H2-RAID：A-Novel-Hybrid-RAID-Architecture-Towards-H

H2-RAID:ANovelHybridRAIDArchitectureTowardsHighReliabilityTianyuWang1,ZhiyongZhang1,MengyingZhao1,KeLiu1,ZhipingJia1(B),JianpingYang2,andYangWu3,41SchoolofComputerScienceandTechnology,ShandongUniversity,QingDao,[e-mail]2StateGridShanghaiMunicipalElectricPowerCompany,Shanghai,China3NARIGroupCorporation(StateGridElectricPowerResearchInstitute),Nanjing,China4BeijingKedongElectricPowerControlSystemCo.Ltd.,Beijing,ChinaAbstract.Withtherapiddevelopmentofstoragetechnology,SolidStateDrive(SSD)hasreceivedextensiveattentionsfromindustryandacademia.AsapromisingalternativeoftheconventionalHardDiskDrive(HDD),SSDshowsitsadvantagesintermsofI/Operformance,powerconsumptionandshockresistance.ButthenaturalconstraintofwriteendurancelimitstheuseofSSDsinlarge-scalestoragesystems,espe-ciallyforscenarioswithhighreliabilityequirements.TheRedundantArraysofIndependentDisks(RAID)technologyprovidesamechanismofdevice-levelfaulttolerance.Toguaranteetheperformance,currentRAIDstrategiesusuallyevenlydistributestheI/Orequeststoalldisks.However,differentfromHDD,thebiterrorrate(BER)ofSSDincreasesdramaticallywhenitgetsolder.Therefore,simplyintroducingRAIDtechnologyintoSSDarraywouldresultinthe“correlatedSSDfailure”problem,thatis,alltheSSDsinarraywearoutatapproximatelythesametime,seriouslyaffectingthereliabilityofthearray.Inthispaper,weproposeaHybridHighreliabilityRAIDarchitecturenamedH2-RAID,whichcombinesSSDswithHDDstoachievethehigh-performanceofSSDsandthehigh-reliabilityofHDDs.Tominimizetheperformancedegradationcausedbythelow-performanceHDDs,wedesignanHDD-awarebackupstrategytocoalescethesmallwritesrequests.Weimple-menttheproposedstrategyonthesimulatorbasedonDisksim.Theexperimentalresultsshowthatwereducetheprobabilityofdatalossfrom11.31%to0.02%withonly5%performanceloss,inaverage.Keywords:Solidstatedrive·Harddiskdrive·RAIDreliabilityHybridarchitecture1IntroductionRecentdevelopmentofNANDflashshowsgreatpotentialtoreplacetraditionalharddrives(HDDs)withflash-basedsolidstatedrives(SSDs)inbothpersonalcSpringerNatureSwitzerlandAG2018J.VaidyaandJ.Li(Eds.):ICA3PP2018,LNCS11337,pp.617–627,2018.[1].Withoutthemechanicalparts,SSDsdeliveroneorderofmagnitudegreaterthroughput,andtwoordersofmagnitudegreaterI/Ooperationspersecond(IOPS),thanconventionalHDDs.Moreover,SSDsexhibittheadvantagesoflowerpowerconsumptionandhighershockresistance.AllthesefeaturesmakeSSDsplayanimportantroleintoday’slargescalestoragesystemsandadoptedbymanycommercialcompanies.AlthoughSSDsoutperformHDDsinmanyaspects,theysufferfromthelim-itationofwriteendurance.Moreover,toincreasethecapacityofSSDs,thetech-nologiesofNANDflasharefromSLC(Single-LevelCell)toMLC(Multi-LevelCell)eventheTLC(Triple-LevelCell),whichfurtherreducestheenduranceoftheSSDs.Therefore,thedevice-levelfaulttolerancebecomesanecessaryprob-lemtobeaddressed.Parity-basedredundantarrayofinexpensivedisks(RAID)[2]providesagoodoptiontoensuredevice-levelfaulttolerance,andhasbeenwidely-usedfordecades.RAIDisadatastoragevirtualizationtechnologythatcombinesmultiplephysicaldisksintooneormorelogicalunitsforthepurposesofdataredundancy,performanceimprovement,orboth.Dependingontherequiredlevelofredundancyandperformance,dataisdistributedacrossthedisksinoneofseveralways,referredtoasRAIDlevels.Onceadiskhaswornout,theRAIDcontrollercanrecoverthedatabymeansofsuchasredundancychecktechnolo-gies.Doinglikethis,thedataerrorisfixedandthereliabilityofthestoragesystemisenhanced.However,thetraditionalRAIDarchitectureisdesignedforHDD.Fortradi-tionalHDDarrays,theBitErrorRate(BER)ofadiskisnotcorrelatedwithitsage.ButfortheSSDarrays,thebiterrorraterapidlyincreasesastheP/EcyclesofNANDflashprogress[3,4].AccordingtoKimandLee[5],theUncorrectablePageErrorRate(UPER)ofcommercialSSDsreaches1E-15whenreceived75Tbyteswrittenandexceeds1E-11whenreceived200Tbyteswritten,whiletheUPERofHDDsmaintains1E-15duringtheirwholelifetime.SincetheprojectedlifetimeofmanycommercialSSDsismorethan200TBW(TotalteraBytesWrit-ten),astheSSDsgetolder,theyshowworsereliability.DuetotheabovefeatureofSSD,simplyapplyingtheRAIDtechnologytotheSSDarraywillencounteraseriousproblem,called“correlatedfailure”.TotakeRAID-5(themostwidelyusedRAIDarchitecture)asanexample,itbalanceswritesacrossdevicesbyevenlydistributingparitychunksintothem.Inthisway,alltheSSDsinthearraywillwearoutatapproximatelythesametime[6].SincetheRAID-5cantolerateonlyonediskfailure,thecontinuousfailuresofSSDswillresulttolargescaledataloss,whichseriouslydamagesthereliabilityoftheSSDarray.ToaddresstheproblemofcorrelatedfailureinSSDarray,thispaperproposesahybridRAIDarchitecturenamedH2-RAID,whichcombinesSSDswithHDDstoachievethehigh-performanceofSSDsandthehigh-reliabilityofHDDs.Inthispaper,foreachSSD,weinsertanextraHDDwiththesamecapacitytoprotectdatainSSD.BothHDDsandSSDsareconnectedtoRAIDcontroller.H2-RAID:ANovelHybridRAIDArchitectureTowardsHighReliability619Therefore,evenifalltheSSDsinthearrayarefailedatthesametime,westillhavetheHDDstoprovideonlineservices.Althought