatesetofsourcenodeswasalwaysThecandidatesetofsourcenodeswasalwaysthewholesetofedgenodes.thewholesetofedgenodes.
OnlythelinksfromthosedestinationnodeswereOnlythelinksfromthosedestinationnodeswereassignedabandwidthcapacityof1000Mb.assignedabandwidthcapacityof1000Mb.
TheCostandPerformanceGainofTheCostandPerformanceGainofUsingTry&ScanPhasesUsingTry&ScanPhases
TheCostandPerformanceGainofTheCostandPerformanceGainofUsingTry&ScanPhases(cont.)UsingTry&ScanPhases(cont.)
ScanPhaseusesextratimeandScanPhaseusesextratimeandmessagesoverTryPhase.messagesoverTryPhase.
OursimulationsshowedthatthetimetoOursimulationsshowedthatthetimetocompleteaTryfollowedbyaScanisthreecompleteaTryfollowedbyaScanisthreetimesthetimeittakestocompletetheTrytimesthetimeittakestocompletetheTryphasealone.phasealone.
GradualDeploymentwithinRSVPGradualDeploymentwithinRSVPFrameworkFramework
GradualDeploymentwithinRSVPGradualDeploymentwithinRSVPFramework(cont.)Framework(cont.)
Atglance,thereisnoinherentlimitationintheprotocolAtglance,thereisnoinherentlimitationintheprotocolthatprohibitsitsuseinanincrementalmanner.thatprohibitsitsuseinanincrementalmanner.
The``RSVPonly''routerswereselectedbasedontheirThe``RSVPonly''routerswereselectedbasedontheirdistancefromtheacore.distancefromtheacore.
Theedgeroutershaveabetterchancetobechosenas``RSVPTheedgeroutershaveabetterchancetobechosenas``RSVPonly''routers.only''routers.
Fromthelearningmechanismperspective,theavailableFromthelearningmechanismperspective,theavailablecapacityofthelinksbetweens``RSVPonly''routersiscapacityofthelinksbetweens``RSVPonly''routersisignored.ignored.
FutureworkcanfocusondeploymentmethodsfortheFutureworkcanfocusondeploymentmethodsfortheOPsARprotocolthatwillmaintainthegainobtainedfromOPsARprotocolthatwillmaintainthegainobtainedfromthelearningmechanism.thelearningmechanism.
FutureWorkFutureWork MachinelearningimprovingMachinelearningimproving
TheoverallschemeofourprotocolisanintelligentthechoiceofroutesfromafullBreadthFirstSearchTheoverallschemeofourprotocolisanintelligentthechoiceofroutesfromafullBreadthFirstSearchalgorithm(BFS).algorithm(BFS).
FutureonresearchcanfocusonimprovingtheeducatedchoiceofrouteswhilelimitingtheoverheadinFutureonresearchcanfocusonimprovingtheeducatedchoiceofrouteswhilelimitingtheoverheadinmemory.memory.
Weexpecttofindwaystousemachinelearningtechniquestoachievethatgoal.Weexpecttofindwaystousemachinelearningtechniquestoachievethatgoal. KSAggregationsKSAggregations
Saveinmemorybyaggregatingtheinformation,usingtechniqueslikelongestprefixmatchingonSaveinmemorybyaggregatingtheinformation,usingtechniqueslikelongestprefixmatchingontransactionsdestination.transactionsdestination.
Packetlosses,link/nodefailuresPacketlosses,link/nodefailures Shouldberelativelyeasyusingtimersandretriesformessages,andusingsoftstatereservation.Shouldberelativelyeasyusingtimersandretriesformessages,andusingsoftstatereservation.
AdditiveresourcesAdditiveresources Handlingadditiveresources,likedelay,requiresminorchangesdtotheprotocolspresented.Handlingadditiveresources,likedelay,requiresminorchangesdtotheprotocolspresented.
TuningtheKSparametersTuningtheKSparameters LinearincreasingthefreshneighborgroupdidnotincreasetheperformanceandsometimescauseittobeLinearincreasingthefreshneighborgroupdidnotincreasetheperformanceandsometimescauseittobe
degraded.degraded. Increasingtheageoutthreshold–doesnotimprovedtheperformanceeithereventhoughitincreasetheIncreasingtheageoutthreshold–doesnotimprovedtheperformanceeithereventhoughitincreasethe
totalmemoryrequirements.totalmemoryrequirements. FurtherresearchmustbeconductedinordertoexploretheinterdependenciesamongthevariousvariablesFurtherresearchmustbeconductedinordertoexploretheinterdependenciesamongthevariousvariables
ofOPsAR,andtoautomaticallylearnandchoosetheoptimalvalues,possiblylusingmachinelearningofOPsAR,andtoautomaticallylearnandchoosetheoptimalvalues,possiblylusingmachinelearningtechniques.techniques.