torontoedu Yashar Ganjali Department of Computer Science University of Toronto Toronto Canada yganjalicstorontoedu ABSTRACT Simplicity is a prominent advantage of SoftwareDe64257ned Networking SDN and is often exempli64257ed by implement ing a compli ID: 49527
Download Pdf The PPT/PDF document "Beehive Towards a Simple Abstraction for..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
HotNets-XIII,October27Ð28,2014,LosAngeles,CA,USA.Copyright2014ACM978-1-4503-3256-9/14/10...$ 1 Collectuponreceiv-ingStatReplys,andRouteevery1second.Toinvokeafunction,oneexplicitlyrequestsallorsomeentriesinstatedictionaries(denotedbywith),orinvokesthefunctionforeachentry(usingforeach).1app Collect(reply,mEntry)22onTimeOut(1sec):23withSandT:24Route(S,T)//Detailsonhandlingswitchandlinkdiscoveryisomittedforbrevity.Figure2:ThissimpleTra!cEngineeringappli-cationstoresßowstatsandthenetworktopology InournaiveTEexample,QueryandCollectaccesstheßowstatisticsdata(i.e.,theSdictionary)onaperswitchbasis,whereasRouteaccessesthewholedictionary.IftherewasnoRoute,theßowstatisticsdictionarycouldbedistributedamongcontrollersbyassigningtheßowstatisticsofeachswitch(i.e.,anentryinS)toonecontroller.Insuchasetting,QueryqueriesaswitchononecontrollerandCollectupdatestheßowstatisticsofthatswitchonthesamecontroller.Havingsaidthat,sinceRouterequiresthewholedictionary,wehavetocollocateallkeysonthesamecontroller.ThisessentiallymeansthatRouteandanyfunctionthatsharesstatewith val).Forinstance,ourTEapplication(Figure2)hasacellfortheßowstatisticsofeachswitchSWiintheformof(S,SWi,StatSWi behaviorisidenticaltowhentheyaredeployedonacentralizedcontroller,eventhoughtheymightbephys-icallydistributedoverdi!erentcontrollers(i.e.,hives). ,...,(Dn,Kn)}.Wecallthisset,the AsshowninFigure3,topreserveconsistency,Bee-hiveguaranteesthatallmessageswith respectivelybyacon-trolapplication.Sincethesetwomessagessharethe essentiallyindicatesthecorrelationofeachswitchtoeachvirtualnetwork.Wealsostoreprovenanceandcausationdataformessages.Forexample,westorethatpacketoutmessagesareemittedbythelearningswitchapplicationuponreceiving80%ofpacketinÕs.Wemeasureruntimemetricsoneachhivelocally,andperiodicallyaggregatethemonasinglehive.ThismergedinstrumentationdataisfurtherusedtoÞndtheoptimalplacementofbeesandisalsoutilizedforapplicationanalytics.Weimplementedthismechanismusingtheproposedabstractionasacontrolapplication.OnOptimalPlacement.FindingtheoptimumplacementofbeesisNP-Hard,asthefacilitylocationproblemcanbereducedtoit.Inadditiontothecomputationalcomplexities,therearelotsofsubtleperformanceside-e! WemigrateB1runningon earlierinSection3,forsuchafunction,Beehiveguaranteesthatthewholestate,i.e.,allcellsofthatapplication,areassignedtoonebee.Itisimportanttonotethat,sinceapplicationsdonotsharestate,theplatformmayplacedi!erentcentralizedapplicationsondi!erenthivestosatisfyextensiveresourcerequirements(e.g.,alargestate).Kandoo.Attheotherendofthespectrum,therearelocalcontrolapplicationsproposedinKandoo[]thatusethelocalstateofasingleswitchtoprocessfrequentevents.ThefunctionsofalocalcontrolapplicationuseswitchIDsasthekeysintheirstatedictionariesand,tohandlemessages,accesstheirstateusingasinglekey.Assuch,Beehiveconceivesacellforeachswitchandallocatesonebeeforeachcell.Inpractice,thisresultsinlocalfunctionsbeingreplicatedonallcontrollerstohandleswitcheslocaltothatcontroller. 10 ,thislivemigrationofcontrolplanefunctionslocalizesmessageprocessingandresultsinconsiderableimprovementincontrolchannelconsumption.ThelargestspikeinFig-ure4fcorrelatestoreplicatingcellstotheotherhives.Notethatthisisalldoneautomaticallyatruntimewithnomanualinterventionand,afteroptimization, ofcontrolapplicationsmanagingthenetworkasacohesivewhole.Forthemostpart,theseapplications 7.CONCLUSIONInthispaper,wehavepresentedaframeworkthatsimpliÞesthedesignandimplementationprocessof ,pages7Ð12,2012.[9]C.Kim,M.Caesar,andJ.Rexford.Floodlessinseattle:Ascalableethernetarchitectureforlargeenterprises.InProceedingsoftheSIGCOMMÕ08,pages3Ð14,2008.[10]T.Koponen,K.Amidon,P.Balland,M.Casado,A.Chanda,B.Fulton,I.Ganichev,J.Gross,N.Gude,P.Ingram,E.Jackson,A.Lambeth,R.Lenglet,S.-H.Li,A.Padmanabhan,J.Pettit,B.Pfa!,R.Ramanathan,S.Shenker,A.Shieh,J.Stribling,P.Thakkar,D.Wendlandt,A.Yip,andR.Zhang.Networkvirtualizationinmulti-tenantdatacenters.In