Академический Документы
Профессиональный Документы
Культура Документы
JeffreyDoto BrandonKrakowsky
th April15 ,2007
Abstract
Withinformationgenerationanddatatransferspeedatanalltimehigh,datastorageis fastbecomingoneofthefastestgrowingindustriesintheworld.Enterpriselevel corporations,endusers,andthegovernment,allneedtodealwithaneverincreasing plethoraofdata.RAID,orRedundantArrayofIndependentDisks,providesavery flexibleandreliablestoragesystemwhichcanbothenhanceperformanceandleverage faulttolerance.WediscussthevariousRAIDtypesandtheiruses,aswellastheir individualdrawbacks.FutureaspectsofRAID,aswellascurrententerpriseclass storageissuesarementioned.
TableofContents
Motivation InformationOverload. 3 ProliferationoftheInternet.3 SarbanesOxleyandHIPAA5 DataManagementfortheUser6 MicroprocessorTechnologicalAdvances7 Introduction MagneticDiskTechnology..7 RAIDDefined..8 History..8 Reliability.9 Performance.10 RAIDTechnology RAIDLevel0..10 RAIDLevel1..10 RAIDLevel2..11 RAIDLevel3..11 RAIDLevel4..12 RAIDLevel5..13 RAIDLevel6..14 HybridRAID RAIDX+Yvs.RAIDY+X.15 FutureRAID RAIDDP.15 RAIDZ16 RAIDControllers InterfacingwithRAID.16 SoftwareBasedRAID.16 HardwareBasedRAID17 ExternalHardwareRAID.17 Implementation.17 Conclusion.18 Appendix20 References..................................................................................................................21
Motivation
InformationOverload Theissueofdatastorageisbecomingincreasinglyimportant,asmoreandmore informationisstoredindigitalform.Assumingthattheaveragedigitalfilegets replicatedthreetimes,thetechnologyresearchfirmIDCdeterminedthattheworld generated161billiongigabytes161exabytesofdigitalinformationlastyear(Brian Bergstein 2007).Datamanagementisoneofthefastestgrowingindustriesintheworld. AccordingtoWorldWideLearn,Theinformationsectorhasthesecondhighest projectedjobgrowthrate. Publishing,theInternet,andtelecommunicationsshouldsee realoutputclimbto$1.6trillionby2014(WendyCroix).
SarbanesOxleyandHIPAA In2002,PresidentGeorgeW.BushsignedintoofficetheSarbanesOxley(SOXor Sarbox)act.Thisactchangedthewaypubliclyheldbusinesseswereheldresponsiblefor theiraccountingpractices,whichdirectlyaffectedtheirInformationTechnologystaff. Thebusinessesnowhadtoadheretostrictregulationsrequiringlongerdataretention time,whichwouldallowauditorstolookbackthroughthebusinessesfinancialhistories, aswellastheiremailrecords(RandolphKahn).Itisamazingwhenyouthinkofthe amountofdocumentsabusinessorinstitutioncangenerate:businesstransactions, invoices,purchaseorders,contracts,paymentsmade,websiterecords,internal documents.TheSarboxlawspecificallysites:recordsrelevanttotheauditor reviewmemoranda,correspondence,communications(includingelectronicrecords) (Jones2005)
SimilartotheSarbanesOxleyAct,TitleIIoftheHealthInsurancePortabilityand AccountabilityAct(HIPAA)of 1996hasreformedtheelectronicaspectofHealthcare insuranceplans,serviceproviders,andemployers.Theseregulationshavetheintentof promotingElectronicDataInterchange,orEDI,throughouttheUShealthcaresystem. Again,longerdataretention timeswillberequired,whichgreatlybenefitsthestorage sectoroftheIT industry (Consulting2005).
Introduction
MagneticDiskTechnology BeforewecanfullyunderstandRAID,wemustfirstunderstandtheinnerworkingsofa magneticharddisk,andtheproblemsoftenassociatedwithadisk.
crashes.Itisimportanttofirstunderstandalittlebackgroundabouthowaharddisk workstounderstandhowitcanfail.
Aharddiskisdifferentfromthealsowellknownfloppydiskbecauseithasafixed, rigidplatterthatiscoatedinamagneticmaterial,whichstorestheactualinformation. Thisrigidplatterhasamotorthatcausesittospinatspeedsvaryingfrom5400rotations perminute(RPM),upto15,000RPM.Anactuatorarmwitharead/writehead somethingakintothestylusofarecordplayer,movesinandout,readingandwriting datatoandfromthedisk(Lobur2006).Thus,thereareseveralobviouspointsoffailure inaharddisk:movingmechanicalpartscandeteriorateandbreak,ortherecanbe manufacturingdefects,suchassurfaceimperfections,ordustcontamination.The chambertothediskmustbehermeticallysealedsonodustcaninterferewiththesurface ofthedisk.Thus,overtime,themechanicsoftheharddriveareboundtofail,andall harddrivesaresoldwithaMeanTimeToFailurerating,orMTTF,whichmeansthat ultimately,everyusermustnotfacethequestionofwillmydatabelost,butwhenwill mydatabelost.Fortheaveragehomeuser,thismaynotbeofterribleimportance,but foramultimilliondollarbusiness,thiscouldbecatastrophic.Theneedforfailsafe, highlyreliablestorageisomnipresentinbigbusinessesofallkinds,government, academicinstitutions,andjustaboutanylargeinstitutiononecanthinkof.
RAIDDefined RAIDwasoriginallycoinedbyDavidA.Patterson,GarthGibson,andRandyH.Katz oftheUniversityofCaliforniaatBerkeley,intheir1988paper,ACaseforRedundant ArraysofInexpensiveDisks(DavidAPatterson1988). Theoriginalmeaningofthe acronymwasRedundantArray ofInexpensiveDisks,butbecausetheterm inexpensiveisrelative,RedundantArrayof IndependentDisksisnowgenerally acceptedasthepropermeaning(Lobur2006).
theparticularimplementationofRAID.RAIDtechnologywasdevelopedtoaddressthe faulttoleranceandperformancelimitationsofconventionaldiskstorage(Technick.net).
ThesetwopointsareconsideredandaddressedbasedonthetypeofRAID.
RAIDTechnology
RAIDLevel 0 AlsoknownasNonRedundant,RAIDLevel 0usesatechniquecalledstripingfor distributinginformationacrossthedisks.Dataisbrokendownintoindividualblocks, eachblockwrittentoaseparatediskdrive.Redundancyisnotemployedandthereisno paritygenerated,resultinginnodataprotection.Ifonestripeddrivefails,theyallfail.In fact,asthenumberofdisksincreases,sodoestheprobabilityof failure(PeterM.Chen 1994Lobur2006Corporation2007).
RAIDLevel0providesexcellentwriteperformancesincetheloadofstoringdatais spreadoutacrossmultipledisksandthereisnooverheadinupdatingredundant
10
RAIDLevel0requiresatleast2drivesandistypicallyveryinexpensivetoimplement sinceitusestheminimalnumberofdisks(Corporation2007).
RAIDLevel1requiresatleast2drivesandistypicallyexpensivetoimplementsinceit usestwicethenumberofdisks(Corporation2007).
11
information.Multipledisksarerequiredtoidentifyafaileddisk(PeterM.Chen1994 Lobur2006Corporation2007).
RAIDLevel 3requiresatleast3drives(Corporation2007).
12
NewParity=(olddataXORnewdata)XORoldparity
13
thisproblembydistributingtheparityoveralldisks,thuseliminatingthededicateddisk bottleneck.
RAID5:
Disk Sector 0 1 2 3 4 0 0 5 10 15 P4 1 1 6 11 P3 16 2 2 7 P2 12 17 3 3 P1 8 13 18 4 P0 4 9 14 19
RAID4:
Disk Sector 0 1 2 3 4 0 0 4 8 12 16 1 1 5 9 13 17 2 2 6 10 14 18 3 3 7 11 15 19 4 P0 P1 P3 P4 P5
Figure2 Optimal RAID5paritysetup. ObservehowintheRAID5depiction,theparity(PBlocks) aredistributedthroughoutalldisks,insteadofbeinglocatedonjustonedisk,asinthecasewithRAID4. Thismeansthatifthemachineneededtowritetosector0ofdisk3,andsector4ofdisk1,theoperation canoccursimultaneouslybecausetheparityforthedisksisindependent.Incontrast,noticehowinthe diagramforRAID4,thereisaconflictifyoutrytoservicethesectors0and4simultaneously:because the parityiscontainedononedisk,youcanonlywriteonegroupatatime.Thisisthecauseforthe aforementionedbottleneck.
14
RAIDLevel 6:BlockinterleavedStripingwithDualErrorProtection Patterson,etal.envisionedthefirst5levelsofRAID,butadditionalresearchhasbeen donetodevelopadditionalRAIDschemesthatovercomesomeweaknessesofthe aforementionedschemes.RAID6protectsagainstmultiplediskfailuredatalossby employingbothParity(P),andReedSoloman(Q)codes.Thisalsorequiresadditional storagecomparedtotheRAID5schemeingeneral,ifyoucanstorethedataonNdisks, RAID6requiresN+2disks,withaminimumof4disks(PeterM.ChenLobur2006). TheadditionaloftheReedSolomoncodingschemeaddsinanadditionaloverheadinto thesystem,insteadof4readsandwrites,therearenow6:the4standardreadwrites,plus anadditionalupdateonthePandQblocks(PeterM.Chen).Thisadditionalcomputation maynotbeavailableonallRAIDcontrollersinfact,AppleComputerspopularXServe RAID,doesNOTofferaRAID6configuration (Apple2007).Ingeneral,RAID6is lesscommonthan 0,1, 3, 5,and01. Wefoundonlyonearraycurrentlyofferedthat supportsRAID6,theAC&NCJetStor416iS(Guide2001).
15
inmirroring.ThemorecommonimplementationisRAID0+1,andisfoundinmost enterpriseclassRAIDdevices(AC&NC2007Computer2007).
FutureRAID
RAIDDP ConventionalRAIDschemeswhichuseasingleparitydrive(RAID3forexample)can easilyrecoveronefaileddisk.Butwhatifaseconddiskfailsbeforetheprevioushas beenrebuilt?Thisisverycommonconsideringtheamountoftimeitcantaketorebuild anentiredrive(SunstarCompany2007).
16
RAIDControllers
A RAIDcontrolleractsasaninterfacebetweenthehostsystemandtheRAIDsystem. It servestomanagethephysicaldisksanddeliverthedatatothehostsystem (Freeman 2004).
17
Implementation
EnterpriseClassStoragevs.PersonalStorage Tothestorageneophyte,everyharddrivelooksthesame,whethertheharddriveis sittinginyourhomecomputer,orinamilliondollardiskarrayatNASA.Thefactis,an enterprise,orbigbusinesslevelharddrive,hasagreatdealofadvancedconsiderations thatmustgointomakingitenterpriseworthy.Theenvironmentrequiredbythedrives functionality,andthepurposesforwhichthedriveswillbeused,dictatespecial engineeringconsiderations.
Mostenterprisedrivesaretypicallyruninanarray,orgroupedfashion,muchlikewe havediscussedwiththeRAIDdiskarrays.Inthissortof environment,thereisanadded amountofheatandvibrationfromthemultipledriveswhichcandetrimentallyaffectthe performanceofasingleharddrive(Anderson2003).Enterpriselevelsystemsmustalso runconstantly,24hoursaday,7daysaweek,withlittleorpreferablynodowntime. Thiscanmeananenterpriseharddrivewitha1,000,000HourMTTFrating,comparedto apersonalharddrivewithanMTTFof300,000hours.Theextraengineeringthatmust gointothiscanaddasignificantcost.Thismayincludeaddinginextragasketstokeep unwantedvaporsout,desiccantsandactivatedcharcoaltoremovehumidityorvolatile solventsfromtheair,andchassismodificationstoabsorbvibrationsandkeeptheunit cool (Anderson2003).
18
Historically,SCSIhasbeenthechoiceforhighlevelstorageapplications,withmostof theaforementionedengineeringgoingintoSCSIdrives(Anderson2003).Sinceabout 2003,though,thereseemstobeatrendtowardSATAreplacingSCSIinstallations.Null andLobursaidthat80%oftheenterpriseleveldrivesinuseattheirtimeofwritingwere SCSI,but,anInfoTechTrendspollfrom2005showedthat51%ofindividualsmaking nondesktoppurchasesselectedSATA,versusonly10%choosingParallelSCSI,and 11%SerialAttachedSCSI.SoitseemsthetrendisgoingtowardsSATAdrives,with theirmuchcheaperpricepercapacity.SCSIdriveshaveacurrentstoragelimitof300 GB,andatthatcapacitylevelcancostcloseto$900,whereasonecanpurchaseaSATA 500GBdrivefor$145.00.FastSCSIdrivesrunningat15,000RPMcancostover$1000 fora147GBdrive! Ingeneral,SCSIdrivesareanywherefrom2to4timesmore expensivethanSATAdrives,forafractionofthestoragespace.Thiscosthasits benefits,though,asseektimeona15,000rpmSCSIdriveisjust3.5ms,whilea comparableSATAdrivehasan89msseek.
Conclusion
Informationisbeinggeneratedatanalarminglyfastrate.Theworldreliesheavilyon datastorage.RAIDprovidesareliableandsafewaytostoreandmanagedata,butthere aremanyoptions.Dependingonimplementation,RAIDcanbeconfiguredforenhanced performance,increasedreliability,orsomecombinationofboth.
AnewsarticlereleasedthisweekdetailedhowtheWhiteHouserecentlylost5yearsof confidentialgovernmentrelatedemail.Thisiscertainlyboundtogenerateagreatdealof
19
controversyinthepress(Steiner2007).Imaginetherepercussions!Storageismorethan aconvenienceitisanecessity.
20
Appendix
JeffreyDoto Abstract Motivation SarbanesOxleyandHIPAA MagneticDiskTechnology RaidTechnology RAIDLevel4:BlockInterleaved Parity RAIDLevel5:BlockInterleaved DistributedParity RAIDLevel6:BlockInterleaved StripingwithDualError Protection HybridRAID:RAIDX+Yvs. RAIDY+X RAIDDP RAIDZ Implementation EnterpriseClassStoragevs. PersonalStorage Conclusion
BrandonKrakowsky Abstract Motivation InformationOverload ProliferationoftheInternet DataManagementfortheUser MicroprocessorTechnological Advances Introduction RAIDDefined History Reliability Performance RAIDTechnology RAIDLevel0 RAIDLevel1 RAIDLevel2 RAIDLevel3 RAIDDP RAIDZ RAIDControllers InterfacingwithRAID SoftwareBasedRAID HardwareBasedRAID ExternalHardwareRAID Conclusion
21
References
AC&NC.(2007)."http://www.acnc.com/02_01_jetstor_sata_416s.html."from http://www.acnc.com/02_01_jetstor_sata_416s.html. Adaptec.(2007)."Let'sTalkAboutRAID."RetrievedApril7,2007,from http://www.adaptec.com/en US/products/raid_tech/_education/talk_about_raid.htm. Anderson,D.(2003)."YouDon'tKnowJackaboutDisks."Storage1(4). Apple.(2007)."AppleServers."RetrievedApril15,2007,from http://www.apple.com/xserve/raid/performance.html. Brian Bergstein,A.T.W.(2007)"TechResearchersCalculateDigitalInfo."Wired Volume,DOI: Cauley,L.(2006)"NSA hasmassivedatabaseofAmericans'phonecalls."USATODAY Volume,DOI: Computer,A.(2007)."http://www.apple.com/xserve/raid/management.html."from http://www.apple.com/xserve/raid/management.html. Consulting,K.(2005)."HIPAAComplianceBrief." Consulting,K.(2005)."SarbanesOxleyComplianceBrief." Corporation,A.C.N.(2007)."RAIDTutorial."RetrievedApril8,2007,from http://www.acnc.com/raid.html. Corporation,H.P.(2005)"HPSolutionsforSarbanesOxleyCompliance." Volume, DOI: DavidA.Patterson,G.G.,andRandyH.Katz(1988)."ACaseforRedundantArraysof InexpensiveDisks(RAID)."ACM. DavidAPatterson,G.G.,andRandyHKatz(1988).ACaseforRedundantArraysof InexpensiveDisks(RAID). Dell(2007)"DellDataArchivingSolutions." Volume,DOI: Economist,T.(2004)."FileThat."Economist(March 4,2004). Engine,G.S.T.E.S."AboutRAIDControllers."RetrievedApril15,2007,from http://industrial computers.globalspec.com/LearnMore/Industrial_Computers_Embedded_Comput er_Components/RAID_Products/RAID_Controllers. Freeman,B.(2004)"StorageBasics:ChoosingaRAIDController." EnterpriseStorageForum.com Volume,DOI: Guide,P.(2001,April17,2001)."Multiple(Nested)RAIDLevels."RetrievedApril15, 2007,from http://www.pcguide.com/ref/hdd/perf/raid/levels/mult.htm. Guide,P.(2001,April17,2001)."RAIDLevel6." RetrievedApril15,2007,from http://www.pcguide.com/ref/hdd/perf/raid/levels/singleLevel6c.html. JohnWoelbern,d.o.O.m.s.andS.J.SonyElectronics'TapeStorage SolutionsDiv (2003)"DoesTapeBackupHaveaFuture? SAIT." Volume,DOI: Jones,K.C.(2005)."Survey:IT,OtherExecsWantBetterSarboxTools."Information Week. Lallande,A.(2005)."AddingUpSarbanesOxleyCostsMillions,countlessstaffhours spentoncompliance."BusinessMatters Lobur,L.N.a.J.(2006). TheEssentialsofComputerOrganizationandArchitecture, JonesandBartlettPublishers. Lobur,L.N.a.J.(2006)."TheEssentialsofComputerOrganizationandArchitecture." 22
Microsystems,S.(2007)."ZFS:Thelastwordinfilesystems."from http://www.sun.com/20040914/feature/. PeterM.Chen,E.K.L.,GarthA.Gibson,RandyH.Katz,DavidA.Patterson"RAID: HighPerformance,ReliableSecondaryStorage." ACMComputingSurveys:62. PeterM.Chen,E.K.L.,GarthA.Gibson,RandyH.Katz,DavidA.Patterson(1994). "RAID:HighPerformance,ReliableSecondaryStorage." ACMComputing Surveys26(2):145185. Planet,E.N.(2003,January 27th,2003)."SerialATA."RetrievedApril15,2007,from http://networking.webopedia.com/TERM/S/Serial_ATA.html. Planet,E.N.(2004,September20th,2004)."SCSI."RetrievedApril15,2007,from http://networking.webopedia.com/TERM/S/SCSI.html. Poelker,C.(2005,December9,2005)."RAIDexplained."AskTheStorageExpert: Questions&Answers RetrievedApril7,2007,from http://searchstorage.techtarget.com/expert/KnowledgebaseAnswer/0,289625,sid5_ gci1126527,00.html. RandolphKahn,E."TheSarbanesOxley ActUnderstandingtheImplicationsfor InformationandRecordsManagement." STATS,P.(2004)"BeginnersGuides:InstallingRAIDonDesktopPCs." Volume, DOI: Steiner,N.S.(2007)."WithoutATrace:TheMissingWhiteHouseEmailsand ViolationsofthePRA."CitizensForResponsibilityandEthicsinWashington. Sullivan,L.(2006)."ComplianceSpendingToReach$28Billionby2007."Information Week. SunMicrosystems,I.(2007)."Solaris10."ZFSFAQs RetrievedApril15,2007,from http://www.sun.com/software/solaris/faqs/zfs.xml. SunstarCompany,I.(2007)."RAIDDP."RetrievedApril15,2007,from http://www.sunstarco.com/Storage/StoreVault/Software%20&%20Technologies/ RAID_DP.htm. Technick.net."RAIDTechnology."RetrievedApril7,2007,from http://www.technick.net/public/code/cp_dpage.php?aiocp_dp=guide_raid. WendyCroix,P.D.,freelancewriter,universityprofessor,andformerretailmanager "QuicklyGrowingEmploymentSectorsOfferJobOpportunities WhichCareersareProjectedtoIncreaseinDemand?"WorldWideLearn TheWorld'sPremierOnlineDirectoryofEducation Volume,DOI:
23