会员注册 | 登录 | 微信快捷登录 支付宝快捷登录 QQ登录 微博登录 | 帮助中心 人人文库renrendoc.com美如初恋!
站内搜索 百度文库

热门搜索: 直缝焊接机 矿井提升机 循环球式转向器图纸 机器人手爪发展史 管道机器人dwg 动平衡试验台设计

   首页 人人文库网 > 资源分类 > PDF文档下载

30-Structural aspects of the System 360 Model 85, Part II The cache.pdf

  • 资源星级:
  • 资源大小:412.87KB   全文页数:7页
  • 资源格式: PDF        下载权限:注册会员/VIP会员
您还没有登陆,请先登录。登陆后即可下载此文档。
  合作网站登录: 微信快捷登录 支付宝快捷登录   QQ登录   微博登录
友情提示
2:本站资源不支持迅雷下载,请使用浏览器直接下载(不支持QQ浏览器)
3:本站资源下载后的文档和图纸-无水印,预览文档经过压缩,下载后原文更清晰   

30-Structural aspects of the System 360 Model 85, Part II The cache.pdf

Thecache,ahighspeedbufferestablishingastoragehierarchyintheModel85,isdiscussedindepthinthispart,sinceitrepresentsthebasicorganizationaldeparturefromotherSYSTEM/BOcomputers.Discussedareorganizationandoperationofthecache,includingthemechanismsusedtolocateandretrievedataneededbytheprocessor.Theinternalperformancestudiesthatledtouseofthecachearedescribed,andsimulatedperformanceofthechosenconfigurationiscomparedwiththatofatheoreticalsystemhavinganentire80nanosecondmainstorage.Finally,theeffectsofvaryingcacheparametersarediscussedandtabulated.StructuralaspectsoftheSystem/360Model8511ThecachebyJ.S.LiptayAmongtheobjectivesoftheModel85isthatofprovidingaSYSTEM/compatibleprocessorwithbothhighperformanceandhighthroughput.OneoftheimportantingredientsofhighthroughputisalargemainstoragecapacityseetheaccompanyingarticleinPartI.However,itisnotfeasibletoprovidealargemainstoragewithanaccesstimecommensuratewiththe80nanosecondprocessorcycleoftheModel85.Alongeraccesstimecanbepartiallycompensatedforbyanincreaseinoverlap,greaterbuffering,deeperstorageinterleaving,moresophisticationinthehandlingofbranches,andotherimprovementsintheprocessor.Allofthesefactorsonlypartiallycompensatefortheslowerstorage,and,therefore,wedecidedtouseastoragehierarchyinstead.Thestoragehierarchyconsistsofa1.04microsecondmainstorageandasmall,faststorecalledacache,lwhichisintegratedintotheCPU.Thecacheisnotaddressablebyaprogram,butratherisusedtoholdthecontentsofthoseportionsofmainstoragethatarecurrentlybeingused.Mostprocessorfetchescanthenbehandledbyreferringtothecache,sothatmostofthetimetheprocessorhasashortaccesstime.Whentheprogramstartsoperatingondatainadifferentportionofmainstorage,thedatainthatportionmustbeloadedintothecacheandthedatafromsomeotherportionremoved.Thisactivitymusttakeplacewithoutprogramassistance,sincetheModel85mustbecompatiblewiththerestofThispaperdiscussesorganizationofthecacheandthestudiesthatledtoitsuseintheModel85andtoselectingofvaluesforitsparameters.theSYSTEM/360line.ZBMSYSTEMSJOURNALVOL.7NO.1196815Figure1AssignmentofcachesectorstomainstoragesectorsMAINSTORAGECacheorganizationThemainstorageunitsthatcanbeusedontheModel85aretheIBM23655andthe2385.Theyhavea1.04microsecondcycletimeandmakeavailablecapacitiesfrom512Kbytesto4096KbytesK1024.Thecacheisa16Kbyteintegratedstorage,whichiscapableofoperatingeveryprocessorcycle.Optionally,itcanbeexpandedto24Kbytesor32Kbytes.Boththecacheandmainstoragearelogicallydividedintosectors,eachconsistingof1Kcontiguousbytesstartingon1Kbyteboundaries.Duringoperation,acorrespondenceissetupbetweencachesectorsandmainstoragesectorsinwhicheachcachesectorisassignedtoasingledifferentmainstoragesector.However,becauseofthelimitednumberofcachesectors,mostmainstoragesectorsdonothaveanycachesectorsassignedtothemseeFigure1.Eachofthecachesectorshasa14bitsectoraddressregister,whichholdstheaddressofthemainstoragesectortowhichitisassigned.Theassignmentofcachesectorsisdynamicallyadjustedduringassigningoperation,sothattheyareassignedtothemainstoragesectorsthatcachearecurrentlybeingusedbytheprogram.Iftheprogramcausesasectorsfetchfromamainstoragesectorthatdoesnothaveacachesectorassignedtoit,oneofthecachesectorsisthenreassignedtothatmainstoragesector.Tomakeagoodselectionofacachesectortoreassign,enoughinformationismaintainedtoorderthecachesectorsintoanactivitylist.Thesectoratthetopofthelististheonethatwasmostrecentlyreferredto,thesecondoneisthenextmostrecentlyreferredto,andsoforth.Whenacachesectorisreferredto,itismovedtothetopofthelist,andtheinterveningonesaremoveddownoneposition.Thisisnotmeanttoimplyanactualmovementofsectorswithinthecache,butratherreferstoalogical16J.8.LIPTAYorderingofthesectors.Whenitisnecessarytoreassignasector,theoneselectedistheoneatthebottomoftheactivitylist.Thiscachesectoristheonethathasgonethelongestwithoutbeingreferredto.Whenacachesectorisassignedtoadifferentmainstoragesector,thecontentsofallofthe1Kbyteslocatedinthatmainstoragesectorarenotloadedintothecacheatonce.Rather,eachsectorisdividedinto16blocksof64bytes,andtheblocksareloadedonademandbasis.Whenacachesectorisreassigned,theonlyblockthatisloadedistheonethatwasreferredto.Iftheyarerequired,theremainingblocksareloadedlater,oneatatime.Eachblockinthecachehasabitassociatedwithittorecordwhetherithasbeenloaded.Thisvaliditybitisturnedonwhentheblockisloadedandoffwhenthesectorisreassigned.Storeoperationsalwayscausemainstoragetobeupdated.Ifthemainstoragesectorbeingchangedhasacachesectorassignedtoit,thecacheisalsoupdatedotherwise,noactivityrelatedtothecachetakesplace.Therefore,storeoperationscannotcauseacachesectortobereassigned,ablocktobeloaded,ortheactivitylisttoberevised.Sinceallofthedatainthecacheisalsoinmainstorage,itisnotnecessaryonacachesectorreassignmenttomoveanydatafromthecachetomainstorage.Allthatisrequiredistochangethesectoraddressregister,resetthevaliditybits,andinitiateIoadingofablock.Theprocessoriscapableofbufferingoneinstructionrequestingthestoringofinformationinmainstorage,sothatitcanproceedwithsubsequentinstructionsevenifexecutionofthestoreinstructioncannotbeinitiatedimmediately.Twoprocessorcyclesarerequiredtofetchdatathatisinthecache.Thefirstcycleisusedtoexaminethesectoraddressregistersandthevaliditybitstodetermineifthedataisinthecache.Thesecondcycleisthenusedtoreadthedataoutofthecache.However,requestscannormallybeoverlapped,sothatonerequestcanbeprocessedeverycycle.Ifthedataisnotpresentinthecache,additionalcyclesarerequiredwhiletheblockisloadedintothecachefrommainstorage.ThestoragewordsizeonwhichtheModel85operatesinternallyis16bytes.Thisisthewidthofthedatapathstoandfromthestorageunits,andistheamounttheprocessorcanstoreorfetchwithasinglerequest.Becauseasingle23655storageunitoperatesonan8bytewideinterface,twounitsarepairedtogetherandoperatedsimultaneously.Exceptforthe512Kconfiguration,mainstorageisinterleavedfourways.Sinceablockis64bytes,fourfetchestomainstoragearerequiredtoloadoneblockintothecache.Withfourwayinterleaving,thismeansonerequesttoeachbasicstoragemodule.Toimproveperformance,thefirstbasicstoragemodulereferredtoduringeachblockloadistheonecontainingthe16byteswantedbytheprocessor.Inadditiontobeingloadedintothecache,thedataissentdirectlytotheprocessor,sothatexecutioncanproceedassoonaspossibleseeFigure2.OntheModel85,channelsstoreandfetchdatabywayoftheMODEL85CACHEFigure2TimingforablockloadMAINSTORAGENttUtUBSMZACCESSTIMEIBSMPIBSMOACCESSTIMEwBSMBASICSTORAGEMODULEprocessor.Channelfetchesareprocessedbygettingtherequireddatafrommainstoragewithoutreferringtothecache.Channelstoresarehandledthesamewayasprocessorstores.Inthisway,ifachannelchangesdatathatisinthecache,thecacheisupdatedbutthechannelsdonothaveanypartofthecachedevotedtothem.PerformancestudiesAmongthequestionsthathadtobeansweredtodeterminewhetherthecacheapproachshouldbetakenwere1howeffectiveisit,and2doesitseffectivenessvarysubstantiallyfromoneprogramtoanotherTheprincipaltoolsusedtoanswerthesequestionsarethetracingandtimingtechniquesreferredtoinPartI.ThetracingtechniqueproducesaninstructionbyinstructiontraceofaprogramoperatingundertheSYSTEM/OOperatingSystem.Theoutputisasequenceoftracetapes,whichcontaineveryinstructionexecuted,whetherintheproblemprogramortheoperatingsystem,andthenecessaryinformationtodeterminehowlongittakestobeexecuted.Thesetracetapescontainabout250,000instructionseachandareusedasinputtoatimingprogram,whichdetermines,cyclebycycle,howtheModel85wouldexecutethatsequenceofinstructions.Thesetechniquesareintendedtodetermineinternalperformanceanddonotprovideanyinformationconcerningthroughput.Anintensiveinvestigationprecededselectionoftheprogramsusedinthisstudy.Inordertomeasuretheeffectivenessofthecache,wepostulatedcacheasystemidenticaltotheModel85exceptthatthestoragehierarchyeffectivenessisreplacedbyasinglelevelstorageoperatingatcachespeed.TheperformanceofsuchasystemisthatwhichwouldbeachievedbytheModel85ifitalwaysfoundthedataitwantedinthecacheandifitneverencounteredinterferenceinmainstorageduetostores.Therefore,itrepresentsanupperlimitontheperformanceoftheModel85howclosetheModel85approachesthisidealcanserveasameasureofhoweffectivethecacheis.NineteentracetapesI18J.s.LIPTAYFigure3Model85performancerelativetosinglelevelstorageoperatingatcachespeedMEAN8175.79808490.94PERCENTAGEOFIDEALPERFORMANCEFlgure4ProbabilityoffindingfetcheddataincachePROBABILITYweretimedforboththeModel85andthepostulatedsystem,andtheperformanceoftheModel85wasexpressedasapercentageoftheperformanceoftheidealsystem.Figure3showsthedistributionofperformancedataobtained,Theaveragewas81percentoftheperformanceoftheidealsystem,witharangebetween66and94percent.Animportantstatisticrelatedtocacheoperationistheprobabilityoffindingthedatawantedforafetchinthecache.Figure4showsthedistributionofthisprobabilityforthesame19tracetapesusedforFigure3.Theaverageprobabilitywas0.968.Itisworthnotingthat,iftheaddressesgeneratedbyaprogramwererandom,theprobabilityoffindingthedatawantedinthecachewouldbemuchlessthan0.01.Therefore,itcanbesaidthatwhatmakesthecacheworkisthefactthatrealprogramsarenotrandomintheiraddressingpatterns.MODEL85CACHETable2AverageperformancerelativetoanidealsystemwithcachesizeandnumberofbytesperblockvariedNumberofsectors16NumberofNumberofbytesperblockcachebytes64128266,8K0.74416K0.8250.8100.78132K0.8910.8850.870Table3Comparativeperformanceusingdifferentcachesectorreplacementalgorithmsalgorithmperformance1partition1.0002partitions0.9904partitions0.9878partitions0.97916partitions0.933usagebits0.931ReplacementalgorithmchosenfortheModel85Table3summarizestheresultsobtained.Thechoiceoftheactivitylistwasmadebecauseitprovidedthebestbalancebetweencostandperformance.SummarycommentTheinclusionofastoragehierarchyrepresentsoneofthemajoradvancesinsystemorganizationpresentintheModel85.Althoughtheconceptofastoragehierarchyisnotnew,thesuccessfulimplementationofananosecond/microsecondlevelofhierarchywasinhibiteduntilnowbythelackofasuitabletechnology.AsimplementedintheModel85,thefastmonolithicstoragephysicallyintegratedwiththeCPUlogicyieldsthedesiredmachinespeed,whilethelargecorestorageyieldsthedesiredstoragecapacity,thecombinationbeingtransparenttotheuser.Itislikelythatwithfutureprogressintechnologythisnanosecond/microsecondhierarchyisnotmerelyaninnovationthatworkedoutwellfortheModel85,butratheritisafundamentalstepforwardthatwillbeincorporatedintomostlargesystemsofthefuture.CITEDREFREENCEANDFOOTNOTE1.Thetermcacheissynonymouswithhighspeedbuffer,asusedinotherModel85documentation.2.D.H.Gibson,Considerationsinblockorientedsystemsdesign,AFZPSConferenceProceedings,SpringJointComputerConference30,AcademicPress,NewYork,NewYork,75801967.MODEL85CACHE21

注意事项

本文(30-Structural aspects of the System 360 Model 85, Part II The cache.pdf)为本站会员(baixue100)主动上传,人人文库网仅提供信息存储空间,仅对用户上传内容的表现方式做保护处理,对上载内容本身不做任何修改或编辑。 若此文所含内容侵犯了您的版权或隐私,请立即通知人人文库网([email protected]),我们立即给予删除!

温馨提示:如果因为网速或其他原因下载失败请重新下载,重复下载不扣分。

copyright@ 2015-2017 人人文库网网站版权所有
苏ICP备12009002号-5