group When thesetwo parmcsagree the faded bu was a 0 othcrwtseit wasa 1 If the checkdrskISthe fadure Justreadall the datadrsksand store the grouppanty in the replacementdrsk Reducmgthecheckd s toone per group C l reducesthe overhead cost to between4 and 10 for the group stzesconsideredhere The performancefor the thud level RAID systemis the sameas the Level 2 RAID but theeffectrveperformanceper dtskmcreasessmceit needsfewer check d s This reductronm total d s also increasesrelrabdtty but since It is shll larger than the useful hfehme of disks this IS a minor pomt One advantageof a level 2 systemover level 3 is that the extra check mformattonassocratedwith eachsectorto correctsoft errorsISnot needed mcreasmgthe capactty per dtsk by perhaps 10 Level 2 also allows all soft errorsto be corrected on the fly wnhouthavmgto rereada sector Table IV summarizesthe thud level RAID charactensncsand Figure 3 comparesthe sectorlayout andcheckd s for levels 2 and 3 CS 152 Computer Architecture and Engineering Lecture 18 Real Processor Walkthru I ExceedsUsefulLrfenme MlTF G lO 820 000 hrs 2004 11 02 or 90 years 1 1OD 10 91 Total Number of D s owrhcad cost Useable Storage Capacity Dave Patterson www cs berkeley edu patterson John Lazzaro www cs berkeley edu lazzaro EventslSec Full RAID vs Single Disk LargeRecu D S Large Writes D S Large R M W D S Small Readr DISC Small Vyrites D 2sG Small R M W DISC G 25 346 000 hrs or 40 years 104D 4 96 96 S 112 142 041s 112 3 02 S 112 041s 112 3 5 Table IV Characterrstrcsof a Level 3 RAID The L3lL2 column gives the performance of L3 tn terms of L2 and the L3ILl column give it in terms of LI loO means L3 ISfaster The performance for the full systems IS the same m RAID levels 2 and 3 but since there are fewer check dtsks the performance per dnk tmproves Park and Balasubramamanproposed a thud level RAID system without suggestmga partrcular applicauon park861 Our calculattons suggesttt 1sa much better match to supercomputerapphcatronsthan to www inst eecs berkeley edu cs152 transacuonprocessing This year two disk manufacturershave systems announcedlevel 3 RAIDS for suchapphcanonsusmgsynchronized5 25 mch disks with G 4 and C l one from IvIaxtorand one from Mtcropohs Magmms871 UC level Regents Fall 2004the UCB This thud hasbrought rehabrhtyoverheadcost to its lowest level so in the last two levels we Improveperformanceof small accesses 1 w out changmgcost or rehabrlny CS 152 L18 Real Processor Walkthru I 10 Fourth Level RAID Independent ReadsbVrltes Spreadmg a transfer across all sks wuhm the group has the followmgadvantage Large or grouped transfer ttme IS reduced becausetransfer bandwulthof theentuearraycanbe exploned But it hasthe followmg drsadvantagek as well ReadmgAvnhngto a disk m a grouprequuesreadmg wnhngto all the d s m a group levels 2 and3 RAIDScan performonly one I O at a Pme per group If the disks are not synchromzed you do not seeaverageseek androtattonaldelays the observeddelaysshouldmove towards the worstcase hencethe S factorm the equatronsabove This fourth level RAID improvesperformanceof small transfersthrough parallehsm theabrhty to do more than one I O per group at a ume We no longer spreadthe mdtvtdual transferinformanonacrossseveral sks but keep eachmdrvrdualunit ma smgledisk The vutue of bit mterleavmg1sthe easycalculatronof the Hammmg codeneededto detector correcterrorsin level 2 But recall that m the thud level RAID we rely on the drsk controller to detecterrorswnhm a single drsksector Hence rf we storean mdrvrdualtransferumt in a single sector we candetecterrorson an mdtvtdualreadwithoutaccessingany otherdrsk Frgure3 showsthe differentways the mformatronis storedin a sectorfor This Week A Real Processor Walkthru Leon An open source SPARC CPU Last Time A Case for RAID 4 Tran er UIlllS a b c d RAID 4 Level 4 Sector 0 accesses la can proceedm parallel the old dataand old panty can be readat the same ume Disk 1 and the new dataand new panty can be wntten at the same nme The performanceof small writes ISalso d ndedby G becausethe smgleSecwr check0disk m a group must be read and wntten with every A small wnte m that Datagroup thereby hmmng the number of writes thatT can be 2 A performed Diskat2a time to the numberof groups a The check sk 1sthe bouleneck and the fmal level RAID removes thusbottleneck MlTF EIficclencyPer Disk Eflctency Per Disk L3 WIL2 WILl w Lx2 WILI 91 S 127 91 96 S 112 96 91 S 121 182 96 S 112 192 91 S 127 136 09 S 127 8 05 S 127 8 09 S 127 11 RAID levels 2 3 and4 By stormga whole transferumt m a sector reads can be mdependentand operate at the maxrmum rate of a disk yet sull detect errors Thus the primary changebetweenlevel 3 and 4 ISthat WC mterlcavedata RAID 5 Check IDataD Disk 5 D s contamng Data and Checks Sectar 0 L a Dtsk 3 BxceedsUsefulhfetune G O 6 25 820 ooohrs 346 000 hrs or 90 years or 40 years 11OD 104D 10 4 Useabk Storage Capacy 91 96 aEcc0 ECCa Sector 0 Events x Full RAID ECCbEfitency Per Dtsk Eficwncy Per Dark Check bECC0 CECCO vs Smgk ECCc LA L4lL3 L4ILl IL4 L4iL 3 L4lLl Disk 5Dtsk dECC0 ECCd Large R 91 S 100 91 961 3100 96 DIS Only91 S one 100 Each tier s 100 192 Large Writes 182 Sector 0 aEcc1DIS Sk 100 umt136 1splaced 96 S tnto 100 146 Large R M W bECC1D S check91 s Check tn level a single D 96 SmallReads 91 3 1200 91 sector 96 3OCKI Ask 6 cECC1D Check Note 9 that the 02 check 120 I Small WrttesdEcc1 05rnfo120 4 ts calculated 0 14 ISnow calculated s Small R M W 09 120 04 120 6 Sector 0 aEcc2 avereach over a pece of each Check bECC2 lran er10 1 tran er urut L gwes TableDisk V Charactenstrcs of a Level 4 RAID The L4lL3 columt 7 cECC2 dECC2 the P Wn0nCe LA an terms of L3 and the L4lLl column gwes it in II Sector 0 D of D s Total Number overheadDisk cost4 II II cl B al Check rnforrnarron for Level 4 RAID for G 4 and C I The sectors are shown below the d s The checkedarem u ate the check mformatwn Wrues tosoofdtsk2andsl of aisk 3 unply writes to So and sl of dtsk 5 The check dtsk 5 becomes the write bottleneck terms of Ll 100 means L4 is faster Small reads improve because they no longer trc up a whok group of at adata timeandSmall writes and R M Ws Frgure 3 Comparrson of locatton check mformatlon In sectors for RAID levelswe 2 make 3 andthe 4 for G 4 Not shownasISwe themade small tn improve some because same assumpttons amount mformatton sectorIIOs added disk controller to Table II of thecheck slowdown for twoper related canbybethe ignored because only detect correct soft errors wlthm a Walkthru sector Remember
View Full Document
Unlocking...