基于多核机群环境的并行程序设计方法研究--mpiopenmp混合编程-research on parallel programming method based on multicore cluster environment - mpi openmp mixed programming.docxVIP

基于多核机群环境的并行程序设计方法研究——MPI+OpenMP混合编程专业：计算机应用技术硕士生：王杰指导老师：王文义教授摘要高性能计算HPC(HighPerformanceComputing)是验证计算机系统处理能力和计算速度的一种有效手段。如今，高性能计算在国内外受到高度重视，其地位与作用已被广泛接受，它已成为解决国家级重大尖端问题的必由之路。目前，高性能计算在分子材料研究、地质勘探、气象、海洋气候、基因工程和军事技术等领域均已取得了巨大的成就。本文把搭建的多核心PC机为节点的机群系统作为平台，成功地完成了对MPI+OpenMP混合编程模型的运行环境配置，并在单节点多核计算机上分别完成了对串行程序、MPI并行程序和OpenMP并行程序的运行时间和加速比的对比分析。进而又重点在由多节点所组成的基于Windows的机群系统上，尝试完成了MPI+OpenMP两种程序的混合编程实验，通过对获得的程序运行时间、加速比和效率的数据结果的分析，得到了在多节点机群上采用MPI+OpenMP的细粒度的混合编程模式要比单用MPI设计的并行程序更适合多核PC机群的特点，从而更能充分发挥出多节点机群系统效率的结论。最后，论文总结了所做的工作和下一步的努力方向。关键词：高性能计算、多核机群、MPI+OpenMP混合编程TheresearchondesignmethodbasedonparallelprogramforwindowsenvironmentsSpecialty：TechnologyofComputerApplicationName：WangJieSupervisor：Prof.WangWenYiAbstractHighperformancecomputingscienceHPC(HighPerformanceComputing)isaneffectivemeanstoimprovecomputersystemprocessingcapacityandspeedofcalculation.Nowadays,thepositionandfunctionofhighperformancecomputinghavebeenwidelyrecognizedathomeandabroad.Ithasbecometheonlyroadtosolvenationalmajoradvancedproblems.Atpresent,ithasmadetremendousachievementsinmolecularmaterialsresearch,geologicalexplorationfield,climateareas,geneengineeringandmilitarytechnology.InthispaperthestructuresisconstitutedbynodesoftheclustersystemofmultiplenuclearPCmachineforhardwarecondition,andtheclustersystemsuccessfullycompletedrunningenvironmentconfigurationMPI+OpenMPwhichisthehybridprogrammingmodel,andcompletedcomparativeanalysisofrunningtimeandspeedupinserialprogram,MPIparallelprogramandOpenMPparallelprograminthemulti-corecomputerofsinglenodes.ThenthepapertrytocompleteMPI+OpenMPmixedprogrammingexperimentoftwoprogramsfocusingonWindowsclustersystemconstitutedbymultiplenetworks,basedonthedataanalysisoftheresultsoftherunningtime,speedupandefficiency,concludingthatitismoresuitableforthecharacterofmulti-corePCclustersysteminamultiplenodeMPI+OpenMPfine-grainedhybridprogrammingmodeprogramthanMPIinmultiplenodesonaparallelprogram.Asaresult,thepaperreachaconclusionthatwecanmakefulluseofefficiencyofclustersystemofmulti