communicatep2pover

ConsistencyofStatesofManagementDatainP2P-BasedAutonomicNetworkManagementJefersonCamposNobreandLisandroZambenedettiGranvilleInstituteofInformatics–FederalUniversityofRioGrandedoSulAv.
BentoGoncalves,9500-PortoAlegre,RS,Brazil{jcnobre,granville}@inf.
ufrgs.
brAbstract.
Autonomicnetworkmanagementisavisionthatbringsautonomiccomputingprinciplestonetworkmanagement.
Besides,itisnecessarysomelevelofdecentralizationtoenablebroadautonomiccapabilities.
Aninterestingalter-nativeofinfrastructureforthisunionistheutilizationofpeer-to-peer(P2P)over-lays.
However,theconsistencyofstatesofmanagementdataamongpeersisanimportantchallenge.
Traditionalmechanismstomaintainconsistencyofthesestatesaresupportedbysomecentralizationwhichwastessomedesirableproper-tiesofP2Pfacilities.
Incontrasttothesemechanisms,weproposeadistributed,scalableandrobustmechanismtomaintaintheconsistencyofstatesofmanage-mentdata.
Inthispaperweintroducemulti-agenttruthmaintenancefeaturesinP2P-basedautonomicnetworkmanagementasamechanismtobringconsistencymaintenanceofthesestates.
WedevelopedamodelofaP2P-basedautonomicnetworkmanagementsystemonPeersimtoperformsimulationexperiments.
Be-sides,theutilizationofP2P-basedautonomicnetworkmanagementsystemsinaccessnetworksispresentedasacasestudy.
1IntroductionTheincreasingcomplexityofcomputernetworksrequiressophisticatedsolutionstomanagetheunderlyingcommunicationinfrastructureandhelpnetworkhumanadmin-istratorsintheirdailytasks[1].
Theapplicationofautonomiccomputing(AC)princi-plesinnetworkmanagement,normallyrefereedasAutonomicNetworkManagement(ANM),hasbeenproposedasawaytoaddresssomedemandsfacedbytraditionalnet-workmanagement,suchascontrollinghighlydynamicenvironmentslikead-hocandpeer-to-peer(P2P)networks[2].
ANMsystemsincreasetheefciencyofnetworkhu-manadministratorsbydecreasingthenumberofmanualinterventions.
ThisefciencyimprovementisdonebyANMsystemsthroughautomationand/oroptimizationofsomeoperationaldetailsofmanagementtasks,suchasfaulthandlingandperformanceman-agement.
ANMsystemscanbedeployedusingdifferentmodelsofmanagementdistribution,fromtotallycentralizedmodelsuptohighlydecentralizedones.
Today,therearenoclearevidencesthatcouldlinkthequalityofautonomicactionsofanANMsystemwiththedistributionmodeladopted.
Someauthors,however,claimthatsomelevelofdecentralizationisrequiredtoachieveamoreadequateANM[3].
Inthisscenario,thetypicaldesignofdecentralizedANMsystemsisbasedonasetofAutonomicManage-mentElements(AME)thatexecutemanagementtasksandinteractwithoneanothertoC.
BartoliniandL.
P.
Gaspary(Eds.
):DSOM2009,LNCS5841,pp.
99–110,2009.
cIFIPInternationalFederationforInformationProcessing2009100J.
C.
NobreandL.
Z.
GranvilleformanAutonomicManagementDomain(AMD).
MultiplesAMDscouldbeintegratedtoformanAutonomicManagementEnvironment.
DifferenttechnologiescanbeemployedtoprovideaninfrastructurefordecentralizedANMsystems.
Aninterestingpossibilityisusingpeer-to-peer(P2P)overlays,whichincorporatecharacteristicsofP2PnetworksintoANMsystems,suchasthesupportforcollaborativemanagement,robustnessinconnectingmanagemententities,andloadbalancingofmanagementtasks[4].
TherearesomeinitiativesinvestigatingP2P-basedANM[5][6]and,intheseinitiatives,peershavesomepropertiesfoundinAMEs.
Be-sides,peerwithsimilarproperties(e.
g.
,managingthesamedevices)canbeorganizedintogroups.
ThesepeergroupshavesomepropertiesfoundinAMDs.
Amanagedelement(e.
g.
,anetworkrouter)canbecontrolledbyoneormultiplepeersinP2P-basedANMsystem,forexample,forthepurposeofmanagementrobust-ness.
Inthiscase,thestatusofthemanagedelementbecomesadistributed,replicatedinformationamongthecontrollingpeers.
InconsistenciesinthisinformationcouldleadtoanimproperoperationoftheANMsystem.
Forexample,thestatusofarouter'slink(e.
g.
,ethernetinterface)canbeadvertiseddifferentlybythecontrollingpeerstoexter-nalrequesters(e.
g.
,managementstation).
Thisinconsistencycanbecausedbyfaultsinnetwork(e.
g.
,lossesinmessagesexchangedamongpeers)andcomputationalre-sources(e.
g.
,incorrectoperationofpeers).
Besides,inconsistencyinmanagementdatacanoccurevenintheregularoperationofP2P-basedANMsystem,duetolackofinfor-mationsynchronizationamongpeers.
Thisinformationsynchronizationisachallengeinunrealibleasynchrounousdistributedsystems,suchasanunstructuredP2Poverlay.
MechanismstosupportconsistencyofmanagementinformationincurrentP2P-basedANMsystemsare,ingeneral,stillsupportedbysomecentralizationthrough,forinstance,externalrepositories[7]orsuperpeers[5][6],missingopportunitiesofunstructuredP2Pfacilities.
Thiscentralizationcomplicatestheachievementofgoodscalabilityandrobustnessfeatures,thus,inspiteofbeingP2P-basedANMsystems,theseexamplesshowundesirablecharacteristicsofclient-serversystems.
Inthispaper,weintroduceMulti-AgentTruthMaintenancefeatures[8]intheP2P-basedANMtoimprovetheconsistencyofstatesofmanagementdata.
TheintroductionofthesefeaturesmaintainsdesirablescalabilityandrobustnesscharacteristicsofP2P-basedANM.
Itisalsoproposedacommunicationstrategyformessageexchangeamongpeerstosupportthistruthmaintenance.
Thisstrategyusesbiology-inspiredprocesses(e.
g.
,replication),whichhavewellknowscalabilityandrobustnessfeatures[9].
Theremainderofthispaperisorganizedasfollows.
Section2discussesrelatedworks.
Section3describesourproposalanditsimplicitconcepts.
Section4showsacasestudy.
Section5showsevaluationsofourproposal.
ConclusionsandfutureworkarenallyprovidedinSection6.
2BackgroundInthissectionwerstdiscussaboutthecurrentstate-of-the-artonP2P-basedauto-nomicnetworkmanagement.
Afterwards,themainconceptsbehindtruthmaintenancesystemsarepresented.
Finally,someimportantservicesrelatedtothemaintenanceofconsistencyofsharedinformationindistributedapplicationsarediscussed.
ConsistencyofStatesofManagementDatainP2P-BasedANM1012.
1P2P-BasedAutonomicNetworkManagementTheutilizationofP2PoverlaysisidentiedasanemergingapproachforAutonomicNetworkManagement(ANM).
Thiscanbespeciallyobservedintheincreasingnumberofresearchworktowardsthisutilization[10].
PBMAN[5]mergetraditionalPBNMwithP2PoverlaystoautonomicallymanageAmbientNetworks(AN).
PBMANenablesscalablemechanismsfornetworkcomposi-tioninsidetheAN,aswellaspoliciesdistributionandretrieval.
Throughthisapproachitispossibletoestablishpoliciestomanagedevicesorservices.
PBMANisstructuredusingsuperpeers,inahierarchicalarchitecture.
Thesesuperpeersareresponsibleforconsistencyofstatesofmanagementdata,amongotherfunctions,ineachhierarchicallevel.
TheMadeiraplatform[6]isanapproachtoANMthatusestheconceptofAdaptiveManagementComponents(AMC),whicharecontainersthatrunonmanagedelements.
AMCscancommunicatewithotherAMCsrunningonothermanagedelementsthroughP2Pcommunicationservices.
AMCsformmanagementclusterswithsuperpeersactingasclusterheads.
Thesesuperpeersareresponsibleforconsistencyofstatesofmanage-mentdata,amongotherfunctions.
ManP2P[11]isaP2P-basednetworkmanagementsystemthatisevolvingtoanau-tonomicconceptionthroughtheimplementationofautonomicmodulesinpeers[7].
ManP2PispartiallyinspiredbytheManagementbyDelegation(MbD)modelandbasedonaservice-orientedapproach.
Thereisnointernalmechanismforconsistencyofstateofmanagementdata,thus,authorsproposetheutilizationofexternalrepositories.
DespitemanyimprovementsbroughtbytheutilizationofP2P-basedANMsystems,therearestillissuestobeaddressed.
Theconsistencyofstateofmanagementdataamongthepeersisusuallyaddressedwithsomecentralization,usingsuperpeers[5][6]orexternalrepositories[7],whichmissesopportunitiesofP2Poverlays(e.
g,robustness).
2.
2TruthMaintenanceSystemsTruth-MaintenanceSystems(TMS)wereproposedtokeeptheintegrityofKnowledgeBases(KB).
Theoriginofthesesystemswasproposedinthe1970s,forresolutionsinmono-agentsystems[12].
ATMSkeepstrackoflogicalstructureofthesetofbeliefsofagents.
Abeliefisamemberofthecurrentsetofbeliefsifithasvalidreasons.
TMSsprovideconsiderablepowerusingfewcomputationalresources[13].
Thus,althoughnotbeingwellknownoutsidearticialintelligencecommunity,TMSsareusedindifferentcontexts,suchaspolicysystems[13]andnetworkmanagementsystems[14].
TMSshavebeenextendedforMulti-AgentSystem(MAS)versions,DistributedTruth-MaintenanceSystems(DTMS)[8].
InaMAS,agentsmustbeabletomaintaintheintegrityoftheirKBs,despitemessageexchangewithotheragents.
Thismaintenaceofintegritycanbedonebyamulti-agentTMS.
Inananalogousway,duringtheoperationofP2P-basedANMsystem,peersmustbeabletomaintaintheintegrityofstatesofmanagementdata,despitemessageexchangewithotherpeers.
Thissimilarityindicatestheuseofmulti-agentTMSinP2P-basedANMasaninterestingpossibility[14].
102J.
C.
NobreandL.
Z.
Granville2.
3ServicesforConsistencyofSharedInformationinDistributedSystemsServicesforconsistencyofsharedinformationcanbeusedasabasicbuildingblockfordistributedapplications.
AP2P-basedANMsystem,asadistributedapplication,canappropriatesomegoodcharacteristicsfoundinthesesystems.
ZooKeeper[15]isacoordinationservicefordistributedapplications.
ItexposesasimpleAPIthatdistributedapplicationscanbebuiltupontoimplementhigherlevelservicesforsynchronization,datadiffusion,andpublish-subscribeschemes.
ZooKeeperusesdistributedserverdatabasesforreadoperations,however,writeoperationsusea"leader"server(i.
e.
,centralizeddatabase)toassuretheconsistencyofthedatabase.
Astrolabe[16]isadistributedinformationmanagementservice.
Itworkslocatingandcollectingthestatusofasetofserversandreportingsummariesofthisinformation.
AstrolabeisimplementedusingaP2Poverlay,whereeverypeerrunanAstrolabeagent(i.
e.
,inaMASfashion).
However,Astrolabewasdevelopedprimarilyusingsimpledatamodels.
Besides,itsoperationisaimedatread-orientedapplications.
ScalableDistributedInformationManagementSystem(SDIMS)[17]isaservicetoaggregateinformationaboutlarge-scalenetworksystems.
TheserviceisbuiltusingideasfromAstrolabe[16]andDistributedHashTables(DHT).
However,asinmostDHTapproaches,consistencyandreplicationissuesareaknownchallenge.
Thepresentedeffortsshowinterestingcharacteristicsforconsistencyofsharedinfor-mationindistributedsystems.
However,theseeffortshavevulnerabilitieswhichmakethemnotappropriateforP2P-basedANMsystems,suchascentralization[15],simpledatamodels[16],andreplicationissues[17].
TheconsistencyofstatesofmanagementdataisstillanissuetobeadressedinP2P-basedANMsystems.
ItisnecessarytomaintainthisconsistencykeepingscalabilityandrobustnessfeaturesofP2Poverlays.
ThemaintenanceofintegrityofKBsamongagentsbroughtbymulti-agentTMSseemstobeavaliddirectiontointroducethiscon-sistencymaintenance[14].
Besides,itisinterestingtobringcharacteristicsofservicesforconsistencyofsharedinformation.
3ProposalInP2P-basedAutonomicNetworkManagement(ANM)systems,peersmustsharemanagementdata.
Inthiswork,managementdatumisdenedasamanagementin-formationdescribedinadenedform(i.
e.
,usingaspeciclanguage).
Besides,inthesesystems,managementdatamustallowtheiruseinautomationand/oroptimizationpro-cedures.
Itisalsoexpectedthatsourcesofmanagementdata(e.
g.
,highlydynamicen-vironments)imposechallengestotheANMsystem.
Despitetheserequirements,itisnecessarytoavoidpotencialinconsistenciesinstateofmanagementdataamongpeers.
Ourproposalisaimedatmeetingrequirementsofconsistencyofstatesofmanage-mentdatainaP2P-basedANMsystem.
Theproposedmechanismintroducesmulti-agenttruthmaintenance[8]featuresthroughaconsistencymaintenancemodulethatrunsineachpeer.
Asfarasweareawareof,theonlystudythatincorporatesmulti-agenttruthmaintenancefeaturesinanetworkmanagementsystemwascarriedoutbyNobreandGranville[14],andonlysomeresultswerepublishedsofar.
ConsistencyofStatesofManagementDatainP2P-BasedANM103Multi-agenttruthmaintenanceisamulti-agentextensiontoTruth-MaintenanceSys-tems(TMS).
ThegoalofTMSutilizationistokeeptheintegrityofknowledgebases.
Inthiswork,aknowledgebaseisagroupofintegratedmanagementdata.
Inamulti-agentTMS,therearemultipleagentsandeachonehasitsownTMS.
TMSskeepintegrityperformingbeliefrevisionandexchangeinasetofbeliefs.
Abeliefisamemberofthecurrentsetofbeliefsifithasvalidreasons.
Theexchangeofbeliefsaboutmanagementdataisdoneasynchronouslyandwedonotconsiderthemessageexchangetobereliable.
Unfortunately,itiswellknownthattheutilizationofasynchronousunreliabledistributedsystemsimposeschallengestoachieveconsistencyinshareddata.
Thus,theconsistencymodelusedisnon-deterministic,inotherwords,itusesa"weak"notionofconsistency.
Thismodelisadoptedforscalability,robustness,andupdatedisseminationissues.
GivenabeliefXthatdependsonsomeotherbeliefY,whenanupdateismadetoY,itiseventuallyreectedinX.
Someauthorscallthisnotionas"eventualconsistency"[16].
3.
1JusticationsforManagementDataMulti-agentTMSisakindofjustication-basedTMS.
Inajustication-basedTMS,adatumisbelievedwhenithasvalidjustications(i.
e.
,validreasons).
Thisbeliefcanbesharedbydifferentpeers(whichplaytheagentrole)inaP2P-basedANMsystem(whichaggregatessomeMAScharacteristics).
Thus,justicationsimprovethealign-mentofpeerswithsystem-wideobjectives(i.
e.
,objectivesoftheP2P-basedANMsys-tem).
ThedatumanditslistofpossiblejusticationsmustbeprovidedbynetworkhumanoperatorsorexpertsystemsfortheP2P-basedANMsystem.
Theassociatedstatesofadatumare"in"(believed)or"out"(disbelieved),accordingtoitsjustications.
ThesejusticationscanbegeneratedbyprocessesinsidethepeerorreceivedthroughtheP2Poverlaycommunicationservices.
Thus,the"in"statecanassumetwoadditionalstates:"internal",wherethedatumhasonlyvalidinternaljusti-cations,and"external",wherethedatumhassomevalidexternaljustication(providedbyotherpeer).
Adatumislabeled"out"whenitlacks,atleast,oneofitsassociatedjustications[8].
Forinstance,theactivation(belief)ofaQoSpolicy(datum)canbejustiedbyanet-workhumanadministratorcommand(justication)andanasynchronoussignalfromamanageddevice(justication).
Thecodebelowshowsapossibleinternalrepresentationofthisdatumanditsjustications.
Intheexample,eachdatumorjustication("just"inthecode)isrepresentedwith2elds,"name"and"desc"(description).
Thelastlineofthecodedenesthatifthejustications"admcmd"and"asyncsig"arepresent,thedatum"qospol"isbelieved.
datum:nameqos_poldescQoSpolicyjust:nameadm_cmddescnetworkhumanadministratorcommandjust:nameasync_sigdescasynchronoussignalfromamanageddevicetms:qos_pol(adm_cmdasync_sig)104J.
C.
NobreandL.
Z.
Granville3.
2ArchitectureofPeersPeersarecommonlyviewedascomposedbyoneormoremanagementservicemod-ules.
Managementservicemodulesperformregularmanagementtasks(e.
g.
,collectingstatistics)ineachpeer,and,inthesetasks,modulesproducemanagementdata,buildingmanagementknowledgebases.
Weintroducetheconsistencymaintenancemoduletoregistratethesetofbeliefaboutmanagementdataineachpeer.
Thismoduleworksassociatingmanagementdataandtheirrespectivejustications.
Whenthereisabeliefchange(i.
e.
,justiticationchange),theconsistencymaintenancemoduleusestheP2Poverlaycommunicationservicestospreadthechange.
Figure1showstherelationbetweentheconsistencymaintenancemodule,managementservicemodules,andP2Poverlaycommunicationservices.
Themanagementservicemodulesshouldinformconsistencymaintenancemoduleabouttheirinternalbeliefsofmanagementdata.
Themanagementservicemodulesarealsoresponsibleforqueryingandrequiringservicesfromtheconsistencymaintenancemodule,possiblyupdatingtheirinternalbeliefs.
Thesebeliefsareexchangedthroughjusticationusingasimpleinterfacethatmanagementservicemodulesandconsistencymaintenancemodulemustuse.
Whenthepresenceofajusticationismodied,theconsistencymaintenancemod-uleperformsthefollowingsteps:unlabelsmanagementdatum,includes(orremoves)thepresenceofjusticationandlabelsdatumagainaccordingtonewrestrictions.
TheP2Poverlaycommunicationservicesareusedtospreadchanges,whichcanchangebeliefsofotherpeers.
Itisimportanttostressthatthereisonlyoneconsistencymaintenancemodulein-sideapeer,thus,itisnotspecicofamanagementservicemodule.
Therefore,everymanagementservicemoduleinapeerinteractswiththesameconsistencymaintenancemodule.
Thisfactcanbeexploredfortheintegrationofdifferentmanagementservices.
Fig.
1.
PeerarchitectureConsistencyofStatesofManagementDatainP2P-BasedANM105Forinstance,apolicyprocessingmodule,afaulthandlingmodule,andacongurationmanagementmodule(possiblyusingdifferentlanguagesforrepresentingmanagementdata)couldbeintegratedbytheconsistencymaintenancemodulethroughjustications.
3.
3CommunicationwithinaPeerGroupTheconsistencymaintenancemodulehandlesthemessageexchangethroughP2Pover-laycommunicationservices.
Inthisprocess,requestsareadaptedinmessagestobespreadamongpeersandvice-versa.
TheP2Poverlayismodelledasanunstructuredoverlaynetworks,thusthereisnorelationbetweentheinformationstoredatapeeranditspositionintheoverlaytopology.
Weusethepremisethatthereissupportforgrouporganization(i.
e.
,peergroups)throughmanagementservicesmodules.
Thus,peersthathaveaspecicmanagementservicemoduleareorganizedintoagroup(withouthumanintervention)andpeerscanparticipateofseveralgroupsaccordinglytomodulesthattheyhave.
Themethodsusedformessageexchangeinsidethepeergrouparemodeledusingconceptsfrombiology-inspireddistributedcomputingmodels[9].
Amongthesemod-els,proliferation-basedonesareaninterestingchoiceforcommunicationrequirementsofourproposal.
Allpeersinthepeergrouprunexactlythesamecommunicationalgo-rithm,whichcanbeinitiatedfromanypeerinthepeergroup.
Wehavechosenreplicationastheinitialproliferationmechanisminthepeergroup.
Thismechanismcansupportanumberofdifferentstrategies[9].
Inourproposal,peersspreadmessagestoreplicatechangesinjusticationsamongtheparticipatingentities(i.
e.
,peersofaspecicpeergroup).
Thisunbridledreplicationisrestrictedtopeergroup,fulllingthecriterionofrobustnessandcontrollingthenumberofmessageswithintheP2Poverlay.
4CaseStudyThecasestudypresentedisanillustrationofthecollaborativefaultmanagementoflinksinaccessnetworksthroughfailurenoticationsentbydevicesandhumanknowl-edgeaboutthesenotications.
Theintegrationoftheseinformation(failurenoticationinadditiontohumanknowledge)producesamanagementdatum,whichcanassumedifferentstates.
ThisdatumcanbeusedagainstaService-LevelAgreements(SLAs)tosupportorclarifyservicelevelclaims.
Amongaccessnetworktechnologiesinmetropolitannetworks,Ethernetisoneofmostinterestingandpromisingchoice,thus,wechooseittobuildourcasestudy.
Inthiscontext,anaccessnetworklinkisanEthernetVirtualConnection(EVC)[18].
FaultmanagementinthislinkisdonethroughAlarmIndicationSignal(AIS)messages[18].
Thesemessagesaretriggeredwhenafailurebetweentwonodesoccurs.
Thus,AISmessagesprovideasynchronousnoticationtootherelementsinthenetworkthatthereisafaultintheEthernetnetwork.
Theeffortstomanagelayer2EthernetservicemustconsideranoverlayedIPinfrastructure[19].
AmanagementservicemodulecollectsAISmessagesandanothermodulecollectsinformationfromhumanadministrators.
Theconsistencymaintenancemoduleis106J.
C.
NobreandL.
Z.
Granvilleresponsibletointegratetheinformationfrombothmanagementservicemodulesandmaintaintheconsistencyofthestateofmanagementdatuminthepeergroup.
Thelinkfailuredetected(datum)isjustiedbyhumanadministratorcommandsfrombothdomains,networkoperatordetectionandnetworkconsumerdetection,andade-vicenoticationreceived(AISmessage).
Thecodebelowshowstherepresentationofthisdatumanditsjustications.
Thesejusticationsareprovidedbymanagementser-vicemodulesandkeptinsidethepeergroupthatoffersthismanagementservice.
datum:namelink_fail_detdesclinkfailuredetectionjust:namenet_opt_detdescnetworkoperatordetectionjust:namenet_con_detdescnetworkconsumerdetectionjust:nameais_msgdescaismessagetms:link_fail_det(net_opt_detnet_con_detais_msg)Initially,thedatumhasthe"out"state,sincethejusticationsarenotpresent.
Thisstatecanbechangedafterthevericationofnewbeliefs(i.
e.
,justicationchanges)byconsistencymaintenancemodule.
Thesebeliefsareinformedbymanagementservicemodulesorreceivedasamessageinthepeergroup.
Ifeveryjusticationofthedatumispresent,thestatechangesto"in".
This"in"stateassumestwooptionsaccordingthesourcesofjustications:"internal"(everyjusticationwasgeneratedinternally)and"external"(somevalidexternaljustication).
Forinstance,humanadministratorcommands(networkoperatordetectionandnet-workconsumerdetection)canbegeneratedinternallyandthepresenceofanAISmes-sagecanbereceivedasajusticationchangemessage.
Inthisexample,the"in"stateassumestheoption"external".
Thecodebelowshowstheanswerfromtheconsistencymaintenancemoduleinthissituation.
tms:link_fail_det:external(net_opt_det:modnet_con_det:modais_msg:msg)Thismanagementoperationistraditionallyperformedthroughseparatecentralizedsys-tems,anetworkmanagementsystem(collectingnoticationsfromdevices),andatrou-bleticketsystem(collectinginformationfromhumanadministrators).
Thetraditionalprocedurebringsconcernsinscalabilityandrobustnessandimposesdifcultiesintheintegrationoftheinformation.
Besides,justicationscanbeusedtoprovideexplana-tionstotheuser[13],improvingtheunderstandingofmanagementdata.
5EvaluationTheevaluationofourproposalcanbeperformedindifferentways.
Toenableafullycontrolledenvironmentfortheevaluation,wechosetodevelopsomesimulationexper-iments.
Intheseexperiments,wepresentsimulationresultsthatsupportourscalabilityandrobustnessclaims.
Scalabilityandrobustnessaresomeofthemostimportantmo-tivationsforusingdecentralizationintheinfrastructureofdifferentsystems[20][9],suchasP2Pfacilities.
ConsistencyofStatesofManagementDatainP2P-BasedANM107Aspreviouslystated,weexpectthattheintroductionofmulti-agentTMSfeatureskeepsdecentralizationpropertiesofaP2P-basedANMsystem,maintainingeachpeerasanindependentandself-sustainableentity.
Asmanysystemshavedemonstrated,asystemthatdoesnotshareresourcescanscalealmostinnitelysimplybyaddingconstitutiveelements(e.
g,peersinaP2P-basedANMsystem).
Besides,maintainingtheindependenceofeachpeer,singlepointsoffailureareeliminated.
ThesimulationexperimentswereimplementedinJavausingPeerSim[21],anopensourceevent-basedsimulatorofP2Psystems.
Thepreliminarysystemversionusedhastheabilitytosimulatefailuresinpeersandmessageexchange,andtheoverlayisbuiltrandomly.
Theexperimentsuseasimplemodeloftransportlayerthatcanemulatesomecharacteristics,suchaslossanddelayprobabilities.
Allpeersinpeergroupsrunexactlythesamealgorithm.
Intheexperiments,wevariedthenumberofpeersofthepeergroupfrom4to14(wedonotexpectlargepeergroupsinP2P-basedANMsystems).
Besides,thesepeergroupsizesseemreasonableforthecasestudyprovidedinSection4.
Inaddition,apeerischosenrandomlyastheprimarysourceofchangestonotaffectmeasurementsandmessagedelayiscontrolled.
Eachexperimentwasconductedatleast10times.
Intheexperiments,thevarianceobservedwaslow.
Intherstexperiment,itismeasuredthenumberofmessagesexchangedtospreadjusticationchangesinthepeergroup.
Thisnumbermustbeconsideredasanimportantcostofthepeergroupoperation,thus,itisimportantforscalabilityanalysis.
Besides,weconsiderthenumberoftransmittedmessagesasindicativeofnetworkload.
Inthisexperimenttherewerenofaultsinpeersorinmessageexchange.
WeshowtheresultsinFigure2.
051015202530468101214#messages#peers1justification2justificationsFig.
2.
Messageexchangeduetojusticationchanges108J.
C.
NobreandL.
Z.
GranvilleOurproposalshowsacceptablescalabilitycharacteristicsonnumberofexchangedmessages,sincethisoperation(messageexchange)isrestrictedtoeachpeergroup.
Theexperimentshowsthatoursystembehaveslikeweexpected,withoutstabilityandconvergenceproblems.
Networkloadgrowslinearlywiththenumberofparticipatingpeers,thuswecaninferthebehaviortrendofpeergroupswithlargernumberofpartic-ipatingpeers.
Ofcourse,anefcientoperationoflargepeergroupsneedsmodicationsincommunicationstrategies,suchastheutilizationofgossip-basedprotocols.
Inthesecondexperiment,wedeterminedtheinuenceofmessagelossonthedis-seminationofajusticationchange.
Inthisexperiment,wevariedthemessagelossprobabilitywithfollowingvalues:25%,50%,and75%(respectively,0.
25,0.
5,and0.
75asindicatedinFigure3).
Usingourcasestudy,wewouldprobablyobservesuchmes-sageloss(specially75%)duetofaultyoroverloadednetworkequipment(e.
g.
,ethernetinterfaces)and/ornetworklinks(e.
g.
,ethernetlines).
Sinceourcasestudyisaimedatfaultmanagement(consideringanoverlayedIPinfrastructure),oursystemmustbehaveacceptablyeveninbadnetworkconditions.
InFigure3,weshowtheaveragepercentageofcoherent(andcorrect)peersaftermessageexchangetocease.
Theexperimentshowstheinuenceofmessagelossinthereplicationprocess.
AscanbeseenfromtheresultsinFigure3,highlossprobabilitiesdoleadtolessconsis-tencyinpeergroup,but,evenwithafewparticipatingpeers,thepercentageofcoherentpeersissubstantial.
Besides,moreparticipatingpeersinpeergroupdecreasetheinu-enceoflossprobability.
Theresultsshowsomefault-tolerancefeatures,sincethepeergroupoperationisnothighlysensitivetopeercrashesandmessagelosses.
Butanincreaseinnumberof2030405060708090100110468101214%coherentpeers#peersloss=0.
25loss=0.
50loss=0.
75Fig.
3.
CoherentpeersafterajusticationchangeConsistencyofStatesofManagementDatainP2P-BasedANM109peersalsoleadstoanincreaseinthenumberofexchangedmessages,sotherobustnessadvantagescomeatsomecost.
However,messagesareexchangedonlywithinthepeergroup,andahighnumberofpeersinapeergroupisnotexpected.
Thus,thenumberofexchangedmessagesdoesnotimposeaissueinscalability.
6ConclusionsandFutureWorkThesupportofnewdemandsfacedbytraditionalnetworkmanagementisakeyresearchissueinnetworkmanagementarea.
Oneofthesedemandsisthesupportforcollabo-rativemanagementtasksoveraccessnetworklinks.
Inordertoenableefcientlythesetasks,itisnecessarysomedecentralizationandautomationfeatures.
ThesefeaturescanbesupportedbyP2P-basedautonomicnetworkmanagement(ANM)systems.
However,theconsistencyofstatesofmanagementdataimposeschallengesforthesesystems.
Inthispaperwehaveintroducedmulti-agentTMSfeaturestoimprovethemainte-nanceofconsistencyofstatesofmanagementdatainP2P-basedANMsystems.
Ourproposalaimsattheintegrationofdatausedbytheentitiesthatformthesesystems(i.
e.
,peers),throughtheutilizationofjustications.
Wehavealsopresentedevaluationsofthisproposalthroughsimulationexperiments.
Inaddition,wehavedescribedacasestudyoffaultmanagementinaccessnetworkstoshowthepossibilitiesofourproposal.
Althoughtheproposalshowsgoodresultsinevaluationsperformeduntilthepresentmoment,itisnecessarytoevaluatemorecomplicatedcases,innumberofpeersandpeergroups,andintheparticipationofapeerindifferentpeergroups.
Wearealsolookingatadditionalsettingsthatcouldleadtoimportanteffects,suchasnetworkpartitions.
Thus,wearecurrentlypursuingnewexperimentswithPeerSim.
References1.
Kind,A.
,Dimitropoulos,X.
,Denazis,S.
,Claise,B.
:Advancednetworkmonitoringbringslifetotheawarenessplane.
IEEECommunicationsMagazine46(10),140–146(2008)2.
Pras,A.
,Schoenwaelder,J.
,Burgess,M.
,Festor,O.
,Perez,G.
M.
,Stadler,R.
,Stiller,B.
:Keyresearchchallengesinnetworkmanagement.
IEEEcommunicationsmagazine45,104–110(2007)3.
Jennings,B.
,VanDerMeer,S.
,Balasubramaniam,S.
,Botvich,D.
,Foghlu,M.
O.
,Donnelly,W.
,Strassner,J.
:Towardsautonomicmanagementofcommunicationsnetworks.
IEEECom-municationsMagazine45(10),112–121(2007)4.
Granville,L.
Z.
,daRosa,D.
M.
,Panisson,A.
,Melchiors,C.
,Almeida,M.
J.
B.
,Tarouco,L.
M.
R.
:Managingcomputernetworksusingpeer-to-peertechnologies.
IEEECommunica-tionsMagazine43(10),62–68(2005)5.
Kamienski,C.
,Fidalgo,J.
,Sadok,D.
,Lima,J.
,Pereira,L.
,Ohlman,B.
:PBMAN:APolicy-basedManagementFrameworkforAmbientNetworks.
In:IEEEWorkshoponPoliciesforDistributedSystemsandNetworks,POLICY2006,pp.
79–83(2006)6.
Fallon,L.
,Parker,D.
,Zach,M.
,Leitner,M.
,Collins,S.
:Self-formingNetworkManagementTopologiesintheMadeiraManagementSystem.
In:Bandara,A.
K.
,Burgess,M.
(eds.
)AIMS2007.
LNCS,vol.
4543,pp.
61–72.
Springer,Heidelberg(2007)7.
Marquezan,C.
C.
,dosSantos,C.
R.
P.
,Nobre,J.
C.
,Almeida,M.
J.
B.
,Tarouco,L.
M.
R.
,Granville,L.
Z.
:Self-managedservicesoverap2p-basednetworkmanagementoverlay.
In:Proceedings.
2ndLatinAmericanAutonomicComputingSymposium,LAACS2007(2007)110J.
C.
NobreandL.
Z.
Granville8.
Huhns,M.
N.
,Bridgeland,D.
M.
:Multiagenttruthmaintenance.
IEEETransactionsonSys-tems,ManandCybernetics21(6),1437–1445(1991)9.
Babaoglu,O.
,Canright,G.
,Deutsch,A.
,DiCaro,G.
A.
,Ducatelle,F.
,Gambardella,L.
M.
,Ganguly,N.
,Jelasity,M.
,Montemanni,R.
,Montresor,A.
,etal.
:Designpatternsfrombi-ologyfordistributedcomputing.
ACMTransactionsonAutonomousandAdaptiveSystems(TAAS)1(1),26–66(2006)10.
EMANICS:Activities-WorkPackage9-AutonomicManagement(2009),http://emanics.
org/content/view/63/109/11.
Panisson,A.
,daRosa,D.
M.
,Melchiors,C.
,Granville,L.
Z.
,Maria,L.
:DesigningtheAr-chitectureofP2P-BasedNetworkManagementSystems.
In:ISCC2006:Proceedingsofthe11thIEEESymposiumonComputersandCommunications,pp.
69–75.
IEEEComputerSo-ciety,LosAlamitos(2006)12.
Doyle,J.
:Atruthmaintenancesystem.
Computation&intelligence:collectedreadings,529–554(1979)13.
Kagal,L.
,Hanson,C.
,Weitzner,D.
:UsingDependencyTrackingtoProvideExplanationsforPolicyManagement.
In:IEEEWorkshoponPoliciesforDistributedSystemsandNetworks,POLICY2008,pp.
54–61(2008)14.
Nobre,J.
C.
,Granville,L.
Z.
:Towardsconsistencyofpolicystatesindecentralizedautonomicnetworkmanagement.
In:IEEEWorkshoponPoliciesforDistributedSystemsandNetworks,POLICY2009(toappear,2009)15.
Hunt,P.
:ZooKeeper:ADistributedCoordinationServiceforDistributedApplications(2008),http://wiki.
apache.
org/hadoop/ZooKeeper16.
VanRenesse,R.
,Birman,K.
,Vogels,W.
:Astrolabe:Arobustandscalabletechnologyfordistributedsystemmonitoring,management,anddatamining.
ACMTransactionsonCom-puterSystems(TOCS)21(2),164–206(2003)17.
Yalagandula,P.
,Dahlin,M.
:Ascalabledistributedinformationmanagementsystem.
ACMSIGCOMMComputerCommunicationReview34(4),379–390(2004)18.
McFarland,M.
,Salam,S.
,Checker,R.
:Ethernetoam:keyenablerforcarrierclassmetroethernetservices.
IEEECommunicationsMagazine43(11),152–157(2005)19.
Ryoo,J.
,Song,J.
,Park,J.
,Joo,B.
S.
:Oamanditsperformancemonitoringmechanismsforcarrierethernettransportnetworks.
IEEECommunicationsMagazine46(3),97–103(2008)20.
Mccann,J.
A.
,Huebscher,M.
C.
:EvaluationIssuesinAutonomicComputing.
In:Jin,H.
,Pan,Y.
,Xiao,N.
,Sun,J.
(eds.
)GCC2004.
LNCS,vol.
3252,pp.
597–608.
Springer,Heidelberg(2004)21.
Jelasity,M.
,Montresor,A.
,Jesi,G.
,Voulgaris,S.
:ThePeersimSimulator(2008).
http://peersim.
sf.
net