Fastoracle数据库学习

ORACLEDATABASEHIGHAVAILABILITYSTRATEGY,ARCHITECTUREANDSOLUTIONSDOAGNuremberg-17/09/2013KirillLoifmanOracleCertifiedProfessionalDBAwww:dadbm.
comTwitter:@loifmkir14/03/2012Footer2ELEMENTSOFHIGHAVAILABILITYHighavailability–Eliminatingsinglepointsoffailurethroughredundancy-Examples:redundantHW,SAN/ASM,RACdatabasesDisasterrecovery–Recoveringfromfailure-Examples:Extendedclusters,standbydatabasesOracleMaximumAvailabilityArchitecture(MAA)–CombiningHAandDRelementsDowntime–Howmuchtimeyoucantolerateyoursystembeingunavailable–Includesunexpectedaswellasplanneddowntime-99.
9%availability~8hoursand45minutesofdowntimeperyear14/03/2012Footer3CAUSESOFDOWNTIME2waysofdealingwiththepotentialfordisaster:a)expectandplanforitb)donothing,andhopeforthebest14/03/2012Footer4UNDERSTANDINGSERVICELEVELAGREEMENTS(SLA)1.
UnderstandingServiceLevelAgreements-customer'sservicerequirements2.
UnderstandingelementsofSLAs:-SystemAvailabilitye.
g.
Thedatabaseserversmustbeavailable7daysaweek,from6amtomidnight-AcceptableDataLosse.
g.
Nomorethan15minutesofdataentrycanbelost-RecoveryTimee.
g.
Intheeventofadisaster,thesystemsshouldbebackupandrunningwithinonehour.
-Performancee.
g.
Transactionresponsetimeshouldnotexceed2seconds.
3.
MapyourSLAstoappropriateHighAvailabilityLevels/Tiers4.
DesignsystemsandprocessestomeetSLAexpectations5.
Donotforgetaboutcost!
14/03/2012Footer5AVAILABILITYTARGETVSCOSTHowmany9sdoyouneed-Sharpdecreaseindowntimeforeachadditional9intheavailabilitytarget.
Whatwillitcost-CostcomparisonofoptionshelpsinachievingrealisticexpectationswhendevelopingSLAsAvailabilityTargetDowntimePerYear(Approx.
)90%36days98%7.
3days99.
7%26hours99.
99%52minutes99.
999%5minutesItemSANSnapshotbackupNativeDBbackupLicensing28,000Training14,000Storage45,00010,000TotalCost87,00010,000RecoveryTimeabout5minutesabout45minutesBenefitsNear-instantrecoveryReducedcomprexity14/03/2012Footer6LEVELSOFHIGHAVAILABILITY-EXAMPLELevel0:Out-of-the-box(nospecifichigh-availabilityelements)Level1:Storage-levelprotection(databaseinstancewithprotectedstorage)Level2:CFCor/andStandby(ColdFailoverClusteror/andStandbyDBatthesamephysicallocation)Level3:Recoveryviaredundantcomponents(HotFailoverwithmulti-nodeRACatthesamephysicallocation)Level4:Activeandpassiverecovery(2siteconcept,MAA,RAC,StandbyDB,Storagemirroring)Note:HALevelcombinationspossible;*percentagesshownareforillustrativepurposesonly.
14/03/2012Footer7ORACLE'SSOLUTIONTODOWNTIMEOracleCluster-ware&ASMRAC,RAC1NodeOracleRestartFlashbackTechnologyRecoveryManager(RMAN)DataGuardStreamsandGoldenGateFast-StartFaultRecoveryDataRecoveryAdvisorOnlinereorganization&redefinitionPatchinginrollingfashionOnlinepatchingDynamicdatabasereconfigurationAutomaticDiagnosticRepository(ADR)HARD-OracleHardwareAssistedResilientData…14/03/2012Footer8ORACLEDATABASEHIGHAVAILABILITY:BESTPRACTICES14/03/2012Footer9STORAGE-LEVELPROTECTIONCONSIDERATION2filesystems(DATA&FRA)areenoughtosharestoragebetweenmultipleDBsWhichRAIDConfigurationforHAtouse-basedonbusinessrequirements,costs,etc.
a)ASMmirroring(bestchoiceforlowcoststorage;enablesextendedclusteringsolutions)b)HardwareRAID1(mirroring;bestperformance;bestchoiceformodernSAN)c)RAID5(parityprotection,moreeconomicalsolution,notforwriteintensiveworkloads)d)BothASMmirroringandHardwaremirroringWhatTypeofStripingWorksBesta)ASMonlystriping-manyLUNstomanageatthestoragelevel-LUNsshouldbeequalandlimitedinsizeb)RAID0andASMstriping(stripe-on-stripe)DiskI/OdesingbestpracticesUseexternalRAIDprotectionwhenpossibleUseLUNswiththesameperformancecharacteristicsUseLUNswiththesamecapacityMaximizethenumberofspindlesinyourASMdiskgroup-painforSANadmin14/03/2012Footer10ASMSTRIPINGONLY14/03/2012Footer11HARDWARERAID–STRIPEDLUNS14/03/2012Footer12SINGLE-NODE,NON-CLUSTEREDORACLEDATABASEHALevel1~95-98%AvailabilitySingle-instance,standalone(noncluster)databaseProtectedstorageviaSAN/ASMOracleRestarttoenhanceHAOtherHAfeatureslike:FlashbackDatabase,OnlineRedefinition,RecoveryManager,andOracleSecureBackup,etc.
2filesystems/ASMdatagroups14/03/2012Footer13ORACLEDATABASEWITHCLUSTERWARE(COLDFAILOVER)HALevel2~98-99%Availability2ormultiple-nodesclusterSingleDBinstanceincoldfailovermodeOracleGRIDInfrastructureOracleRACOneNodeasOraclerecommendedcoldfailoversolutionSharedstorageonASM,OCFS2,NFS,…OracleRAConeNodeprosandcons:+betterDBavailabilitythantheothers+Onlinemigration&patching+Singlevendorsolution+ReadytoscaletofullRAC+SupportsDataGuard+Host-basedmirroringwithASM-Worksonlyfrom11gR2-Licensedseparately(+25%onCPU)14/03/2012Footer14ORACLEDATABASECOLDFAILOVER-OTHEROPTIONSOracleGridInfrastructurewithOracleclusterwarescripts+NotlicensedseparatelyincaseyouclusteranOraclelicensedproduct(database)+WorksforOracleClusterwareandDatabasebefore11gR2+Host-basedmirroringwithASMpossible-NoonlinerelocationpossibleaswithRACOneNode-Oracledoesnotdirectlysupportactionscriptsascustomcodefragments.
RedHatFailoverCluster+Inexpensive+MakesenseusingLinuxRedHat+Simpletomanagebutamajorupgradecanbeapain-Storagenotsharedbutmountedtoactivenode-Host-basedmirroringinastretchedclusterrequiresdifferentLinuxLVMwithmoreoverheadHPServiceguard+Complexbutstable+AvailableonHP-UXandLinux+Host-basedmirroringwithHP-UXLVMpossible-Expansive14/03/2012Footer15ORACLERACDATABASEHALevel3~99,5%Availability2ormultiple-nodesclusterOracleGRIDInfrastructure+OracleRACDBOracleRACprosandcons:+Hotfailoverfunctionality+Scalabilityacrossdatabaseinstances+Abilityofincreasingcapacitywithoutdowntime+Optimalcomputerresourceusage+Rollingupgradesandpatching+DBconnectionload-balancing-MightaffectDMLintensiveperformance-Licensedseparately(~50%ontopofCPUlicense)-Strictrequirementsfortheclusterinterconnect14/03/2012Footer16ORACLERACDATABASEONEXTENDEDCLUSTERSHALevel4~99,7%AvailabilitySameconfigurationaslocalOracleRAC2DataCenterlocationsDistancesovertenkilometersrequiredarkfiberRedundantstorageThirdvotingdiskonaninexpensiveNFSProsandCons:+GreateravailabilitythanlocalOracleRACcluster+Disasterrecoverypossibilitiesforsomedisasters+Extremelyrapidrecoveryifonesitefails-Cannotprotectagainstalldatacorruptionsorcomprehensivedisasters14/03/2012Footer17EXTENDEDRACDISKMIRRORINGNeedcopyofdataateachlocationTwooptions:1)Host-basedmirroring+WorkswithOracleASM,LRHHA-LVM,HP-UXLVM,…+Nodowntimeincaseofstoragefailure-Extraoverheadonserverresources-Morecomprexityforclustermaintenance2)Remotearray-basedmirroring+WorkswithHPCLX,EMCSRDF,HITACHIHAM,…+Doneonhardwarelevel+Nooverheadonserver-Outageincaseofprimarystoragefailureforsome(HPCLX)-StorageHAsoftwarelicenserequiredDiskmirroringisvendorspecificRedHat:Whenusingastretchedclustersolution,notallstorageconfigurationsaresupportedbyRedHat,i.
e.
clvmandcmirrorarenotsupported,butHA-LVM(withoutclvmd)maybeused14/03/2012Footer18ACHIEVINGQUORUMWITHEXTENDEDRACEveryClusternodemustaccessmorethanhalfofthevotingdisksatanytimeorthatnodewillbeevictedfromthecluster*Thesitethathousesthemajorityofthevotingdisksisapotentialsinglepointoffailurefortheentirecluster.
14/03/2012Footer19ORACLERACANDDATAGUARDCOMPLEMENTARITYRACaddressessystemfailuresbyprovidingrapidandautomaticrecoveryfromfailures,suchasnodefailuresandinstancecrashes.
DataGuardaddressessitefailuresanddataprotectionthroughtransactionalconsistentprimaryandstandbydatabasesthatdonotsharedisks,enablingrecoveryfromsitedisastersanddatacorruption.
14/03/2012Footer20ORACLEMAXIMUMAVAILABILITYARCHITECTURE(MAA)MAAisasetofbestpracticeblueprintsfortheintegrateduseofOracleHAtechnologiesMAAprovidesguidanceonoptimalHAarchitecturesdesignedtoachievethedifferentservicelevelobjectivesMAAresources:oracle.
com/goto/maa14/03/2012Footer21ORACLEMAAEXAMPLE–RAC&DATAGUARDHALevel4~99.
9%availabilityMAAprovidesthemostcomprehensivearchitectureforreducingdowntimeforscheduledandunscheduledoutagesTwoidenticalsitesPrimarysitecontainstheRACdatabaseSecondarysitecontainsPhysicalStandbydatabaseorbothPhysicalandLogicaloneonRACActiveDataGuardcanbeuseforonlinereporting(extralicensed)DataGuardswitchoverandfailoverfunctionsallowtherolestobetradedbetweensites14/03/2012Footer22ACTIVE-ACTIVEHAANDZERO-DOWNTIMEUPGRADES:GOLDENGATE14/03/2012Footer23DATABASEHASOLUTIONS–ANOTHEREXAMPLEStretchedRHCSwith2NodesSingleDBinstance(s)incoldfailovermode1OraclebinarysetperDBinstanceStoragemirroringwithDRBDHALevel3-99.
7%14/03/2012Footer2414/03/2012Footer24ORACLEEXADATAVERSION3ExadataX3DatabaseforHighAvailabilitySamearchitectureandpriceasX24xlargerflashmemory->22TBPCIflashmemoryperrack20xmorewriteperformance->ExadataSmartFlashWriteCaching33%moredatathroughput->100GB/secrunningSQL33%fasterDBCPUs->8-CoreXeonSandyBridgeE5-269075%morememory->2-4TBDRAMFull10GbEthernettoDataCentre10-30%lowerpower->Upto3KilowattReductionperRackNewlowercost1/8racksolutionNewOraclePlatinumServicesonX3AppliedtoOracleEngineeredSystemsDeliveredatnoadditionalcostsforPremierSupportcustomers24/7oraclemonitoringandremotediagnostics5MinuteFaultNotification15MinuteRestorationorEscalationtoDevelopment30MinuteJointDebuggingwithDevelopmentQuarterlysystemupdatesExadatalimitations-Cannotinstallothersoftware-Cannotspreadclusternodes-Cannotmirrorstorage-…14/03/2012Footer25ORACLE12CHIGHAVAILABILITYFEATURESOracleMultitenantarchitectureApplicationContinuity-improveapplicationrecovery(formerTAF)GlobalDataServices-supportglobaldatabaseservicesActiveDataGuardFarSync-extendzero-data-lossprotectionFlexASM-enableinter-nodestoragefailoverandreduceASM-relatedresourceconsumptionSYSDG,SYSBACKUP–newprivilegestoavoidgrantingSYSDBAActiveDuplicateCloning–rmanfeatureMoreonlineoperations:datafilemove,partitionmoveDatabaseRollingUpgradesusingActiveDataGuardOEM12cCloudControl:-OracleSiteGuard–extendsautomationofdisasterrecoverytotherestoftheOraclestack-HighAvailabilityConsole-integratesmonitoringofvariousHAareas-MAAConfigurationAdvisor-AutomationofmigrationadatabasetoASMandRAC….
THANKYOU!
KirillLoifmanwww.
dadbm.
com