基于Python的MPI-豆柴文库

您所在位置：网站首页 / 基于Python的MPI.docx / 文档详情

在线预览结束，喜欢就下载吧，查找使用更方便

5 金币

下载文档

/ 2

下载提示

如果您无法下载资料，请参考说明：

1、部分资料下载需要金币，请确保您的账户上有足够的金币

2、已购买过的文档，再次下载不重复扣费

3、资料包下载后请先用软件解压，在使用对应软件打开

文本预览

基于Python的MPI
Introduction
MPI(MessagePassingInterface)isastandardsetofmessage-passinglibrarieswidelyusedinparallelcomputing.Python,beingarelativelyhigh-levelprogramminglanguage,hasamatureMPIAPIwithmultipleimplementationlibrariessuchasmpichandopenmpi.ThispaperaimstodiscussthecapabilitiesandusagepatternsofMPIinPython-basedprogramming.
MPICapabilitiesinPython
MPIcanbeusedinPythonviamultiplelibrariesdependingonthespecificimplementationchosen.TheMPIlibraryprovidescommunicationprimitivestomovedataamongdistributedcomputingnodes,allowingforparallelcomputing.Theseprimitiveoperationsinclude:
-SendingandreceivingmessagesviaSendandRecvfunctions
-Blockingandnon-blockingcommunicationviaBlockingSend,BlockingRecv,Non-blockingSendandNon-blockingRecvfunctions
-Collectiveoperationssuchasbroadcast,scatter,gatherandreducefunctions
Tofacilitatetheparallelcomputingenvironment,manyMPIfunctionshavemultipleversionsthatcanoperateoneachprocessoronsubsetsofprocesses.Forexample,MPI_Barriersynchronizesallprocessesandguaranteesthatnoneofthemwillproceedfurtheruntilallothershavereachedthebarrier.
MPIalsosupportsprocess-levelcommunicationbetweengroupsofprocesses,calledcommunicators.MPIcomm_worldisthedefaultcommunicatorthatincludesallprocessesinthegroup.MPI_Comm_splitandMPI_Comm_createcancreatenewcommunicatorswhichcanbeusedtoseparateprocessesbasedontheirtaskorfunctionality.
UsagePatterns
MPIcanbeusedforparallelprogramminginmultipleways.Someofthemostpopularusagepatternsare:
1.SharedMemoryProgramming
MPIcanbeusedtowriteparallelcodethatutilizessharedmemory.Toachievethis,collectiveoperationsareutilizedincombinationwithprocess-baseddatadistributiontoensurethateachprocesshasacopyofshareddata.
2.DistributedMemoryProgramming
MPIcanbeusedtowriteparallelcodethatutilizesdistributedmemory.Thisinvolvespartitioningtheinputdataintosmallerchunksanddistributingthesechunkstoindividualprocesses.Processesthenworkindependentlyontheirsubsetsofdataandcommunicatewithotherprocessesthroughcollectiveoperations.
3.HybridProgramm