双机知识胶片(vcd)

MC/ServiceGuardIntroduction别洪威HP企业客户服务部2001.4.16MC/ServiceGuardConceptsHardwareClusterManagementPackageManagementNetworkTroubleshootingConceptHighAvailabilitySolutionsHighlyAvailableSystemMC/ServiceGuardMC/LockManagerOnLineJFSProcessResourceManagerProtectedDataMirrorDisk/UXHPDiskArraysJFSReliableSystemsHP9000SystemHPperipheralsHP-UXHardwareConceptsLeadingidea:–NoSinglePointOfFailureTheMCinMC/ServiceGuardstandsfor“MultipleComputers”andanMC/ServiceGuardclusterconsistsofbetween2toupto16nodesdependingontheMC/ServiceGuardversion.HardwareConceptsAsupportedMC/ServiceGuardclusterrequiresredundanthardware–ToprotectagainstdiskfailuresLVMdiskmirroringordiskarraysmustbeused;–TwodiskbusesarealsostronglyrecommendedandshouldbeeitherconfiguredasPVGroupsorPVlinkstoprotectagainstbusordeviceadapterfailures.–AtleasttwoLANinterfacespernodearerequiredtoprotectagainstLANcardfailures.–ToallowMC/ServiceGuardtohandleLANfailuresatleastonebridgemustbepresent.SharedVolumeGroupPackageConceptsApplicationResource:–VolumeGroup–LogicalVolume–FileSystem–IPAddress–ShellScript–FunctionRule:–Resourcegiventoapackagecannotbegiventootherpackages.RelocatableIPAddressStationaryIP–StationaryIPaddressesneverchange.ThesearenormalIPaddressesassignedinthesamewayasbefore.RelocatableIP–RelocatableIPaddressesareassignedonaper-applicationbasis.Ifyouwanttoaccessanapplication,youconnecttotherelocatableIPaddresses.Thiswillconnectyou,regardlessofwhichsystemorinterfaceithasbeenassignedto.ClusterConcepts2to16Nodes.Statusandheartbeatsareexchanged.Packagestransferredafterfailure.Noreboot.MajorstepsinapplicationrecoveryafteranodefailureNodefailsClusterreformationAdoptivenodetakesoverFsckphaseApplicationstartup&recoveryApplicationavailabilitySimpleoperatorcommandsMovepackagestoalternatenodesPerformhardwareorsoftwareupgrades,maintenanceMovepackagesbackBackwardcompatibility:-OperatingSystem-MC/ServiceGuardPkgEPkgDPkgAPkgGPkgFPkgBPkgIPkgHPkgCHardwareorsoftwareupgrades...PkgEPkgDPkgAPkgGPkgFPkgBPkgIPkgHPkgC...thenrollthroughtheclusterMC/ServiceGuard:MinimizingPlannedDowntimeFastapplicationfailoverLANfailureprotectionNoidlesystemsRotatingStandby/AutomaticFailbackWorkloadbalancingfeaturesManagesplanneddowntimeSurvivesmulti-nodefailuresAdvancedTapeSharingMC/ServiceGuardHAClustersSimple,flexibleframeworkformanagingapplicationresourcesProcesses:App_Process_1App_Process_2Middleware_1Middleware_2Disks:lvol_data1lvol_data2Network:IP56.23.101.44Node1PkgBPkgAPkgCMC/ServiceGuard:ApplicationPackagesBalanceworkloadafteranodefailureMinimizeimpactonremainingnodesNode4PkgCPkgHPkgINode2PkgAPkgDPkgENode3PkgBPkgFPkgGNode1IfNode1fails...PkgAPkgBPkgCMC/ServiceGuard:WorkloadBalancingPRMwithMC/ServiceGuard:SuperiorLoadBalancingDynamicallocationofprocessingproceduresLoadbalancingfornormalandpost-failureoperationIfNode1failsNode2PkgCPkgBPkgA70%20%10%Node1PkgCNode1PkgC100%80%20%Node2PkgAPkgBEliminatingperformanceimpactofanodefailurePackagecanrunonanynodeinthecluster.AllnodesareequalAllnodesconnectedtoalldisksFailednodebecomesthefuturestandbyValuableforrotatingupgradeUsefulifyouwanttominimizeapplicationre-starts/interruptionsPkgAPkgBPkgDPkgCStandbyNode1Node1Node5Node4Node3Node2MC/ServiceGuard:RotatingStandbyPkgAStandbyPkgDPkgCPkgBNode1Node1Node5Node4Node3Node2PkgAPkgBPkgDPkgCStandbyNode1Node1Node5Node4Node3Node2PkgAPkgDPkgCPkgBNode1Node1Node5Node4Node3Node2UnderRepairCampusClusters:fast,flexibleHA&localdisasterprotectionHADiskArrayorXP51210km10kmHeartbeatFCHUBFCHUBFCHUBFCHUBBuildsonMC/SGcapabilitiesSinglecluster,multiplesitesContinuoussite-to-sitedatamirroringBasedonFibreChannelforspeed&upto10kmdistanceHADiskArrayorXP512HardwareGeneralDiskConceptsEachnodemusthaveitsownbootdisksNon-LVMdisksarenotsupportedforHAapplicationsDiskswhichdonotsupportLVMMirrorDisk/UX(HP-PB)arenotsupportedMirrorDisk/UXisnotrequiredbutishighlyrecommendedServiceGuarddoesnotrespondtodiskfailuresbutassumesthatthesearehandledbyLVMandMirrorDisk/UXDiskswhicharenotneededbyHAapplicationscanbenon-LVMGeneralBoot/RootDiskConceptsEachnodemusthaveitsownbootdisksNoneedtohaverootonsharedbusRootdiskmaybeconnectedtoasharedbusbutthisisnotrecommendedRootdiskcanbeoninternaldisks,externaldisksoracombinationofbothRootdiskscanbemirrored(stronglyrecommended)orunmirroredTheRootVolumeGroupThisVGmustnotcontainanyHAapplicationpackageLVMMirrorDisk/UXisrecommendedTherootVGmaybeonasharedbusoranon-sharedbusKeeptherootVGassimpleaspossible:1bootdiskplusthemirrorbootdiskisidealDonotuseexoticnetworkbasedbackuputilitiessuchasOmnibacktobackuptherootVG.Thesetakealongtimetogetinstalledandworking.Fbackup/frecoveraregoodenoughUsetheavailableutilitiestocollectinformationontheLVMconfiguration:LVMcollect,cmgetconfig,etcDonotthinkaboutstripingontherootVGGeneralDataDiskConceptsRecommendedtobemirroredMaycontainServiceGuardpackagesornon-HAservice/data/applicationMustbeonashared,externalbusthatisconnectedtoallnodesthatarepotentialownersoftheassociatedpackagesLVMandServiceGuardNotwodataVGsontheclustercanh