多处理器系统中的数据局部性及其优化技术研究
国防科技大学计算机学院,长沙 410073
下一篇 上一篇
摘要
数据局部性是多处理器系统中的重要研究方向之一。结合该领域目前国内外研究现状和我们近一阶段的研究进展,讨论了多处理器系统中的数据局部性及其优化问题。针对现有局部性度量模型存在的不足,提出了一种增强的可用于层次式并行计算机体系结构的局部性度量模型。在静态和动态局部性优化技术方面,分别探讨了基于投影分层的数据变换框架和基于瞬时访问信息的动态页迁移策略,并展开了系列相关的讨论。另外,针对利用数据局部性时必须解决的一个关键问题——存储一致性问题,进行了深入的研究,提出了以操作系统为中心的线程存储一致性模型。
参考文献
[ 1 ] WolfM , LamM .Adatalocalityoptimizingalgorithm[A].ProcSIGPLANConferenceonProgrammingLan guageDesignandImplementation[C], 1991.30~44
[ 2 ] MckinleyKS , CarrS , TsengCW .Improvingdatalo calitywithlooptransformation[J].ACMTransactionsonProgrammingLanguagesandSystems, July1996, 18 (4) :424~453
[ 3 ] KandemirM , ChoudharyA , RamanujamJ, etal.Amatrix basedapproachto globallocalityoptimization[J].JournalofParallelandDistributedComputing, 1999, 58:190~235
[ 4 ] SalisburgC , ChenZ , MelhemR .ModelingCommuni cationLocalityinMultiprocessors[J].JournalofParal lelandDistributedComputing5, 1999, 6:71~98
[ 5 ] TanakaA .Extensionoftheworkingsetformodelingspatiallocalityinprogrambehavior[A].Proc6thIn ternationalSymposiumonModeling, AnalysisandSimu lationofComputerandTelecommunicationSystems[C], 1998.27~35
[ 6 ] 夏 军, 杨学军, 曾丽芳, 等.基于投影分层技术的嵌套循环空间局部性优化方法[R].长沙:国防科技大学计算机学院, 2001
[ 7 ] EggersS , JeremiassenT .Eliminatingfalsesharing[A].ProcInternationalConferenceonParallelProcess ing[C], 1991, I :377~381
[ 8 ] TorrelasJ , LamM , HennesseyJ.Falsesharingandspatiallocalityinmulti processorcache[J].IEEETransactionsonComputers, 1994, 43 (6) :651~663
[ 9 ] ChowJ , SarkarV .Falsesharingeliminationbyselectionofruntimeschedulingparameters[A].Proc26thInter nationalConferenceonParallelProcessing[C], 1997
[10] CierniakM , LiW .Recoveringlogicaldataandcodestructures[R].TechniqueReport591, DepartmentofComputerScience, UniversityofRochester, 1995
[11] BlackD , GuptaA , WeberWD .Competitivemanage mentofdistributedsharedmemory[A].ProceedingsofCOMPCON [C], March1989.184~190
[12] VergheseB , DevienS , GuptaA , etal.Operatingsys temsupportforimprovingdatalocalityoncc numacom puteservers[A].ProceedingsoftheSeventhInterna tionalConferenceonArchitectualSupportforProgram mingLanguagesandOperatingSystems (ASPLOSVII) [C], October1996.279~289,
[13] BaylorS , EkanadhamK , JannJ, etal.Lazyhomemi grationfordistributedsharedmemorysystems[A ].ProceedingsofInternationalConferenceonHighPerfor manceComputing[C], December1997
[14] LaudonJ , LenoskiD .TheSGIorigin:AccNUMAhighlyscalableserver[A].Proceedingsofthe24thAn nualInternationalSymposiumonComputerArchitecture[C], May1997
[15] ChandraR , DevineS , VergheseB , etal.Schedulingandpagemigrationformultiprocessorcomputeservers[A ].ProceedingsArchitectureSupportforProgram mingLanguagesandOperatingSystems[C], October1994.12~24
[16] NikolopoulosD , PapatheodorouT , PolychronopoulosC , etal.Userleveldynamicpagemigrationformulti programmedshared memorymultiprocessors[A].Pro ceedingsofthe29thInternationalConferenceonParallelProcessing[C], August2000
[17] CarterJB , BennettJK , ZwaenepoelW .Techniquesforreducingconsistency relatedinformationindistribut edsharedmemorysystems[J].ACMTransactinosonComputerSystems, August1995, 13 (3) :205~243
[18] YangXuejun, DaiHuadong.Operationsystem centricmemoryconsistencymodel———threadconsistencymodel[A], TheFourthInternationalWorkshoponAdvancedParallelProcessingTechnologies (APPT’01) [C ], September2001.26~36