简介: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.