Mobile and Cloud Computing srirama/talks/MobileandCloudComputingLaboratory... · Mobile and Cloud Computing…

  • Published on
    09-Jul-2018

  • View
    212

  • Download
    0

Embed Size (px)

Transcript

<ul><li><p>MobileandCloudComputingLaboratory</p><p>MajorResearchInterests</p><p>SatishSriramasatish.srirama@ut.ee</p></li><li><p>WhoamI</p><p> HeadofMobile&amp;CloudLab,InstituteofComputerScience,UniversityofTartu,Estonia</p><p>http://mc.cs.ut.ee</p><p>2/24/2015 SatishSrirama 2</p></li><li><p>TARTU</p><p>Pop:100,000</p><p>Estoniapop:1,300,000</p><p>2/24/2015 SatishSrirama 3</p></li><li><p>Academicexcellencesince16322/24/2015 4</p></li><li><p>CS@UT KeyNumbers</p><p>600 students600 students</p><p>60 staff60 staff 40academic(professors,associateprofessors) 20fulltimeresearchers</p><p>60 industryguestlecturers/year60 industryguestlecturers/year</p><p>60 PhDstudents60 PhDstudents</p><p>HighlyinternationalprofileHighlyinternationalprofile 60% ofacademicstaffhaveinternationalexperience 60% internationalPhDandMastersstudents</p></li><li><p>6Curricula</p><p>Bachelors(3years)</p><p>ComputerScience</p><p>Masters(2years)</p><p>ComputerScience</p><p>SoftwareEngineering(jointTUT)</p><p>PhD(4y)</p><p>ComputerScience</p><p> MastersofSecurityandMobileComputing(withAalto,KTH,DTU,NTNU) MastersofCyberSecurity(coordinatedbyTUT)</p></li><li><p>6KeyCompetenceAreas</p><p>SoftwareEngineering</p><p>http://sep.cs.ut.ee</p><p>Distributed&amp;MobileandCloud</p><p>Computinghttp://mc.cs.ut.eehttp://ds.cs.ut.ee</p><p>DataMining,Bioinformatics&amp;Neuroscience</p><p>http://biit.cs.ut.ee</p><p>ProgrammingLanguages</p><p>http://lambda.cs.ut.ee</p><p>Security,Crypto&amp;Codinghttp://crypto.cs.ut.ee/</p><p>EstonianLanguageTechnology</p><p>SoftwareTechnologyCompetenceCentre(STACC.ee)</p></li><li><p>Mobile&amp;CloudLab MainResearchActivities</p><p>2/24/2015 SatishSrirama 8</p></li><li><p>Outline</p><p> Cloudcomputing Migratingenterpriseapplicationstothecloud Scientificcomputingonthecloud MobileCloud &amp;InternetofThings</p><p>2/24/2015 SatishSrirama 9</p></li><li><p>WhatisCloudComputing? Computingasautility</p><p> Utilityservicese.g.water,electricity,gasetc Consumerspaybasedontheirusage</p><p> CloudComputingcharacteristics Illusionofinfiniteresources Noupfrontcost Finegrainedbilling(e.g.hourly)</p><p>2/24/2015 SatishSrirama 10</p></li><li><p>CloudComputing Services SoftwareasaService SaaS</p><p> Awaytoaccessapplicationshostedonthewebthroughyourwebbrowser</p><p> PlatformasaService PaaS Providesacomputingplatform</p><p>andasolutionstack(e.g.LAMP)asaservice</p><p> InfrastructureasaServiceIaaS Useofcommoditycomputers,</p><p>distributedacrossInternet,toperformparallelprocessing,distributedstorage,indexingandminingofdata</p><p> Virtualization</p><p>SaaS</p><p>Facebook,Flikr, Myspace.com,Googlemaps API,Gmail</p><p>PaaS</p><p>Google App Engine, Force.com, Hadoop, Azure, </p><p>Heroku, etc</p><p>IaaSAmazon EC2, Rackspace, GoGrid, SciCloud, etc.</p><p>LevelofAbstraction</p><p>2/24/2015 SatishSrirama 11</p></li><li><p>CloudComputing Themes</p><p> Massivelyscalable Ondemand&amp;dynamic Onlyusewhatyouneed Elastic</p><p> Noupfrontcommitments,useonshorttermbasis AccessibleviaInternet,locationindependent Transparent</p><p> Complexityconcealedfromusers,virtualized,abstracted</p><p> Serviceoriented EasytouseSLAsSLA ServiceLevelAgreement</p><p>2/24/2015 SatishSrirama 12</p></li><li><p>CloudComputingProgress</p><p>[ArmandoFox,2010]</p><p>2/24/2015 SatishSrirama 13</p></li><li><p>MIGRATINGENTERPRISEAPPLICATIONSTOTHECLOUD</p><p>ResearchChallenges</p><p>2/24/2015 SatishSrirama 14</p></li><li><p>Enterpriseapplicationsonthecloud</p><p> EnterpriseapplicationsaremostlybasedonSOAandcomponentizedmodels</p><p> Faulttolerance,highavailability&amp;scalability Essentialprerequisitesforanyinformationsystem</p><p> Cloudwithitspromiseofvirtuallyunlimitedresourcescanoffertheaboveprerequisites Availabilityzones Elasticityandhorizontalscaling Utilitycomputing</p><p>2/24/2015 15SatishSrirama</p></li><li><p>REMICS</p><p> Reuseandmigrationoflegacyapplicationstothecloud</p><p>2/24/2015</p><p>Requirements</p><p> Track Track</p><p>Recover</p><p> BlueAge Modelio TALE</p><p> BlueAge Modelio TALE</p><p>Migrate</p><p> Modelio RedSeeds D2CM</p><p> Modelio RedSeeds D2CM</p><p>Validate</p><p>MetrinoFokusMBTRSLTesting</p><p>MetrinoFokusMBTRSLTesting</p><p>ControlandSupervision</p><p>Models@RuntimeModelDrivenInteroperabilityPerformancemonitoring</p><p>Models@RuntimeModelDrivenInteroperabilityPerformancemonitoring</p><p>Text</p><p>SoA MLPIM4Cloud</p><p>SoA MLClould MLUML/Code</p><p>UML/RSL</p><p>http://www.remics.eu/</p><p>SatishSrirama 16</p></li><li><p>CloudML Developedtotamecloudheterogeneity Domainspecificlanguage(DSL)for</p><p>modellingtheprovisioninganddeploymentatdesigntime Nodes,artefactsandbindingscanbe</p><p>defined DifferentmeanstomanipulateCloudML</p><p>models ProgrammaticallyviaJavaAPI Declaratively,viaserializedmodel(JSON)</p><p> Models@Runtime DynamicdeploymentofCloudMLbased</p><p>models</p><p>2/24/2015 [Ferry etal,Cloud 2013]SatishSrirama 17</p></li><li><p>AutoScalingenterpriseapplicationsonthecloud</p><p> AutoScaling Scalingpolicy&gt;WhentoScale Resourceprovisioningpolicy&gt;Howtoscale</p><p> Thresholdbasedscalingpoliciesareverypopularduetotheirsimplicity ObservemetricssuchasCPUusage,diskI/O,networktrafficetc.</p><p> E.g.AmazonAutoScale,RightScale etc. However,configuringthemoptimallyisnoteasy</p><p>2/24/2015 18/18SatishSrirama</p></li><li><p>OptimalResourceProvisioningforAutoScalingEnterpriseApplications</p><p> Cloudprovidersoffervariousinstancetypeswithdifferentprocessingpowerandprice Canitbeexploitedindecidingtheresourceprovisioningpolicy?</p><p> Makesthepolicytobeawareofcurrentdeploymentconfiguration</p><p> Anotherchallenge:Cloudproviderschargetheresourceusageforfixedtimeperiods E.g.HourlypricesofAmazoncloud</p><p> DevelopedanLPbasedoptimizationmodelwhichconsidersboththeissues[SriramaandOstovar,CloudCom 2014]</p><p>2/24/2015 19/18SatishSrirama</p></li><li><p>Scalingenterpriseapplicationwiththeoptimizationmodel</p><p>2/24/2015 20/18SatishSrirama</p><p>IncomingloadandscalingcurvesofOptimizationmodel</p><p>InstancetypeusagecurvesofOptimizationmodel</p><p>ScalingwithAmazonAutoScale</p><p>[SriramaandOstovar,CloudCom 2014]</p></li><li><p>OptimizationModelIntuitionbehindinstancelifetimeconsideration</p><p> Consider2instancetypes Smallinstance(PW=6r/s,Price=$0.25/h), Mediuminstance(PW=12r/s,Price=$0.4/h)</p><p>2/24/2015 21/18SatishSrirama</p><p>Loadis6r/s</p><p>Loadincreasesto12r/s=&gt; ?</p><p>Cost=(costoftwosmallinstances) (10minprofitofasmallinstance)=0.5 0.04=$0.46</p><p>Solution1</p><p>Cost=(costofamediuminstance)+(10mincostofasmallinstance)=0.4+0.04=$0.44</p><p>Solution2</p><p> Savedcostwithsolution2:0.46 0.44=0.02$ Themodelcanfindthisautomatically</p></li><li><p>CurrentInterests Remodelingenterpriseapplicationsforthecloudmigration Cloudhashugetroubleswithcommunication/transmissionlatencies[Sriramaetal,SPJ2011]</p><p> Intuition:Reduceinternodecommunicationandtoincreasetheintranodecommunication</p><p> AutoscalethembasedonoptimizationmodelandCloudML</p><p>2/24/2015 22SatishSrirama</p><p>[SriramaandViil,HPCC2014]</p></li><li><p>SCIENTIFICCOMPUTINGONTHECLOUD</p><p>ResearchChallenges</p><p>2/24/2015 SatishSrirama 23</p></li><li><p>ScientificComputingontheCloud Publiccloudsprovideveryconvenientaccesstocomputingresources Ondemandandinrealtime Aslongasyoucanaffordthem</p><p> Highperformancecomputing(HPC)oncloud Virtualizationandcommunicationlatenciesaremajorhindrances[Sriramaetal,SPJ2011;Batrashev etal,HPCS2011]</p><p> Thingshaveimprovedsignificantlyovertheyears Researchatscale</p><p> Costtovalueofexperiments DesktoptoCloudMigration(D2CM)toolfordomainscientists[Sriramaetal,HPCS2013]</p><p>2/24/2015 SatishSrirama 24</p></li><li><p>MigratingScientificWorkflowstotheCloud</p><p> Workflowcanberepresentedasweighteddirectedacyclicgraph(DAG)</p><p> Partitioningtheworkflowintogroupswithgraphpartitioningtechniques [SriramaandViil,HPCC2014] Suchthatthesumoftheweightsoftheedgesconnectingtoverticesindifferentgroupsisminimized</p><p> UtilizedMetismultilevelkwaypartitioning SchedulingtheworkflowswithtoolslikePegasus</p><p> Consideredpeertopeerfilemanager(Mule)forPegasus</p><p>2/24/2015 25SatishSrirama</p></li><li><p>EconomicsofCloudProviders</p><p> CloudComputingprovidersbringashiftfromhighreliability/availabilityserverstocommodityservers Atleastonefailureperdayinlargedatacenter</p><p> Why? Significanteconomicincentives</p><p> muchlowerperservercost Caveat:Usersoftwarehastoadapttofailures</p><p> Veryhardproblem! Solution:Replicatedataandcomputation</p><p> MapReduce&amp;DistributedFileSystem</p><p>2/24/2015 SatishSrirama 26</p></li><li><p>MapReducemodel</p><p>2/24/2015 SatishSrirama 27</p><p>https://tedwon.atlassian.net/wiki/display/SE/Apache+Hadoop</p></li><li><p>ApacheHadoop MapReduce</p><p> MostprominentOpenSourcesolution TheuseronlyhastowriteMapandReducefunctions</p><p> Frameworkhandleseverythingelsefortheuser Scheduling,datadistribution,synchronization,errorsandfaults</p><p> ParallelismisachievedbyexecutingMapandReducetasksconcurrently</p><p>2/24/2015 SatishSrirama 28</p></li><li><p>AdaptingComputingProblemstoCloud</p><p> ReducingthealgorithmstocloudcomputingframeworkslikeMapReduce[Sriramaetal,FGCS2012]</p><p> DesignedaclassificationonhowthealgorithmscanbeadaptedtoMR Algorithm singleMapReducejob</p><p> MonteCarlo,RSAbreaking Algorithm nMapReducejobs</p><p> CLARA(Clustering),MatrixMultiplication Eachiterationinalgorithm singleMapReducejob</p><p> PAM(Clustering) Eachiterationinalgorithm nMapReducejobs</p><p> ConjugateGradient ApplicableespeciallyforHadoop MapReduce</p><p>2/24/2015 SatishSrirama 29</p></li><li><p>IssueswithHadoopMapReduce</p><p> Itisdesignedandsuitablefor: Dataprocessingtasks Embarrassinglyparalleltasks</p><p> Hasseriousissueswithiterativealgorithms Longstartupandcleanuptimes~17seconds NowaytokeepimportantdatainmemorybetweenMapReducejobexecutions</p><p> Ateachiteration,alldataisreadagainfromHDFSandwrittenbackthereatthe end</p><p> Resultsinasignificantoverheadineveryiteration</p><p>2/24/2015 SatishSrirama 30</p></li><li><p>AlternativeApproaches</p><p> Restructuringalgorithmsintononiterativeversions CLARAinsteadofPAM [Jakovits&amp;Srirama,Nordicloud 2013]</p><p> AlternativeMapReduceimplementationsthataredesignedtohandleiterativealgorithms[Jakovits andSrirama,HPCS 2014]</p><p> E.g.Twister,HaLoop,Spark Alternativedistributedcomputingmodels</p><p> BulkSynchronousParallelmodel[Valiant,1990] [Jakovitsetal,HPCS2013]</p><p> BuiltafaulttolerantBSPframework(NEWT)[Kromonovetal,HPCS2014]</p><p>2/24/2015 SatishSrirama 31</p></li><li><p>MOBILECLOUDResearchChallenges</p><p>2/24/2015 SatishSrirama 32</p></li><li><p>2/24/2015</p><p>[TomiTAhonen]</p><p>SatishSrirama 33</p></li><li><p>MobileApplications</p><p> Onecandointerestingthingsonmobilesdirectly Todaysmobilesarefarmorecapable Locationbasedservices(LBSs),mobilesocialnetworking,mobilecommerce,contextawareservicesetc.</p><p> Itisalsopossibletomakethemobileaserviceprovider Mobilewebserviceprovisioning[Sriramaetal,ICIW2006;Srirama</p><p>andPaniagua,MS2013]</p><p> Challengesinsecurity,scalability,discoveryandmiddlewarearestudied[Srirama,PhD2008]</p><p> MobileSocialNetworkinProximity[Changetal,ICSOC2012;PMC2014]</p><p>2/24/2015 SatishSrirama 34</p></li><li><p>However,westillhavenotachieved</p><p> Longerbatterylife Batterylastsonlyfor12hoursforcontinuouscomputing</p><p> Samequalityofexperienceasondesktops WeakerCPUandmemory Storagecapacity</p><p> Stillitisagoodideatotakethesupportofexternalresourcesforbuildingresourceintensivemobileapplications</p><p>2/24/2015 35SatishSrirama</p></li><li><p>MobileCloudApplications</p><p> Bringthecloudinfrastructuretotheproximityofthemobileuser</p><p> Mobilehassignificantadvantagebygoingcloudaware Increaseddatastoragecapacity Availabilityofunlimitedprocessingpower PClikefunctionalityformobileapplications Extendedbatterylife(energyefficiency)</p><p>2/24/2015 SatishSrirama 36</p></li><li><p>MobileCloud Ourinterpretation</p><p> WedonotseeMobileCloudtobejustascenariowheremobileistakingthehelpofamuchpowerfulmachine!!!</p><p> Wedonotseecloudasjustapoolofvirtualmachines</p><p> MobileCloudbasedsystemshouldtakeadvantageofsomeofthekeyintrinsiccharacteristicsofcloudefficiently Elasticity&amp;AutoScaling Utilitycomputingmodels Parallelization(e.g.,usingMapReduce)</p><p>2/24/2015 37SatishSrirama</p></li><li><p>MobileCloudBindingModels</p><p>2/24/2015</p><p>TaskDelegation CodeOffloading[Flores&amp;Srirama,JSS2014]</p><p>SatishSrirama 38</p><p>[Floresetal,MoMM 2011]</p><p>MobileCloud</p><p>[FloresandSrirama,MCS2013]</p></li><li><p>MCM enables</p><p> InteroperabilitybetweendifferentCloudServices(IaaS,SaaS,PaaS)andProviders(Amazon,Eucalyptus,etc)</p><p> ProvidesanabstractionlayerontopofAPI CompositionofdifferentCloudServices AsynchronouscommunicationbetweenthedeviceandMCM [Warren etal,IEEEPC2014]</p><p> MeanstoparallelizethetasksandtakeadvantageofClouds intrinsiccharacteristics</p><p>2/24/2015 SatishSrirama 39/25</p><p>[Floresetal,MoMM 2011]</p></li><li><p>MCMapplications CroudSTag [Sriramaetal,MobiWIS 2011]</p><p> SocialgroupformationwithpeopleidentifiedinPictures/Videos Zompopo [Sriramaetal,NGMAST2011]</p><p> Intelligentcalendar,byminingaccelerometersensordata Bakabs [Paniaguaetal,iiWAS2011]</p><p> ManagingtheCloudresourcesfrommobile Sensordataanalysis</p><p> Humanactivityrecognition Contextawaregaming MapReduce basedsensordataanalysis[Paniaguaetal,MobiWIS 2012]</p><p> SPiCa:ASocialPrivateCloudComputingApplicationFramework[Changetal,MUM2014]</p><p>2/24/2015 40SatishSrirama</p></li><li><p>CodeOffloading MajorComponents</p><p> Majorresearchchallenges What,when,whereandhowtooffload?</p><p> Mobile Codeprofiler Systemprofilers Decisionengine</p><p> Cloudbasedsurrogateplatform</p><p>2/24/2015 41SatishSrirama</p><p>[FloresandSrirama,MCS2013]</p></li><li><p>Challengesandtechnicalproblems Inaccuratecodeprofiling</p><p> Codehasnondeterministicbehaviourduringruntime Basedonfactorssuchasinput,typeofdevice,executionenvironment,CPU,memoryetc.</p><p> Somecodecannotbeprofiled(e.g.REST) Integrationcomplexity</p><p> DynamicbehaviourvsStaticannotations E.g.Staticannotationscauseunnecessaryoffloading</p><p> Dynamicconfigurationofthesystem Offloadingscalabilityandoffloadingasaservice</p><p> Surrogateshouldhavesimilarexecutionenvironment ShouldalsoconsideraboutresourceavailabilityofCloud</p><p>2/24/2015 42SatishSrirama</p><p>[Floresetal,IEEECommunicationsMag2015]</p></li><li><p>Practicaladaptabilityofoffloading</p><p>Applicationsthatcanbenefitbecamelimitedwithincreaseindevicecapacities</p></li><li><p>Waytoproceed?</p><p> Codeoffloadingisnotyetareality!!! Takeadvantageofcrowdsourcing</p><p> Computationaloffloadingcustomizedbydataanalytics Byanalysinghowaparticularappbehavesinacommunityofdevices</p><p> E.g.Caratdetectsenergyanomalies[Oliner etal,2013] Bystudyingover~328,000appsgetsanideaonwhatisresourceintensiveapp</p><p> Determinesenergydraindistributionofanapp Decisionmodelscanalsobenefitfromcrowdsourcing</p><p> Analysisofcodeoffloadingtraces[FloresandSrirama,MCS2013]</p><p>2/24/2015 44SatishSrirama</p><p>[Floresetal,IEEECommunicationsMag2015]</p></li><li><p>Sensors Tags MobileThings</p><p>Appliances&amp;Facilities</p><p>InternetofThings</p><p>Howtointeractwiththingsdirectly?</p><p>Howtoprovideenergyefficient</p><p>services?</p><p>2/24/2015 SatishSrirama 45/31</p></li><li><p>MobileResourceComposition</p><p>MediationFramework(MRCMF)</p><p>MobileHostedThingsMiddleware</p><p>(MHTM)</p><p>MobileHostedCloudMiddleware</p><p>(MHCM)</p><p>SurroundingThings</p><p>CloudServices</p><p> MachinetoMachine(M2M)Communication ConstrainedApplicationProtocol(CoAP) 6LoWPAN/BLE/WiFi Direct RFID/NFC/QRCodeReader/EPC SensorML /SSI/EXI Etc.</p><p> Service/ResourceBus ContextAwareness QoS SemanticReasoning</p><p>MobileIoTMiddleware</p><p> ResourceAwareness BusinessProcessModel CloudServiceAdaptors MobilitySupport</p><p>LightweightServiceProvisioningMiddleware</p></li><li><p>ResearchResults</p><p> ParticipatedinanumberofEUfundedprojects PartnerintheEstonianCenterofExcellenceinComputerScience</p><p> PartnerinSoftwareTechnologyandApplicationsCompetenceCentre(STACC) AnR&amp;Dcenterthatconductsindustrydrivenresearchprojectsinthefieldsofsoftwareengineeringanddatamining</p><p> OutputresultedinseveralSMEs Plumbr [Sor andSrirama,JSS2014;Sor etal,SPE2015],ZeroTurnaround etc.</p><p>2/24/2015 SatishSrirama 47</p></li><li><p>Garage48, Startups, SME-s, ..#estonianmafia</p><p>2/24/2015 SatishSrirama 48</p></li><li><p>THANKYOUFORYOURATTENTIONsrirama@ut.ee</p><p>2/24/2015 SatishSrirama 49</p></li></ul>

Recommended

View more >