您好,欢迎访问三七文档
当前位置:首页 > 商业/管理/HR > 质量控制/管理 > LMS-Virtual.Lab声学计算加速
LMSVirtual.LabREV8AAcousticTrainingLMSVirtual.LabREV8AHowtospeedupyourcomputation?2copyrightLMSInternational-2008SYSNOISERev5.6usesadynamicmemoryallocationDuringaresolution(schematic!)MEMORYisaSYSNOISEEnvironmentVariable,tosetthemaximummemorywhichcanbeused,youcanmodifyit:Tools→Options→AcousticsandinthetabLMSSysnoisespecifythemaximummemory.NodesPropsB.C.MatricesResultsMEMORYMemoryAllocation(1)3copyrightLMSInternational-2008Memoryisreservedintotalvirtualmemory,i.e.thesumofphysicalRAMmemoryandswapspace:accesstophysicalRAMmemoryisextremelyfastswapspaceisadedicatedmemoryareaontheharddisk,withmuchsloweraccessonUNIXsystems,aprocesscanneverallocatemorememorythantheswapspacesizerule-of-thumb:swapspacesize=2*PhysicalRAMsizeMemoryAllocation(2)4copyrightLMSInternational-2008IN-COREsolverworkspaceissufficientlylargetostoresystemmatricesduringassemblyanddecompositionstagesoftheanalysistwoin-coresituationsexistworkspacecompletelyinphysicalRAMmemoryworkspaceinphysicalRAMmemoryandswapareaOUT-OF-COREsolverworkspaceisnotsufficientlylargetostoresystemmatricesduringassemblyanddecompositionstagesoftheanalysissystemmatricesareassembledinblockswhicharewrittentoascratchfile(*.TMP)ondisk;decompositionwillalsoproceedblockbyblockandreadblocksfromthescratchfileworkspacemustbesufficientlylargetocontaintheassemblyanddecompositionblocks(seelater)In-CoreversusOut-of-CoreSolvers(1)5copyrightLMSInternational-2008Solutionefficiencyin-coresolverusingphysicalRAMmemoryonlyin-coresolverusingphysicalRAMmemoryandswapspace:muchtimecanbelostin“swapping”fromRAMmemorytotheharddisk(largeelapsedorwall-clocktimecomparedtoCPUStatsinEchoWindow)=inefficient!out-of-coresolveralsoinvolvesdiskaccess,butwritingorreadingofeachblockrequiresasingleoperation:efficientuse!RecommendationsletSYSNOISEdecideuponin-coreorout-of-coresolversetMEMORYtoavaluethatdoesnotexceedphysicalRAMmemoryIn-CoreversusOut-of-CoreSolvers(2)6copyrightLMSInternational-2008In-CoreversusOut-of-CoreSolvers(3)IN-COREsolverspeedinRev5.6thiswillalwaysbethefastestmethodIN-COREsolverlimitsForthe32-bitversionofSYSNOISE,absolutelimitforin-corememory:2GBytesRAMForthe64-bitversionofSYSNOISE,nopracticallimitforin-corememory7copyrightLMSInternational-2008OUT-OF-COREsolverlimitspracticallimittoproblemsizeimposedbyavailableharddiskspaceforblockstorageRev5.3A•single*.TMPscratchfilecontainingtheblocks•absolutelimitformanyUNIXsystems:2GBytessizeRev5.4–5.6•I-BEMsolverallowsupto8different*.TMPfiles•environmentvariablesTMPDIR1,TMPDIR2,…•effectivelimit:16GBytesIn-CoreversusOut-of-CoreSolvers(4)8copyrightLMSInternational-2008OUT-OF-COREmemoryallocationtipsMemoryallocationmustbesufficientlylargeforsimultaneousstorageofmultipleassemblyordecompositionblocksBEMDIRECT:1assemblyblockor2decompositionblocksBEMINDIRECT:2assemblyblocksor2decompositionblocksInsufficientmemoryallocationforblockstoragewillleadtoanabortedanalysisIn-CoreversusOut-of-CoreSolvers(5)9copyrightLMSInternational-2008SelectionofsolverlibraryisgovernedbytheenvironmentvariableSOLVERSOLVER=1⇒LINPACKsolverSOLVER=2⇒userdefinedsolverSOLVER=3⇒reservedforfutureiterativesolversSOLVER=4⇒LAPACKsolverSOLVER=7⇒LMSsolverSOLVER=8⇒Matrix-leveldistributedsolverSettingofSOLVERcanbemodifiedpriortostartingtheresponseanalysisDefaultsettingin‘SYSNOISE.STP’file:SOLVER=7SolverLibraries(1)10copyrightLMSInternational-2008Distributedcomputingprinciplesnetworked¶llelcomputingfrequencylevel&matrixleveldistributionNetSolverlicensestructureNetSolvertechnicalprinciplesMPImechanismprocessorscheduling&loadbalancingRunningSYSNOISENetSolvercommandstructureMPIhostfilestructureSYSNOISEParallelonIBMSP-2DistributedComputing-NETSOLVER11copyrightLMSInternational-2008LargeSYSNOISEjobscanbeprocessedinparallelonmultipleprocessors:distributedcomputingsplitsthelargejobinsub-jobswhichrunsimultaneouslyoneachprocessorRev5.6supportstwotypesofparallelcomputing:on”real”multi-processorparallelmachines(e.g.IBMSP-2;SGIOrigin;HP.-class;DECAlphaServer)onasetofnetworkedsingle-ormulti-processorUNIXworkstations,actingasavirtualparallelmachine(SGIR4K,R8K&R10K;HPB-class,C-class&J-class;IBMRS/6000;DECALPHA;SUNUltraSparc)DistributedComputingPrinciples(1)12copyrightLMSInternational-2008SYSNOISERev5.6:distributedcomputingisalsosupportedonNTplatforms(networkedsingle-ormulti-processorPC’s)Hybridparallelconfigurations(UNIX+NT)arenotsupportedDistributedComputingPrinciples(2)13copyrightLMSInternational-2008FrequencyleveldistributioneachprocessorrunsadifferentsubsetofallrequestedanalysisfrequenciesavailableforbothnetworkandparallelconfigurationsMatrixleveldistributiontosolververylargeproblems,thesystemmatrixissplitintoblocks,andeachprocessorassemblesandsolvesadifferentblockinRev5.6availableforgeneralUNIXorNTnetworkconfigurationsthroughtheMPImechanismDistributedComputingPrinciples(3)14copyrightLMSInternational-2008NetSolverLicense(1)DistributedcomputingrequiresNetSolverLicense:NumberofKernellicensesaSYSNOISEsessionalwaysuses1Kernellicensepre-andpost
本文标题:LMS-Virtual.Lab声学计算加速
链接地址:https://www.777doc.com/doc-5892205 .html