摘要
Withthegrowingpopularityofcloud-baseddatacenternetworks(DCNs),taskresourceallocationhasbecomemoreandmoreimportanttotheefficientuseofresourceinDCNs.Thispaperconsidersprovisioningthemaximumadmissibleload(MAL)ofvirtualmachines(VMs)inphysicalmachines(PMs)withunderlyingtree-structuredDCNsusingthehosemodelforcommunication.Thelimitationofstaticloaddistributionisthatitassignstaskstonodesinaonce-and-for-allmanner,andthusrequiresaprioriknowledgeofprogrambehavior.Toavoidloadredistributionduringruntimewhentheloadgrows,weintroducemaximumelasticityscheduling,whichhasthemaximumgrowthpotentialsubjecttothenodeandlinkcapacities.Thispaperaimstofindtheschedulewiththemaximumelasticityacrossnodesandlinks.Wefirstproposeadistributedlinearsolutionbasedonmessagepassing,andwediscussseveralpropertiesandextensionsofthemodel.Basedontheassumptionsandconclusions,weextendittothemultiplepathscasewithafattreeDCN,anddiscusstheoptimalsolutionforcomputingtheMALwithbothcomputationandcommunicationconstraints.Afterthat,wepresenttheprovisionschemewiththemaximumelasticityfortheVMs,whichcomeswithprovableoptimalityguaranteeforafixedflowschedulingstrategyinafattreeDCN.Weconducttheevaluationsonourtestbedandpresentvarioussimulationresultsbycomparingtheproposedmaximumelasticschedulingschemeswithothermethods.Extensivesimulationsvalidatetheeffectivenessoftheproposedpolicies,andtheresultsareshownfromdifferentperspectivestoprovidesolutionsbasedonourresearch.
出版日期
2019年01月11日(中国期刊网平台首次上网日期,不代表论文的发表时间)