并行程序設(shè)計(jì) 中文課件 08 基于消息傳遞的程序設(shè)計(jì)技術(shù)

ParallelProgrammingInstructor:ZhangWeizhe(張偉哲)ComputerNetworkandInformationSecurityTechniqueResearchCenter,SchoolofComputerScienceandTechnology,HarbinInstituteofTechnologyProgrammingUsingtheMessage-PassingParadigm3IntroductionProgramingwithMPIProgrammingwithPVMComparisionwithMPIandPVMOutline4AParallelMachineModelInterconnect…TheclusterAnodecancommunicatewithothernodesbysendingandreceivingmessagesoveraninterconnectionnetwork可以通過在互連網(wǎng)絡(luò)上發(fā)送和接收消息來與其他節(jié)點(diǎn)進(jìn)行通信的節(jié)點(diǎn)ThevonNeumanncomputer5Principlesof

Message-PassingProgrammingEachprocessorinamessagepassingprogramrunsaseparateprocess(sub-program,task)Thelogicalviewofamachinesupportingthemessage-passingparadigmconsistsofpprocesses,eachwithitsownexclusiveaddressspace.AllvariablesareprivateEachdataelementmustbelongtooneofthepartitionsofthespace;hence,datamustbeexplicitlypartitionedandplaced.CommunicateviaspecialsubroutinecallsAllinteractions(read-onlyorread/write)requirecooperationoftwoprocesses-theprocessthathasthedataandtheprocessthatwantstoaccessthedata.6Principlesof

Message-PassingProgramming消息傳遞程序中的每個(gè)處理器都運(yùn)行一個(gè)單獨(dú)的進(jìn)程（子程序，任務(wù)）支持消息傳遞范例的機(jī)器的邏輯視圖由p個(gè)進(jìn)程組成，每個(gè)進(jìn)程都有自己的獨(dú)占地址空間。所有變量都是私有的每個(gè)數(shù)據(jù)元素必須屬于該空間的一個(gè)分區(qū);因此，必須明確分區(qū)和放置數(shù)據(jù)。通過特殊的子程序通話所有交互（只讀或讀/寫）都需要兩個(gè)進(jìn)程的協(xié)作，這兩個(gè)進(jìn)程是具有數(shù)據(jù)訪問數(shù)據(jù)的進(jìn)程。7Principlesof

Message-PassingProgrammingSPMDSingleProgramMultipleDataSameprogramrunseverywhere EachprocessonlyknowsandoperatesonasmallpartofdataMPMDMultipleProgramMultipleData Eachprocessperformadifferentfunction(input,problemsetup,solution,output,display)8MessagesMessagesarepacketsofdatamovingbetweenprocessesThemessagepassingsystemhastobetoldthefollowinginformation:SendingprocessSourcelocationDatatypeDatalengthReceivingprocess(es)DestinationlocationDestinationsize9MessagePassingMessage-passingprogramsareoftenwrittenusingtheasynchronous異步orlooselysynchronous松散同步paradigms.Asynchronouscommunicationdoesnotcompleteuntilthemessagehasbeenreceived.Anasynchronouscommunicationcompletesassoonasthemessageisonitsway

10IntroductionProgramingwithMPIProgrammingwithPVMComparisionwithMPIandPVMOutline11WhatisMPI?ThedevelopmentofMPIstartedinApril1992.MPIwasdesignedbytheMPIForum(adiversecollectionofimplementors,librarywriters,andendusers)quiteindependentlyofanyspecificimplementation

MPI由MPI論壇（實(shí)施者，圖書館作家和最終用戶的多樣化集合）設(shè)計(jì)，完全獨(dú)立于任何具體的實(shí)現(xiàn)Website

//mpi/12WhatisMPI?MPIdefinesastandardlibraryformessage-passingthatcanbeusedtodevelopportablemessage-passingprogramsusingeitherCorFortran.Afixedsetofprocessesiscreatedatprograminitialization,oneprocessiscreatedperprocessor

mpirun–np5programEachprocessknowsitspersonalnumber(rank)EachprocessknowsnumberofallprocessesEachprocesscancommunicatewithotherprocessesProcesscan’tcreatenewprocesses(inMPI-1)13WhatisMPI?MPI定義了消息傳遞的標(biāo)準(zhǔn)庫(kù)，可用于使用C或Fortran開發(fā)便攜式消息傳遞程序。在程序初始化時(shí)創(chuàng)建一組固定的進(jìn)程，每個(gè)處理器創(chuàng)建一個(gè)進(jìn)程每個(gè)進(jìn)程知道其個(gè)人號(hào)碼（等級(jí)）每個(gè)進(jìn)程知道所有進(jìn)程的數(shù)量每個(gè)進(jìn)程都可以與其他進(jìn)程進(jìn)行通信進(jìn)程無法創(chuàng)建新進(jìn)程（在MPI-1中）14MPI:theMessagePassingInterfaceTheminimalsetofMPIroutines.MPI_InitInitializesMPI.MPI_FinalizeTerminatesMPI.MPI_Comm_sizeDeterminesthenumberofprocesses.MPI_Comm_rankDeterminesthelabelofcallingprocess.MPI_SendSendsamessage.MPI_RecvReceivesamessage.15StartingandTerminatingtheMPILibrary

MPI_InitiscalledpriortoanycallstootherMPIroutines.ItspurposeistoinitializetheMPIenvironment.MPI_Finalizeiscalledattheendofthecomputation,anditperformsvariousclean-uptaskstoterminatetheMPIenvironment.Theprototypesofthesetwofunctionsare:

intMPI_Init(int*argc,char***argv) intMPI_Finalize()

MPI_InitalsostripsoffanyMPIrelatedcommand-linearguments.MPI_Init也剝離任何與MPI相關(guān)的命令行參數(shù)。AllMPIroutines,data-types,andconstantsareprefixedby“MPI_”.ThereturncodeforsuccessfulcompletionisMPI_SUCCESS.所有MPI例程，數(shù)據(jù)類型和常量都以“MPI_”作為前綴。成功完成的返回碼為MPI_SUCCESS。16CommunicatorsAcommunicatordefinesacommunicationdomain-asetofprocessesthatareallowedtocommunicatewitheachother.InformationaboutcommunicationdomainsisstoredinvariablesoftypeMPI_Comm.CommunicatorsareusedasargumentstoallmessagetransferMPIroutines.Aprocesscanbelongtomanydifferent(possiblyoverlapping)communicationdomains.MPIdefinesadefaultcommunicatorcalledMPI_COMM_WORLDwhichincludesalltheprocesses.17Communicators通信者定義通信域-允許彼此通信的一組進(jìn)程。有關(guān)通信域的信息存儲(chǔ)在MPI_Comm類型的變量中。通信器用作所有消息傳輸MPI例程的參數(shù)。進(jìn)程可以屬于許多不同（可能重疊）的通信域。MPI定義了一個(gè)名為MPI_COMM_WORLD的默認(rèn)通訊器，包括所有進(jìn)程。18QueryingInformationThe

MPI_Comm_size

and

MPI_Comm_rank

functionsareusedtodeterminethenumberofprocessesandthelabelofthecallingprocess,respectively.Thecallingsequencesoftheseroutinesareasfollows:

intMPI_Comm_size(MPI_Commcomm,int*size) intMPI_Comm_rank(MPI_Commcomm,int*rank)

Therankofaprocessisanintegerthatrangesfromzerouptothesizeofthecommunicatorminusone.進(jìn)程的等級(jí)是從零到通信者的大小減一的整數(shù)。19OurFirstMPIProgram#include<mpi.h>main(intargc,char*argv[]){ intnpes,myrank;

MPI_Init(&argc,&argv); MPI_Comm_size(MPI_COMM_WORLD,&npes); MPI_Comm_rank(MPI_COMM_WORLD,&myrank); printf("Fromprocess%doutof%d,HelloWorld!\n", myrank,npes);

MPI_Finalize();}20ParallelProgrammingWithMPICommunication通訊Basicsend/receive(blocking)基本發(fā)送/接收（阻塞）Non-blocking非阻塞Collective集體Synchronization同步Implicitinpoint-to-pointcommunication隱含的點(diǎn)對(duì)點(diǎn)通信Globalsynchronizationviacollectivecommunication通過集體交流進(jìn)行全球同步ParallelI/O(MPI2)21Basic

SendingandReceivingMessagesThebasicfunctionsforsendingandreceivingmessagesinMPIaretheMPI_SendandMPI_Recv,respectively.Thecallingsequencesoftheseroutinesareasfollows:

intMPI_Send(void*buf,intcount,MPI_Datatype datatype,intdest,inttag,MPI_Commcomm) intMPI_Recv(void*buf,intcount,MPI_Datatype datatype,intsource,inttag, MPI_Commcomm,MPI_Status*status)

MPI_Send22MPI_SendThemessagetobesentisdeterminedbypointingtothememoryblock(buffer),whichcontainsthemessage.Thetriad,whichisusedtopointtothebuffer(buf,count,type),isincludedintotheparametersofpracticallyalldatapassingfunctions,Theprocesses,amongwhichdataispassed,shouldbelongtothecommunicator,specifiedinthefunctionMPI_Send,Theparametertagisusedonlywhenitisnecessarytodifferentiateamongthemessagesbeingpassed.Otherwise,anarbitraryintegernumbercanbeusedastheparametervalue23MPI_Send要發(fā)送的消息通過指向包含該消息的存儲(chǔ)器塊（緩沖器）來確定。用于指向緩沖區(qū)（buf，count，type）的三元組被包含在幾乎所有數(shù)據(jù)傳遞函數(shù)的參數(shù)中，數(shù)據(jù)傳遞的過程應(yīng)屬于MPI_Send函數(shù)中指定的通信器，僅當(dāng)需要區(qū)分正在傳遞的消息之間時(shí)才使用參數(shù)標(biāo)簽。否則，可以使用任意整數(shù)作為參數(shù)值24MPI_Recv2526SendingandReceivingMessages

MPIallowsspecificationofwildcardargumentsforbothsourceandtag.IfsourceissettoMPI_ANY_SOURCE,thenanyprocessofthecommunicationdomaincanbethesourceofthemessage.IftagissettoMPI_ANY_TAG,thenmessageswithanytagareaccepted.Onthereceiveside,themessagemustbeoflengthequaltoorlessthanthelengthfieldspecified.27SendingandReceivingMessages

MPI允許為源和標(biāo)簽指定通配符參數(shù)。如果source設(shè)置為MPI_ANY_SOURCE，則通信域的任何進(jìn)程都可以是消息的源。如果標(biāo)簽設(shè)置為MPI_ANY_TAG，則會(huì)接受帶有任何標(biāo)簽的消息。在接收端，該消息的長(zhǎng)度必須等于或小于指定的長(zhǎng)度字段。28MPIDatatypes

MPIDatatypeCDatatypeMPI_CHARsignedcharMPI_SHORTsignedshortintMPI_INTsignedintMPI_LONGsignedlongintMPI_UNSIGNED_CHARunsignedcharMPI_UNSIGNED_SHORTunsignedshortintMPI_UNSIGNEDunsignedintMPI_UNSIGNED_LONGunsignedlongintMPI_FLOATfloatMPI_DOUBLEdoubleMPI_LONG_DOUBLElongdoubleMPI_BYTEMPI_PACKED29Point-to-pointExampleProcess0 Process1#defineTAG999floata[10];intdest=1;MPI_Send(a,10,MPI_FLOAT,dest,TAG,MPI_COMM_WORLD);#defineTAG999MPI_Statusstatus;intcount;floatb[20];intsender=0;MPI_Recv(b,20,MPI_FLOAT,sender,TAG,MPI_COMM_WORLD,&status);MPI_Get_count(&status,MPI_FLOAT,&count);30Non-blockingCommunicationInordertooverlapcommunicationwithcomputation,MPIprovidesapairoffunctionsforperformingnon-blocking非阻塞sendandreceiveoperations.intMPI_Isend(void*buf,intcount,MPI_Datatypedatatype, intdest,inttag,MPI_Commcomm, MPI_Request*request)intMPI_Irecv(void*buf,intcount,MPI_Datatypedatatype, intsource,inttag,MPI_Commcomm, MPI_Request*request)Theseoperationsreturnbeforetheoperationshavebeencompleted.FunctionMPI_Testtestswhetherornotthenon-blockingsendorreceiveoperationidentifiedbyitsrequesthasfinished.intMPI_Test(MPI_Request*request,int*flag, MPI_Status*status)

Non-blockingCommunicationThefollowingschemeofcombiningthecomputationsandtheexecutionofthenonblockingcommunicationoperationispossible:組合計(jì)算和非阻塞通信操作的執(zhí)行的以下方案是可能的：31

ProgramsEvaluatingofMPIProgramExecutionTime

Theexecutiontimeneedstoknowforestimatingtheobtainedspeedupofparallelcomputation執(zhí)行時(shí)間需要知道估計(jì)獲得的并行計(jì)算加速Obtainingthetimeofthecurrentmomentoftheprogramexecutionisprovidedbymeansofthefollowingfunction:Theaccuracyoftimemeasurementcandependontheenvironmentoftheparallelprogramexecution.Thefollowingfunctioncanbeusedinordertodeterminethecurrentvalueoftimemeasurementaccuracy:時(shí)間測(cè)量的準(zhǔn)確度可以取決于并行程序執(zhí)行的環(huán)境。可以使用以下功能來確定時(shí)間測(cè)量精度的當(dāng)前值：3233MPICollectiveCommunicationRoutinesthatsendmessage(s)toagroupofprocessesorreceivemessage(s)fromagroupofprocesses將消息發(fā)送到一組進(jìn)程或從一組進(jìn)程接收消息Potentiallymoreefficientthanpoint-to-pointcommunication潛在地比點(diǎn)對(duì)點(diǎn)通信更有效率ExamplesBroadcastReductionBarrierScatterGatherAll-to-allCollectiveCommunication-BroadcastThereistheneedfortransmittingthevaluesofthevectorXtoalltheparallelprocesses,AnevidentwayistousetheabovediscussedMPIcommunicationfunctionstoprovideallrequireddatatransmissions:Therepetitionofthedatatransmissionsleadstosummingupthelatenciesofthecommunicationoperations,Therequireddatatransmissionscanbeexecutedwiththesmallernumberofiterations34CollectiveCommunication-Broadcast需要將矢量X的值發(fā)送到所有并行進(jìn)程，一個(gè)明顯的方法是使用上述討論的MPI通信功能來提供所有必需的數(shù)據(jù)傳輸：數(shù)據(jù)傳輸?shù)闹貜?fù)導(dǎo)致總結(jié)通信操作的延遲，所需的數(shù)據(jù)傳輸可以用較少的迭代次數(shù)執(zhí)行3536CollectiveCommunication-BroadcastTheone-to-allbroadcastoperationis:

intMPI_Bcast(void*buf,intcount,MPI_Datatypedatatype,intsource,MPI_Commcomm)SendsdatafromroottoallothersinagroupCollectiveCommunication-Broadcast3738CollectiveCommunication–ReduceTheall-to-onereductionoperationis:

intMPI_Reduce(void*sendbuf,void*recvbuf,intcount, MPI_Datatypedatatype,MPI_Opop,inttarget, MPI_Commcomm)Combinesdatafromallprocessesingroup-Performs(associative)reductionoperation(SUM,MAX)-ReturnsthedatatooneprocessCollectiveCommunication–Reduce39CollectiveCommunication–ReduceTheBasicMPIOperationTypesforDataReductionFunctions…4041CollectiveCommunication–synchronizationThebarriersynchronizationoperationisperformedinMPIusing:

intMPI_Barrier(MPI_Commcomm)

Abarrieroperationsynchronisesanumberofprocessors.屏障操作同步多個(gè)處理器。42CollectiveCommunication–ScatterThecorrespondingscatteroperationis:

intMPI_Scatter(void*sendbuf,intsendcount, MPI_Datatypesenddatatype,void*recvbuf, intrecvcount,MPI_Datatyperecvdatatype, intsource,MPI_Commcomm)

SendseachelementofarrayinroottoseparateprocessCollectiveCommunication–GatherGatheringDatafromAlltheProcessestoaProcess…Gatheringdatafromalltheprocessestoaprocessisreversetodatascattering.ThefollowingMPIfunctionprovidestheexecutionofthisoperation:4344CollectiveCommunication–GatherThegatheroperationisperformedinMPIusing:

intMPI_Gather(void*sendbuf,intsendcount, MPI_Datatypesenddatatype,void*recvbuf, intrecvcount,MPI_Datatyperecvdatatype, inttarget,MPI_Commcomm)

Collectsdatafromsetofprocesses45OtherCollectiveCommunicationMPIalsoprovidestheMPI_Allgatherfunctioninwhichthedataaregatheredatalltheprocesses.

intMPI_Allgather(void*sendbuf,intsendcount, MPI_Datatypesenddatatype,void*recvbuf,intrecvcount, MPI_Datatyperecvdatatype,MPI_Commcomm)Iftheresultofthereductionoperationisneededbyallprocesses,MPIprovides:

intMPI_Allreduce(void*sendbuf,void*recvbuf,intcount,MPI_Datatypedatatype,MPI_Opop,MPI_Commcomm)Tocomputeprefix-sums,MPIprovides:

intMPI_Scan(void*sendbuf,void*recvbuf,intcount,MPI_Datatypedatatype,MPI_Opop,MPI_Commcomm)

46OtherCollectiveCommunicationTheall-to-allpersonalizedcommunicationoperationisperformedby:

intMPI_Alltoall(void*sendbuf,intsendcount, MPI_Datatypesenddatatype,void*recvbuf, intrecvcount,MPI_Datatyperecvdatatype, MPI_Commcomm)Usingthiscoresetofcollectiveoperations,anumberofprogramscanbegreatlysimplified.使用這套核心集體操作，可以大大簡(jiǎn)化一些程序。4748TopologiesandEmbeddings

MPIallowsaprogrammertoorganizeprocessorsintologicalk-dmeshes.TheprocessoridsinMPI_COMM_WORLDcanbemappedtoothercommunicators(correspondingtohigher-dimensionalmeshes)inmanyways.Thegoodnessofanysuchmappingisdeterminedbytheinteractionpatternoftheunderlyingprogramandthetopologyofthemachine.MPIdoesnotprovidetheprogrammeranycontroloverthesemappings.

49TopologiesandEmbeddings

MPI允許程序員將處理器組織成邏輯k-d網(wǎng)格。MPI_COMM_WORLD中的處理器ID可以通過多種方式映射到其他通訊器（對(duì)應(yīng)于高維網(wǎng)格）。任何此類映射的優(yōu)點(diǎn)由底層程序的交互模式和機(jī)器的拓?fù)浣Y(jié)構(gòu)決定。MPI不提供程序員對(duì)這些映射的任何控制。50TopologiesandEmbeddingsDifferentwaystomapasetofprocessestoatwo-dimensionalgrid.and(b)showarow-andcolumn-wisemappingoftheseprocesses,(c)showsamappingthatfollowsaspace-llingcurve(dottedline)(d)showsamappinginwhichneighboringprocessesaredirectlyconnectedinahypercube.51CreatingandUsing

CartesianTopologies

Wecancreatecartesiantopologiesusingthefunction:我們可以使用以下功能創(chuàng)建笛卡爾拓?fù)洌?/p>

intMPI_Cart_create(MPI_Commcomm_old,intndims, int*dims,int*periods,intreorder,MPI_Comm*comm_cart) Thisfunctiontakestheprocessesintheoldcommunicatorandcreatesanewcommunicatorwithdimsdimensions.該功能采用舊通信器中的進(jìn)程，并創(chuàng)建一個(gè)具有dims維度的新通信器。Eachprocessorcannowbeidentifiedinthisnewcartesiantopologybyavectorofdimensiondims.每個(gè)處理器現(xiàn)在可以在這個(gè)新的笛卡爾拓?fù)溆沙叽鏒IMS的向量標(biāo)識(shí)。Example:CalculatingπThevalueofconstantπcanbecomputedbymeansoftheintegralTocomputethisintegralthemethodofrectanglescanbeusedfornumericalintegration52Example:CalculatingπCyclicschemecanbeusedtodistributethecalculationsamongtheprocessorsPartialsums,thatwerecalculatedondifferentprocessors,havetobesummed5354Calculatingπ–SequentialProgram

intnum_steps=1000;doublewidth;voidmain() { inti; doublex,pi,sum=0.0; width=1.0/(double)num_steps;

for(i=1;i<=num_steps;i++){ x=(i-0.5)*width; sum=sum+4.0/(1.0+x*x); } pi=sum*width;}55MPIExample–Calculatingπ56MPIExample–Calculatingπ57CollectiveCommunication–Calculatingπ58CollectiveCommunication–Calculatingπ59IntroductionProgramingwithMPIProgrammingwithPVMComparisionwithMPIandPVMOutline60WhatisPVM?

ThedevelopmentofPVMstartedinsummer1989atOakRidgeNationalLaboratory(ORNL).Isasoftwarepackagethatallowsaheterogeneouscollectionofworkstations(hostpool)tofunctionasasinglehighperformanceparallelmachine(virtual)PVM,throughits

virtualmachine

providesasimpleyetusefuldistributedoperatingsystemIthas

daemon

runningonallcomputersmakingupthe

virtualmachine61WhatisPVM?PVM的發(fā)展始于1989年夏天在橡樹嶺國(guó)家實(shí)驗(yàn)室（ORNL）。是一個(gè)軟件包，允許異構(gòu)收集的工作站（主機(jī)池）作為單個(gè)高性能并行機(jī)器（虛擬）PVM通過其虛擬機(jī)提供了一個(gè)簡(jiǎn)單而有用的分布式操作系統(tǒng)它具有在組成虛擬機(jī)的所有計(jì)算機(jī)上運(yùn)行的守護(hù)程序62PVMResourcesWebsite

/pvm/pvm_home.html

Book PVM:ParallelVirtualMachine

AUsers'GuideandTutorialforNetworkedParallelComputing

AlGeist,AdamBeguelin,JackDongarra,WeichengJiang,RobertManchek,VaidySunderam

/pvm3/book/pvm-book.html

63HowPVMisDesigned64BasicPVMFunctionsEnrollsthecallingprocessintoPVMandgeneratesauniquetaskidentifierifthisprocessisnotalreadyenrolledinPVM.IfthecallingprocessisalreadyenrolledinPVM,thisroutinesimplyreturnstheprocess'stid.將調(diào)用進(jìn)程注冊(cè)到PVM中，并生成唯一的任務(wù)標(biāo)識(shí)符，如果此進(jìn)程尚未注冊(cè)在PVM中。如果調(diào)用進(jìn)程已經(jīng)注冊(cè)在PVM中，這個(gè)例程只需返回進(jìn)程的tid。 tid=pvm_mytid();StartsnewPVMprocesses.Theprogrammercanspecifythemachinearchitectureandmachinenamewhereprocessesaretobespawned.開始新的PVM流程。程序員可以指定要生成進(jìn)程的機(jī)器體系結(jié)構(gòu)和機(jī)器名稱。numt=pvm_spawn("worker",0,PvmTaskDefault,"",1,&tids[i]);65BasicPVMFunctionsTellsthelocalpvmdthatthisprocessisleavingPVM.ThisroutineshouldbecalledbyallPVMprocessesbeforetheyexit.Addhoststothevirtualmachine.Thenamesshouldhavethesamesyntaxaslinesofapvmdhostfile.pvm_addhosts(hostarray,4,infoarray);Deleteshostsfromthevirtualmachine.

pvm_delhosts(hostarray,4);66BasicPVMFunctionsImmediatelysendsthedatainthemessagebuffertothespecifieddestinationtask.Thisisablocking,sendoperation.Returns0ifsuccessful,<0otherwise.

pvm_send(tids[1],MSGTAG);Multicasts組播amessagestoredintheactivesendbuffertotasksspecifiedinthetids[].Themessageisnotsenttothecallereveniflistedinthearrayoftids.

pvm_mcast(tids,ntask,msgtag);67BasicPVMFunctionsBlocksthereceivingprocessuntilamessagewiththespecifiedtaghasarrivedfromthespecifiedtid.Themessageisthenplacedinanewactivereceivebuffer,whichalsoclearsthecurrentreceivebuffer.

pvm_recv(tid,msgtag);Sameaspvm_recv,exceptanon-blocking非阻塞receiveoperationisperformed.Ifthespecifiedmessagehasarrived,thisroutinereturnsthebufferidofthenewreceivebuffer.Ifthemessagehasnotarrived,itreturns0.Ifanerroroccurs,thenaninteger<0isreturned.

pvm_nrecv(tid,msgtag);68BasicPVMFunctionspvm_barrier("worker",5);pvm_bcast("worker",msgtag);pvm_gather(&getmatrix,&myrow,10,PVM_INT,msgtag,"workers",root);pvm_scatter(&getmyrow,&matrix,10,PVM_INT,

msgtag,"workers",root);pvm_reduce(PvmMax,&myvals,10,PVM_INT,msgtag,"workers",root);69PVMExample:HelloWorld!#include<stdio.h>#include"pvm3.h"main(){intcc,tid;charbuf[100];printf("i'mt%x\n",pvm_mytid());cc=pvm_spawn("hello_other",0,0,"",1,&tid);if(cc==1){ cc=pvm_recv(-1,-1); pvm_bufinfo(cc,0,0,&tid); pvm_upkstr(buf); printf("fromt%x:%s\n",tid,buf);}else printf("can'tstarthello_other\n");pvm_exit();exit(0);}#include"pvm3.h"main(){ intptid; charbuf[100]; ptid=pvm_parent(); strcpy(buf,"hello,worldfrom"); gethostname(buf+strlen(buf),64); pvm_initsend(PvmDataDefault); pvm_pkstr(buf); pvm_send(ptid,1); pvm_exit(); exit(0);}70SetuptoUsePVMSetPVM_ROOTandPVM_ARCHinyour.cshrcfileBuildPVMforeacharchitecturetypeCreatea.rhostsfileoneachhostlistingallthehostsyouwishtouseCreatea$HOME/.xpvm_hostsfilelistingallthehostsyouwishtouseprependedbyan``&''71StartingPVMBefore

wegooverthestepstocompileandrunparallelPVMprograms,youshouldbesureyoucanstartupPVMandconfigureavirtualmachine.OnanyhostonwhichPVMhasbeeninstalledyoucantype%pvmandyoushouldgetbackaPVMconsolepromptsignifyingthatPVMisnowrunningonthishost.Youcanaddhoststoyourvirtualmachinebytypingattheconsolepromptpvm>addhostnameAndyoucandeletehosts(excepttheoneyouareon)fromyourvirtualmachinebytypingpvm>deletehostnameIfyougetthemessage``Can'tStartpvmd,''thencheckthecommonstartupproblemssectionandtryagain.72StartingPVM在完成編譯并運(yùn)行并行PVM程序的步驟之前，您應(yīng)該確?？梢詥?dòng)PVM并配置虛擬機(jī)。在任何安裝了PVM的主機(jī)上，您可以鍵入%pvm并且您應(yīng)該返回PVM控制臺(tái)提示符，表示PVM正在此主機(jī)上運(yùn)行。您可以通過在控制臺(tái)提示符下鍵入來將主機(jī)添加到虛擬機(jī)pvm>addhostname并且您可以通過鍵入從虛擬機(jī)中刪除主機(jī)（您所在的除外）pvm>deletehostname如果您收到消息“無法啟動(dòng)pvmd”，請(qǐng)檢查常見的啟動(dòng)問題部分，然后重試。73StartingPVMToseewhatthepresentvirtualmachinelookslike,youcantypepvm>confToseewhatPVMtasksarerunningonthevirtualmachine,youtypepvm>ps–aOfcourseyoudon'thaveanytasksrunningyet;that'sinthenextsection.Ifyoutype``quit"attheconsoleprompt,theconsolewillquitbutyourvirtualmachineandtaskswillcontinuetorun.AtanyUnixpromptonanyhostinthevirtualmachine,youcantype%pvmandyouwillgetthemessage``pvmalreadyrunning"andtheconsoleprompt.Whenyouarefinishedwiththevirtualmachine,youshouldtypepvm>haltThiscommandkillsanyPVMtasks,shutsdownthevirtualmachine,andexitstheconsole.ThisistherecommendedmethodtostopPVMbecauseitmakessurethatthevirtualmachineshutsdowncleanly.YoushouldpracticestartingandstoppingandaddinghoststoPVMuntilyouarecomfortablewiththePVMconsole.AfulldescriptionofthePVMconsoleanditsmanycommandoptionsisgivenattheendofthischapter.74StartingPVM要查看當(dāng)前虛擬機(jī)的外觀，可以鍵入pvm>conf要查看虛擬機(jī)上運(yùn)行的PVM任務(wù)，請(qǐng)鍵入pvm>ps–a當(dāng)然你還沒有任何運(yùn)行的任務(wù);這在下一節(jié)。如果您在控制臺(tái)提示符下鍵入“quit”，則控制臺(tái)將退出，但您的虛擬機(jī)和任務(wù)將繼續(xù)運(yùn)行。在虛擬機(jī)中的任何主機(jī)上的任何Unix提示符下，可以鍵入%pvm您將收到消息“pvmalreadyrunning”和控制臺(tái)提示。完成虛擬機(jī)后，您應(yīng)該鍵入pvm>halt該命令可以殺死任何PVM任務(wù)，關(guān)閉虛擬機(jī)，并退出控制臺(tái)。這是阻止PVM的推薦方法，因?yàn)樗_保虛擬機(jī)完全關(guān)閉。您應(yīng)該練習(xí)啟動(dòng)和停止并將主機(jī)添加到PVM，直到您對(duì)PVM控制臺(tái)感到滿意為止。在本章末尾給出了PVM控制臺(tái)及其許多命令選項(xiàng)的完整說明。75IntroductionProgramingwithMPIProgrammingwithPVMComparisionwithMPIandPVMOutline76PVMandMPIGoalsPVMAdistributedoperatingsystemPortabilityHeterogeneityHandlingcommunicationfailuresMPIAlibraryforwritingapplicationprogram,notadistributedoperatingsystemportabilityHighPerformanceHeterogeneityWell-definedbehavior77PVMandMPIGoalsPVM分布式操作系統(tǒng)可移植性異質(zhì)性處理通信故障MPI用于編寫應(yīng)用程序的庫(kù)，而不是分布式操作系統(tǒng)可移植性高性能異質(zhì)性明確的行為78WhatisNotDifferent?Portability

–sourcecodewrittenforonearchitecturecanbecopiedtoasecondarchitecture,compiledandexecutedwithoutmodification(tosomeextent)Support

MPMD

programsaswellas

SPMDInteroperability–theabilityofdifferentimplementationsofthesamespecificationtoexchangemessagesHeterogeneity(tosomeextent)

PVM&MPIaresystemsdesignedtoprovideuserswithlibrariesforwritingportable,heterogeneous,MPMDprograms79WhatisNotDifferent?可移植性-為一個(gè)架構(gòu)編寫的源代碼可以復(fù)制到第二個(gè)架構(gòu)，編譯并執(zhí)行，無需修改（在某種程度上）支持MPMD程序以及SPMD互操作性-相同規(guī)范的不同實(shí)現(xiàn)交換消息的能力異質(zhì)性（某種程度上）PVM和MPI是為用戶提供寫入便攜式，異構(gòu)的MPMD程序的庫(kù)的系統(tǒng)80ProcesscontrolAbilitytostartandstoptasks,tofindoutwhichtasksarerunning,andpossiblywheretheyarerunning.PVM

containsallofthesecapabilities–itcanspawn/killtasksdynamicallyMPI-1

hasnodefinedmethodtostartnewtask.

MPI-2

containfunctionstostartagroupoftasksandtosendakillsignaltoagroupoftasks81Processcontrol能夠啟動(dòng)和停止任務(wù)，查找正在運(yùn)行的任務(wù)，以及可能運(yùn)行的位置。PVM包含所有這些功能-

它可以動(dòng)態(tài)地生成/刪除任務(wù)MPI-1沒有定義的方法來啟動(dòng)新任務(wù)。MPI-2包含啟動(dòng)一組任務(wù)并向一組任務(wù)發(fā)送殺死信號(hào)的功能82ResourceControlPVM

isinherently

dynamic

innature,andithasarichsetofresourcecontrolfunctions.HostscanbeaddedordeletedloadbalancingtaskmigrationfaulttoleranceefficiencyMPI

isspecificallydesignedtobe

static

innaturetoimproveperformance83ResourceControlPVM本質(zhì)上是動(dòng)態(tài)的，它具有豐富的資源控制功能。主機(jī)可以添加或刪除負(fù)載均衡任務(wù)遷移容錯(cuò)效率MPI是專門設(shè)計(jì)為靜態(tài)的，以提高性能84Virtualtopology

-onlyfor

MPIConvenientprocessnamingNamingschemetofitthecommunicationpatternSimplifieswritingofcodeCanal