基于虚拟机迁移的虚拟机集群资源调度_周文煜陈华平杨寿

3920116()J.HuazhongUniv.ofSci.&Tech.(NaturalScienceEdition)Vol.39Sup.Jun.20112011202228.(19872),,,E2mail:[e-mail].(2006AA01A110);(60673172,60273041);.周文煜陈华平杨寿保方君(中国科学技术大学计算机科学与技术学院,安徽合肥230026),,,,,,,,.,.;;;;TP315A167124512(2011)S120130204ResourceschedulinginvirtualmachineclusterbasedonlivemigrationofvirtualmachineZhouWenyuChenHuapingYangShoubaoFangJun(SchoolofComputerScienceandTechnology,UniversityofScienceandTechnologyofChina,Hefei230026,China)AbstractTosolvetheproblemofunbalancingresourceloadstatusinvirtualmachinecluster,thispa2perproposesaresourceschedulingschemebasedonlivemigrationtechnology.Itclassifiesresourcesofvirtualmachine,describesitviavector,andpresentsthemeasurementsofthequantityandloadstatusofeachcategoryofresource.Itusesbest2fitalgorithmtoseekthedestinationhostofavirtualmachinetobemigratedto,andsettlethecoexistenceofloadbalancingandpowersavingbyusingup2perlimitandlowerlimittoconstrictresourceutilizationinnormalrange.Thefrequentmigrationcasedbyunstableresourceutilizationisalsopreventedbytheconceptofminimummigrationcycle.Thisschemeactstransparentresourceschedulingforapplicationsdeployedinvirtualmachine,addi2tionallybringsintheloadbalancingandpowersavingofvirtualmachinecluster.Keywordscloudcomputing;virtualization;cluster;resourcescheduling;self2adapting[1],,.(HA)[224],;LiveMigration[224],.,.,,.,,,;,2011-6-1712:45,,,/;,.,,,,.1,,,.,[5].;,,,.(),VMMCPU,.[6]CPU;[7];[829].,,CPU.,VMWareDRS[10],DRS.[11]CPU.,,,,,.2,,,,,.2.1,CPUI/O,CPU,Memory,NetworkI/O.4.CPUMemory,,;I/O,,,,.VMMI/O,,,.I/O,,I/O.2.22.,CPU,.CPU:a.VCPU,CPU;b.Weight,CPUVMMCPU;c.Cap,CPUCPU.MemorySize,.,CPU.I/O.2.31,,,.CPU,,#131#,:,I/O,,,I/O.13,,,,,,;,,;,,.3.1,.,,,.CPU3CPU,Mem,Net,3,.,RVM=RVM-CPU,RVM-Mem,RVM-Net,RPM=RPM-CPU,RPM-Mem,RPM-Net;3LVM=LVM-CPU,LVM-Mem,LVM-Net,LPM=LPM-CPU,LPM-Mem,LPM-Net;DVM=RVM#LVMT.CPU,Xen,6s,;.:K,K\5,K$tL,$tL,.,(),.,.,,,:foreveryVMin()do(1)VM;RVM,LVM,DVM;endforforeveryPMin()do(2)PM;RPM,LPM;endforforeveryVMin()do(3)ifnot(LVMLVM-up)thengetPMin()whichhasmax{LcPM|LcPMLPM-up};migrateVMtoPM;endifendforLavg=sum{LPM|LPMin()}/size();(4)if(not(LavgLPM-up))and(()isnotnull)thenresumeaPMfrom();elseifLavgLPM-frthengetPMin()whichhasmin{LPM};foreveryVMruninPMdoperform(3);endforgetPMoutfrom()andhaltit;#132#()39endiftmig.(5),RVM,LVMDVM$t,,RPMLPM$t.,,LPM-upLPM-fr,LVM-upLVM-fr.,.tmig,K$t.3.23.2.1CPUCPURPM-CPUCPUfNCPU,RPM-CPU=fNCPU.CPURVM-CPUCPUfCPUNVCPUVMMCPUWVM,RVM-CPU=fNVCPUWVM.CPUDVM-CPUCPURVM-CPU,DVM-CPU=RVM-CPULVM-CPU.3.2.2,,.3.2.3,.,VMM,,.,,.,,,.7@24.,,.[1]BarhamP,DragovicB,FraserK,etal.Xenandtheartofvirtualization[C]MProceedingsofthe19thACMSOSP.NewYork:ACMPress,2003:1642177.[2]LoweS.MasteringVMwarevSphere4[M].Indian2apolis:WileyPublishing,2009.[3]MatthewsJN,DowEM,DeshaneT,etal.Run2ningXen:ahands2onguidetotheartofvirtualization[M].Boston:PrenticeHall,2008.[4]WilliamsDE,BuytaertK,GarciaJR,etal.Virtu2alizationwithXenincludingXenEnterprise,XenServ2er,andXenExpress[M].Burlington:SyngressPub2lishing,2007.[5],,.[J].,2004,27(6):8032811.[6]Menasc†DA,BennaniMN.Autonomicvirtualizedenvironments[C]MProceedingsofICAS.06.Wash2ington:IEEEComputerSociety,2006:20230.[7]ZhaoWeiming,WangZhenlin,LuoYingwei.Dy2namicmemorybalancingforvirtualmachines[C]MProceedingsofVEE.09.NewYork:ACMPress,2009:21230.[8]SongYing,LiYaqiong,WangHui,etal.Aservice2orientedpriority2basedresourceschedulingschemeforvirtualizedutilitycomputing[C]MProceedingsofHiPC.08.Heidelberg:IEEEComputerSociety,2008:2202231.[9]SongYing,WangHui,LiYaqiong,etal.Multi2tieredon2demandresourceschedulingforVM2baseddatacenter[C]MProceedingsofCCGrid.09.Wash2ington:IEEEComputerSociety,2009:1482155.[10]VMwareInc.ResourcemanagementwithVMwareDRS[EB/OL].http:Mwww.vmware.com/vmtn/resources/401.[11]HuLiting,JinHai,LiaoXiaofei,etal.Magnet:anovelschedulingpolicyforpowerreductioninclus2terwithvirtualmachines[C]MProceedingsof2008IEEEInternationalConferenceonCusterCompu2ting.Tokyo:IEEEComputerSociety,2008:13220.#133#,:3920116()J.HuazhongUniv.ofSci.&Tech.(NaturalScienceEdition)Vol.39Sup.Jun.20112011202228.(19812),,,E2mail:[e-mail].(2006AA01A116,2006AA01A117);/0(INFO21152B01);(CNIC-QN-10004);(0815011103).曹宗雁1,2赵毅1牛铁1朱鹏1迟学斌1(1中国科学院计算机网络信息中心,北京100190;2中国科学院研究生院,北京100049),,.,LSF.2,.;;;;TP393A167124512(2011)S120134205PriorityschedulingofclusterjobsbasedonuserevaluationCaoZongyan1,2ZhaoYi1NiuTie1ZhuPeng1ChiXuebin1(1ComputerNetworkInformationCenter,ChineseAcademyofSciences,Beijing100190,China;2GraduateUniversityofChineseAcademyofSciences,Beijing100049,China)AbstractStandingontherequirementsofsupercomputingcenterconstructionandserviceproviding,aclusterjobschedulingstrategyisproposedoutwiththeaimofprovidingmoreavailableCPUhoursforscientificresearch,thatusersareevaluatedanddynamicallysetwithcorrespondingprioritytothejobscheduler.Someaspectsforuserevaluationarediscussed,includinguserbasicproperties,userac2tivitycharacters,applicationcategory,jobscale,runtimeestimationandfaulttoleranceofuserjobs.Thedetailofapplyingthepriorit