sizeubuntu12.04lts

C.
-H.
Hsuetal.
(Eds.
):NPC2013,LNCS8147,pp.
118–129,2013.
IFIPInternationalFederationforInformationProcessing2013ITC-LM:ASmartIteration-TerminationCriterionBasedLiveVirtualMachineMigrationLiangweiZhu,JianhaiChen,QinmingHe,DaweiHuang,andShuangWuCollegeofComputerScience,ZhejiangUniversiy,ZhedaRd.
38,Hangzhou310027,China{zhulw,chenjh919,hqm,tossboyhdw,catting}@zju.
edu.
cnAbstract.
Livemigrationofvirtualmachines(VMs)playsanimportantroleingrids,cloudsanddatacenters,andhasbecomethecornerstoneofresourcemanagementinvirtualizedsystems.
Theefficiencyoflivemigrationdependsonthedowntime,totalmigrationtimeandtotaltransferreddata.
However,whilemigratingamemory-intensiveVM,XEN/KVMoftendomanyuselessiterationsofmemorycopyinordertoreachexpecteddowntimewhichcanneverbereached,leadingtoagreatdealofuselessdatatransferringandinsufferabletotalmigrationtime.
ItconsumesmassofnetworkbandwidthandCPUresourcewhentransferringmemoryfromonetoanothernode.
Hence,acriticaltaskistodeterminetheoptimaltimetoterminatethecopyiterationforlivemigration.
Inthispaper,weproposeasmartiteration-terminationcriterionbasedlivemigrationwhichistermedasITC-LM,toselfadaptivelycontrolwhentoterminateiteration.
WehaveimplementedITC-LMintoKVM/QEMU.
Theimprovementissignificant,especiallywhenmigrateamemory-intensiveVM.
Theexperimentalresultsshowthat,ourapproachcandecrease50.
33%oftotaltransferreddataonaveragewithoutimpairingmigrationdowntime.
Keywords:Virtualization,Livemigration,Iteration-TerminationCriterion,TerminatingConditions.
1IntroductionVirtualizationtechnologyplaysanimportantroleincloudcomputing[1,2],inwhichalargenumberofvirtualmachines(VMs)aredynamicallyallocatedtomultiplephysicalmachines(PMs).
VirtualizationenableslivemigrationofVMs,whichprovidesaflexiblewaytorelocateVMsfromonephysicalnodetoanother,leadingtoefficientresourcemanagement[3],suchasloadbalancingandpower-savingetc.
Livemigrationofvirtualmachineistheessentialmechanismofvirtualization,whichisincludedinallcurrentmainstreamvirtualizationplatformssuchasKVM[4],XEN[5],VMware[6],etc.
Pre-copyasdefaultlivemigrationmethodoftheseplatformsisthemostpopularalgorithmoflivemigration,whichfirstsendsVM'smemorytothedestinationhostandthenresumesVMinit.
Inordertoimprovemigrationperformance,someresearchemploysmemorycompressiontoreducethetransferredpagesduringlivemigration.
ITC-LM119However,theefficiencyofcurrentlivemigrationmethodsarenotalwayssatisfying,especiallyinamemory-intensivescenario.
Nowmethodskeepstheiterationsofcopyingmemoryuntilthedowntimeisshortenough,andiftheexpecteddowntimeisalwaystoohigh,itstopsmemorycopyiterationsafterafixednumberofmemory-copyiterations.
Asaresult,thesemethodsusuallystopmemory--copyiterationtoolateinmemory-intensivescenariosandwillconsumealotofnetworkresourceofadatacenter,evenresultingintheperformancedegradationofdatacenter.
Unfortunately,theexistinglivemigrationapproachescommonlyignoretheproblemofdeterminationaboutwhentoterminatetheiterations.
Inpre-copy,memoryistransferredfromthesourcenodetothedestinationnodewhileVMisstillrunningonsource.
Modifiedpageswhicharegeneratedintheiterationaretransferredinsubsequentiteration.
ThisprocessbasedonthattheVMremainingdirtypagescanconvergetoasmallvalue.
However,therearestilllotsofremainingdirtypagesaftermultipleiterationswhenlow-bandwidthorVMwithahighworkload.
Inotherwords,moreiterationsofcopymemorycannotdecreasetheremainingdirtypagesandnomorebenefittotheservicedowntime.
Inthispaper,weproposeasmartiteration-terminationcriterion(ITC)basedlivemigrationmethod,termedasITC-LM.
WeuseITCvaluetodeterminewhentoterminatetheiterationinlivemigration.
Actually,ITCvaluechangesdynamicallyduringthelivemigrationprocess.
Inourmethod,weterminatetheiterationwhentheITCvalueislessthanagiventhreshold.
Themaincontributionsofthispaperareasfollows:First,weproposedasmartITC-LMtechniquetodecidewhentostopiterationduringlivemigration.
Tothebestofourknowledge,ourmethodisthefirsttoconsiderboththeiterationroundsandtheconvergenceofremainingdirtypages.
Second,wehaveimplementedourproposedalgorithminrecentstablereleaseofKVM/QEMU[9,10]andshowthatourmethodscanbeconvenientlydeployedinvirtualizationplatform.
Eventually,wedemonstratetheeffectivenessofITC-LMmethodinourexperimentsbyfourdifferentkindsofworkloads.
TheresultsshowthatITC-LMdecreases53.
35%oftotalmigrationtimeand50.
33%oftotaltransferreddataonaverage.
2BackgroundandMotivationTherearevariousalgorithmsoflivemigration.
Mostofthesestudiesaremainlybasedonthepre-copylivemigrationalgorithm.
Itbasicallyworks[21]asfollows:1)TheresourcesofmemoryandVCPUsarereservedonthedestination.
2)MemoryofVMissenttodestinationandusingbitmaptologthedirtypageswhichrewriteduringthememorycopy.
3)ThesourcecontinuouslycopiesVM'smemorydirtypagestothedestination.
Anumberofiterationsareperformedtoretransferthepageswhicharedirtiedinpreviousiteration.
120L.
Zhuetal.
4)SuspendtherunningVMatthesource,andcopyremainingpagestothedestination.
5)TheVMisresumedonthedestination.
Inthebestcase,theapproachofpre-copycanachieveanexpecteddowntimebyseveraliterativecopyoperations.
2.
1TerminatingConditionsTherearesomecommonconditionstodecidethetimetoterminatetheiterativecopyoperations.
RemainingDirtyPages.
Ideally,thedefaultsizeofremainingdirtypagescanbereachedusingiterativecopyoperations.
Howeverthisiscompletelydependingontheassumptionofremainingdirtypagescanconvergeexpectedsize.
Someapplicationrewritememoryfrequencythatremainingdirtypagesstillverylargeovermultiplerounds[17].
MaximumNumberofIterations.
Theterminativeconditionsofexactvaluesforthemaximumnumberofiterationarearguable,sometimeswhenthedirtyrateislowcomparedtothetransferrate,theremainingpageswilldecreasequicklyanddowntimewillnothavebenefitfrommoreiteration.
HybridTerminatingCondition.
Someofapproachesuseterminativeconditionssimultaneouslyofabovetwoasshowninfigure1.
Eachiterationchecktheremainingpageswhetherbelowtheexpectedremainpages,andcheckthenumberofiterationismorethanthemaximumnumberofiterations.
ThisconditionhasbeenusedinXEN,KVM/QEMUetc.
Fig.
1.
AtypicalalgorithmofusetwoterminativeconditionssimultaneouslyITC-LM1212.
2TheProblemsofCommonTerminatingConditionsThehybridterminatingconditioncanfinishthelivemigrationbutthedefaultremainingpagessizeandmaximumnumberofiterationsishardtoestimate.
Beside,wecanknowthatinordertominimizethesizeofremainingdirtypagesthemaximumiterationsneedtobesettoavaluethatlargerenoughformostcasesoflivemigration.
However,alotofuselessiterationsofpre-copycausedlargeamountoftransferreddata,especiallywhenmigrateamemory-intensiveVM.
(a)Remainingdirtypages(b)TransferreddataFig.
2.
NumberofremainingdirtypagesandtransferreddataofeachmigrationiterationroundswhenmigrateaVMwhichrunningworkloadofRUBiSwebserverworkload.
VMsizeis2GB.
Fromfigure2weobservethattheremainingpagesconvergefromthefifthiteration.
Buttheremainingdirtypagessizestillbiggerthanthedefaultsize30MB(someapproachesuseremainingpagessizedividetransferrate).
Soiterationsofcopywillcontinuousperformuntilthenumberofiterationsexceedsthemaximumnumberwhichissetasthirty-seventhatlargerenoughformostcasesoflivemigrationinourexperiments.
3AlgorithmDesignAlthough,thelivemigrationcompleteafternumberofiterationexceedsthemaximumnumber,themoreiterationthemorenetworkbandwidthandCPUresourceitconsumesanditwouldprolongthetotalmigrationtime.
Moreover,itmakesapplicationsinmigratedVMsufferlongertimeofperformancedegradation.
Theiterationsofpre-copyarecompletelydependingontheassumptionoftheremainingpagescanconvergetoasmallvalue.
Howeveriterationofpre-copymayneverconvergetoasmallvalueoreventheconvergenceofiterationneverhappenwhentheVM'sworkloadishighorbandwidthinstability.
Sothereshouldbesomesmartthresholdtoforcethefinaliterationofalivemigrationwhichdoesnotconverge.
122L.
Zhuetal.
3.
1TerminatingConditionBasedonRemainingDirtyPagesFirst,somenotationsaredefinedasfollows:irM:Theremainingdirtypagesafterithiterationinthemigratedvirtualmachine.
itM:Thedataistransferredtodestinationofithiteration.
iT:Thetimeofcompleteithiterationused.
ipageR:Theaveragedirtypagesrateofithiteration.
itranR:Theaveragetransferredrateofithiteration.
TheaveragedirtypagerateofithiterationcanbecalculatedasitrrpageTMMMRiiii)(1=.
(1)Theaveragetransferredrateofithiterationcanbepresentedas:ittranTMRii=.
(2)OnlywhenitranRisgreaterthanipageR,theiterationcanreducetheremainingpages.
Fromformula(1)and(2)wecanget:1=<=0,00,1iiDDα,trustc_isaconstantvaluewhichdenotesthetrustvalue,anddistrustc_isaconstantvaluewhichdenotesthedistrustcoefficient.
Accordingtoourexperiments,thevalueoftrustc_anddistrustc_isempiricalandcorrelative.
Forexample,whenwesettrustc_to1andsetdistrustc_to2,thenweobtainagoodperformanceoflivemigrationinourexperimentsandwewilldomoreresearchaboutvaluesoftrustc_anddistrustc_inourfuturework.
3.
3ITC-LMAlgorithmWehavedesignedtheITC-LMAlgorithmandimplementeditinrecentstablereleaseofKVM/QEMUusingITC.
QEMUisanopensourcemachineemulator.
KVM(forKernel-basedVirtualMachine)isafullvirtualizationsolutionforLinuxonx86hardwarewhichcontainingvirtualizationextensions.
WhenQEMUuseKVMforitsvirtualizationacceleration,cangetabetterperformance.
ThemajorpartofITC-LMwasimplementedinQEMU.
Weusebitmapofmemorytocalculatethesizeofdirtypages.
ThepseudocodeofITC-LMalgorithmislistedinfollows:PseudoCodeofITC-LMalgorithm.
ITC=0;c_trust=1;c_distrust=2;pre_remain_pages_size=full_memory_size();while(true){copyiteration()if(remain_pages_size()Inthefollowing,wefirstintroducetheexperimentalenvironment,andthenwepresenttheresultsofdifferentbenchmarks.
4.
1ExperimentalEnvironmentAlllivemigrationexperimentsareperformedontwoidenticalhostsassourceanddestinationhostrespectively.
EachhosthasdualIntel(R)Xeon(R)CPUE5606@2.
13GHZwithatotalofeightcores.
Eachonehas16GBRAM.
ThehostrunsUbuntu12.
04LTSwithKVMmoduleandQEMU-1.
4.
0.
ThesourceanddestinationhostsareconnectedviaGigabitswitchedEthernet.
TheOSofLinuxVMareallUbuntu12.
04LTSwithKernel-3.
2.
0andallVMimagesarestoredinaNetworkFileSystem(NFS).
WeperformexperimentsVMsareconfiguredwithtwovirtualCPUs.
IneachexperimentthenodesofsourceandtargetonlyrunthelivemigrationVM.
Besides,weuseanotheridenticalhosttodeploymenttheclientemulatorofRUBiS.
4.
2OverviewofWorkloadWeperformourexperimentswiththefollowingVMworkloads:1.
Kernel-complication.
LinuxkernelcompilationisabalancedworkloadtousethesourceofVM.
TwoparallelthreadswereusedtorunLinux3.
8.
5kernel[12]compilation.
2.
ParallelBenchmarks.
TheNASparallelbenchmarks(NPB)[14]areasetofprogramswhichevaluatetheperformanceparallelperformance,availableincommonly-usedprogramingmodelslikeMPIandOpenMP.
Inourexperiments,weuseEmbarrassinglyParallel(EP)ofNPB(NPB-EP)tosimulationtheparallelcomputingworkload.
3.
SPECjbb2005[13].
ItisaSPEC'Sbenchmarkforevaluatingtheperformanceofserversidejava.
Itprovidesenhancedworkloadwithamoreobject-orientedmannertoreflectreal-worldapplications.
4.
DynamicwebServer.
RiceUniversityBiddingSystem(RUBiS)[11]whichisaprototypemodeledaftereBay.
comusedtoevaluatepatternsandapplicationserversperformance.
Itcontainsaclient-browseremulator,andweimplementitinathirdphysicalhost.
4.
3ExperimentResultsInthissection,theresultsfromthefourworkloadsarepresent.
Theevaluationmetricofexperimentsprimarilyincludestotalmigrationtime,totaltransferreddataanddowntimeduringlivemigrationofvirtualmachine.
Werunlivemigrationfivetimesforeachworkloadandusethearithmeticmeanforeachmetric.
126L.
Zhuetal.
(a)Kernel-Compiling(b)NPB-EP(c)SPECjbb(d)RUBiSFig.
4.
RemainingdirtypagesofeachiterationcomparedwithITC-LMandPre-defaultforeachworkload.
(Memory1GB)TotalTransferredData.
Figure4respectivelyshowstheeachiterationremainingdataofduringlivemigrationofVMrunning.
TheiterationroundsofITC-LMfarlessthanthepre-copydefaultalgorithminKVM/QEMUundertheworkloadofkernel-compile,SPECjbbandRUBiS.
Asfigure4(b)showsthebenchmarkofNPB-EPisacompute-intensiveworkloadandVMproducelessdirtypages,anditerationendingsoon.
Thustwoapproachesbothinlessiteration.
Experimentalresultsinfigure5(a)showthatcomparewithKMV/QEMU'sdefaultmigrationalgorithm,ITC-LMcanreducetotaltransferreddata50.
73%,7.
74%,69.
55%,73.
29%respectivelyindiverseworkloadofabove,anaverageof50.
33%.
Thiswilllightengreatlynetworkloadsofdatacenter.
TotalMigrationTime.
Theresultsinfigure5(b)showthatbenefitfromlessiteration,ITC-LMcanreducetotalmigrationtime50.
54%,17.
70%,70.
03%,75.
14%,anaverageof53.
35%.
Forthis,theservicerunninginlivemigrationVMcansufferlesstimeofthedecreaseofquality.
Downtime.
InordertoevaluatetheinfluenceofITC-LMondowntime,experimentsareperformedwithtwomemorysize:1GBand2GB.
Figure5(c)and(d)showsITC-LMcangetagoodperformanceintotaltransferreddataandtotalmigrationtimewhilehasslightinfluenceonthedowntimeoflivemigration.
ThisisbecauseITC-LMbasedonasmartthresholdnotadefaultremainingdirtypagesormaximumiterationrounds.
ITC-LM127(a)Totaltransferreddata(Memory1GB)(b)Totalmigrationtime(Memory1GB)(c)Downtime(Memory1GB)(d)Downtime(Memory2GB)Fig.
5.
Totaltransferreddata,totalmigrationtimeanddowntimeofITC-LMandPre-defaultduringlivemigrationforfourdifferentkindsofworkloads5RelatedWorkThetechnologyoflivemigrationiswidelyusedinvirtualization.
Atpresent,thereareseveraltypesoflivemigrationmethods.
Thepre-copyapproachisamainmigrationmethodinthemainstreamvirtualizationplatformsuchasKVM[4],XEN[5],VMware[6],etc.
Someresearchhasbeendonetoimprovetheperformanceoflivemigrationbasedonpre-copyinwhichthewidespreadusedmethodismemorycompression.
Zhangetal.
[15]proposedanovelapproachMMDtofindidenticalandsimilarmemorypagestoredundantmemorydata.
Deltacompressiontechnique[16]appliedXORonthecurrentpageswithkeptpreviouslysentpagesinsourcehost.
Jinetal.
[17]designedanadaptivememorycompressionbasedonmemorypagecharacteristics.
ME2[18]identifiedusefulpagesandthenusedRLEalgorithmtocompressdata.
Pagerewritingfrequencyisrelatedtodirtypagerate.
Microwiper[19]ordereddirtymemorypagesaccordingtheirrewritingrate.
Petteretal.
[20]designedapageprioritymapontopofthedirtypagebitmapandproposeddynamicpagetransferreorderingbasedonit.
CR/TR-Motion[22]logexecutiontraceonsourceandreplayitontargethost.
Chiangetal.
[27]proposedabootstrappingVMintrospectiontechniquetogetthe128L.
Zhuetal.
informationofmemorypool,andskipsfreememorypagesduringmigration.
Joetal.
[28]firstsentthememory-to-diskmappingtothehost,andthenfetchthecontentsdirectlyfromthesharedstorage.
Post-copy[7]resumesrunningVMonthetargethostwithonlyitsCPUstatebeforecopyingtheVM'smemoryfromsourcehosttotarget.
Adaptivepre-paginganddynamicself-ballooning[8]canimprovethepost-copyperformance.
Hirofuchietal.
[23]throughalightweightextensionimplementpost-copytoKVM.
Besides,someresearchesproposedhybridlivemigration[24,25,26]approachthatusepre-copyandpost-copymethodssimultaneously.
Theydosomememorycopyiterationofpre-copybeforethestageofpost-copymethod.
6ConclusionsandFutureWorkInthispaper,wehavepresentedthedesignandimplementationofasmartITC-LMtechniqueforlivemigrationofvirtualmachines.
Wechoosefourrepresentativeserverapplicationsinmoderndatacentertoverifyouralgorithm.
TheresultsshowthatITC-LMhasagoodperformanceindifferentkindsofworkload.
Infuturework,wewillstudytrustvalueanddistrustcoefficientofITC-LMtomakeourapproachmoreeffective.
Furthermore,wewillimplementourapproachtoothervirtualizationplatforms.
Acknowledgments.
ThisworkwaspartlysupportedbytheNationalKeyTechnologiesR&DProgramunderGrantsNo.
2011BAD21B02.
References1.
Armbrust,M.
,Fox,A.
,Griffith,R.
,etal.
:Aviewofcloudcomputing.
CommunicationsoftheACM53(4),50–58(2010)2.
Fox,A.
,Griffith,R.
,Joseph,A.
,etal.
:Abovetheclouds:ABerkeleyviewofcloudcomputing.
Dept.
ElectricalEng.
andComput.
Sciences,UniversityofCalifornia,Berkeley,Rep.
UCB/EECS,28(2009)3.
Wang,X.
,Du,Z.
,Chen,Y.
,etal.
:Virtualization-basedautonomicresourcemanagementformulti-tierWebapplicationsinshareddatacenter.
JournalofSystemsandSoftware81(9),1591–1608(2008)4.
Kivity,A.
,Kamay,Y.
,Laor,D.
,etal.
:kvm:theLinuxvirtualmachinemonitor.
In:ProceedingsoftheLinuxSymposium,vol.
1,pp.
225–230(2007)5.
Barham,P.
,Dragovic,B.
,Fraser,K.
,etal.
:Xenandtheartofvirtualization.
ACMSIGOPSOperatingSystemsReview37(5),164–177(2003)6.
Nelson,M.
,Lim,B.
H.
,Hutchins,G.
:Fasttransparentmigrationforvirtualmachines.
In:ProceedingsoftheAnnualConferenceonUSENIXAnnualTechnicalConference,p.
25(2005)7.
Hines,M.
R.
,Deshpande,U.
,Gopalan,K.
:Post-copylivemigrationofvirtualmachines.
ACMSIGOPSOperatingSystemsReview43(3),14–26(2009)8.
Hines,M.
R.
,Gopalan,K.
:Post-copybasedlivevirtualmachinemigrationusingadaptivepre-paginganddynamicself-ballooning.
In:Proceedingsofthe2009ACMSIGPLAN/SIGOPSInternationalConferenceonVirtualExecutionEnvironments,pp.
51–60.
ACM(2009)ITC-LM1299.
KernelBasedVirtualMachine,KVM,http://www.
linux-kvm.
org/page/Main_Page10.
QEMU,http://wiki.
qemu.
org/Main_Page11.
RiceUniversityBiddingSystem,RUBiS,http://rubis.
ow2.
org/12.
Linux-kernel,https://www.
kernel.
org/13.
StandardPerformanceEvalutionCorporation,SPECJbb2005,http://www.
spec.
org/jbb2005/14.
NASParallelBenchmarks,NPB,http://www.
nas.
nasa.
gov/publications/npb.
html15.
Zhang,X.
,Huo,Z.
,Ma,J.
,etal.
:Exploitingdatadeduplicationtoacceleratelivevirtualmachinemigration.
In:2010IEEEInternationalConferenceonClusterComputing(CLUSTER),pp.
88–96.
IEEE(2010)16.
Svrd,P.
,Hudzia,B.
,Tordsson,J.
,etal.
:Evaluationofdeltacompressiontechniquesforefficientlivemigrationoflargevirtualmachines.
VirtualExecutionEnvironments(VEE)46(7),111–120(2011)17.
Jin,H.
,Deng,L.
,Wu,S.
,etal.
:Livevirtualmachinemigrationwithadaptive,memorycompression.
In:IEEEInternationalConferenceonClusterComputingandWorkshops,CLUSTER2009,pp.
1–10.
IEEE(2009)18.
Ma,Y.
,Wang,H.
,Dong,J.
,etal.
:ME2:EfficientLiveMigrationofVirtualMachinewithMemoryExplorationandEncoding.
In:2012IEEEInternationalConferenceonClusterComputing(CLUSTER),pp.
610–613.
IEEE(2012)19.
Du,Y.
,Yu,H.
,Shi,G.
,etal.
:Microwiper:EfficientMemoryPropagationinLiveMigrationofVirtualMachines.
In:201039thInternationalConferenceonParallelProcessing(ICPP),pp.
141–149.
IEEE(2010)20.
Svard,P.
,Tordsson,J.
,Hudzia,B.
,etal.
:Highperformancelivemigrationthroughdynamicpagetransferreorderingandcompression.
In:2011IEEEThirdInternationalConferenceonCloudComputingTechnologyandScience(CloudCom),pp.
542–548.
IEEE(2011)21.
Clark,C.
,Fraser,K.
,Hand,S.
,etal.
:Livemigrationofvirtualmachines.
In:Proceedingsofthe2ndconferenceonSymposiumonNetworkedSystemsDesign&Implementation,vol.
2,pp.
273–286.
USENIXAssociation(2005)22.
Liu,H.
,Jin,H.
,Liao,X.
,etal.
:Livevirtualmachinemigrationviaasynchronousreplicationandstatesynchronization.
IEEETransactionsonParallelandDistributedSystems22(12),1986–1999(2011)23.
Hirofuchi,T.
,Nakada,H.
,Itoh,S.
,etal.
:Enablinginstantaneousrelocationofvirtualmachineswithalightweightvmmextension.
In:201010thIEEE/ACMInternationalConferenceonCluster,CloudandGridComputing(CCGrid),pp.
73–83.
IEEE(2010)24.
Sahni,S.
,Varma,V.
:AHybridApproachtoLiveMigrationofVirtualMachines.
In:2012IEEEInternationalConferenceonCloudComputinginEmergingMarkets(CCEM),pp.
1–5.
IEEE(2012)25.
Shribman,A.
,Hudzia,B.
:Pre-Copyandpost-copyVMlivemigrationformemoryintensiveapplications.
In:Caragiannis,I.
,etal.
(eds.
)Euro-ParWorkshops2012.
LNCS,vol.
7640,pp.
539–547.
Springer,Heidelberg(2013)26.
Chen,Y.
,Huai,J.
P.
,Hu,C.
M.
:Livemigrationofvirtualmachinesbasedonhybridmemorycopyapproach.
ChineseJournalofComputers34(12),2278–2291(2011)27.
Chiang,J.
H.
,Li,H.
L.
,Chiueh,T.
:Introspection-basedmemoryde-duplicationandmigration.
In:Proceedingsofthe9thACMSIGPLAN/SIGOPSInternationalConferenceonVirtualExecutionEnvironments,pp.
51–62.
ACM(2013)28.
Jo,C.
,Gustafsson,E.
,Son,J.
,etal.
:Efficientlivemigrationofvirtualmachinesusingsharedstorage.
In:Proceedingsofthe9thACMSIGPLAN/SIGOPSInternationalConferenceonVirtualExecutionEnvironments,pp.
41–50.
ACM(2013)