A Case for Redundant Arrays of Inexpensive Disks (RAID) Davtd A Patterson, Garth Gibson, and Randy H Katz Computer Saence D~v~smn Department of Elecmcal Engmeermg and Computer Sclencea 571 Evans Hall Umversity of Cabforma Berkeley. CA 94720 (partrsl@WF -kY du) Abstract Increasmg performance of CPUs and memorres wrll be squandered lf not matched by a sunrlm peformance ourease m II0 Whde the capactty of Smgle Large Expenstve D&T (SLED) has grown rapuily, the performance rmprovement of SLED has been modest Redundant Arrays of Inexpensive Disks (RAID), based on the magnetic duk technology developed for personal computers, offers an attractive alternattve IO SLED, promtang onprovements of an or&r of mogm&e m pctformance, rehabdlty, power consumption, and scalalnlrty Thu paper rntroducesfivc levels of RAIDS, grvmg rheu relative costlpetfotmance, and compares RAID to an IBM 3380 and a Fupisu Super Eagle 1 Background: Rlsrng CPU and Memory Performance The users of computers are currently enJoymg unprecedented growth m the speed of computers Gordon Bell said that between 1974 and 1984. smgle chip computers improved m performance by 40% per year, about twice the rate of mmlcomputers [Bell 841 In the followmg year B111 Joy predicted an even faster growth [Joy 851 Mamframe and supercomputer manufacturers, havmg &fficulty keeping pace with the rapId growth predicted by “Joy’s Law,” cope by offermg m&processors as theu top-of-the-lme product. But a fast CPU does not a fast system make Gene Amdahl related CPU speed to mam memory s12e usmg this rule [Siewmrek 821 Each CPU mnstrucaon per second requues one byte of moan memory, If computer system costs are not to be dommated by the cost of memory, then Amdahl’s constant suggests that memory chip capacity should grow at the same rate Gordon Moore pr&cted that growth rate over 20 years fransuforslclup = 2y*-1%4 AK predzted by Moore’s Law, RAMs have quadrupled m capacity every twotMoom75110threeyeaFIyers861 Recently the rauo of megabytes of mam memory to MIPS ha9 been defti as ahha [Garcm 841. vvlth Amdahl’s constant meanmg alpha = 1 In parl because of the rapti drop of memory prices, mam memory we.9 have grownfastexthanCPUspeedsandmanymachmesare~ppedtoday~th alphas of 3 or tigha To mamtam the balance of costs m computer systems, secondary storage must match the advances m other parts of the system A key meas- Pemuswn to copy mthout fee all or w of &IS matcnal IS granted pronded that the COP!S zzrc not made or lstnbuted for dwct commernal advantage, the ACM copyright notIce and the tltk of the pubbcatuon and IW da’, appear, and notxe IS @“en that COPYI"K IS by pemtrs~on of the Association for Computing Machtnery To COPY otherwIse, or to repubbsh, requres B fee and/or spenfic perm~ss~o” 0 1988 ACM 0-89791~268-3/88/~/OlOP $1 50 ure of magneuc tik technology 1s the growth m the maxnnum number of bits that can be stored per square mch, or the bits per mch m a track umes the number of tracks per mch Called MA D , for maxunal area1 density, the “Fmt Law m Disk Density” predicts ~rank87] MAD = lo(Year-1971)/10 Magnettc dd technology has doubled capacity and halved pnce every three years, m hne with the growth rate of semiconductor memory, and m practice between 1967 and 1979 the dtsk capacity of the average IBM data processmg system more than kept up with its mam memory [Stevens81 ] Capacity IS not the o~rty memory charactensuc that must grow rapidly to mamtam system balance, since the speed with which msuuctions and data are delivered to a CPU also determmes its ulamdte perfarmanceThespeedof~mem~has~tpacefoPtworeasons (1) the mvenuon of caches, showmg that a small buff= can be managed automamzally to contain a substanttal fractmn of memory refaences. (2) and the SRAM technology, used to build caches, whose speed has lmpmvedattherateof4O%tolOO%peryear In umtmst to pnmary memory technologres, the performance of single large expensive ma8netuz d&s (SLED) has improved at a modest rate These mechamcal devu~ are dommated by the seek and the rotahon delays from 1971 to 1981, the raw seek tune for a high-end IBM disk improved by only a factor of two whllt the rocstlon hme did not cbange[Harkex811 Greater denslty means a lugher transfer rate when the mformatmn 1s found. and extra heads can educe the aveaage seek tnne, but the raw seek hme only unproved at a rate of 7% per year There 1s no reasontoexpectafasterratemthenearfuture To mamtam balance, computer systems have been usmg even larger mam memones or solid state d&s to buffer some of the I/O acttvlty This may be a fine solutron for apphcattons whose I/O actrvlty has locality of reference and for which volatlltty 1s not an issue. but appbcauons dommated by a high rate of random muests for small peces of data (such BS tmmact~on-pmcessmg) or by a low number of requests for massive amounts of data (such as large simulahons nmnmg on supercomputers) are facmg a sermus p&mnance hmuatmn 2. The Pendrng I/O Crisw What t3 the Impact of lmprovmg the performance of sOme pieces of a problem while leavmg others the same? Amdahl’s answer IS now known asAmdahl'sLaw[Amdahl67] 1 S z (1-n +flk Whae S = the effecttve speedup, f=fractmnofworkmfastermode,and k = speedup whde m faster mode I/G Suppose that some current appbcatmns spend 10% of thev ume In Then when computers are 10X faster--accordmg to Bdl Joy m JUSt Over thtte years--then Amdahl’s Law predicts efQcove speedup wdl be only 5X When we have computers lOOX faster--vm evolutmn of umprcuzessors or by multiprocessors-&s applrcatlon will be less than 10X faster, wastmg 90% of the potenhal speedupWhde we can lmagme improvements m software file systems via buffcrmg for near term 40 demands, we need mnovaUon to avoid an J./O crms [Boral83] 3 A Solution: Arrays of Inexpensrve Disks RapId unprovements m capacity of large disks have not been the only target ofd& designers, smce personal computers have created a market for inexpensive magnetic disks These lower cost &sks have lower perfor- mance as well as less capacity Table I below compares the top-of-the-lme IBM 3380 model AK4 mamframe dtsk, FUJ~$U M2361A “Super Eagle” muucomputer disk, and the Conner Penpherals CP 3100 personal computer d& ChoroctensacS IBM FUJUSU Canners 3380 v 2361 v 3380 M2361A CP3100 3100 31Go (>I mmrr 3100 Is tt?tter) D&c dmmeter (mches)
View Full Document
Unlocking...