Scribd
Upload
54 views

Lesson 3 4 Cache Oblivious Algorithms: The Ideal Cache Model

This document discusses cache-oblivious algorithms and the ideal cache model. It introduces some key concepts: 1. Cache-aware algorithms are optimized for a specific cache size L, while cache-oblivious algorithms do not reference the cache parameters (z,L) and aim to perform well on any cache. 2. The ideal cache model assumes the cache size L is equal to the transfer size, with a fully associative cache and optimal replacement policy. 3. Under this model, accessing a value is a cache hit if in cache and miss if not, requiring a full line transfer from slow to fast memory on a miss.

Uploaded by

Projectlouis
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
54 views

Lesson 3 4 Cache Oblivious Algorithms: The Ideal Cache Model

This document discusses cache-oblivious algorithms and the ideal cache model. It introduces some key concepts: 1. Cache-aware algorithms are optimized for a specific cache size L, while cache-oblivious algorithms do not reference the cache parameters (z,L) and aim to perform well on any cache. 2. The ideal cache model assumes the cache size L is equal to the transfer size, with a fully associative cache and optimal replacement policy. 3. Under this model, accessing a value is a cache hit if in cache and miss if not, requiring a full line transfer from slow to fast memory on a miss.

Uploaded by

Projectlouis
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 4

Lesson34CacheObliviousAlgorithms

Inacacheawarealgorithm,thevalueofLisdeterminedbythecachesize.
Q
(nzL)= (n/L)
aware

Ifthealgorithmmakesnoreferencetofastmemoryoritsparameters(z,L)thenthealgorithm
isoblivious(tofastmemory).

Q
(nzL)=O(n/L)
oblivious

Therearetwoquestionsthatneedtobeasked.

1. Howdowemodelautomaticfastmemory?
2. CananobliviousalgorithmmatchI/Operformanceofanawarealgorithm?

TheIdealCacheModel

Thisistheidealcachemodelthatwillbeusedin
thelesson.

TheListhesameasthetransfersize.


Inaprogramloadsandstoresareissuedfortheslow
memoryaddress.
Forexample:
Theprogramwantstoloadavaluefrommemorya.
Firstitcheckstoseeifthevalueofaisinfastmemory.
Ifitis,itjustusesitthisisacachehit.

Acachemissoccurswhenthevalueisnotinthe
cache.Thenitmustberetrievedfromslowmemory
andacopyisstoredinfastmemory.

Thehardwaremuststoreanentirecachelineofdata,soothermemoryvaluesaroundthe
desiredmemorylocationwillalsoberetrievedandstoredfrommemory.

Forastoreinstruction:
Ifthevalueisinthecache,itisacachehit.Thevaluewillbeupdatedinthecacheandstoredin
slowmemory.

Ifthestorevalueisnotinthecache,thisisacachemiss.
Foracachemisstheentirememorylinemustbetransferredfromslowmemorytothecache.

Theidealcacheisfullyassociative.
Thismeansthetransferfromslowmemoryisallowedtogo
toanylineinthecache.
Atsomepoint,thecachewillbefull.Sotoloadanewline,anoldlinewillhavetobeevicted
fromthecache.
Ifthelinethatistobeevictedhasnotbeenwrittentomemory,itisdirty.Adirtylinemustbe
writtenbeforeitcanbeevicted,thisiscalledastoreeviction.

Optimalreplacementevictlineorblockaccessedfurthestinthefuture.

SummaryofIdealCacheModel
1. Programissuesloadandstoreops
2. HardwaremanagesZ/Lcachelines
3. SlowmemorycachetransfersinlinesorblocksofsizeL
4. Accessingavalueincache=cachehit,otherwiseitisacachemiss
5. Cacheisfullyassociative
6. Assumeoptimalreplacementpolicy

Q(nZ,L)=#misses+storeevictions

LRUleastrecentlyused
Thisreplacementpolicylooksfortheleastrecentlyusedaddresstoevictfromthecache.
LRUwillcausemoreevictionsthatoptimalreplacement.

Howidealistheidealcache?
Keyassumptions:
1. optimalreplacement
2. twolevelsofmemory
3. Automaticreplacement
4. fullyassociative

TakeacachemodelthatusesaLRUpolicy.Compare
ittoacachethatusesanoptimalreplacementpolicy
thathasthecachesize.

Thislemmacanbeusedinthefollowingway:
Designanalgorithmonanidealmodelwithoptimalreplacement.Itshouldbeasymptotically
closetoamodelwithanLRUpolicyandtwicethecache.

Ifyoudesignanalgorithmforanideal
model,ifthemissesareregular,itwill
performinasimilarfashiontotheideal
model.

TheTallCacheAssumption
Thecacheshouldbetaller(withregardstothenumberoflinesz)thanitiswide(thesizeof
thelines,L).
Thismeansanefficientalgorithmmightbelinkedtothechoiceofdatastructure.

You might also like