Issuu on Google+

9RO,VV<HDU

,QWO-UQORQ&RPSXWHU6FLHQFHDQG7HFKQRORJLHV

LoadBalancingGridSchedulerforthe ComputationalGridEnvironment Mr.R.Rajkumar1,Dr.S.ThamaraiSelvi2,Mr.G.Kannan3

w w w .a sd fjo ur na ls .c om

1Student,2ProfessorandPrincipleInvestigator,3ResearchAssociate DepartmentofIT,MITCampus,AnnaUniversity,Chennai,India

Abstract Ǧ Distributed computing paradigm is a wide spread technology, where computational grid provideshugecomputationpowerforthelargescaledistributedapplication.Oneofthechallengingissue in computational grid is load balancing. This paper, proposed a new load balancing scheduler algorithm which can not only increases the utilization of the resource and throughput, but also realize the load balancingwithinthegridenvironment.Theupdatedtopologyandloadinformationisacquireddynamically fromtheresourceusingtheeventnotificationapproach.Inordertomaximizetheutilizationofresources andtoincreasetheperformanceofthesystemapplicationlevelloadbalancingisneededfortheindividual parallel jobs. In many approaches load balancing is done only at the local scheduler level, which is applicable to small application and leads to more communication overhead between the nodes. For the largescaleapplicationloadbalancingatthelocalschedulerlevelwillnotprovidethefeasiblesolution.So thenovelloadbalancingalgorithmisproposed,whichprovidestheloadbalancingatthemetaǦscheduler level.Toinitiatetheloadbalancingtriggeringpolicyisused,whichdeterminestheappropriatetimeperiod timetostarttheloadbalancingoperation.Triggeringpolicycanbeinitiatedbyusingtwoapproachessuch as threshold and boundary value app oach. These approach increases the performance in the large scale applicationbysubmittingthejobtotheleastloadedmachinetoreducetheelapsedtimeandwaitingtime ofjob,andtomaximizetheutilizationoftheresourceswhichareidleorleastloaded.

fro m

Index Terms – Load balancing, topology information, load information, event notification, triggering policy,thresholdvalueandboundaryvalueapproach.

1INTRODUCTION

D ow nl oa de d

A computational grid is a hardware and software infrastructure that provides dependable, consistent, pervasive, and inexpensive access to highǦend computational capabilities. Oneof the challenging issue in grid environment is Load Balancing, which makes the coordination and scheduling of the resource dynamically to execute the tasks more quickly [1], [14]. Grid technology provides the best toolkit and the system platform to deal with these issues. A vast array of Globus Toolkit (GT4) middleware has been developedtosupportapplicationsinGrid[2],[15].

TheimpactofusingLoadBalancingScheduler(LBS)algorithmistochoosethebestfitresourcefromthe GridEnvironment,whichexecutesthetaskmorequicklythantheotherresource.Inthegridenvironment, eachclusterisconsideredasaresource[19].TherearemanyGridSchedulingalgorithmsexist,inwhichthe resourcewithhighCPUspeedandfreememoryisselectedasthebestresourcefortaskexecution.These typesofalgorithmmayleadtooverheadinthecase,wherethenewlyarrivedtaskrequiresmoreCPUspeed andfreememory.Thissituationleadstooverloadingoftheparticularresourceduetodemandandsome times,thechancefortheunavailabilityoftheresource. So far the load balancing algorithm is implemented by exploiting the static or dynamic resource information,topologyinformationandloadinformationoftheresource,suchasloadedorunloaded[12], [13].Theyhaveexploitedtheinformationasaseparatemodel,buttheydidn’tintegratealltheinformation inthesinglealgorithm.Inordertoavoidoverloadingandunavailabilityoftheresourcewehaveproposeda

MRXUQDOV#DVGIUHVLQ

ZZZDVGIMRXUQDOVFRP

3DJH


9RO,VV<HDU

,QWO-UQORQ&RPSXWHU6FLHQFHDQG7HFKQRORJLHV

newLBSalgorithmbyexploitingbothtopologyandloadinformation,whichisobtaineddynamicallyfrom theGridresource.

w w w .a sd fjo ur na ls .c om

Ofcoursetherearemanyloadbalancingalgorithms,whichactmoreefficientlyingridenvironment.Since thejobsaredynamicallysubmittedtoscheduler,theremaybeapossibilitytohavesamerequirementfor manyjobs.AsaresultGridschedulerwillsubmitmorenumberofjobstothesingleresource,whichleads to overloading. In case, if the local users fire the job to the resource without the knowledge of metaǦ scheduler,thentheresourceisforcedtohandletoomanynumberofjobsfrombothmetaǦscheduleraswell aslocaluserandthissituationleadstooverloadingofresource. In order to avoid the overloading and job waiting time, load in the particular resource is needs to be balanced. So the Load Balancing is a preferable solution to provide application level load balancing for individualjobs[9].TheimpactofusingLBSalgorithmsinthemetaǦschedulerwillreducethetotalelapse time,waitingtimeofthejobandmaximizetheutilizationoftheresourcewhichisidleorleastloaded[9], [17],[19].

2RELATEDWORK

The most important issues in grid environment, is the performance degradation caused due to load imbalanceandachievingminimumresponsetimefortheclientjobrequest[16].ThereforeLoadBalancing isindispensablefortheheterogeneouscluster,toassurefinedistributionofworkloadoneachcluster.All these approaches, broadly implements load sharing algorithms, which can be static or dynamic and also usesthecentralizedordistributedcontrol[3].Thereferenceshowsthatahybridofbothstaticanddynamic strategy for the resource selection provides various Load Balancing policies for providing a good performance[13].

D ow nl oa de d

fro m

IntheliteraturedynamicloadbalancingtechniqueforgridapplicationbasedonGraph partitioning, which exploits knowledge of the topology of the Grid environment to partition the communicationgraphinsuchawayastoreducethecrossǦsitecommunication[12].Forthedynamicload balancing algorithm, it is unacceptable to frequently exchange state information because of high communication overhead. In [9], sender processor collects the status information about neighboring processors by communicating with them at every load balancing instant. For the large scale Grid Environmentwherecommunicationlatencyisverylarge,thestatusexchangeateachloadbalancinginstant canleadstolargecommunicationoverhead[9],[12]. In our approach the problem of frequent exchange of information is alleviated by estimating the load informationondemandbyusingtheeventnotificationapproach.Herethetriggeringpolicyisconsidered whichmakesthedecisionatwhattimetheloadbalancingoperationistobeinitiated[3].

3DYNAMICLOADBALANCINGGRIDSCHEDULER 3.1LoadBalancingpolicy

Inordertomaximizetheutilizationofresourceandtoreducethewaitingtimeofthejob,applicationlevel loadbalancingisneededforindividualparalleljobs.Asperthegridenvironmentisconcern,loadbalancing istobedonebyconsideringanyoneoftheloadbalancingpoliciessuchastransferpolicy,selectionpolicy, locationpolicy,informationpolicyandtriggeringpolicy[3],[13].

Inourproposedmodelwehaveconsideredonlyfourpolicies.Firstisthetransferpolicy,whichisalsocalled as threshold policy and the thresholds are expressed in units of load [7], [28]. Suppose new job or task originatesatthenode,whentheloadatthenodeexceedsathresholdT,thetransferpolicydecidesthatthe node is a sender. In case the node falls below threshold T, then the transfer policy decides that the

MRXUQDOV#DVGIUHVLQ

ZZZDVGIMRXUQDOVFRP

3DJH


9RO,VV<HDU

,QWO-UQORQ&RPSXWHU6FLHQFHDQG7HFKQRORJLHV

w w w .a sd fjo ur na ls .c om

particularnodecanbeareceiverfortheremotetask.Secondisthelocationpolicy,theresponsibilityofthis policyistofindsuitablenodestoshareload.Awidelyusedmethodforfindingasuitablenodeisthrough polling. In polling, a node polls another node to find out whether it is a suitable node for load sharing. Nodescanbepolledeitherseriallyorparallel.Analternativetopollingistobroadcastaquerytofindoutif any node is available for load sharing. Third is the information policy, which is responsible for making decisionsuchaswheninformationaboutthestatesofothernodesinthesystemshouldbecollected,where it should be collected from, and what information should be collected. Of course there are many information policies available, but we have considered the sender initiated demandǦdriven policy. In the sender initiated policy, sender will look for the receiver to transfer their loads, and the reverse one is receiverinitiatedpolicy[13]. Finally we considered the triggering policy, which determines the appropriate period to start a load balancing operation. The triggering policy can be initiated by any one of the two approaches. In the thresholdapproach,thresholdvalueissetfortheresourceorclusterloadpresentinthegridenvironment. If the resource load exceeds the particular threshold value, then the load balancing policies such as information,selectionandlocationpolicyareconsidered,tomigratethejobtootherresourceswhichare belowthethresholdvalue[8].Theboundaryvalueapproachisused,ifnojobarrivestothemetaǦscheduler for certain time interval, then the average load of the resources is calculated. The upper and lower boundary value is set for the resource, suppose if the resource load exceeds the upper boundary value meansthentheloadbalancerwillmigratethejobtotheresourcewhichisbelowthelowerboundaryvalue [5].Herethejobmigrationtakesplaceuntiltheresourcecomestomoderateload.

3.2LoadBalancingGridSchedulermodel

D ow nl oa de d

fro m

In this paper, Load Balancing Grid Scheduler (LBGS) model is proposed using Load Balancing Scheduler (LBS),LoadBalancer(LB)andJobMigration(JM)algorithms.ThisschedulermodelisshowninFig1.The userswillsubmittheirjobstothemetaǦschedulerwhichfallsinthequeueofrequesthandler.TheRequest HandlerserviceprovidesauserinterfacethroughwhichaclientcansubmitthejobsdescribedusingJSDL specificationtotheunderlyingmetaǦscheduler.Thisservicewillobtainthejobsfromuserandstoresitina queueofrequesthandler.Thedispatchmanager,whichispresentintheCAREResourceBroker(CRB) obtains the submitted job information periodically from the queue, which is implemented in the request handler component [4]. It sends the jobs to LBS for discovering suitable resources for every scheduling intervalitmaintains. 

 Fig1:StructureofLoadBalancingGridscheduler

MRXUQDOV#DVGIUHVLQ

ZZZDVGIMRXUQDOVFRP

3DJH


9RO,VV<HDU

,QWO-UQORQ&RPSXWHU6FLHQFHDQG7HFKQRORJLHV

w w w .a sd fjo ur na ls .c om

Thisschedulerwillperformtheloadbalancingandjobmigration,byexploitingtheinformationgathered fromloadmonitorandinformationmanager,alongwiththetopologyandloadcost.Itwillallocatethejob totheresourcewhichishavinglesstopologycosttoreachandlessloadcost.TheLoadMonitorcontains theinformationaboutalltheloadagents,andpreparessystemloadinformationaboutalltheresource.This load information is automatically updated from the load agent when there is any change in the resource loadbyusingeventnotificationapproach.ThestructureoftheLoadMonitorisshowninFig2.Finallythe loadinformationisgiventotheLBSforactualLoadBalancingandjobmigration.TheLoadAgentservice actsasaserverandacopyofloadagentservicehastorunonallresourceorclusterwheretheuserswantto runtheirapplications.Loadagentprovidessystemloadinformationsuchasjobqueuelength,CPUspeed andthenumberofnodecountavailableintheresource. The information manager will query the Monitoring and Discovery Service (MDS) and sends the host informationtotheLBS.Basedonthemonitoringintervalitkeepstrackofthehoststatusandupdatesthe hostinformationtotheLBS.Ifanynewresourcesareaddedthoseinformationarealsoupdatedperiodically [18].TheTransferManagerisinvokedbythedispatchmanagerwiththejobǦidandthematchingresourceǦ idasinput.Onceitisinvoked,thetransfermanagercreatesaremotedirectoryforthegivenpathnamein user input. Transfer manager gives the permission rights for the execution of given job in the remote directory.Oncethisprocessisover,itinformsthedispatchmanagerthroughmessages. The execution manager is invoked by the dispatch manager when the transfer manager completed the creation of directory in the remote host. The dispatch manager will dispatch the job for execution. Executionmanagerwillkeepsupdatingthejobstatustothescheduler.Finallyitreportsthecompletionor failureofjobtothescheduler. Beforerepresentingtheexactproblemstatement,wefirstdescribethenotationsandterminologythatare usedthroughoutthepaperas showninTable1.

fro m

Table1

Listofnotationsandterminology

D ow nl oa de d

M N Di TFi

TCi JUi ɐi(T) ɑi(T) Ʉ Ʌ LT Llow Lhigh Lmoderate Li(T) TLR(T) ALR(T) LUBR LLBR

MRXUQDOV#DVGIUHVLQ

Numberofheterogeneousresources(R1,R2,…,RM) Numberofjobstobeprocessed DelaytoreachtheresourceRi TopologyFactoroftheresourceRi (Numberofhoptoreachtheresource) Topologycostoftheresource JobunitofthetaskorjobJi(sizeofjob) EstimatedarrivalratefortheresourceRiattimeT EstimatedserviceratefortheresourceRiattimeT Boundaryvalueestimationfactor Thresholdvalueestimationfactor Thresholdloadvalueoftheresource Resourceisoflowload Resourceisofhighload Resourceisofmoderateload EstimatedloadoftheresourceRiattimeT TotalloadofalltheresourceR=(R1,R2,R3…,RM) AverageloadofalltheresourceR LoadupperboundaryfortheresourceR LoadlowerboundaryfortheresourceR

ZZZDVGIMRXUQDOVFRP

3DJH


9RO,VV<HDU

,QWO-UQORQ&RPSXWHU6FLHQFHDQG7HFKQRORJLHV

  We will now introduce certain key performance metrics considered in this paper such as Topology and Loadcost,regardingtheGridSchedulingandLoadBalancingintherealGridEnvironment.

3.3EstimationofTopologyCost

w w w .a sd fjo ur na ls .c om

Topology is regarded as one of the outmost important factors since it could strongly affect the Grid performance [12]. In our proposed approach, LBS will take the decision to submit the job to best fit resource,whichishavinglesstopologycosttoreach,byexploitingtopologyinformation.Herethetopology information, such as delay and number of hop count is gathered by using the Network Weather Service Tool.Thetopologycostiscalculatedasfollows, TCi=2*Di*TFi(1)

HerethedelayDirepresentsthelatencyortimetoreachtheparticularresourceRi.Andthetopologyfactor TFirepresentsthenumberofhoptoreachtheparticularclusterorresourceRi.Theimpactofconsidering topologycostwillreducetherequestandresponsetime(elapsetime)ofjob[11].

3.4EstimationofLoadCost

One of the key performance metrics considered for load balancing is Load Cost. The major impact of considering this metrics will help in choosing the best resource. For the estimation of load cost, the resource information such as CPU speed, free memory and number of node count is pulled by using the Load Agent. The important factor which needs to be the part of the scheduling and load balancing algorithmistheLoadCost.FirstwecalculatethejobunitJUiofthejobJiasfollows, JUi=NC(Ji)(2)

fro m

WhereNCdenotesthenumberofnodecountrequiredbythejobJi.Thencalculatetheestimatedarrival rateɐi(T)oftheresourceRiatthetimeTasfollows, ɐi(T)=ȳ(JU1,JU2,JU3,...,JUN)(3)

D ow nl oa de d

whereJU1,JU2,JU3,...,JUNrepresentsthenumberofjobsunitspresentinthequeueoftheresourceRi[21]. AndtheEstimatedservicerateɑi(T)oftheresourceatthetimeTisexpressedas, ɑi(T)=ȳ(C1,C2,C3,...,Ck)*CPUspeedi(4)

WhereC1,C2,C3,...,CkindicatesthenumberofcomputingnodepresentintheresourceRi,andtheservice rate vary with respect to CPU speed as well as the node count of the resource [20]. Finally the load cost Li(T)oftheresourceRiisestimatedbytakingtheratiooftheestimatedarrivalratetotheestimatedservice rateusingequation(3)and(4)[8],[9],[18]. Li(T)=ɐi(T)/ɑi(T)(5)

TheimpactofconsideringLoadCost,willhelpsinchoosingthebestresourcewhichisleastloaded,reduces thewaitingtimeofthejobandalsoavoidtheoverloadingofresources.

 

MRXUQDOV#DVGIUHVLQ

ZZZDVGIMRXUQDOVFRP

3DJH


9RO,VV<HDU

,QWO-UQORQ&RPSXWHU6FLHQFHDQG7HFKQRORJLHV

4LOADBALANCINGSCHEDULERALGORITHM



Inthispaper,LBSalgorithmisproposedwhichperformsthejobtoresourcematchinginthedynamically varyinggridenvironment.ThejobinformationisplacedintheJobPoolandtheresourceinformationis placedintheResourcePool.

w w w .a sd fjo ur na ls .c om

Table2 JobinformationpresentinJobPool  Required CPUspeed 1.4 2.0 3.0 1.4 3.0

RequiredFree Memory 80 60 60 20 20

JobId 1 2 3 4 5

Required Nodecount 1 7 7 1 3

 IntheJobPool,thejobinformationsuchasjobid,requiredfreememory,requiredCPUspeedandrequired numberofnodecountispresentasshowninTable2.Basedonthejobinformation,jobisclassifiedasdata intensiveandcomputationintensivejobs.SimilarlyintheResourcePool,theresourceandtopologyrelated informationsuchasresourceid,availablefreememory,CPUspeed,delaytoreachtheresource,numberof hopcounttoreachtheresourceandthenumberoftaskremaininginresourceorclusterqueuearepresent asshowninTable3.

Table3

1

2 3

4 5

Resource

No.ofHope

Available Free Memory

CPUSpeed

No.ofNode Count

No.ofJobs Unitinthe Queue

0.160

3

3000

1.4

5

10

0.305

5

8000

2.0

15

30

0.160

3

2000

3.0

5

10

0.305

5

9000

3.2

15

50

0.243

7

5000

1.4

5

20

Delayto Reach

D ow nl oa de d

Id

fro m

ResourceinformationpresentinResourcePool

 In the LBGS, matchmaking is a process in which, each job in the Job Pool is matched with the set of resourceavailableintheResourcePool.Assumethereare‘N’numberofjobsand‘M’numberofresource availableinJobPoolandResourcePoolrespectively.

MRXUQDOV#DVGIUHVLQ

ZZZDVGIMRXUQDOVFRP

3DJH


9RO,VV<HDU

,QWO-UQORQ&RPSXWHU6FLHQFHDQG7HFKQRORJLHV

TheLoadBalancingalgorithmsuchasLBSworksasshowninAlgorithm1.Thisalgorithmwilltakesthejob andresourceinformationasinputandgivestheoutputasbestresourceforeachjob[10],[18].  Algorithm1 LoadbalancingScheduler(LBS)Algorithm

D ow nl oa de d

fro m

w w w .a sd fjo ur na ls .c om

GetthelistofjobspresentinJobPool for(allthejobs) { IdentifythetypeofJob If(resource==1) { SubmittheJob } elseif(JobisDataintensive) { Identifythesetofmatchedresource for(Setofmatchedresource) { FindtheTopologyCostforallthenodes ChooseresourcehavingminimumTopologyCost if(lessTopologyCostresource==1) { SubmittheJob } else { CalculatetheLoadCostforallthenodes ChooseresourcehavingminimumLoadCost ThensubmittheJob } } } elseif(Jobiscomputationintensive) { Identifythesetofmatchedresource for(setofmatchedresource) { FindtheLoadcostforallthenodes ChooseresourcehavingminimumLoadcost if(minimumLoadcostresource==1) { SubmittheJob } else { CalculatetheTopologycostforallthenodes ChooseresourcehavingminimumTopology ThensubmittheJob } } }

MRXUQDOV#DVGIUHVLQ

ZZZDVGIMRXUQDOVFRP

3DJH


9RO,VV<HDU

,QWO-UQORQ&RPSXWHU6FLHQFHDQG7HFKQRORJLHV

w w w .a sd fjo ur na ls .c om

Else(Jobisbothdataandcomputationintensive) { Identifythesetofmatchedresource for(Setofmatchedresource) { Calculatetopologyandloadcostforallthenodes Calculatethetotalcostforallthenodes Totalcost=Topologycost+Loadcost ChooseresourcehavingminimumTotalcost ThensubmittheJob } } }  Inthisalgorithm,classificationisdonefirstsuchaswhetherdataintensiveorcomputationintensive.Ifthe jobrequiresmorefreememorybutrequirelesscomputationnodemeans,thenthejobisconsideredtobea Dataintensivejob.Suppose,ifthejobrequiresmorecomputationnode,butrequiresonlylessamountof free memory means, then the job is considered to be a computation intensive job. Otherwise the job is considered to be both data and computation intensive. Based on its classification it calls the other three methodsforitscomputationandbestresourceselection.Suppose,ifthejobisdataintensive,itcompute the topology cost as the foremost process for the choosing the best resource. Here the Load cost is calculatedonlyondemand.Incase,ifthejobiscomputationintensivethenitcomputetheloadcostand choosethebestresourceforsubmittingthejob.Thetopologycostiscomputedonlyiftoomanyresources werematchedwithminimumloadcostrange.Supposeifthejobdoesn’tfallsundereitherdataintensiveor computation intensive, it concludes that the job is both data and computation intensive, and it compute totalcostforallthenodesandfinallyitchoosestheresourcehavingminimumoftotalcost.

5LOADBALANCERANDJOBMIGRATIONALGORITHM

D ow nl oa de d

fro m

Ofcourse,theremaybelotofschedulingalgorithmswereproposedbydifferentauthors,whichworkswell inthespecifiedenvironment.Insomecases,toomanynumbersofjobsmayhavethesamerequired,which willdefinitelyleadstooverloadingofparticularresource.Sotheapplicationlevelloadbalancingisneeded fortheindividualparalleljobs.Intheproposedmodel,manyloadbalancingpoliciessuchasInformation, Selection, Location and Triggering policy is considered. And the Triggering policy determines the appropriatetimeperiodtostartaloadbalancingoperation. Suppose,iftheLoadBalancingisdoneveryoftenmeans,thenitleadstotheperformancedegradationand drawbacktosystem.So,thisloadBalanceralgorithmcanbeinitiatedbasedontwoapproachesbyusingthe triggeringpolicy.Inthefirstapproach,boundaryvalueisusedwhennojobsarrivedtothemetaǦscheduler for the specified time interval. In the second approach threshold value is used when the Load in the resourceexceedstheparticularthresholdvalue.

TheLoadBalancer(LB)algorithmusingboundaryvalueapproachwillworksasshowninAlgorithm2.This algorithm will get the resource load information from the Load Monitor service, which is present in the metaǦscheduler. First thisLB algorithmcalculates the total load ofthe resourceTLR(T) by adding all the resourceloadinformationattimeTasfollows, TLR(T)=ȳL1(T),L2(T),…,Lm(T)(6) WhereL1,L2,L3,...,LmrepresentstheLoadoftheresourceR1,R2,R3,...,Rmrespectively.Thenitcompute theaverageofloadresourceALR(T)atthetimeT,bytakingtheratiooftheTLR(T)tothetotalnumberof resourceavailable’m’usingtheequation(6).

MRXUQDOV#DVGIUHVLQ

ZZZDVGIMRXUQDOVFRP

3DJH


9RO,VV<HDU

,QWO-UQORQ&RPSXWHU6FLHQFHDQG7HFKQRORJLHV

ALR(T)=TLR(T)/m(7)  Thensettheupperandlowerboundarybyusingtheequation(7)andtheparameterɄ.Theloadupper boundaryoftheresourceRisexpressedas, LUBR=ALR(T)+Ʉ(8)

w w w .a sd fjo ur na ls .c om

AndtheloadlowerboundaryoftheresourceRisexpressedas, LLBR=ALR(T)–Ʉ(9)

where Ʉ is the parameter denotes the boundary value estimation factor and its value is based on the multiprocessorsystem[5].

Algorithm2

LoadBalancer(LB)AlgorithmusingBoundaryvalue

D ow nl oa de d

fro m

GettheLoadinformationforallresourceusingLoadMonitorservice CalculateLoadCostforallresource CalculatethetotalLoadCostTLR(T) ComputeaverageloadcostALR(T)=totalloadcost/n SettheupperboundaryasLUBR=ALR(T)+& SetthelowerboundaryasLLBR=ALR(T)Ǧ& for(alltheresource) { if(resourceLoadCost>LUBR) { Nodeisoverloaded UseAlgorithm4 } elseif(resourceloadcost<LLBR) { Nodeisleastloaded GettheJobforexecution } else { Resourceismoderatelyload Noneedforjobmigrationhere } }  ResourcewhichhavetheloadgreaterthenLUBRissaidtobeoverloadedandtheresourcewhichhavethe loadlesserthenLLBRissaidtoleastloadedbyexploitingequation(8),(9).Aresourceismoderatemeans theresourceloadisinthestatebetweenLUBRandLLBR[7].Suppose,iftheloadinformationexceedsthe boundary value, then the load balancing policies such as Selection, Information and Location policy are considered to migrate the job to other resources which are below the boundary value. If the resource is overloadedmeansJobMigration(JR)algorithmisusedwhichworksasshowninAlgorithm4. 

MRXUQDOV#DVGIUHVLQ

ZZZDVGIMRXUQDOVFRP

3DJH


9RO,VV<HDU

,QWO-UQORQ&RPSXWHU6FLHQFHDQG7HFKQRORJLHV

TheLoadBalancer(LB)algorithmusingthresholdvalueapproachwillworksasshowninAlgorithm3.This algorithm will get the resource load information from the Load Monitor service, which is present in the metaǦscheduler.FirstthisLBalgorithmassignsthethresholdvalueLTtotheresourceasfollows,

w w w .a sd fjo ur na ls .c om

LT=Ʌ(10) where Ʌ is the parameter denotes the threshold value estimation factor and its value is based on the resourceservicerate[5].IftheloadLioftheresourceRifallsbelowthethresholdloadLT,thentheresource loadisassignedas, Llow=Li(11)

Suppose, ifthe loadLi of the resource Ri is greater than the threshold load LT,then the resource load is assignedas, Lhigh=Li(12)

Incase,iftheloadLioftheresourceRiisequaltothethresholdloadLT,thentheresourceloadisassigned as, Lmoderate=Li(13)

WhereLidenotestheloadoftheresourceRi.

Algorithm3

LoadBalancer(LB)AlgorithmusingThresholdvalue

D ow nl oa de d

fro m

GettheLoadinformationofallresourceusingLoadMonitorservice CalculateLoadCostforallresource SettheThresholdloadvalueoftheresourceasLT if(theloadLioftheresource<LT) { SetRiaslowloadresourceLlow } elseif(theloadLioftheresource>LT) { SetRiashighloadresourceLhigh } else { SetRiasmoderateloadresourceLmoderate } for(alltheresource) { if(theloadLioftheresourceRi==Lhigh) { Nodeisoverloaded UseAlgorithm4 } elseif(theloadLioftheresourceRi==Llow) { Nodeisleastloaded GettheJobforexecution

MRXUQDOV#DVGIUHVLQ

ZZZDVGIMRXUQDOVFRP

3DJH


9RO,VV<HDU

,QWO-UQORQ&RPSXWHU6FLHQFHDQG7HFKQRORJLHV

} else { Resourceismoderatelyload Noneedforjobmigrationhere } }

w w w .a sd fjo ur na ls .c om

Using equation (11), (12) and (13) identify the overloaded resource and then finally use load balancing policies such as Selection, Information and Location policy to migrate the job to other least loaded resources which are below the threshold value using the Job Migration (JR) algorithm as shown in Algorithm4. 

Algorithm4

JobMigrationAlgorithm

D ow nl oa de d

fro m

Getthelistofoverloadedresource Getthelistofleastloadedresource for(alltheoverloadedresource) { While(overloadedresourceLoadCost>LUBR) { TaketheoneJobfromResourceQueue for(allleastloadedresource) { FindtheLoadcostforallthenodes ChooseresourcehavingminimumLoadCost if(minimumLoadcostresource==1) { ThenMigratetheJobtotheresource } else { FindtheTopologycostforallthenodes ChooseresourcehavinglessTopologyCost ThenMigratetheJobtotheresource } } Oncethejobmigrationalgorithmisexecutedsuccessfullymeans,wecanrealisetheloadbalancingandalso the work load is get distributed among all the resource. As a result of load balancing and job migration, waiting time of the job is reduced, maximize the utilization of the resource which is least loaded and increasestheoverallperformanceoftheresource.

6EXPERIMENTALRESULTANDPERFORMANCEEVALUATION

In this result phase the main focus is to show the result, as the proposed scheduler performs well, when comparingtothenormalschedulerbyconsideringallthechallengingissues.Wehavesimulatedtheresult byexploitingtenresourcesandhundredsofjobs.TheresultofboththeschedulingandLBGSiscompared andtheperformancemeasureisalsorepresentedasgraph.TheperformanceoftheproposedLBGSismore efficientinelapsetimeofjob,withrespecttodelayasshowninFig2.

MRXUQDOV#DVGIUHVLQ

ZZZDVGIMRXUQDOVFRP

3DJH


,QWO-UQORQ&RPSXWHU6FLHQFHDQG7HFKQRORJLHV

w w w .a sd fjo ur na ls .c om

9RO,VV<HDU

    Fig2:PerformanceevaluationofNormalSchedulerVsLoadBalancingGridSchedulerwithrespecttodelay andelapsetime.

D ow nl oa de d

fro m

LoadVsWaitingtimeofJob 



WaitingTimeofJob Load NormalScheduler LoadBalancingGridScheduler

Fig3:PerformanceevaluationofNormalSchedulerVsLoadBalancingGridSchedulerwithrespecttoload andwaitingtimeofjob.

Similarly,theperformanceevaluationofthisschedulerisdonewithrespecttothewaitingtimeofjoband the load of the resource as shown in the Fig 3. Since the performance of the system vary with respect to delay,aswellastheload.So,theparametersuchasdelay,load,waitingtimeandelapsetimeisconsidered fortheperformanceevaluation.  

MRXUQDOV#DVGIUHVLQ

ZZZDVGIMRXUQDOVFRP

3DJH


9RO,VV<HDU

,QWO-UQORQ&RPSXWHU6FLHQFHDQG7HFKQRORJLHV

7CONCLUSIONANDFUTUREWORK

w w w .a sd fjo ur na ls .c om

urproposedmodelexhibitsdifferentfeaturesliketopologyawareanddynamicloadbalancingduringthe job firingto list of matched resource.As far as thedeployment issues are concern, it providessupportfor multiple users, flexible and extensible architecture. Hence the metaǦscheduler is developedwith all the features, in which the load balancing is realized during the scheduling. We have evaluated the proposed LBS with the simulation and traces from real time Grid environment in CARE laboratory. The result obtained with performance evaluation, can balance the load, decreases the elapse time and increase the utilizationoftheresource,whichareidleorleastloaded.Sotheobtainedresultshows,theproposedLBGS modelwithloadbalancingperformsbetter,thennormalschedulingintermsofdelayandload. In feature, we are working towards Load Balancing and job migration between the metascheduler in the realGridEnvironment.

8ACKNOWLEDGEMENTS

We wish to express our thanks to Centre for Advance Computing Research and Education (CARE) team members in the research laboratory, for their helpful and constructive suggestion, which considerably improvethequalityofthepaper.ThispaperhasbeensupportedbytheCARE,forimplementingtheGrid SchedulingandLoadBalancingalgorithmintherealGridEnvironment. 9REFERENCES

[1] I. Foster and C. Kesselman, “The Grid: Blueprint for a Future Computing Infrastructure”, Morgan Kaufmann,1999. [2]I.Foster,andC.Kesselman,“Globus:ametacomputinginfrastructuretoolkit,”InternationalJournalof HighPerformanceComputingApplications,vol.2,pp.115–128,1997.

fro m

[3] George Coulouris, Jean Dolimore and Tim Kindberg, “Distributed Systems: Concepts and Design”, AddisonWesleyLongman,1994.

D ow nl oa de d

[4] Thamarai Selvi Somasundaram, Balachandar R. Amarnath, R. Kumar, P. Balakrishnan, K. Rajandar, R. Rajiv, G. Kannan, G. Rajesh Britto, E. Mahendran and B. Madusudhanan, “CARE Resource Broker: A framework for scheduling and supporting virtual resource management”, Journal: Future Generation ComputerSystem,2010.

[5]HANXiangchun1,CHENDuanjun1,CHENJing,“Onecentralizedschedulingpatternfordynamicload balanceingrid”,IEEEInternationalForumonInformationTechnologyandApplications,2009.

[6] Ivan Rodero, Francesc Guim, Julita Corbalan, “Evaluation of Coordinated Grid Scheduling Strategies”, 11thIEEEInternationalConferenceonHighPerformanceComputingandCommunications,2009. [7]BinLuandHongbinZhang“GridLoadBalancingSchedulingAlgorithmBasedonStatisticsThinking”, 9thIEEEInternationalConferenceforYoungComputerScientists,2008.

[8]NaveenKumarGondhiandDr.DurgeshPant,“AnEvolutionaryApproachforScalableLoadBalancing inClusterComputing”,IEEEInternationalAdvanceComputingConference,2009.

[9] Ruchir Shah, Bhardwaj Veeravalli and Manoj Mistra, “On the Design of Adaptive and Decentralized LoadǦBalancing Algorithms with Load Estimation for Computational Grid Environments”, IEEE TransactionsonParallelandDistributedSystems,December2007.

MRXUQDOV#DVGIUHVLQ

ZZZDVGIMRXUQDOVFRP

3DJH


9RO,VV<HDU

,QWO-UQORQ&RPSXWHU6FLHQFHDQG7HFKQRORJLHV

[10] Ashiq Anjum, Richard McClatchey, Arshad Ali and Ian Willers, “Bulk Scheduling With the DIANA Scheduler”,IEEETransactionsOnNuclearScience,December2006. [11]IoannisPsoroulas,IoannisAnagnostopoulos,VassiliLoumosAndEleftheriosKayafas,“AStudyOfThe ParametersConcerningLoadBalancingAlgorithms”,IjcsnsInternationalJournalOfComputerScienceAnd NetworkSecurity,April2007.

w w w .a sd fjo ur na ls .c om

[12] Gregory A.Koenig And Laxmikant V.Kale, “Optimizing Distributed Application Performance Using DynamicGridTopologyǦAwareLoadBalancing”,DepartmentOfComputerScience,UniversityOfIllinois AtUrbanaǦChampaign,2007IEEE. [13]BelabbasYagoubi,HadjTayebLilaandHalimaSiMoussa,“LoadBalancinginGridComputing”,Asian JournalsofInformationTechnology,2006. [14] Quan Liu and Yeqing Liao, “Grouping based Fine grained Job Scheduling in Grid Computing”, IEEE FirstInternationalWorkshoponEducationTechnologyandComputerScience,2009. [15]Yigang,HuiDongandZiranZhang,“AnOpenSourceToolkitforGridResourceSchedulinginDigital Library”,IEEENinthInternationalConferenceonHybridIntelligentSystems,2009. [16]SHILeiandXUHuiǦhui,“GridresourceSchedulingAlgorithmBasedonQoSGuideGA”,IEEE,2009.

[17] XiuǦmei Wen, Wei Zhao and FanǦxing Meng, “Research of Grid Scheduling Algorithm based on P2P GridModel”,IEEEInternationalConferenceonElectroniccommerceandBusinessIntelligence,2009. [18] Yi Hu and Bin Gong, “MultiǦobjective optimization approach using a CEǦACO inspired strategy to improveGridJobScheduling”,IEEEFourthChinaGridAnnualConference”,2009.

fro m

[19] KuoǦChan Huang, “On Effects of Resource Fragmentation on Job Scheduling Performance in Computing Grids”, IEEE 10th International Symposium on Pervasive Systems, Algorithms and Networks, 2009.

D ow nl oa de d

[20]MennoDobber,RobVanDerMeiAndGerKoole,“DynamicLoadBalancingAndJobReplicationInA GlobalǦScale Grid Environment: A Comparison”, IEEE Transactions On Parallel And Distributed Systems, February2009. [21]DanielGrosuandAnthonyT.Chronopoulos,“AlgorithmicMechanismDesignforLoadBalancingin DistributedSystem”,IEEETransactionsonSystems,manandCybernetics,February2004.

MRXUQDOV#DVGIUHVLQ

ZZZDVGIMRXUQDOVFRP

3DJH


Ijcst004