基于MPI的集群計(jì)算系統(tǒng)設(shè)計(jì)與實(shí)現(xiàn).pdf

1、哈爾濱工程大學(xué)碩士學(xué)位論文基于MPI的集群計(jì)算系統(tǒng)設(shè)計(jì)與實(shí)現(xiàn)姓名：姚成浪申請(qǐng)學(xué)位級(jí)別：碩士專業(yè)：計(jì)算機(jī)應(yīng)用技術(shù)指導(dǎo)教師：李靜梅20070101AbstractT11etedmologyofcomputerhaspromptedtherapidgrowthofthecomputationalscienceThenewgenerationofcomputershouldbebetterthallthefo珊ereitherinthepowe

2、rorthespeedofcomputingFurthermorethedemalldofcomputingisincreasingHoweverinpracticesingleprocessorstillcan’tsatlsfythedemandofsomeengineeringcalculationInadditiontoimprovingthecoreputingcapabilityofprocessor’parallelcomp

3、utingisalsoaPotentwaYtoincrcasethecapabilityofcomputingTheparallelcomputingcanbewellimplementedindustercomputingsystem，whichisnecessarytoorganlzeseVeral卿uterStoworktogetherinordertOgetpowerfulparallelcomputingTheseheavyt

4、asksaredistributedtoallofcomputingsystemexistingcomputationresourcesbyclusterInthethesis，thestructureandfunctionofLinuxdusterarefirstlydiscussedindetails，andthenthestateofdusterandtheparallelprogrammingmodel1nparallelcom

5、putingarcanalyzedItisresearchedonthesinglesystemlmageandtheclustermiddlewareAnewtypeofparallelclustersystemisdesignedusmgdustermiddlewareAfterimplementingfunctionsofsomemodulesinclustersVstem，thissystemhasatestingcapabil

6、ityfornetworkoverload，andltcandynamicallvrcflectthechangefeaturesfornodesandrationallydistributetasks，accordingtotheloadofnodeTheseresearchesandimplementationsarepopresearchsubjectsonparallelcomputingsystem，andprovideref

7、erencestorrelativeresearchesAttheendofthisthesis，twoparallelapplications，usingstaticdispatchingalgorithmanddynamicweightedroundrobindispatchingalgonthms印arately，aredesignedbasedonthisclustersystemTheexternalperformanceof