欢迎来到天天文库
浏览记录
ID:34482879
大小:558.15 KB
页数:41页
时间:2019-03-06
《advancement proposal storage embedded networks (sen) and adaptive caching using multiple exnew》由会员上传分享,免费在线阅读,更多相关内容在教育资源-天天文库。
1、AdvancementProposalStorageEmbeddedNetworks(SEN)andAdaptiveCachingusingMultipleExperts(ACME)IsmailAriari@cs.ucsc.eduUCSC-CRL-03-01June2,2003DepartmentofComputerScienceUniversityofCalifornia,SantaCruz0ABSTRACTThegapbetweenCPUspeedsandthespeedofthetechnologiesprovidingthedataisincr
2、easing.Thiscausestheperformanceofprocessestobelimitedbytheperformanceofthestoragedevices,thenetworksandthebuses.Furthermore,thenumberofCPUsthatsharethesedataaccessresourcesisgrowingexponentially.Caching,prefetchingandparallelismaresomeofthetechniquesusedtodaytocopewithI/Olatency
3、andsystemscalabilitytosupportmoreusers.Thispaperdescribesthetwomajorcontributionsofourongoingresearchondistributeddataaccess.ThefirstcontributionisthedesignoftheStorageEmbeddedNetworks(SEN)architecturethataimstoimproveuserresponsetimesandscalabilityontheInternetbybetterdistributi
4、onofcaches.SENarchitectureiscomposedoftrustedroutersembeddedwithvolatileandnon–volatilestoragethatsnoopbypassingobjectsforcaching.Requestsarecheckedbyeveryhop,thusensuringthetransmissionoftheclosestcopyonthedatapathandloadreductionattheupstream.Thetwomaincontroloverheadsofothera
5、rchitectures,connectionestablishmentandcontinuouscachecommunications,donotexistinSEN.Thesecondcontributionisthedesignofadaptivecachingschemesusingmultipleex-perts,calledACME,thatmanagetheSENcachesandfurtherimprovethehitratesoverstaticcachingtechniques.Machinelearningalgorithmsar
6、eusedtorateandselectthecur-rentbestpoliciesormixturesofpoliciesviaweightupdatesbasedontheirrecentsuccess.Eachadaptivecachenodecantuneitselfbasedontheworkloaditobserves.Sincenocachedatabasesorsynchronizationmessagesareexchangedforadaptivity,theclusterscomposedofthesenodeswillbeex
7、ceedinglyscalableandmanageable.Weproposetoextendourpreliminarydesignsandanalysisintwodirections.ThefirstistocomparetheStorageEmbeddedNetworks(SEN)withtheexistinghierarchicalanddistributedcacheclustersintermsofuserresponsetimes,networkbandwidthusage,serverloadreductionsandscalabil
8、ity.Forthispartwewillrunlargescalesimulationsov
此文档下载收益归作者所有