Affinity based virtual machine migration (AVM) approach for effective placement in cloud environment by IRJET Journal

Affinity based virtual machine migration (AVM) approach for effective placement in cloud environment Gokuldhev M 1 , Thanammal Indu V 2

Words

Key : Cloud computing, virtual machine migration, AVM approach, Affinity based grouping algorithm, multi objective optimization, genetic algorithm

2Assistant Professor, Department of Computer Science and Engineering, Amrita College of Engineering and Technology, Nagercoil, India ***

1Assistant Professor, Department of Computer science and Engineering, Vel Tech Rangarajan Dr. Sagunthala R & D institute of Science and Technology, Chennai, India

1. INTRODUCTION

BasedonthisMatrix,ifa VM PMpairhashigheraffinity then the corresponding VM is placed in that PM, if the Resource requirements of that VM is satisfied by that PM. Theallocationpatternisfedasoneofthecandidatesofthe initialpopulation,andhencetheoptimalsolutionisreached faster based on this affinity based solution pattern. Then, basedonthetrafficratebetweentheVMsafterplacement, VMmigrationtakesplacesuchthattheoverheadduetothe migrationisnegligiblewhencomparedtothereductionin the traffic rate and communication cost due to migration [4].VMmigrationonlyconsidersdynamicconsolidation,for which one or some of the hosts of VMs can be changed in response to the change in workloads. Swarm intelligence algorithmsincludesuchalgorithmsassimulatedannealing, tabu search, iterated local search, GA, ACO, PSO, and ABC optimization.Theuseofswarm intelligence algorithms to solve many combinatorialoptimizationproblemshasbeenasubjectof researchforalongtime.TheyarewidelyusedtosolveNP completeproblems[5].Otherthantheenergyconsumption andefficiencyproblems,networkingandcommunicationare twogreatchallengesincloudcomputing.Thecostbetween any two communicating VMs is defined as the number of switchesorhopsontheroutingpathoftheVMpair. Communication sensitive applications that involves frequent access to some particular links are identified to haveanaffinityorlikeliness,andhencefollowscolocation VM placement pattern while ensuring their resource requirementconstraint[6]. At present, the major objectives of the proposed VM placementalgorithmare:

International Research Journal of Engineering and Technology (IRJET) e ISSN: 2395 0056 Volume: 09 Issue: 02 | Feb 2022 www.irjet.net p ISSN: 2395 0072 © 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page1

Abstract - The VM placement problem has been studied widely based on various metrics like migration cost, communication cost, traffic rate in order to provide optimization in terms of network traffic, energy consumption and resource management etc. The cost between any two communicating VMs is defined as the number of networking devices (e.g., switches) on the end to end routing path of the VM pair between the source machine and the destination machine. The proposed AVM approach exploits a meta heuristic optimization algorithm for VM placement and migration problem. The system proposes an advancement in the existing algorithm by reducing the search space and time required to find an optimal VM placement solution as it imposesnon completemigrationoadjustmenttheNebula,privatesetupresourcesusersCloudgeneticthecommunicationcommunicationtheavailabilitycoidentifiedapplicationstheoffspringcommunicationFurthermore,randomnessintheselectionofinitialpopulation.mutationbetweenthetrafficawareandcostawareVMplacementpatternsensuresthatreducethecommunicationcostforVMstoreachdestination.Inthissystemsomecommunicationsensitivethatinvolvefrequentaccesstothelinksaretohaveanaffinityorlikeliness,andhencefollowslocationVMplacementpatternensuringtheresourceconstraint.So,theaimofthissystemistogroupVMsbasedontheiraffinitymetricsliketrafficrateandcost.Byimplementingthiswork,thecostisreducedby15%.Theexecutiontimeofsystemhasalsoreducedwhencomparedtotheexistingalgorithm.Computingdeliversenormousnumberofservicestothewithunlimitedscalableandreliablecomputinginapayasyougobasis.ManyorganizationshavetheirpubliccloudandotherenterpriseshavesetuptheircloudsmanagedbyframeworkslikeOpenStack,OpenVMwarevCloudandCloudStack[1].ThestrategyforselectionofVMisakeystepofmultiVMresourceinasinglephysicalmachine(PM).FindinganptimalsolutioninpolynomialtimeisnotfeasiblesinceVMandserverconsolidationproblemsaremostlyNPoptimizationproblems.

Geneticalgorithmsaimtofindanapproximationsolution through multiple iterations, which are good methods for solvingNP completeproblems.Hence,theycanbeefficiently used for the VM migration and server consolidation problems. The Traditional Genetic Algorithm does not evaluateorfiltertheinitialpopulation[2].Thesearchspace andtimetofindtheoptimalplacementpatternishigherif the initial population is not filtered. Hence, we use the affinitybetweenthevirtualmachinestominimizetheinitial populationandtogroupthemtogether.

1) To place the Virtual Machines (VM) based on the degreeofaffinitywiththephysicalmachines. 2) To reduce the traffic rate upon virtual machines migration. 3) To reduce the communication cost for the virtual machine’splacement. Therearevariousresearchchallengesinvolvedin this work. Since VM migration and server consolidation problemsare mostlyNP completeoptimization problems, findinganoptimalsolutioninpolynomialtimeisnotfeasible. AsthenumberofVMstobeallocatedincreases,thesearch spaceandsearchtimetofindthenearoptimalsolutionalso increases.Sinceweusegeneticalgorithmtoimplementthe solution,itisessentialtoavoidtheriskofbeingtrappedina localminimalsolution. TwoormoreVMsrunningthesamecommunication sensitiveapplicationmighthavehugetrafficrateprevailing betweenthem.Thistrafficratewilldefinitelyhaveanimpact on the overall operational cost of the system [7]. The proposedsystemworksinthisscenarioandhelpstoreduce theoveralltrafficrateandcommunicationcostduringVM PlacementandMigration.Therearecertainnodes/PMsina networkofphysicalmachinesthattheotherVMsarehighly interestedin communicatingwith. Thesesmall number of cylindrical shaped high capacity nodes are functionally differentwhencomparedtotherestofthenodes.Herewe representcertainspecialhigh capacitynodesthatconnect thephysicalmachinesofonenetworkwiththeotherassink nodes.So,anyVMthatistobeplacedinoneoftheordinary PMswillhavecertaindependencywiththesesinknodes[3]. Thus,themajortaskistoallocatetheVMtothePMwhich ensuresminimalcommunicationcosttoreachthesinknode that the VM is highly dependent. Physical machine is an underlyinghardwareresource,whichcanbelogicallyused as similar partitions running different resources. Physical machineprovidestheresourcesforvirtualization. Virtualizationplaysanimportantroleinthesuccess ofcloudcomputingbyprovidingtheessentialinfrastructure for cloud. Virtualization is an abstraction of the logical resourcesfromtheirunderlyingphysicalresourcesbyhiding their heterogeneity, geographical distributions etc. It abstractstheresourcesofthephysicalserversandprovides multiplecompleteandisolatedexecutionenvironments[5]. Itispossibletorunmultipleapplicationsinseparate VMs havingdifferentconfigurationsduetodynamicsharingand physical resource reconfiguration facilities provided by virtualization. It is possible to run multiple operating systems and multiple applications on the same machine/serveratthesametime,increasingtheutilization and flexibility of the software. Resource utilization of the physicalserverscanalsobeimprovedbyvirtualization.Asa whole,virtualizationhelpsinimprovingresourceutilization, energy resources.availability,management,dynamicresourcesharingandprovidesscalabilityandreliabilityofthecloudcomputing

IfthesamePMcouldaccommodatetwoormorenumber ofVMs,thenitiscollocationaffinity[3].Iftheaffinityvalue between the PM VM pair is positive (i.e., higher than a particularthreshold),thenthatparticularVMcanbeplaced in that PM. So, all those VMs that exceed this positive threshold of affinity with the PM, which is satisfying their resourcerequirementscanbeplacedinthesamePM.Iftwo VMs are placed in different PMs for reasons other than resource constraints, then it is no colocation placement affinity.Dispersedly/separatelyplacingtheVMsinseparate PMs[3]canbetermedasNo colocationplacementaffinity.

When wedeploy the VMs forcommunicationsensitive applications, the sink nodes connect the VMs to different regionsofthedistributedcloudregion.So,eachVMwillhave a demand to each of the sink nodes of its region that connectsitselftoothercloudregions[8]. The demand for eachoftheVMtothesinknodessumsupto1.0,tomakeit easy for comparison purpose. Thus, the PM that connects itselfwiththesinknodethatishighlydemandedbytheVM, willbeselectedforallocationofthatVM.

VM placement is a strategy to allocate the virtual machinetothecorrespondingphysicalmachinesuch that, theoverallrequirementsoftheVMaresatisfiedbythePM. VMplacementmustalsoensurethattheplacementstrategy [4]enforcesminimaloperationalcost,communicationcost and traffic rate. VM migration helps the servers to be reconsolidatedorreshuffledtoreducetheoperationalcosts like energy, communication cost, traffic demand and bandwidthdemands[7].Networkcommunicationconsumes alargeportionofthetotaloperationalcostofdatacentres. Hence,toreducethetrafficbyreducingthecommunication cost,whilesatisfyingtheresourcerequirementsoftheVM and compensating the data transfer overhead due to migrationtoitsminimumisachallengingtask.Dependency betweenoneormoreVMswithaspecialplacementpattern runningaspecialapplicationcanbetermedasaffinity[3]. Affinitycanbederivedfromthecommunicationtrafficorthe datatransmissiontrafficthataregeneratedbetweenthetwo VMs whilerunningtheapplications.On simpleterms,it is the relationship between the VMs. Fixed VM Placement patternisusedif,weneedtoplacesomeVMsintoafixedPM. If the customer insists on placing their VM in a particular server,thenthispatternisused.

International Research Journal of Engineering and Technology (IRJET) e ISSN: 2395 0056

Communicationcostcanbedefinedasthetotalnumber of networking devices (e.g., switches) between the host machinestothedestinationmachine.Ouraimistoreduce the communication cost to its minimum upon migration. Communicationcostalsohasanimpactontheoveralltraffic in the network. Let a, b be two physical machines, then Communication cost (a, b) α Traffic Rate. As the communicationcostdecreases,thetrafficratealsodecreases [10].Cloud deployment models might include public cloud, communitycloud,privatecloudandhybridcloud.Whilethe public cloud provides services to everyone via internet,

Volume: 09 Issue: 02 | Feb 2022 www.irjet.net p ISSN: 2395 0072

2. RELATED WORKS 2.1 VM Placement Overview Mohammad Masdari et al. in [1] explain about the VM placement and migration in detail and in step by step process.TheVMsarebeingmigratedtotheoptedphysical machines based on the demand. In an intra cloud VM placement,firsttheindividualcloudmembershouldbefound tohosttheVM,whichPMshouldhosttheVM.Someofthe objectivesfactorsistochooseabetterphysicalmachinetofit in are reducing intra data center traffic, minimizing inter datacentertrafficinfederatedcloud,preventingcongestion indatacenter'snetwork,distributingtraffic evenlyindata center'snetwork,mitigatingenergyconsumption,reducing cost for cloud providers and increasing the return on investment(ROI),increasingsecurity,maximizingresource utilization, improving load balancing, high performance, locality: To increase accessibility, high reliability and availability, minimizing the number of active servers, reducing the number of active networking elements (switches),minimizingtheSLAviolation. In each placement only some of the objectives are met based on the requirements and follows a placement algorithmtomakealowoverhead,efficientVMplacementin clouddatacenters.Mostoftheplacementschemesrefersto the existing algorithm for comparison of the proposed placement algorithm. Some of the algorithms are first fit, secondfit,randomfit,leastfullfirstandmostfullfirst.

2.3 VM Placement based on network topology and communication cost

areprovidedauthorprofileTodecidesplacedencapsulatedinmultipleplacementGLPKsecondtoobjectiveoperatingandinitialplacementisintroducedforeachVMandtheappropriatephysicalmachinechosentomaximizethesatisfactionvalue.TheVMapproachisbasedondifferentobjectivessuchasplacement,throughputmaximization,consolidationservicelevelagreementsatisfactionversusprovidercostminimization.Therearetwomethodsoptedforsolvingthiskindofbasedproblems.Oneistousemathematicalmodelsdefinetheproblemandfeeditintosolversoperating.ThemethodistouseoptimizationtoolssuchasGurobi,etc.,VMplacementisalsobeingclassifiedasVMinsinglecloudenvironmentandVMplacementincloudenvironments.TheplacementisbeingdoneagivensetofphysicalmachineswithasetofdemandsinVMs.Thesefluctuatingdemandsarebeingbyaplacementcontroller.Theplacementcontrollerhowmanyinstancestobegonetofulfiltheobjective.fulfilthespecialobjectivetheavailabilityofVManditsandphysicalmachineanditscapacityaretaken.ThegivesanoptimalplacementalgorithmthatisofflinethecommunicationandflowdemandprofileofVMgivenearlier.

Network awareVMplacementalwaysworksin parwith most of the energy efficient means of VM placement. This techniqueaimsatreducingtheusageofserversbyswitching thestateofidleserverstosleepmodeandtherebyreducing thepowerconsumption[1].Italsoconcentratesonreducing the network communication cost after migration with specialconsiderationstointer VMdependencies.

Themajorgoalofthisresource awareVMplacementisto maximizetheresourceutilization.Itmainlyconcentrateson resourceslikeCPU,storageandRAMcapacity[9].Butasfor now,maximizingtheresourceutilizationisbeingconsidered as a major requirement for any type of VM placement techniques.IndataawareVMplacement,themajoremphasisison optimizingtheVMplacementbymaximumutilizationofthe inter VMcommunicationpatterninthenetwork[11].Italso insists on paying extra attention to the dynamic traffic patternsandthechangesinthenetworkafterVMPlacement.

Rest of the report is organized as follows: Chapter 2 contains the literature survey that discusses the previous worksandtheirscopeinthedomainofdiscussion.Chapter3 givesadetailedunderstandingoftheproposedsystemthat is to be implemented with the system overview, overall systemarchitectureandthealgorithms.Chapter4discusses the implementation steps in detail, tools and techniques involved for implementation and performance and result analysis. Chapter 5 deals with the conclusion and future workassociatedwiththeproject.

2.2 Network and traffic aware VM Placement Amir Rahimzad ehilkhechi et al. [2] focuses on specific objective VMplacementwithreducednetworkandtrafficin a special cloud from a provider’s perspective. The performance ofa VM strictly depends on the ability of the infrastructuretomeetthetrafficofVM.Asatisfactionvalueis

International Research Journal of Engineering and Technology (IRJET) e ISSN: 2395 0056

Volume: 09 Issue: 02 | Feb 2022 www.irjet.net p ISSN: 2395 0072

private clouds are managed internally by a single organization [1]. If two or organizations that have similar requirementsjoinhandstodeployacloudmodel,thenitisa community cloud. Different types of cloud models are interconnectedtogethertoformahybridcloudmodel.

TrafficawareVMplacementmainlyconcentratesonthe bandwidth demands, network topology, dynamic traffic variation,communicationpatternsbetweentheVMsduring the initial VM placement. Such techniques mostly aim at minimizing the network utilization, while maximizing the resourceutilization[8].Thisprojectisbasedonthisnetwork aware VM placement technique in which, the proposed system correlates the relationship between the average traffic/datarateandtheend to endcommunicationcost.

The author deals on the network topology and networkconditionsareconsideredduringtheVMplacement.

Most of the proposed approaches in [1] to fulfil the objectiveuseathree tierinadatacentercloudwhosegoalis tointerconnecttheserverswithahigh speedlink.Itconsists ofarackwhicharebottomlevelserversconnectedtoeither oneortwoTop of Rack(TOR)switches.TheTORswitches arefurtherconnectedtooneortwoswitchesinaggregation tier. Finally, each aggregation switch is connected with multipleswitchesincoretier.

Somespecialnodescalledsinknodesarebeingintroduced whicharefunctionallydifferentfromotherphysical nodes and has the high probability of VM interested to get communicated with these sink nodes. The VM that is dependentonthesinknoderequiresamassiveandend to end traffic. The efficient communication in terms of communication delay and data flow should be considered. The communication between the physical machines is negligible when compared to the communication between thephysicalmachineandthesinknodes.Theentirenetwork issetupintheformofgraphgivenbyG(V,E)whereEisthe linkbetweenthenodes,VthesetofnodesingraphG.Sisthe setofsinknodesinthenetworkwhereS∈V.TheVMwitha demandforsinkshouldnotbeplacedinaphysicalmachine thatsatisfiesthedemandbutofhighcost.TheVMplacement shouldbeinsuchawaythatthedemandforthesinkshould be satisfied also with the low cost. Maximizing the satisfaction factor both the customers and the service providers will be in a win win situation. The win win situation occurs since the VMs will experience a better qualityservicesatisfiescustomerandthelinksgettingless saturatedsatisfiestheproviders.Twodifferentapproaches are being introduced. Both greedy and heuristic based approachesarebeingintroduced.Eachoftheseapproaches hasdifferentvariants.Thesatisfactoryfactoralsovaries.

The simulation results of these algorithms are being testedusingsimulationsandtheresultofthesealgorithmsis found to be nearly optimal when compared to the VM placementregardlessoftheirdemands.JianhaiChenetal[3] dealswithjointaffinityawaregroupingandVMplacement.

2.4 VM placement patterns ThenbasedontheaffinityrelationshipbetweentheVMs an affinity aware resource scheduling framework is being generated.ThegoalofresourceallocationbyJianHai[3]isto maximizetheratiobetweenperformanceandcostandhence thedatacenterrevenueswhichinvolvesbothperformance andenergyefficiency.Theenergyandperformanceefficiency requiresallocationtominimumnumberofphysicalmachines andallocationtominimizetheperformancecostrespectively.

Thevirtualmachinesarebeingplacedinthephysical machinesbasedonrandomgenerationorVMplacementis donebasedoncertaindemandsofthecustomer.Thismay alsoincludetheefficientperformanceofapplicationpresent indifferentvirtualmachinesfromtheproviderperspective. After the placement scheme by Weizhe Zhang et al. [4] of virtual machines in different cloud environments, the communication between different VMs is being continued. Basedonthedatatrafficwhichinfluencesthecommunication costandtheperformanceefficiencyoftheapplications,the VMsarebeingreplacedtoanotherphysicalmachinesothat thedatatrafficgetsreduced.Thisreductioninamountofdata transfer will definitely reduce the communication cost between the VMs and also increase the efficiency of the performance of application hosted in different virtual machines. This replacement of virtual machines after analyzingthedatatransferbetweenthevirtualmachinesis calledthemigrationofvirtualmachines. VMmigrationreducestheoperationalcostastheservers arebeingreshuffled.Accordingtomethodproposedbythe author the demand for the traffic and bandwidth also accountsabitinthetotaltrafficofthesystem.VMmigration also causes additional data transfer overhead. So VM migration should be in such a way that the data transfer overheadshouldbereduced.Thecommunicationcostand theperformancefactorshouldalsobereducedcomparedto the foundofbetweenintelligencecommunicationrelationshipphysicalwhichtakingcommunicationcompleteformulatedpreviousplacementstrategy.Anetworkawareproblemisintoanondeterministicpolynomialtimeproblem.Thisstudyaimstoreducethecost.ThecommunicationisbeingreducedbyintoaccounttheinheritdependenciesamongVMsformsamultitierapplicationandthetopologyofthemachines.ThemigrationisdonesothatagoodismaintainedbetweenthenetworkandVMmigrationcost.Theswarmalgorithmisbeingusedtomaintainthistradeoffthenetworkcommunicationandthemigrationcostthenetwork.Anapproximateoptimalsolutionisbeingfirst.Thennumberofiterationsisbeingperformed.

Certified Journal | Page4

Non affinity aware grouping based resource allocation towardsgeneralVMplacementplacestheVMsintothesame physicalmachinewhicharetobeplacedindistinctphysical machines and vice versa. This leads to performance degradation of applications in different VMs deployed in multiple machines. A joint affinity aware problem is being definedandajointaffinityawaregroupingandbinpacking methodisproposedtoovercometheproblemraisedinnon AffinityawareVMplacement.Firsttheaffinityrelationships between the VMs are being calculated based on the relationships between the VMs such as collocated or dispersedlylocatedandanaffinityproblemisbeingdefined.

International Research Journal of Engineering and Technology (IRJET) e ISSN: 2395 0056

Volume: 09 Issue: 02 | Feb 2022 www.irjet.net p ISSN: 2395 0072

2.5 Virtual Machine Migration in an Over Committed Cloud

Byco locatingtheVMswhicharerequiredtobedispersedly located may require more resource in a single physical machine. This may lead to performance lag in the VM physical machine pair. Dispersed location of the virtual machines which are to be co located may increase the communication between the virtual machines. This may increasethedatatrafficandalsothecommunicationcostin thenetwork.ThisisthesameproblemarisinginrandomVM placement. The challenges put forth are to make detailed analysis of the VMs are to be placed in the appropriate physical machinessothatthecommunicationcostandthe performanceoftheapplicationrunningintheVMdecreases and increases respectively. Also, a heuristic performance evaluation for the placed VMs is being done to check the quality of VM placement. It is also said that solving the resource scheduling and allocation is similar to the bin packing problem which is a NP hard problem. So, making decision of the optimal VM placement becomes more challenging when performance maximization and communication cost reduction guarantee is also being considered. Generalization of dependency since current systems does not focus on the generalization of VMs. This method also helps to obtain and identify the relationship betweenVMs.Thebinpackingproblemalsomaymisplace the colocation placed and dispersedly placed VMs irrespectiveoftheirdependencies.

International Research Journal of Engineering and Technology (IRJET) e ISSN: 2395 0056 Volume: 09 Issue: 02 | Feb 2022 www.irjet.net p ISSN: 2395 0072

ThesystembyFeiTao[5]alsoprovideshighsecurity andisolationwhenrequiredtotheoperatingsystems. Also encapsulatesandcustomizestheapplicationsrunninginthe virtual machines. Thisalsoincludes thesupport forlegacy applications.Migrationisusuallydoneincloudenvironment after the VM placement and also analyzing the data flow between the virtual machines placed in different physical machines.BasedonthedataflowbetweentheVMs,toreduce the communication cost and to increase the performance efficiency of the application running in different virtual machines,migrationofvirtualmachinesisbeingcarriedout. The author solves the VM migration problem but in the dynamicmode.Basedonthetotaloperationalcostoccurring duetothedatatrafficintheintegratedcloudenvironment the migration is done. Then this VM placement scheme is analyzedbyallowingtheflowofdatabetweentheVMsinthe cloudToenvironment.reducethetotal operational cost further, the VM is beingreshuffledtoappropriatephysicalmachinesavailable inthesystem.Thisdynamicallocationofvirtualmachinesto thephysicalmachinesenablestheprovidertochooseabetter placementschemeamongthe availableoptimal placement schemes.Theinefficiencyofthevirtualmachinesisduetothe reasonthatimproperallocationofthevirtualmachines.This leadstoimbalancedresourceschedulingdistribution.Thus,it

3.1 Affinity based grouping

issaidthattheproperuseofVMsreducestheoperatingcost andtheenergyconsumptionbythesystem. Theefficiencyofthevirtualmachinesisbeingimproved byemployingthreeprimarymeasures. Thethreeprimary measures are improving the framework and distribution, schedulingtasksandmigratingVMsdynamically.Otherthan energy consumption and efficiency, networking and communicationarethegreatchallengestobefacedduring theVMplacementanddynamicVMmigration.Itisknown thatthedatatobetransferredtoavirtualmachineinother physicalmachinetakesmuchtimethanthetimetakenbythe VMtocommunicatewiththevirtualmachinepresentwithin thesamephysicalmachine.Thisisbecausedatarequiresless timetotransferinrandommemorythanvianetwork.

The solution is given in the form ofan algorithm. This Non dominated sorting genetic algorithm is being chosen sinceitsolvesmulti objectiveproblems.Thisalgorithmfalls intothecategoryofintelligenceoptimizationalgorithm.This algorithm can yield better results when some parts of the algorithms are made to be deterministic. The algorithm is proposedinsuchawaythatitismoreadaptivetothecloud. ThisalgorithmwhenimplementedproducesasetofPareto optimization solutions. The main contribution is split into three main categories. The multi objective optimization algorithmcontributesthefollowingobjectives.Thedynamic migration of VMs is made efficient by proposing a triple objectiveoptimizationmodel

2.6 Dynamic Migration of Virtual Machines

The proposed work deals with placing the virtual machines in the physical machines by reducing the traffic rateandcommunicationcostuponmigration.Thisproblem canbeconsideredasanoptimizationproblem,sincewehave to reduce the traffic rate and communication cost while considering,whethertheresourcerequirementsoftheVMs can be satisfied by the PMs. This work uses an AVM algorithmic approach to solve this optimization problem. When the applications are deployed on the VMs in a distributedcloudregion,eachapplicationmighthavecertain demand to those nodes which connects other VMs which alsohavesimilarcomputationenvironmentinthenetwork. So,thedemandofeachvirtualmachinetothesinknodesof thatregionistakenastheinputfortheinitialVMplacement algorithm. Based on this demand, the PM satisfying this demandwithminimalcommunicationcostisconsideredas thebestoptiontoallocatetheVM.Thereexists anaffinity betweentheVMandthePMthatsatisfiesitsdemands.Hence thisVM PMpairisgroupedtogether.

Geneticalgorithmisadoptedand changedbytheauthortosuittheVMmigrationproblem.This changeinadoptedalgorithmreducesthecommunicationcost andnetworkcost.Experimentalresultsverifythatnetwork costisreducedbygeneticalgorithmwhentheinstancesare low.Whenthenumberofinstancesishigh,geneticalgorithm does not work efficient. In such cases artificial bee colony algorithmreducesthenetworkcost.So,whenthenumberof instances is high artificial bee colony reduces the communication cost and when the instances are less, networkcostisbeingreducedbygeneticalgorithm.Itisalso advantageousthattherunningtimeofartificialbeecolony algorithm is half than that of running time for genetic algorithm. The most commonly used consolidation in VM migration is energy aware VM migration. The aim of this energy aware server consolidation is to improve the utilizationofeachserverbyconsolidatingasmanynumbers ofVMsintoservers.Theremainingunusedserversremainin low powerstate.Thismethodconsidersonlytheserver side constraintsincludingtheresourcesdemanded.Thenetwork costduetovirtual machinemigrationisbeingoverlooked. Network communication incurs a great part in the total operationalcost.Manyofthestudiesconsideronlyonefactor i.e., on migration cost, communication cost or energy consumption. The solution also solves the data center sufferingfromthelong livedcongestion.Corenetworkover commitmentandunbalancedworkloadcausesthislong lived congestion.

3. AFFINITY AWARE VM MIGRATION (AVM)

Acomprehensiveanalysisoftheabovesolutionis made which is non deterministic polynomial complete problem.

From these iterations a good solution among the optimal solutionsis Artificialchosen.beecolonyand

EachVMhasaspecificdemandforeachofthesinknodesas specifiedintheFigure1.Thesumofthedemandsforeachsink nodeis1.0.

Fig 1: Communication between sink node and VMs Thethreeparametersofthedemandvectorinthefig1 representthedemandofthevirtualmachinetoanyavailable sinknodeinthenetworktoreachthephysicalmachine.So, thedemandof0.7representsthedemandofsink3whichis thehighesttoreachtheexpectedphysicalmachine.Atthe sametimeassumethereare2physicalmachinesavailable withsameworkingenvironmenttheVMswillbeplacedto onethathasasmallernumberofhopstoreach.Henceinthe fig1.PM1willbeselectedtoplacetheVMsasitrequiresone hopcomparedtofivehopsforthePM2. ThecloudusersrequesttopayfortheVMs,toruntheir applications in them. Every application might require different number of VMs for running. In case of communication sensitive applications, they have huge data/trafficratebetweenthem.So,whiledeployingtheVMs itmusttrytominimizethetrafficrate.AffinityGeneratorin the architecture (Fig. 2) is responsible for framing all the affinity relations between the VMs while running the applications.Theinputtotheaffinitygeneratorconsistsof the resource data of VMs, PMs and the workload data are receivedbytheaffinitygeneratorfromthecloudcontroller.

Ifaschedulerequestisreceived,itisimmediatelyforwarded to the VM Optimizer to make VM placement/migration decisions. Final decisions regarding the VM placement, migration,bootorshutdowndecisionsaredonebytheVM optimizer.Itcorrelatestwofitnessfunctionstodecideupon theaffinityBasedgrouping.onthenetworktopology,eachofthephysical machines is assigned a communication cost to each of the sink nodes of the network. In the proposed work a fully connectednetworktopologyisconsideredwith65physical machines and 27 currently requesting virtual machine.

3.2 Affinity Calculation

Fitnessfunctionsareusedforcalculatingthecommunication cost and migration cost, which in turn is a minimization function. Mutation is performed between the resultant individuals of these fitness functions to minimize the communicationcostandtrafficrate.Thepopulationsizeisset to100individualsandthemutationrateissetto0.1.Wetryto mutatebetweentheVMallocationscodeswhicharereferred as bucket codes in the algorithm that are generated using trafficrateaffinity and migrationcostaffinity.Theresultant allocation minimizes the traffic rate, thereby reducing the communicationcost.

TheinitialVMplacementalgorithm(3.5)beginshereandit is used to find the best physical machine to host the virtual machine and the system proceeds to the VM migration algorithm(3.7).TheresultofthisalgorithmisthepossibleVM migration pattern that reduces the communication cost and trafficrateuponmigration.

Thisaffinitycanbecalculatedusingthefollowingequation

International Research Journal of Engineering and Technology (IRJET) e ISSN: 2395 0056

Inordertorecordthedetailedaffinityinformationofthe VMs,VMoptimizerauditsandstorestheaffinityrelations.As a result, analysis of the gathered affinity data is done to decidewhetherthereisaffinitybetweentheVMandPMor not.Resourceschedulingthatisdonebasedontheaffinityis controlled and managed by the affinity scheduler. The schedulingrequestsfromthecloudcontrollerischeckedby the affinity scheduler and it decides on the type of scheduling that is to be done (e.g., initial placement, migration,loadbalancing).

3.3 Proposed AVM Architecture

Volume: 09 Issue: 02 | Feb 2022 www.irjet.net p ISSN: 2395 0072

3.5 Algorithm for Initial VM placement

Based on this demand and communication cost, affinity betweentheVMandPMcanbecalculatedusing(1). Fig 2: Proposed AVM architecture

3.4 Communication Cost Matrix

Sink PM Sink1 Sink2 Sink3 PM1 3 3 7 7 2 2 PM2 4 4 5 5 5 5 PM3 7 7 4 4 7 7 PM4 5 5 8 8 4 4 PM5 3 3 9 9 3 3 (2)

Output: SolutionmatrixindicatingtheassignmentofVMsto PMs. Step-1 Declareinitialaffinity. Step 2 foreachvmi inVMdo Step 3 foreachpmi inPMdo Step 4 foreachsi inSdo Step 5 Calculateusingtheformulae: Step 6 endfor Step 7 assignAff[vm][pm]=affinityandresetaffinitytozero. Step 8 endfor Step 9 ForeachrowinAff[M][N]findthemaximumaffinity toapmi forEachvmj Step-10 IfRA[pm]>RD[vm]thenassignX[vm][pm]=1

Eachphysicalmachinewillhaveacommunicationcost,that iscalculatedbythenumberofhopsornetworkingdevices likeswitches/routers.Thefollowingmatrixrepresentsthe communicationcostbetweeneachPMandSinknodes. Usingthiscommunicationcostanddemandin(Table1),the affinitybetweenaVM PMpairiscalculatedusing(1) Based on this demand and communication cost, affinity between the VM and PM can be calculated using (1). The notationsfortheVMPlacementAlgorithmaredescribedin thefollowingTable3. Table 2. Communication Cost Matrix between PMs and Sink Nodes

Inputs: VMset,PMset,Resourcedemandandavailability vectors,StaticcostbetweenthePM sinkpair,demandof VM sinkpair.

VMswillhavetobemigratedonlyif,thetrafficrateand communicationcostwillgetreducedaftermigration.So, themigrationcostiscalculatedbyconsideringthesizeof theVMandthetrafficratebetweentheminthenetwork. TheVMsthatshowmaximumaffinitytoaparticularPM aregroupedtogether.Thisgroupisthenconsideredasa

The communication cost in this context can be describedasthenumberofhops(Networkingdeviceslike routersorswitches) betweenthesource machineandthe destination machine. The two virtual machines have communication cost affinity, if they run the same communication sensitive application, but with relatively higher communication cost. The following equation is the fitnessfunctionforcalculatingthecommunicationcost,and thesystemminimizesit.Thetrafficrateismeasuredbysome activeorpassivetechniquesofnetworktrafficmeasurement depend on the environment. In this work it is measured usingcloudsimtrafficgenerator.

(A) Communication cost Affinity

ACRONYM DESCRIPTION M NumberofVMs N NumberofPMs VMSet VM[1…….M] PMSet PM[1… .N] mi ith VM pmj jthPM S NumberofSinkNodes SinkNodeSet S[1… .S] si ithsinknode X[M][N] maAssignmenttrix/Solutionmatrix D[M][S] DemandVectorforeach VM Sinkpair RA[N] ResourceAvailability matrixforPMs RD[M] ResourceDemandmatrix forVMs SC[N][S] StaticcostbetweenPM sink pair DC[N][S] Dynamiccostdueto congestioninthenetwork Affinity[M][N] AffinitybetweenVM PM pair fitnesscommn Fitnesscalculationbasedon communicationcost fitnessmig Fitnesscalculationbasedon migrationcost. (3)

Table 3. Notations for Initial Vm Placement Step 11 elseAff[vm][pm]=0,gotoStep9 Step 12 DisplaythesolutionmatrixX[vm][pm]

International Research Journal of Engineering and Technology (IRJET) e ISSN: 2395 0056

Volume: 09 Issue: 02 | Feb 2022 www.irjet.net p ISSN: 2395 0072

(B) Migration Cost Affinity Trafficrateisadynamicindicator.Itchanges fromtimetotimebasedonthetypeofapplicationsthat arerunningontheVM.Higherthetrafficrate,higheristhe operationalcost.Hence,inordertominimizethetraffic rateinparwithreducingthecommunicationcost,the trafficrateisconsideredasanimportantconstraint,while calculatingthemigrationcost.(3)istocalculatethefitness function(minimization)ofmigrationcostaffinity(4)

3.6 Affinity Types

Population size=100 Mutation rate=0.1 Input:bucketcodegeneratedfromalgorithm1 Output:nearoptimalbucketcodeforassignment Step 1Generatearandomizedinitialpopulationofbucket Stepcodes.2Addthebucketcodegeneratedfromalgorithm1to therandomizedpopulations Step 3Calculatefitnesscommn(vm,pm)= Step 4Calculatefitnessmig(vm,pm)= Step 5Choosetwoindividualseachwithmaximum fitnesscommn,fitnessmig andperformtwo pointcrossover. Step6Calculatethefitnessoftheresultantcode,andcheckif itisbetterthanthebestfitnesssofar.

GUI support.Net Beans is an Integrated Development Environment for developing Java applications and also Serverandwebrelatedapplications.Applicationwhichhas databasemanipulationsoperationsalsodevelopedwithhelp this IDE. The Net Beans IDE is primarily used for developmentofJava,butalsosupportsotherlanguages,such asPHP,C/C++andHTML5. The Net Beans Platform provides a framework for the developmentofinteractiveJavaSwingrealtimeapplications. Net Beans plugins and NetBeans Platform based applications;noadditionalSDKisrequired.Applicationscan install plugins and private libraries depending upon the necessarilyoftheapplications.

Cloud Reports is a GUI tool for the simulation of cloud computing in a distributed environment. Cloud Reports is the graphical format of Cloudsim which is user friendly,reliableandcustomizable.Italsoallowscreatenew customers, datacenters, hosts with different bandwidth capacity,Resourceslikestorage,memoryMIPSetc.

The project implementation is done in NetBeans IDE, in which cloudsim 3.0 is installed to analyse the behaviourofthealgorithminreal timecloudenvironment. Cloudsimdoesn’tprovidetheGUItovisualizetheresultof theanalysisingraphicalformat.So,CloudReportsisusedfor

(A) CloudSim CloudsimtoolkitisdevelopedatMelbourneUniversity bytheGRIDSlaboratory.Itisa multi layeredsimulation structurewhichincludesmodelingandsimulationofdata centerenvironments.Italsoprovidesgoodmanagementof VM interfaces, memory, storage etc. It also induces schedulingandallocationpoliciesforcloudplatform.Ithas alsobeenidentifiedasthebesttoolforcloudcomputing sinceitallowsthemodelingofIaaS,PaaSandSaaSservices.

Each VM will have a demand for each of the sink nodes.ThesumofthedemandsofaVMtoallthesinknodes mustbeequalto1.0.ThematrixintheTable3.represents thedemandmatrixbetweenfiveVMsand3Sinknodesthat arepresentinthescenariorepresentedearlier.Theresource requirement of the virtual machines (Table 4) must be

(B) Cloud Reports

ItextendsallthefunctionalitiesofCloudSimlikeVM allocationpolicies,powerconsumptionandbrokerpolicies and other resource allocation and resource utilization models.CloudReportsprovidecapabilityforuserbysetting the number of virtual machines each customer owns, a brokerresponsibleforallocatingthesevirtualmachinesand resourceconsumptionalgorithms. 4.2 Input and Output vectors

Step7Ifyes,thenaddittothepopulationelsediscardand movetostep5 Step8Stopifmatrixisreachedorthebestfitnessdoesn’t changeovermuchiteration.

International Research Journal of Engineering and Technology (IRJET) e ISSN: 2395 0056 Volume: 09 Issue: 02 | Feb 2022 www.irjet.net p ISSN: 2395 0072

4. IMPLEMENTATION AND RESULT ANALYSIS

4.1 Tools and Techniques

© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page9 singleunit,andisallocatedtothatPM,providedthatthe resourcerequirementsaresatisfied.Oncetheresource requirementsoftheparticularunitofVMsaresatisfiedby thePM,thentheVMOptimizerdecidestoallocatethat particularsetofVMstothatPM 3.7 Algorithm for VM migration

Table 5. Affinity Matrix between VMs and PMs

0.18

4.3 Performance Metrics

Theproposedsystemthuscalculatestheaffinitybetweenthe VMsandPMs,usingwhichtheVMshavinggreateraffinityto aparticularPMwillbegroupedtogethertobeplacedinthe particularPM.

PM/VM

Detailedinformationaboutthecreationand imagesforarticlescanbefoundat: Westronglyencourageauthorstocarefullyreviewthe material Fig. 4. Traffic Rate Comparison Id 1 2 3 4 5 20 5 1 5 10 25 2 4 5 9 0.27 0.22 0.19 0.24 0.00 2 0.24 0.20 0.22 0.00 3 0.21 0.21 0.17 0.27 0.00 VM 4 0.26 0.24 0.20 0.00 0.00 VM 5 0.20 0.19 0.16 0.24 0.27 No.Ofmigrations

(A) Reduction in Communication Cost

 TrafficRate  CommunicationCost

Table 4. Resource Availability and Demand Vectors

PM VM PM 1 PM 2 PM 3 PM 4 PM5 VM 1

satisfiedbythephysicalmachineinwhich,theVMistobe Theplaced.twoarraysrepresenttheunitsofresourceavailablein each PM and the units of resource required by each VM. Usingtheabovematrixinputs,theaffinitybetweeneachPM andVMcanbecalculatedusingtheaffinityformula,andthe resultantmatrix(Table5)displaystheaffinitybetweenthe VMs and PMs within the scale of 0.00 to 1.00.for which 0 beingthelowestand1beingthehighest.

Sinceaffinityawaregroupingisusedtogroupthe VMs based on their demand and Communication cost, the traffic rate is observed to be reduced eventually when comparedtotheexistingBGM BLAalgorithm.Even though the number of migrations increases, the traffic rate for the proposedsystemisseentogetreducegradually.

Theperformanceoftheoptimisationproblemcanbe calculatedbythefollowingtwometrics:

Thefollowingbarchartrepresentsthereductionin the communication cost after migration. By including the communicationcostaffinity,itreducesthecommunicationcost by15%whilecomparingwiththeexistingsystemasspecified inthebarchartinFigure3.Here,Consumer1usesBGM BLA andConsumer2usesAVMPlacementSolution. (B) Reduction in the Traffic Rate

5. CONCLUSION AND FUTURE WORK

The initial VM placement algorithm considers the affinity factorasanimportantconstraintingroupingandplacingthe VMs. Some communication sensitive applications, when deployedinVMswithheavytrafficratebetweenthem,incur highoperationalcost.Thisworkdefinedanaffinitymatrix between the PMs and VMs based on the demand of the virtual machine to the corresponding high capacity sink nodeoftheregionandthetrafficratebetweentheVMs.The affinityisusedbetweenthevirtualmachinestominimizethe initialpopulationandtogroupthemtogether.Theaffinity awareVMplacementalgorithmreducesthecommunication cost and traffic rate between the VMs that run communicationsensitiveapplicationsuponmigration.Since theproposedworkreducethesearchspaceandsearchtime byintroducingaffinityawaregrouping,theexecutiontime required for runningthealgorithm alsogetsreduced. The system considers this as an optimization problem and proposesthisnetworktraffic awareVMPlacementSolution. Theproposedsystemthusreducesthetrafficrateandhence reducingthecommunicationcostby15%.Therearemulti objective VM placement Solutions in this sector, that proposes solutions based on network traffic, topology, resource utilisation and energy efficiency. So, we can transformthisoptimizationsolutionintoamulti objective onewithmorethantwoparameterstooptimize.TheAffinity awareVMPlacementtechniquecanfurtherbeimprovised using machine learning algorithms which we planned to implementinfuture.

International Research Journal of Engineering and Technology (IRJET) e ISSN: 2395 0056 Volume: 09 Issue: 02 | Feb 2022 www.irjet.net p ISSN: 2395 0072

REFERENCES

[1] Mohammad Masdari, Sayyid Shahab Nabavi, Vafa Ahmadi, “An overview of virtual machine placement schemesincloudComputing”,JournalofNetworkand ComputerApplications66,2016 [2] AmirRahimzadehIlkhechi,IbrahimKorpeoglu,Ozgur Ulusoy,“Network awarevirtualmachineplacementin cloud data centers with multiple traffic intensive components”,ComputerNetworks91,2016 [3] Weiwei Fang, Xiangmin Liang, Shengxin Li, Luca Chiaraviglio, Naixue Xiong, “VMPlanner: Optimizing virtual machine placement and traffic flow routing to reduce network power costs in cloud data centers”, ComputerNetworks57,2013 [4] JianhaiChen,QinmingHe,DeshiYe,WenzhiChe,Yang Xiang,KevinChiew,LiangweiZhu,“Jointaffinityaware grouping and virtual machine placement”, MicroprocessorsandMicrosystems,2016 [5] Fei Tao, Chen Li, T. Warren Liao, and Yuanjun Laili, “BGM BLA:ANewAlgorithmforDynamicMigrationof Virtual Machines in Cloud Computing”, IEEE transactions on services computing, Vol. 9, No. 6, November/December2016 [6] [6] M.Aibinu,H.BelloSalau,NajeebArthurRahman, M.N. Nwohu, C.M. Akachukwu, “A novel Clustering based Genetic Algorithm for route optimization”, EngineeringScienceandTechnology,an International Journal19, 2016 [7] Alvaro Garcia Piquer, Jaume Bacardit, Albert Fornells, Elisabet Golobardes, “Scaling up multi objective evolutionaryclusteringalgorithmsusingstratification”, PatternRecognitionLetters,2016 [8] Arash Chaghari, Mohammad Reza Feizi Derakhshi, Mohammad Ali Balafar, “Fuzzy clustering based on Forest optimization Algorithm”, Journal of King Saud University ComputerandInformationSciences,2016 [9] A.Azadeh, S. Elahi, M. Hosseinabadi Farahani, B. Nasirian,“Ageneticalgorithm Taguchibasedapproach to inventory routing problem of a single perishable product with transhipment”, Computers & Industrial Engineering,2017 [10] Chao Lung Yang, R. J. Kuo, Chia Hsuan Chien, Nguyen Thi Phuong Quyen, “Non dominated sorting geneticalgorithmusingfuzzymembershipchromosome forcategoricaldataclustering”,AppliedSoftComputing, [11]20

Thecorrespondinggraphmeasurestheefficiencyof the proposed method compared to the existing method.Fromthegraph(Fig.5)itisinferredthat,over the iterations with different datasets the proposed method out performs existing approaches in its efficiency which is measured by the time taken to completethetaskevenaftertheVMmigration NumberofIterations 5. Efficiency of the proposed AVM method comparison

Fig.

Jiangtao Zhanga, Xuan Wanga, Hejiao Huanga, Shi Chena,“Clustering basedvirtualmachinesplacementin distributed cloud computing”, Future Generation ComputerSystems66,2017 [12] Juan Zamora, Marcelo Mendoza, Héctor Allende, “Hashing based clustering in high dimensional data”, ExpertSystemsWithApplications,2016

E F F I C I E N C Y

Journal | Page12

International Research Journal of Engineering and Technology (IRJET) e ISSN: 2395 0056 Volume: 09 Issue: 02 | Feb 2022 www.irjet.net p ISSN: 2395 0072

Dr. Gokuldhev M is currently working as Assistant Professor, Department of Computer science and Engineering, Vel Tech Rangarajan Dr. Sagunthala R & D institute of Science and Technology,Chennai Thanammal Indu V is currently working as Assistant Professor, Department of Computer science andEngineering,Amritacollegeof Engineering and Technology, Nagercoil,India.

BIOGRAPHIES

[13] Satish Chander, P.Vijaya, Praveen Dhyani, “Multi kernel and dynamic fractional lion optimization algorithmfordataclustering”,AlexandriaEngineering Journal,2017 [14] Xiumin Wanga, Xiaoming Chenc, Chau Yuend, Weiwei Wue, Meng Zhangf, Cheng Zhang, “Delay cost tradeoff for virtual machine migration in cloud data centers”,JournalofNetworkandComputerApplications 78,2017 [15] Yongqiang Gao, Haibing Guan, Zhengwei Qi, Yang Houb, Liang Liu, “A multi objective ant colony system algorithm for virtual machine placement in cloud computing”,JournalofComputerandSystemSciences 79,2013