User Tools

Site Tools


federation_research

ClommunityWiki

Original file

Auto-ConfigurationofMultipleCloudDataCentersinCommunityNetworkingCloud

= CommunityNetworkingCloud =

Communitynetworkingcloudistomaintaincloudservicesontopofcommunitynetworks.Clommunity1)isanongoingresearchprojectinvestigatingthisconcept.Theideaisthatallthecloudmachinesareprovidedbythecommunitymembers.Thecloudservicesarehostedovermultiplemachineswhicharegeographicallydistributed.

TypicalcloudcomputingsolutionsarebasedonalargedatacenterfilledwithhardwareandsoftwarerequiredforprovidingdifferentX-as-a-Servicemodels.Tocovermoregeographicalareasandtoincreasethequalityofservices,cloudprovidersutilizedifferentdatacenterslocatedclosertotheusers.MainlyVMsrunninginthesamedatacenterhavelowcommunicationcost,lowlatencyandhighreliabilitywhilethisdoesnotapplyforinter-data-centerVMscommunications.

InClommunitythesemachinesareofferedbycommunitymembersandthereisnocontroloverthetopologyoftheunderlyingnetworkconnectingthemandthemachineshaveheterogeneousresources.Toovercomethisproblem,wehavebeenworkingoncentralizedanddistributedsolutionsbasedoncommunitydetectionmethodstoclusterClommunitymachinesintomultipledatacenters2)inwhicheachdatacentercanprovideabetterqualityofserviceguarantees.

= ProblemDefinition =

ThenetworkismodelledasagraphinwhichnodesofthegraphrepresentstheCloudmachinesandtheedgesarethenetworkconnections.ThegraphisaweighteddirectedgraphG={V,E}duetothenetworkasymmetries.VisthesetofverticesrepresentingthecomputingnodesandEisthesetofedgesrepresentingthenetworkconnectionbetweentwonodes.Inourinitialevaluationweconsiderallthemachineshavingthesameamountofresourcesandreliability3).EveryvertexvhasasetofattributesincludingaUID,latitudeandaltitude.EveryedgeehasattributesUID,bandwidthandlatency.

Wedefineasetofobjectivessuchasdecreasingthelatencybetweenthenodesandincreasingthebandwidthbetweenthenodesofadatacenter.Itcanbeextendedtootherdemandingobjectivesaswell.BasedonthegivengraphGandtheobjectives,weformulatetheproblemasacommunitydetectionproblem,inwhicheverycommunitywillbeanindividualClouddatacenter,inordertoimprovetheobjectives(O={o1,…,on})forthewholesystem.

= Solution =

Tofindthecommunities,wecaneitherusecentralordistributedcommunitydetectionalgorithms.Eachofthemhastheirownconsandpros.Becauseofthedistributednatureofthecommunitynetworkingcloudsweproposeadistributedcommunitydetectionalgorithmwhichcanfindthecommunitiesinaweighteddirectedgraph.Thealgorithmisbasedontheideaofcolordiffusionandeverynodeofthegraphcandecidetowhichcommunityitbelongsonlybasedonitsadjacentnodes.

= Evaluation =

Toevaluateourworkwesimulatedtheenvironment.Inthefollowingsubsectionsweexplainhowwemodeledthenetworkandthecloudcomputingplatformservicesandweanalyseifourapproachcanimprovetheobjectives.

= ModellingTheNetwork =

Asanexample,wemodeledQMPSU4)usingasnapshotoftheirdatasetwhichisdepictedinthefigurebelow.Weusedthisdatasettoevaluateourwork.

= ModellingCloudComputingPlatform =

ToevaluateoursolutionweselectOpenStackasacloudcomputingplatform.Inhere,wediscusshowwemodelledOpenStack.

= OpenStackServicesandtheirinter-communication =

OpenStackisanopen-sourceCloudComputingPlatformwhichcanbeusedtoprovideInfrastructure-as-a-Serviceontopofubiquitoushardwares.FivemajorcomponentsofOpenStackareKeystone(identity),Glance(image),Cinder(blockstorage),Neutron(network)andNova(computing).Eachofthesecomponentsarebuiltofmultipleloosely-coupledserviceswhichcanberunonseparatemachines.

Inthefigurebelow,theOpenStackservicesandtheirinteractionsareshown.

IneveryOpenStackoperation,manycommunicationshappenbetweentheseservices.Asanexample,thecommunicationsbetweentheservicesforbootingaVMarecountedinthefigureabove.

= OpenStackCloudArchitecture =

Forevaluatingouralgorithm,wehavedefinedthreedifferenttypesofnodesineverycloudzonebasedonascalablearchitecturerecommendedforasmallandmedium-sizedOpenStackcloud5).Acontrollernodewhichincludeskeystone,nova,neutron,glance,cinderandimagerepositoryservices,anodetohostmessagequeueanddatabaseservicesandtherestofthenodesinthedatacenterarecomputenodes.

= EvaluationResults =

WeimplementedthealgorithmandthesimulationenvironmentinJava.Inthefigurebelow,thefourdetectedcommunitiesonthenetworkofQMPSUareshown:

EachofthesecommunitiesareconsideredasaClouddatacenter.Wehaveimplementedourownsimulationenvironmenttoevaluateouralgorithm.ThesimulationassignsdifferentOpenStackrolesasmentionedbefore(controller,messagequeueanddatabase,andcomputeroles)andthencomputesthecostofdifferentOpenStackoperations.

1) “ACommunityNetworkingCloudinaBox-CLOMMUNITY.”2013.20Jun.2014<http://clommunity.blog.pangea.org/files/2013/02/Project_Factsheet_Clommunity.pdf>
2) Wecalleachcommunityaclouddatacenter.However,othertermscanbeappliedsuchascloudzonesormicrocloudswhicharefederatedthroughabrokerorotherfederationmethods.
3) Wearecurrentlyworkingonadynamicenvironmentwherenodesjoinandleaveandmachineshaveheterogeneousconfigurationsintermsofcomputingpowerandstorageanditwillbepublishedsoon.
4) Cerdà-Alabern,Llorenç,AxelNeumann,andPauEscrich.”Experimentalevaluationofawirelesscommunitymeshnetwork.”Proceedingsofthe16thACMinternationalconferenceonModeling,analysis&simulationofwirelessandmobilesystems3Nov.2013:23-30.
federation_research.txt · Last modified: 2015/01/15 14:48 by hooman