ComputerArchitectureQuestions.docx

上传人:b****3 文档编号:3806320 上传时间:2022-11-25 格式:DOCX 页数:14 大小:24.65KB
下载 相关 举报
ComputerArchitectureQuestions.docx_第1页
第1页 / 共14页
ComputerArchitectureQuestions.docx_第2页
第2页 / 共14页
ComputerArchitectureQuestions.docx_第3页
第3页 / 共14页
ComputerArchitectureQuestions.docx_第4页
第4页 / 共14页
ComputerArchitectureQuestions.docx_第5页
第5页 / 共14页
点击查看更多>>
下载资源
资源描述

ComputerArchitectureQuestions.docx

《ComputerArchitectureQuestions.docx》由会员分享,可在线阅读,更多相关《ComputerArchitectureQuestions.docx(14页珍藏版)》请在冰豆网上搜索。

ComputerArchitectureQuestions.docx

ComputerArchitectureQuestions

Computerarchitecturequestions

Thesequestionswerecollectedfrompreviousexamsandtests,soyouwillfindanewsetofprocessorspecificationsinsertedatvariouslocations:

thequestionsfollowingusethoseprocessorspecifications.Youwillalsofindsomeessentiallyidenticalquestions!

“Re-use”isawell-establishedsoftwareengineeringprinciple:

weuseitforexamquestionstoo!

Exceptwhereotherwiseindicated,usethefollowingoperatingsystemandprocessorcharacteristicsinallquestions.

Youroperatingsystemuses8kbytepages.Themachineyouareusinghasa4-waysetassociative32kbyteunifiedL1cacheanda64entryfullyassociativeTLB.Cachelinescontain32bytes.Integerregistersare32bitswide.Physicaladdressesarealso32bits.Itsupportsvirtualaddressesof46bits.1Gbyteofmainmemoryisinstalled.

a)Giveoneadvantageofadirectmappedcache.

b)Whatisthemaindisadvantageofadirectmappedcache?

c)Howmanysetsdoesthecachecontain?

d)Howmanycomparatorsdoesthecacherequire?

e)Howmanybitsdothesecomparatorsworkon?

f)Yourprogramisatextprocessorforlargedocuments:

inaninitialcheck,itscansthedocumentlookingforillegalcharacters.Foran8Mbytedocument,whatwouldyouexpecttheL1cachehitratetobeduringtheinitialcheck?

(Youareexpectedtodoacalculationandgiveanapproximatenumericanswer!

g)Yourprogrammanipulateslargearraysofdata.Inordertoconsistentgoodperformance,youshouldavoidonething.Whatisit?

(Beprecise–anumericanswerrelevanttotheprocessordescribedaboveandanexplanationisrequiredhere.)

h)Whatisthealternativetoaunifiedcache?

Whatadvantagesdoesitprovide?

i)Inadditiontodataandtags,acachewillhaveadditionalbitsassociatedwitheachentry.Listthesebitsandaddashortphrasedescribingthepurposeofeachbit(orsetofbits).(Inallcases,makeyouranswersconcise:

simplylistanydifferencesfromaprecedinganswer.)

(i)Aset-associativewrite-backcache

(ii)Aset-associativewrite-throughcache

(iii)Adirectmappedcache

(iv)Afullyassociativecache

j)32processesarecurrentlyrunning.IftheOSpermittedeachprocesstousethemaximumpossibleaddressspace,howmanypagetableentriesarerequired.

(i)Conventionalpagetables

(ii)Invertedpagetables

k)Drawadiagramshowinghowthebitsofavirtualaddressareusedtogeneratea32-bitphysicaladdress.

l)“Aprogramwhichsimplycopiesalargeblockofdatafromonememorylocationtoanotherexhibitslittlelocalityofreference,thereforeitsperformanceisnotimprovedbythepresenceofacache.”Commentonthisstatement.Isitstrictlytrue,mostlytrueornottrueatall?

Explainyouranswer.Assumeyouarerunningprogramsontheprocessordescribedatthebeginningofthissection.

m)Youareadvisingateamofprogrammerswritingalargescientificsimulationprogram.TheteammainlyconsistsofCSgraduateswhoskippedanystudyofcomputerarchitectureintheirdegrees.Performanceiscritical.

Listsomesimplethingsthatyouwouldadvisethemtodowhenwritingcodeforthissystem.Provideaonesentenceexplanationforeachpointofadvice.

(1markforeachvalidpieceofadvice,1forexplainingitand1foraddinganumberthatmakestheadvicespecifictotheprocessordescribedearlier.)

------------------------------------------------------------------------------------------------------------

Exceptwhereotherwiseindicated,usethefollowingoperatingsystemandprocessorcharacteristicsinallquestions.

Youroperatingsystemuses8kbytepages.Themachineyouareusinghasa4-waysetassociative32kbyteunifiedL1cacheanda128entryfullyassociativeTLB.Integerregistersare32bitswide.Physicaladdressesarealso32bits.Itsupportsvirtualaddressesof44bits.Thebusis64bitswide:

themostusualbustransactionhasfourdatacycles.1Gbyteofmainmemoryisinstalled.

n)Whywouldaprocessorexecutebothstatementss1ands2fromacompoundstatement:

if(condition)s1;

elses2;

o)Whatwouldyouexpectthecachelinelengthtobe?

p)Howmanycomparatorsdoesthecacherequire?

q)Howmanybitsdothesecomparatorsworkon?

r)HowmanycomparatorsdoestheTLBrequire?

s)Howmanybitsdothesecomparatorsworkon?

t)Yourprogramisatextprocessorforlargedocuments:

inaninitialcheck,itscansthedocumentlookingforillegalcharacters.Foran8Mbytedocument,whatwouldyouexpectthehitratestobeduringtheinitialcheck?

(Youareexpectedtodoacalculationandgiveanapproximatenumericanswer!

(i)Cache

(ii)TLB

u)Underwhatconditionswouldyouexpecttoachieve100%TLBhits?

(Twoanswersrequired.Forone,youareexpectedtodoacalculationandgiveanapproximatenumericanswer!

v)Whyarecachesbuiltwithlong(iemorethan8byte)lines?

(Tworeasonsneeded.)

w)Yourprogrammanipulateslargearraysofdata.Inordertoconsistentgoodperformance,youshouldavoidonething.Whatisit?

(Beprecise–anumericanswerrelevanttotheprocessordescribedaboveandanexplanationisrequiredhere.)

x)Whatisthemaximumnumberofpagefaultscanbegeneratedbyasinglememoryaccess?

Explainyouranswer.(Assumethepagefaulthandlerislockedinmemoryandnootherpagefaultsaregeneratedforpagesofinstructions.)

y)Asysteminterfaceunitwilloftenchangetheorderinwhichmemoryaccessesgeneratedbytheprogramareplacedonthesystembus.Givetwoexamplesofsuchre-orderingsandexplainwhytheorderischanged.

zz)Whydoesareadtransactioncheckthewritequeueinasysteminterfaceunit?

aa)Ifyouhadonlyalimitednumberoftransistorsavailableforimprovingbranchperformance,whatpredictionlogicwouldyouadd?

Whywillitwork?

bb)Whydoessuccessfulbranchpredictionimprovetheperformanceofaprocessor?

cc)Givetwoexamplesofspeculationinhighperformanceprocessors.Addasentencetoexplainhowthisimprovesprocessorperformance.

dd)Consideringallthe‘caches’presentinahighperformanceprocessor(instructionanddatacache,TLB,branchhistorybuffer,etc),whichonesincreasetheperformanceofaprogramwhichsimplycopiesdatafromonelocationtoanother?

Whichoneshavelittleornoeffect?

Agoodanswerwilllisteachcacheandaddasignforincreaseordecreaseandaddasinglephraseofexplanation.

ee)Onasystemwitha128KbyteL1cacheforaprogramwithaworkingdatasetof2Mbytes:

a.Calculatetheexpectedcachehitratewhennoassumptionscanbemadeaboutdataaccesspatterns.

Hitrate=128x103/2x106=6%

Answerstogreateraccuracyarewrong!

Workingdatasetsizehasbeenspecifiedto1significantfigureonly,soanswerisonlygoodfor1significantfigure!

b.Wouldyouexpecttheactualhitratetobebetterorworsethanthis?

Why?

Better–thisassumesperfectlyrandomaccesstoeverything.Loopvariables,constants,etcarelikelytohavemuchbetterhitrates.

ff)Yoursystemhasa4-waysetassociativecachewith432-bitwordspercacheline.

a.Ifthetotalcachesizeis64kbytes,howmanysetsarethere?

64kbytes/(4ways*4bytes/word*4words/line)=1024

b.Aphysicaladdressonthissystemhas32-bits.Howmuchmainmemorycanitaccommodate?

232=4Gbytes

c.Whatisthetotalnumberofbitsareneededforthecache?

It’sawrite-backcache.

Data:

64kx8bits

2bitsusedtoaddressabytewithinaword+2bitsforawordwithinalinenotneededintag(or,equivalently,4bitsfor16bytes/line)

Addressingasetrequires10bits,sotagshave32-10-4=18bits

Valid+dirty+LRU=3bits

Totaloverheadpercacheline:

18+3=21bits

Numberofcachelines:

64kb/4=16k

Totalbits:

64kx8+16kx21

d.Howmanycomparatorsdoesthiscacherequire?

Oneperway-4

e.Howaretherealaddressesoflinesinthesamesetrelated?

They’reseparatedbyamultipleof1024(numberofsets)*16(numberofbytesinaline)

f.Ifthiscachewasfullyassociative,

i.Howmanycomparatorswouldbeneeded?

Oneperline–16k!

ii.Howmanyoverheadbitswouldberequired?

Tagsneedtobe30bitsnow(32-4)+valid+dirty+LRU(2ormore)=32+bits/line(dependsonhowmanybitsusedforLRUalgorithm)

iii.Whatwouldbetheadvantage(ifany)obtainedfromtheextraresources.

Highertoleranceofnon-uniformdistributionsofaddresses–abilitytotoleratemore‘hits’onthesameset!

gg)YourOSusesapagesizeof8kbytes.

a.Whatisthecoverageofa64entryTLB?

64x8kybtes=512kbytes

b.WhatwilltheTLBhitratebefora2Mbworkingdatasetprogram(makingnoassumptionsaboutaccesspatterns)?

Worstcase:

512k/2Mb=25%

c.Iftheprogramsweepsthroughthedataaccessing64-bitdoublesinsequentialorder,whatwilltheTLBhitratebe?

1kdoubles/page,worstcaseis1TLBmiss/page,somissrateis0.1%andhitrate99.9%

hh)It’snecessaryforaprocessortoprovideaninstructiontoinvalidateallorpartofacache.Why?

ToallowthelocationsintowhichaDMAoperationwillbeperformedtobeflushedfromthecache.

Toflushdatabelongingtoanoldpageoutofthecacheonpageswap.

ii)ThePowerPCallowsyoutomarksomepagesofmemoryas“nottobecached”.Whywouldyouwanttodothis?

IfyoumarkI/Obufferslikethis,thenwritestothemalwaysgodirectlytomemory(notwastingcachespace)andtheydon’tneedtobeinvalidatedwhennewDMAoperationsareperformed.MostOS’swillcopydatafromreadbufferstoauser’saddressspaceimmediatelyafterit’sbeenreadfromthedevice,sothere’snoadvantageincachingit:

it’sonlyeverreadonce.

jj)Whatbenefitwouldyouexpectfromafullyassociativecache(comparedtoothercacheorganizations)?

kk)Despitethis,fullyassociativ

展开阅读全文
相关资源
猜你喜欢
相关搜索

当前位置:首页 > 工程科技 > 能源化工

copyright@ 2008-2022 冰豆网网站版权所有

经营许可证编号:鄂ICP备2022015515号-1