基于网络中心的分布式计算机系统资源控制机制(IJISA-V9-N7-5)

I.J.IntelligentSystemsandApplications,2017,7,41-51PublishedOnlineJuly2017inMECS()DOI:10.5815/ijisa.2017.07.05Copyright©2017MECSI.J.IntelligentSystemsandApplications,2017,7,41-51DistributedComputerSystemResourcesControlMechanismBasedonNetwork-CentricApproachZhenbingHuSchoolofEducationalInformationTechnology,CentralChinaNormalUniversity,No.152LouyuRoad,430079,Wuhan,ChinaE-mail:[e-mail],YaroslavKornagaandYaroslavLavrenkoNationalTechnicalUniversityofUkraine―KievPolytechnicInstitute‖,Pr.Pobedy,37,03056,Kiev,UkraineE-mail:{v.mukhin,y.kornaga,y.lavrenko}@kpi.uaOksanaHerasymenkoTarasShevchenkoNationalUniversityofKiev,VolodymyrskaStreet,64/13,01601,Kiev,UkraineE-mail:[e-mail]—Inthispaper,wepresentthedevelopmentofadecentralizedmechanismfortheresourcescontrolinadistributedcomputersystembasedonanetwork-centricapproach.Intially,thenetwork-centricapproachwasproposedforthemilitarypurposes,andnowitsprinciplesaresuccessfullyintroducedintheotherapplicationsofthecomplexsystemscontrol.Duetothefeaturesofcontrolsystemsbasedonthenetwork-centricapproach,namelyaddingthehorizontallinksbetweencomponentsofthesamelevel,addingthegeneralknowledgecontrolinthesystem,etc.,therearenewpropertiesandcharacteristics.Theconceptofimplementingofresourcecontrolmoduleforadistributedcomputersystembasedonanetwork-centricapproachisproposedinthisstudy.We,basingonthisconcept,realizedtheresourcecontrolmoduleandperformtheanalysisofitsoperationparametersincomparewithresourcecontrolmodulesimplementedonthehierarchicalapproachandonthedecentralizedapproachwiththecreationofthecommunitiesofthecomputingresources.Theexperimentsshowedtheadvantagesoftheproposedmechanismforresourcescontrolincomparewiththecontrolmechanismsbasedonthehierarchicalanddecentralizedapproaches.IndexTerms—Distributedcomputersystem,resourcecontrolsystem,decentralizedresourcecontrol,tasksscheduling,network-centricapproach.I.INTRODUCTIONThewideuseofdistributedcomputersystems(DCS)hasbecomeadailyroutine.TheincreaseinthenumberofusersofsuchsystemsleadstothefactthatDCSareconstantlyincreasinginscale.Inaddition,suchsystemsaredesignedtoservesimultaneouslyalargenumberofusers,whichrequiresefficientresourcecontroltechniquesinsuchsystems[1,2].Asinrecentyears,thereisatendencytocreatealargescaleDCS,whichareintendedtoserveabignumberofusers,theresourcecontrolinsuchsystemsrequirethedevelopmentofnewmethodsandtools.ForthescalableDCSisimportanttoapplythedecentralizedapproachtoresourcecontrol[3,4,5].LetDCSisasetofheterogeneouscomputingnodes(CN)connectedbyheterogeneousdatachannels,andthesechannelshavemultichannelcapability.Channelingcanbebothphysicalandtopological.Physicalmulti-channelismeanttherealizationoffrequencyortimedivisionofasinglephysicaldatachannel.Suchchannelcanbefiber-opticlines,thephysicalnatureofwhichinvolvesthedifferentiationofmodesoffrequencies,includingduplexing.Thetopologicalmulti-channelistheimplementationofmultiplealternativeroutesbetweentheDCSnodes,whichareconnectedbythetopologicallyredundantcommunicationsystem.Wemaycombinethephysicalandtopologicalmulti-channelsforthehigh-speed,reliabledatatransfer.Theschedulerofsuchadistributedsystemshouldconsiderthefeaturesofthissysteminordertoincreaseofefficiencyofitsfunctioning.So,thedevelopmentoftheschedulerforDCSdescribedaboveshouldbebasedonthedecentralizedapproach.LetusdescribesomedetailsofthedecentralizedcontrolapproachforDCSresourcescontrolandoftheschedulersthatareimplementit.II.ANALYSISOFRECENTRESEARCHESANDPUBLICATIONSCurrentlytherearethefollowingmainapproachesintheresourcecontrolofdistributedcomputersystems:centralizedcontrolofresources,decentralizedcontrolandhierarchicalone.Eachoftheseapproacheshasitsadvantagesanddisadvantages.Moredetailsofthefeatures,advantagesanddisadvantagesoftheseapproachescanbefoundin[3,4,6].42DistributedComputerSystemResourcesControlMechanismBasedonNetwork-CentricApproachCopyright©2017MECSI.J.IntelligentSystemsandApplications,2017,7,41-51Oneofthemaincomponentsofresourcecontrolsystem(RCS)ofDCSisaschedulerwhichcoordinatesthesolutionofthetasksthatarriveforprocessingtothedistributedsystembyplacingthemonavailablecomputingresources[7].Itisimportanttomeettherequirementsoftheuserandtoavoidtheidlesforthecomputationalresources[8].In[7]thereareidentifiedthefollowingkeycharacteristicsoftheDCSscheduler:theruntimeofthetask,DCSperformanceandtheresponsetimeofthesystem.Inrecentyears,therewasdevotedmuchattentiontothestudyofdecentralizedresourcecontroltechniquesforthedistributedcomputersystemstoimprovetheperformanceoftheDCSinthecontextofitsscalabilityandreliability[9].In[9]aredescribedthecomparativeanalysisoftheexistingdecentralizeddistributedcomputersystemsandthemechanismsthatareusedforresourcecontrol.ThisstudycontainsacomparativeanalysisofmechanismsofresourcecontrolofDCSindifferentcontexts.Additionally,[10,11,12]presentsmorerecentdevelopmentsofDCSresourcescontrolsystems,thatareimplementedonadecentralizedbasis.Thedevelopmentin[11]usestheexchangesofnetworkmessagesbetweenDCScomp