思科數(shù)據(jù)中心架構(gòu)25代設(shè)計(jì)方案英文guide

CCNA|CCNPCCNA|CCNP歸原作者所有本資料只供試讀資源請攻城獅 CiscoDataCenterInfrastructure2.5DesignGuideCiscoValidatedDesignMay14,2008Americas170WestTasmanDrive:408526-800553-NETSFax:408527-TextPartNumber:OL-11565-歸原作者所有本資料只供試讀資源請攻城獅 CiscoValidatedTheCiscoValidatedDesignProgramconsistsofsystemsandsolutionsdesigned,tested,and edtofacilitatefaster,morereliable,andmorepredictablecustomerdeployments.Formoreinformationvisit ALLDESIGNS,SPECIFICATIONS,STATEMENTS,INFORMATION,AND MENDATIONS(COLLECTIVELY,"DESIGNS")INTHISMANUALAREPRESENTED"ASIS,"WITHALLFAULTS.CISCOANDITS RSDI WARRANTIES,INCLUDING,WITHOUTLIMITATION,THEWARRANTYOFMERCHANTABILITY,FITNESSFORAPARTICULARPURPOSEANDNONINFRINGEMENTORARISINGFROMACOURSEOFDEALING,USAGE,ORTRADEPRACTICE.INNOEVENTSHALLCISCOORITS RSBELIABLEFORANYINDIRECT,SPECIAL,CONSEQUENTIAL,ORALDAMAGES,INCLUDING,WITHOUTLIMITATION,LOSTPROFITSORLOSSORDAMAGETODATAARISINGOUTOFTHEUSEORINABILITYTOUSETHEDESIGNS,EVENIFCISCOORITS HAVEBEENADVISEDOFTHEPOSSIBILITYOFSUCHDAMAGES.THEDESIGNSARESUBJECTTOCHANGEWITHOUTNOTICE.USERSARESOLELYRESPONSIBLEFORTHEIRAPPLICATIONOFTHEDESIGNS.THEDESIGNSDONOTCONSTITUTETHETECHNICALOROTHERPROFESSIONALADVICEOFCISCO,ITS RSORPARTNERS.USERSSHOULDCONSULTTHEIROWNTECHNICALADVISORSBEFOREIMPLEMENTINGTHEDESIGNS.RESULTSMAYVARYDEPENDINGONFACTORSNOTTESTEDBYCISCO.CCVP,theCiscoLogo,andtheCiscoSquareBridgelogoaretrademarksofCiscoSystems,Inc.;ChangingtheWayWeWork,Live,y,andLearnisaservicemarkofCiscoSystems,Inc.;andAccessRegistrar,Aironet,BPX,Catalyst,CCDA,CCDP,CCIE,CCNA,CCNP,CCSP,Cisco,theCiscoCertifiedInternetworkExpertlogo,CiscoIOS,CiscoPress,CiscoSystems,CiscoSystemsCapital,theCiscoSystemslogo,CiscoUnity,Enterprise/Solver,EtherChannel,EtherFast,EtherSwitch,FastStep,FollowMeBrowsing,FormShare,GigaDrive,GigaStack,HomeLink,InternetQuotient,IOS, ,IP/TV,iQExpertise,theiQlogo,iQNetReadinessScorecard,iQuickStudy,LightStream,Linksys,Meeting ce,MGX,NetworkingAcademy,NetworkRegistrar,Packet,PIX,ProConnect,RateMUX,ScriptShare,SlideCast,SMARTnet,StackWise,TheFastestWaytoIncreaseYourInternetQuotient,andTransPathareregisteredtrademarksofCiscoSystems,Inc.and/oritsaffiliatesintheUnitedStatesandcertainothercountries.Allothertrademarksmentionedinthis orWebsitearethepropertyoftheirrespectiveowners.TheuseofthewordpartnerdoesnotimplyapartnershiprelationshipbetweenCiscoandanyother.(0612R)CiscoDataCenterInfrastructure2.5Design?2007CiscoSystems,Inc.Allrights OL-11565- OL-11565-CCNA|CCNP歸原作者所有本資料只供試 資源 攻城獅 CONTENTCHAPTER DataCenterArchitecture 1-DataCenterArchitecture 1- Multi-TierModel ServerClusterModel HPCClusterTypesand 1- 1- 1-CHAPTER DataCenterMulti-TierModel 2-DataCenterMulti-TierDesign 2- 2- tformand 2-Distributed 2-TrafficFlowintheDataCenter 2-DataCenterAggregation tformsand 2-Distributed 2- 2-PathSelectioninthePresenceofServiceModules 2-8ServerFarmTrafficFlowwithServiceModules ServerFarmTrafficFlowwithoutServiceModules Layer2FaultSize2-12SpanningTreeScalability2-1310GigEDensity2-13DefaultGatewayRedundancywithHSRP2-DataCenterAccessLayer2-mendedtformsandModules2-DistributedForwarding2-SharingServicesattheAggregationLayer2-DataCenterServicesLayer2-CiscoDataCenterInfrastructure2.5Design OL-11565- OL-11565-CCNA|CCNP

歸原作者所有本資料只供試讀資源請攻城/ tformsand 2-/Performance 2-TrafficFlowthroughtheService 2- 2-CHAPTER ServerClusterDesignswith 3-Technical 3- 3-Catalyst6500SystemBandwidth EqualCostMulti-PathRouting RedundancyintheServerClusterDesign 3-6ServerClusterDesign—Two-TierModel 3-2-WayECMPDesignwith1RU ServerClusterDesign—Three-Tier 3-mendedHardwareand 3-CHAPTER DataCenterDesign FactorsthatInfluence 4-WhyImplementaDataCenterCoreLayer? WhyUsetheThree-TierDataCenterDesign? WhyDeployServicesSwitch? Determiningum 4- umNumberof 4-Server NIC 4-Pervasive 4-ServerConsolidation TopofRackSwitching BladeServers Importanceof 4-CHAPTER SpanningTree 5-ExtendingVLANsintheData 5- 5- 5-CiscoDataCenterInfrastructure2.5Design OL-11565- OL-11565-CCNA|CCNP/歸原作者所有本資料只供試讀資源請攻城/CalculatingVirtualPortsperLine 5-StepstoResolveLogicalPortCount 5-CHAPTER DataCenterAccessLayer 6-OverviewofAccessLayerDesign 6-ServiceModuleInfluenceon 6-ServiceModule/ApplianceandPath 6- 6-Layer2LoopedAccessLayerModel Layer2LoopedAccessTopologies TriangleLoopedTopology SpanningTree,HSRP,andServiceModule 6-Failure 6-SquareLooped SpanningTree,HSRP,andServiceModule 6-Failure 6-Layer2 AccessLayer 6-Layer2 Access 6-Layer2 U 6-SpanningTree,HSRP,andServiceModule 6-Failure 6-Layer2 InvertedU 6-SpanningTree,HSRP,andServiceModule 6-Failure 6-FlexLinksAccess SpanningTree,HSRP,andServiceModuleDesign ImplicationsRelatedtoPossibleLoopConditions FailureScenarios UsingEtherChannelMin- 6-CHAPTER IncreasingHAintheData 7- 7-Aggregation1CSM Aggregation1OSPFandRouteMapConfigurations AggregationInter-switchLinkConfiguration ServiceModuleFT 7- 7-Possible 7-CiscoDataCenterInfrastructure2.5Design OL-11565- OL-11565-CCNA|CCNP

歸原作者所有本資料只供試讀資源請攻城/ 7-/ 7-SlotUsageversusImproved 7-CHAPTER Configuration IntegratedServicesDesign 8-CoreSwitch 8-AggregationSwitch 8-CoreSwitch AggregationSwitch 8-AccessSwitch4948- 8-AccessSwitch4948- 8-AccessSwitch6500- 8- ServicesSwitchDesign 8-CoreSwitch CoreSwitch DistributionSwitch 8-DistributionSwitch 8-ServiceSwitch 8-ServiceSwitch 8- 8-ACEand 8-FWSM 8-ACE 8-FWSM 8-ACE Additional 8-CiscoDataCenterInfrastructure2.5Design OL-11565- OL-11565-CCNA|CCNP歸原作者所有本資料只供試 資源 攻城獅 CHAPTERDataCenterArchitectureThischapterisanoverviewofprovenCiscosolutionsforprovidingarchitecturedesignsintheenterprisedatacenter,andincludesthefollowingtopics:DataCenterArchitectureDataCenterDesignDataCenterArchitectureThedatacenterishometothecomputationalpower,storage,andapplicationsnecessarytosupportanenterprisebusiness.ThedatacenterinfrastructureiscentraltotheITarchitecture,fromwhichallcontentissourcedorpassesthrough.Propernningofthedatacenterinfrastructuredesigniscritical,andperformance,resiliency,andscalabilityneedtobecarefullyconsidered.Anotherimportantaspectofthedatacenterdesignisflexibilityinquicklydeployingandsupportingnewservices.Designingaflexiblearchitecturethathastheabilitytosupportnewapplicationsinashorttimeframecanresultinasignificantcompetitiveadvantage.Suchadesignrequiressolidinitialnningandthoughtfulconsiderationintheareasofportdensity,accesslayeruplinkbandwidth,trueservercapacity,andoversubscription,tonamejustafew.Thedatacenternetworkdesignisbasedonaprovenlayeredapproach,whichhasbeentestedandimprovedoverthepastseveralyearsinsomeofthelargestdatacenterimplementationsintheworld.Thelayeredapproachisthebasicfoundationofthedatacenterdesignthatseekstoimprovescalability,performance,flexibility,resiliency,andmaintenance.Figure1-1showsthebasiclayereddesign.CiscoDataCenterInfrastructure2.5Design1- OL-11565-1- OL-11565-CCNA|CCNPDataCenter Campus10GigabitEthernetGigabitEthernetorCampus10GigabitEthernetGigabitEthernetorThelayersofthedatacenterdesignarethecore,aggregation,andaccesslayers.Theselayersarereferredtoextensivelythroughoutthisguideandarebrieflydescribedasfollows:Corelayer—Providesthehigh-speedpacketswitchingbackneforallflowsgoinginandoutofthedatacenter.ThecorelayerprovidesconnectivitytomultipleaggregationmodulesandprovidesaresilientLayer3routedfabricwithnosinglepointoffailure.Thecorelayerrunsaninteriorroutingprotocol,suchasOSPForEIGRP,andloadbalancestrafficbetweenthecampuscoreandaggregationlayersusingCiscoExpressForwarding-basedhashingalgorithms.Aggregationlayermodules—Provideimportantfunctions,suchasservicemoduleintegration,Layer2 definitions,spanningtreeprocessing,anddefaultgatewayredundancy.Server-to-servermulti-tiertrafficflowsthroughtheaggregationlayerandcanuseservices,suchasfirewallandserverloadbalancing,tooptimizeandsecureapplications.ThesmallericonswithintheaggregationlayerswitchinFigure1-1representtheintegratedservicemodules.Thesemodulesprovideservices,suchascontentswitching,firewall,SSLoffload,intrusiondetection,networkysis,andAccesslayer—Wheretheserversphysicallyattachtothenetwork.Theservercomponentsconsistof1RUservers,bladeserverswithintegralswitches,bladeserverswithpass-throughcabling,clusteredservers,andmainframeswithOSAadapters.Theaccesslayernetworkinfrastructureconsistsofmodularswitches,fixedconfiguration1or2RUswitches,andintegralbladeserverswitches.SwitchesprovidebothLayer2andLayer3topologies,fulfillingthevariousserverbroadcast administrativerequirements.CiscoDataCenterInfrastructure2.5Design OL-11565- OL-11565-CCNA|CCNPChapter Thischapterdefinestheframeworkonwhichthe mendeddatacenterarchitectureisbasedandintroducestheprimarydatacenterdesignmodels:themulti-tierandserverclustermodels.DataCenterDesignThemulti-tiermodelisthemostcommondesignintheenterprise.Itisbasedontheweb,application,anddatabaselayereddesignsupportingcommerceandenterprisebusinessERPand solutions.Thistypeofdesignsupportsmanywebservicearchitectures,suchasthosebasedon .NETorJava2EnterpriseEdition.ThesewebserviceapplicationenvironmentsareusedbyERPand fromSiebelandOracle,tonameafew.Themulti-tiermodelreliesonsecurityandapplicationoptimizationservicestobeprovidedinthenetwork.Theserverclustermodelhasgrownoutoftheuniversityandscientificcommunitytoemergeacrossenterprisebusinessverticalsincludingfinancial,manufacturing,andentertainment.Theserverclustermodelismostcommonlyassociatedwithhigh-performancecomputing(HPC),parallelcomputing,andhigh-throughputcomputing(HTC)environments,butcanalsobeassociatedwithgrid/utilitycomputing.Thesedesignsaretypicallybasedoncustomized,andsometimesproprietary,applicationarchitecturesthatarebuilttoserveparticularbusinessobjectives.Chapter2,“DataCenterMulti-TierModelDesign,”providesanoverviewofthemulti-tiermodel,andChapter3,“ServerClusterDesignswithEthernet,”providesanoverviewoftheserverclustermodel.Laterchaptersofthisguideaddressthedesignaspectsofthesemodelsingreaterdetail.Multi-TierThemulti-tierdatacentermodelisdominatedbyHTTP-basedapplicationsinamulti-tierapproach.Themulti-tierapproachincludesweb,application,anddatabasetiersofservers.Today,mostweb-basedapplicationsarebuiltasmulti-tierapplications.Themulti-tiermodelusessoftwarethatrunsasseparateprocessesonthesamemachineusinginterprocesscommunication(IPC),orondifferentmachineswithcommunicationsoverthenetwork.Typically,thefollowingthreetiersareused:Web-Multi-tierserverfarmsbuiltwithprocessesrunningonseparatemachinescanprovideimprovedresiliencyandsecurity.Resiliencyisimprovedbecauseaservercanbetakenoutofservicewhilethesamefunctionisstillprovidedbyanotherserverbelongingtothesameapplicationtier.Securityisimprovedbecauseanattackercancompromiseawebserverwithoutgainingaccesstotheapplicationordatabaseservers.Webandapplicationserverscancoexistonacommonphysicalserver;thedatabasetypicallyremainsseparate.CiscoDataCenterInfrastructure2.5Design1- OL-11565-1- OL-11565-CCNA|CCNPDataCenterDataCenterResiliencyisachievedbyloadbalancingthenetworktrafficbetweenthetiers,andsecurityisachievedbycingfirewallsbetweenthetiers.Youcanachievesegregationbetweenthetiersbydeployingaseparateinfrastructurecomposedofaggregationandaccessswitches,orbyusingVLANs(seeFigure1-Figure1- PhysicalSegregationinaServerFarmwithAppliances(A)andServiceModules WebWebWebThedesignshowninFigure1-3usesVLANstosegregatetheserverfarms.Theleftsideoftheillustration(A)showsthephysicaltopology,andtherightside(B)showstheVLANallocationacrosstheservicemodules,firewall,loadbalancer,andswitch.Thefirewallandloadbalancer,whichareVLAN-aware,en theVLANsegregationbetweentheserverfarms.NotethatnotalloftheVLANsrequireloadbalancing.Forexample,thedatabaseintheexamplesendstrafficdirectlytoWebCiscoDataCenterInfrastructure2.5Design OL-11565- OL-11565-CCNA|CCNPChapter Figure1- LogicalSegregationinaServerFarmwith WebWebPhysicalsegregationimprovesperformancebecauseeachtierofserversisconnectedtodedicatedhardware.TheadvantageofusinglogicalsegregationwithVLANsisthereducedcomplexityoftheserverfarm.Thechoiceofphysicalsegregationorlogicalsegregationdependsonyourspecificnetworkperformancerequirementsandtrafficpatterns.Businesssecurityandperformancerequirementscaninfluencethesecuritydesignandmechanismsused.Forexample,theuseofwire-speedACLsmightbepreferredovertheuseofphysicalfirewalls.Non-intrusivesecuritydevicesthatprovidedetectionandcorrelation,suchastheCiscoMonitoring,ysis,andResponseSystem(MARS)combinedwithRouteTriggeredBlackHoles(RTBH)andCiscoIntrusionProtectionSystem(IPS)mightmeetsecurityrequirements.CiscoGuardcanalsobedeployedasaprimarydefenseagainstdistributeddenialofservice(DDoS)attacks.Formoredetailsonsecuritydesigninthedatacenter,refertotheServerFarmSecuritySRNDatthefollowingURL: ServerClusterInthemoderndatacenterenvironment,clustersofserversareusedformanypurposes,includinghighavailability,loadbalancing,andincreasedcomputationalpower.Thisguidefocusesonthehighperformanceformofclusters,whichincludesmanyforms.AllclustershavethecommongoalofcombiningmultipleCPUstoappearasaunifiedhighperformancesystemusingspecialsoftwareandhigh-speednetworkinterconnects.Serverclustershavehistoricallybeenassociatedwithuniversityresearch,scientificlaboratories,andmilitaryresearchforuniqueapplications,suchasthefollowing:Meteorology(weatherSeismology Militaryresearch(weapons,CiscoDataCenterInfrastructure2.5Design1- OL-11565-1- OL-11565-CCNA|CCNPDataCenterDataCenterServerclustersarenowintheenterprisebecausethebenefitsofclusteringtechnologyarenowbeingappliedtoabroaderrangeofapplications.ThefollowingapplicationsintheenterprisearedrivingthisFinancial ysis—Real-timebond ysisandhistoricalFilmanimation—Rendering multi-gigabyteManufacturing—AutomotivedesignmodelingandSearchengines—QuickparallellookuppluscontentIntheenterprise,developersareincreasinglyrequestinghigherbandwidthandlowerlatencyforagrowingnumberofapplications.Thetime-to-marketimplicationsrelatedtotheseapplicationscanresultinatremendouscompetitiveadvantage.Forexample,theclusterperformancecandirectlyaffectgettingafilmtomarketfortheholidayseasonorprovidingfinancialmanagementcustomerswithhistoricaltrendinginformationduringamarketshift.HPCClusterTypesandInthehighperformancecomputinglandscape,variousHPCclustertypesexistandvariousinterconnectareused.Thetop500 puterlistat providesafairlycomprehensiveviewofthislandscape.ThemajorityofinterconnectusedtodayarebasedonFastEthernetandGigabitEthernet,butagrowingnumberofspecialtyinterconnectsexist,forexampleincludingInfinibandandMyrinet.SpecialtyinterconnectssuchasInfinibandhaveverylowlatencyandhighbandwidthswitchingcharacteristicswhencomparedtotraditionalEthernet,andleveragebuilt-insupportforRemoteDirectMemoryAccess(RDMA).10GENICshavealsorecentlyemergedthatintroduceTCP/IPoffloadenginesthatprovidesimilarperformancetoInfiniband.TheCiscoSFSlineofInfinibandswitchesandHostChannelAdapters(HCAs)providehighperformancecomputingsolutionsthatmeetthehighestdemands.FormoreinformationonInfinibandandHighPerformanceComputing,refertothefollowingURL: TheremainderofthischapterandtheinformationinChapter3,“ServerClusterDesignswithEthernet”focusonlargeclusterdesignsthatuseEthernetastheinterconnecttechnology.Althoughhighperformanceclusters(HPCs)comeinvarioustypesandsizes,thefollowingcategorizesthreemaintypesthatexistintheenterpriseenvironment:HPCtype1—Parallelmessagepassing(alsoknownastightlyApplicationsrunonallcomputenodessimultaneouslyinAmasternodedeterminesinputprocessingforeachcomputeCanbealargeorsmallcluster,brokendownintohives(forexample,1000serversover20hives)withIPCcommunicationbetweencomputenodes/hives.HPCtype2—DistributedI/Oprocessing(forexample,search requestisbalancedacrossmasternodes,thensprayedtocomputenodesforparallelprocessing(typicallyunicastatpresent,withamovetowardsmulticast).Thistypeobtainsthequickestresponse,appliescontentinsertion(advertising),andsendsto.CiscoDataCenterInfrastructure2.5Design OL-11565- OL-11565-CCNA|CCNPChapter HPCType3—Parallelfileprocessing(alsoknownaslooselyThesourcedatafileisdividedupanddistributedacrossthecomputepoolformanipulationinparallel.Processedcomponentsarerejoinedaftercompletionandwrittentostorage.Middlewarecontrolsthejobmanagementprocess(forexample,tformlinearfilesystemThetraditionalhighperformancecomputingclusterthatemergedoutoftheuniversityandmilitaryenvironmentswasbasedonthetype1cluster.ThenewenterpriseHPCapplicationsaremorealignedwithHPCtypes2and3,supportingtheentertainment,financial,andagrowingnumberofotherverticalFigure1-4showsthecurrentserverclusterFigure1- ServerCluster+

HPC1–ParallelMessagePassingHPC2–DistributedI/OHPC3–ParallelFileProcessingDB–DataBaseClusterAPP–ApplicationBandwidthBandwidth

LB–LoadBalancingClusterSC–StretchedClustering

HPCToday:Mainlyconsistsofcustom+LatencyThefollowingsectionprovidesageneraloverviewoftheserverclustercomponentsandtheirpurpose,whichhelpsinunderstandingthedesignobjectivesdescribedinChapter3,“ServerClusterDesignswithCiscoDataCenterInfrastructure2.5Design1- OL-11565-1- OL-11565-CCNA|CCNPDataCenterDataCenterFigure1-5showsalogicalviewofaserverFigure1- LogicalViewofaServer(front(front(backCommonFSmaybeconnectedviaEthernetorSANStorageFC(orLogicalThecomponentsoftheserverclusterareasFrontend—Theseinterfacesareusedforexternalaccesstothecluster,whichcanbeaccessedbyapplicationserversorusersthataresubmittingjobsorretrievingjobresultsfromthecluster.Anexampleisan whoissubmittingafileforrenderingorretrievinganalreadyrenderedresult.ThisistypicallyanEthernetIPinterfaceconnectedintotheaccesslayeroftheexistingserverfarmMasternodes(alsoknownasheadnode)—Themasternodesareresponsibleformanagingthecomputenodesintheclusterandoptimizingtheoverallcomputecapacity.Usually,themasternodeistheonlynodethatcommunicateswiththeoutsideworld.Clusteringmiddlewarerunningonthemasternodesprovidesthetoolsforresourcemanagement,jobscheduling,andnodestatemonitoringofthecomputernodesinthecluster.Masternodesaretypicallydeployedinaredundantfashionandareusuallyahigherperformingserverthanthecomputenodes.Back-endhigh-speedfabric—Thishigh-speedfabricistheprimarymediumformasternodetocomputenodeand putenodecommunications.Typicalrequirementsincludelowlatencyandhighbandwidthandcanalsoincludejumboframeand10GigEsupport.GigabitEthernetisthemostpopularfabrictechnologyinusetodayforserverclusterimplementations,butothershowpromise,particularlyInfiniband.CiscoDataCenterInfrastructure2.5Design OL-11565- OL-11565-CCNA|CCNPChapter Computenodes—ThecomputenoderunsanoptimizedorfullOSkernelandisprimarilyresponsibleforCPU-intenseoperationssuchasnumbercrunching,rendering,compiling,orotherfilemanipulation.Storagepath—ThestoragepathcanuseEthernetorFibreChannelinterfaces.FibreChannelinterfacesconsistof1/2/4GinterfacesandusuallyconnectintoaSANswitchsuchasaCiscoMDStform.Theback-endhigh-speedfabricandstoragepathcanalsobeacommontransportmediumwhenIPoverEthernetisusedtoaccessstorage.Typically,thisisforNFSoriSCSIprotocolstoaNASorSANgateway,suchastheIPSmoduleonaCiscoMDStform.Commonfilesystem—Theserverclusterusesacommonparallelfilesystemthatallowshighperformanceaccesstoallcomputenodes.Thefilesystemtypesvarybyoperatingsystem(forexample,PVFSorLustre).PhysicalServerclusterdesignscanvarysignificantlyfromonetoanother,butcertainitemsarecommon,suchasthefollowing:CommodityofftheShelf(CotS)serverhardware—Themajorityofserverclusterimplementationsarebasedon1RUIn -orAMD-basedserverswithsingle/dualprocessors.Thespiralingcostofthesehighperforming32/64-bitlowdensityservershascontributedtotherecententerpriseadoptionofclustertechnology.GigEor10GigENICcards—Theapplicationsinaserverclustercanbebandwidthintensiveandhavethecapabilitytoburstatahighratewhennecessary.ThePCI-XorPCI-ExpressNICcardsprovideahigh-speedtransferbusspeedanduselargeamountsofmemory.TCP/IPoffloadandRDMAarealsousedtoincreaseperformancewhilereducingCPUutilization.Lowlatencyhardware—Usuallyaprimaryconcernofdevelopersisrelatedtothemessage-passinginterfacedelayaffectingtheoverallcluster/applicationperformance.Thisisnotalwaysthecasebecausesomeclustersaremorefocusedonhighthroughput,andlatencydoesnotsignificantlyimpacttheapplications.TheCiscoCatalyst6500withdistributedforwardingandtheCatalyst4948-10Gprovideconsistentlatencyvaluesnecessaryforserverclusterenvironments.Non-blockingorlow-over-subscribedswitchfabric—ManyHPCapplicationsbandwidth-intensivewithlargetiesofdatatransferandinterprocesscommunicationsbetweencomputenodes.GEattachedserveroversubscriptionratiosof2.5:1(500Mbps)upto8:1(125Mbps)arecommoninlargeserverclusterdesigns.Mesh/partialmeshconnectivity—Serverclusterdesignsusuallyrequireameshorpartialmeshfabrictopermitcommunicationbetweenallnodesinthecluster.Thismeshfabricisusedtosharestate,data,andotherinformationbetween puteand puteserversinthecluster.Jumboframesupport—ManyHPCapplicationsuselargeframesizesthatexceedthe1500byteEthernetstandard.Theabilitytosendlargeframes(calledjumbos)thatareupto9Kinsize,providesadvantagesintheareasofserverCPUoverhead,transmissionoverhead,andfiletransfertime.CiscoDataCenterInfrastructure2.5Design1-OL-11565-1-OL-11565-CCNA|CCNPDataCenterDataCenterFigure1-6takesthelogicalclusterviewandcesitinaphysicaltopologythatfocusesonaddressingtheprecedingitems.BackEndHighSpeedComputerComputerFrontBackEndHighSpeedComputerComputerFront mendedserverclusterdesignleveragesthefollowingtechnicalaspectsorEqualcostmulti-path—ECMPsupportforIPpermitsahighlyeffectiveloaddistributionoftrafficacrossmultipleuplinksbetweenserversacrosstheaccesslayer.AlthoughFigure1-6demonstratesafour-wayECMPdesign,thiscanscaletoeight-waybyaddingadditionalpaths.Distributedforwarding—Byusingdistributedforwardingcardsoninterfacemodules,thedesigntakesadvantageofimprovedswitchingperformanceandlowerlatency.L3plusL4hashingalgorithms—DistributedCiscoExpressForwarding-basedloadbalancingpermitsECMPhashingalgorithmsbasedonLayer3IPsource-destinationplusLayer4source-destinationport,allowingahighlygranularlevelofloaddistribution.Scalableserverdensity—Theabilitytoaddaccesslayerswitchesinamodularfashionpermitsaclustertostartoutsmallandeasilyincreaseasrequired.Scalablefabricbandwidth—ECMPpermitsadditionallinkstobeaddedbetweenthecoreandaccesslayerasrequired,providingaflexiblemethodofadjustingoversubscriptionandbandwidthperIntheprecedingdesign,masternodesaredistributedacrossmultipleaccesslayerswitchestoprovideredundancyaswellastodistributeload.Furtherdetailsonmultipleserverclustertopologies,hardware mendations,andoversubscriptioncalculationsarecoveredinChapter3,“ServerClusterDesignswithEthernet.”CiscoDataCenterInfrastructure2.5Design OL-11565- OL-11565-CCNA|CCNP歸原作者所有本資料只供試 資源 攻城獅 CHAPTERDataCenterMulti-TierModelThischapterprovidesdetailsaboutthemulti-tierdesignthatCisco mendsfordatacenters.Themulti-tierdesignmodelsupportsmanywebservicearchitectures,includingthosebasedon.NETandJava2EnterpriseEdition.ThesewebserviceapplicationenvironmentsareusedforcommonERPsolutions,suchasthosefromPeopleSoft,Oracle,SAP,BAAN,andJDEdwards;andsolutionsfromvendorssuchasSiebelandThemulti-tiermodelreliesonamulti-layernetworkarchitectureconsistingofcore,aggregation,andaccesslayers,asshowninFigure2-1.Thischapterdescribesthehardwareanddesign foreachoftheselayersingreaterdetail.Thefollowingmajortopicsareincluded:DataCenterMulti-TierDesignDataCenterCoreDataCenterAggregationDataCenterAccessDataCenterServices Forahigh-leveloverviewofthemulti-tiermodel,refertoChapter1,“DataCenterArchitectureCiscoDataCenterInfrastructure2.5Design2- OL-11565-2- OL-11565-CCNA|CCNPDataCenterhatDataCenterhatDataCenterMulti-TierDesignThemulti-tiermodelisthemostcommonmodelusedintheenterprisetoday.Thisdesignconsistsprimarilyofweb,application,anddatabaseservertiersrunningonvarioustformsincludingbladeservers,onerackunit(1RU)servers,andmainframes.Figure2-1showsthedatacentermulti-tiermodeltopology.Familiarizeyourselfwiththisdiagrambeforereadingthesubsequentsections,whichprovidedetailsoneachlayerofthis Figure2- DataCenterMulti-TierModelCam