linksns

VirtualNetworkEmbeddinginElasticOpticalDataCenterNetworkAidongSu1,a*andYongyiZhang1,b1DalianAirforceCommunicationNCOAcademy,P.
R.
Chinaasuaidong@126.
com,b80834567@qq.
comKeywords:VONembedding;cloudcomputing;elasticopticalnetwork;datacenterAbstract.
Theboominginternetservicesurgetheresearchonthecloudcomputingandthenetworkresourceutilization.
Thecombinationofelasticopticalnetworkanddatacentercansolvethenetworkresourcesdeficientproblemandthecomputingresourcesunbalancedproblem.
Virtualopticalnetworkembeddingprovidesthewayofresourcehigh-efficiency.
Inthispaper,weproposeanovelvirtualnetworksembeddingalgorithmorientingspectrumresource,andthesimulationresultsverifyitssuperiorityonincreasingspectrumresourceutilizationandreducingdemandblockingrate.
IntroductionWiththedevelopmentofInternet,bandwidthdemandisbooming.
Theconstructionofflexibleandlargecapacityopticalnetworkbecomesimportant[1].
WavelengthDivisionMultiplexing(WDM)opticalnetwork,allocatingnetworkresourceinaone-size-fits-allmanner,leadstoinefficientresourceutilizationandlowflexibility.
Byintroducingtheorthogonalfrequencydivisionmultiplexing(OFDM)andbreakingthroughthefixedbandwidthspacingrestrictionbetweenwavelengthtunnels,elasticopticalnetworkutilizesspectrumresourcesefficiently.
Meanwhile,withthewidespreaduseofcloudcomputingandvirtualizationtechnologyindatacenters(DC),thevirtualnetworkembedding(VNE)becomesachallengeintheclouddataDCnetworks(DCNs)[2-5].
Itenablestheco-existenceofmultiplevirtualnetworksonthesamesubstratenetworkbysharingtheavailableresources.
Thus,VNEinelasticopticalDCNsiswidespreadlyconcerned.
ThispaperfirstdescribesVNEprobleminelasticopticalDCNsandpresentsthecorrespondingmathematicalmodel.
Forstatictraffic,wedesignavirtualnetworkembeddingalgorithmbasedonthelayeredauxiliarygraphreferredtoasVNEorientingspectrumresource(VNE-OSR).
Theproposedalgorithmcanintegratefourdifferentserviceorderingstrategies.
Simulationresultsshowthat,intermsofimprovingnetworkresourceutilizationandreducingtheblockingrate,theproposedVNE-OSRalgorithmreflectsgoodperformances.
ElasticOpticalDataCenterNetworkVirtualizationFig.
1SchematicdiagramofvirtualnetworkembeddingOpticalDCNvirtualizationequatesthecombinationofthevirtualnodeembeddingandthevirtuallinkembedding,i.
e.
,themappingfromvirtualopticalnetwork(VON)tophysicalnetworks[6].
Thatincludes1)selectingappropriateservers(orDC)forthecomputingresourcerequestsofvirtualnodes,i.
e.
,themappingfromvirtualnodestosubstratecomputingelements,and2)allocatingappropriatefiberlinksandspectrumforvirtuallinks,i.
e.
themappingfromvirtuallinkstofiberlinks[7].
Concretely,asshowninFig.
1(a),thereare5serversand6fiberlinksinthesubstratenetworks.
Thereexist8spectrumslotsineachfiberlink,whichcanbeexpressedbyaneight-binary-array,where"1"denotesthisspectrumslothasbeenoccupied;otherwise,it's"0".
Thenumberbesideseachserver(orDC)indicatestheremaindercomputingresource.
AsshowninFig.
1(b),thearrivingVONneeds3virtualnodesof4computingresourcesand2virtuallinksof2continuousslots.
Fig.
1(c)showstheresultofVONembedding,i.
e.
,thevirtualnodesa,bandcaremappedaccordinglytoserversD,BandE,andthevirtuallinksabandacaremappedaccordinglytoDBandDE.
TheVNEintheelasticopticalDCNscanneatlydistributespectrumsaccordingtodemands,soitcanrisethespectrumresourceutilization,andmeanwhile,VNEmainlyorientsthescenewheretheDCNpower-systemfailsandthenrecoversgradually.
Inthissituation,thereexistmanyimproperserverssinceapowerfailureandscarceserver-computing-resourcewillleadtomanyblockedVONdemands,thusitisveryvaluabletoresearch.
ProblemDescriptionTheelasticopticalDCNshavetheabstractedsubstratetopology(,)sssGVE,wheresVrepresentsthesetofsubstratenodes,andsErepresentsthesetofbi-directionallinks(eachlinkisconsistoftworeversed-unidirectionalfibers).
EachsubstratenodesnV∈hasacertainamountofavailablecomputingresourcenc.
ThespectrumresourceineachfiberlinkseE∈isdividedintospectrumslotswiththesamebandwidth,andeachspectrumslotcorrespondswithanOFDMsub-carrier,i.
e.
eachfiberlinkconstitutesaseriesofcontinuoussub-carriers.
ThissituationcouldbeexpressedbyabinaryarrayebwithBelements,whereBrepresentsthemaximumsub-carrierquantityineachfiber.
EachVONrequestcouldbeindicatedbynon-directionalgraph(,)rrrGVE,andeachvirtualnoderjV∈hasitscomputingresourcerequestjm.
InthesameVON,anybandwidthsub-requestamongallvirtuallinksisequal,sothebandwidthrequestofeachvirtuallinkrkE∈isindicatedbyrn,i.
e.
,itisthecontinuoussub-carrieramountwhichneedbeassignedtothevirtuallink.
Eachfiberlinkhasthesamequantityofsub-carriers,andasmentionedabove,anyrequiredbandwidthineachvirtuallinkinthesameVONrequesthasthecoincidentamount.
ThecoreofVONproblemistomapaVONrequestintosubstratenetworks,i.
e.
themappingfromvirtualnodesintosubstratenodesandthemappingfromvirtuallinksintothefiberlinks.
ForthestaticVONembeddingproblem,giventhatalltherequestdemandswerenotblocked,thetargetoftheVONembeddingalgorithmisminimizingthemaximumsub-carrierserialnumberusedinallfiberlinks.
VirtualNetworkEmbeddingAlgorithmWeproposeanovelVNEalgorithmbasedonthelayeredauxiliarygraph(LAG)referredtoasVNEorientingspectrumresource(VNE-OSR)forstaticdemands,andittakestwophases:thecomputingresourceallocationforvirtualnodesandthebandwidthresourceallocationforvirtuallinks.
Thealgorithmcanallocateappropriatespectrumresourceaccordingtothedemandactualsize.
VNE-OSRfirsttriestoconstructaLAGaccordingtovirtuallinkbandwidthrequirementsofaVONandtheonline-servicebandwidth-conditionoffiberlinks.
IfaLAGisbuiltsuccessfully,weexecutethemappingofnodesandlinksonthisgraph;otherwise,weblockthedemand.
Table1showsthepseudo-codeofVNE-OSR.
Lines2-7expresstheprocessofconstructingaLAG,anddescribehowtotransportaVONdemandmappingfromsubstratenetworkstoacertainLAG.
Thealgorithmorderlycheckseachfiberwhetherrnavailablecontinuousspectrumslotsexit.
Ifthereexistsufficientspectrumslots,weinsertthefiberintotheLAGi,whereiisthestatingspectrumslotindex.
Whenallfibersarecheckedup,thealgorithmwillcheckinterconnectingelementsonLAGi,andformssomesub-graphs.
Andthenitsortsthesesub-graphsinthedescendingorderbasedonthenodenumber,where()subknodeGdenotesthenodenumberinsubkG.
rVdenotesthevirtualnodenumberinaembeddingrequestrV.
Lines8-11runthenodemappingandthelinkmapping.
Table1Pseudo-codeofVNE-OSRalgorithmVNE-OSRInput:SubstratenetworksG,aVONrequestrG;Output:NodemappingNM,linkmappingLM;1.
backupsGinstG;2.
for1i=to1rBn+do3.
restoresGtostG;4.
foreachconnectedcomponentinsGdo5.
subkG←selectaconnectedcomponentofsG;6.
removesubkGfromsG;7.
sort{,1.
.
.
1}subjGjk=basedon()subjnodeGindescendingorder;8.
for1j=to1kdo9.
applyNMLMalgorithmtoembedrGontosubjG;10.
markrGasblocked;11.
restoresGtostG;SimulationSimulationSetting.
WeadoptNSFNETasthetestingtopology.
Eachfiberlinkconsistsofapairofreversed-unidirectionalfibers.
Themaximumsub-carrierserialnumber(MSSN)occupiedinsubstratenetworksandthemeanblockingprobability(MBP)arethetestmerits.
MSSNiscalculatedbytheequation(1),wheresfisbinary,andifthesub-carrierisoccupied,1sf=;or,0sf=.
maxsMSSNsf=.
(1)ResultsandAnalysis.
Basedonthedifferentservicesequenceofdemands,wecombinetheproposedVONembeddingalgorithmwithfourdifferentorderingstrategies,thatis,firstfitbasedVNE-OSRalgorithm(VNE-FF),bandwidthfitbasedVNE-OSRalgorithm(VNE-BF),computingfitbasedVNE-OSRalgorithm(VNE-CF)andresourcefitbasedVNE-OSRalgorithm(VNE-RF).
Wedothissimulationfortwotargets:1)withefficientbandwidthresource,undertheconditionwherethesystemcanservealldemands,wecomparethefouralgorithmsbyMSSNsinfiberlinks;2)withlimitedbandwidthresource,wecompareMBPs.
Alldemandscanbeservedandthereareefficientcomputingandbandwidthresources.
Wesupposethereare300sub-carriersineachfiber,and300computingresourcecapacityineachphysicalnode(DC).
InFig2,inVONs,thebandwidthrequirementsofthevirtuallinksrangefrom2to4,andthedemandscoperangesfrom10to80.
Withtheincreasingdemands,theoccupiedMSSNsrise.
MSSNofVNE-FFisthehighestanditperformsworst.
Thus,forstaticdemands,thedemandservicesequencecaneffectMSSNs.
ComparedwithVNE-FF,otherthreealgorithmsperformbetter.
InFig.
3,wesupposethere50sub-carriersineachfiberand800computingresourcecapacityineachphysicalnode.
ForVONs,thebandwidthrequirementsofvirtuallinksrangefrom2to5andthedemandscoperangefrom20to200.
Whenthedemandsarelessthan60,allMBPsare0.
Withtheincreasingdemandscope,allMBPsrise.
That'sbecause,underthelimitedbandwidthresourceinfibers,thesmalldemandscopeleavesmorereminderbandwidthresource,whichcanservemoredemandsandreduceMBP,andwhereastheopposite.
Andthen,asshowninFig.
3,VNE-BFgainsthehighestMBPandperformsworst,inversely,VNE-CFperformsthebest.
That'sbecause,VNE-BFfollowsaserviceorderbasedonthebandwidthrequirementsequenceanditfirstlyservesthebiggestbandwidthrequirementdemand,leadingintothemoreoccupiedbandwidthresourceinfibers.
Thus,therestresourcecan'tserveallthesubsequentdemands.
Fig.
2ComparisonofMSSNswithdifferentdemandsamongVNE-FF,VNE-BF,VNE-CFandVNE-RFFig.
3ComparisonofMBPswithdifferentdemandsamongVNE-FF,VNE-BF,VNE-CFandVNE-RFConclusionItisvaluabletoresearchthevirtualopticalnetworkembeddingintodatacenternetworksorasingledatacenter.
ThispaperproposesaVNEalgorithmorientingspectrumresourcemaximumutilization.
Thesimulationresultstestifytheadvantageofouralgorithmintheresourceefficiency.
References[1]S.
Sakr,A.
Liu,D.
M.
Batista,etal.
"ASurveyofLargeScaleDataManagementApproachesinCloudEnvironments",IEEECommunicationsSurveys&Tutorials,2011,13(3):311-336.
[2]C.
Kachris,I.
Tomkos.
"ASurveyonOpticalInterconnectsforDataCentres",IEEECommunicationsSurveys&Tutorials,2012,14(4):1021-1036.
[3]M.
Jinno,H.
TakaraandB.
Kozicki.
"Conceptandenablingtechnologiesofspectrum-slicedelasticopticalpathnetwork(SLICE)",ACP,2009,pp.
1-2.
[4]M.
Jinno,H.
TakaraandB.
Kozicki.
"Spectrum-EfficientandScalableElasticOpticalPathNetwork:Architecture,Benefits,andEnablingTechnologies",IEEECommunicationsMagazine,2009,47(6):66-73.
[5]M.
Jinno,H.
TakaraandB.
Kozicki.
"Dynamicopticalmeshnetworks:drivers,challengesandsolutionsforthefuture",ECOC,2009,pp.
1-14.
[6]L.
Gong,Z.
Q.
Zhu.
"VirtualOpticalNetworkEmbedding(VONE)overElasticOpticalNetworks",JournalofLightwaveTechnology,2014,32(3):450-460.
[7]L.
K.
N.
Georgakilas,A.
Tzanakaki,M.
Anastasopoulos,etal.
"ConvergedOpticalNetworkandDataCenterVirtualInfrastructurePlanning",IEEE/OSAJournalofOpticalCommunicationsandNetworking,2012,4(9):681-691.