A service creation environment based on end to end composition of web services

The demand for quickly delivering new applications is increasingly becoming a business imperative today. Application development is often done in an ad hoc manner, without standard frameworks or libraries, thus resulting in poor reuse of software assets. W

AServiceCreationEnvironmentBasedonEndtoEnd

CompositionofWebServices

VikasAgarwal,KoustuvDasgupta,NeeranKarnik,ArunKumar

AshishKundu,SumitMittalandBiplavSrivastava

{avikas,kdasgupta,kneeran,kkarun,kashish,sumittal,sbiplav}@

IBMIndiaResearchLaboratoryBlock1,IITcampus,HauzKhas

NewDelhi110016,India

ABSTRACT

Thedemandforquicklydeliveringnewapplicationsisincreasinglybecomingabusinessimperativetoday.Applica-tiondevelopmentisoftendoneinanadhocmanner,withoutstandardframeworksorlibraries,thusresultinginpoorreuseofsoftwareassets.Webserviceshavereceivedmuchinterestinindustryduetotheirpotentialinfacilitatingseamlessbusiness-to-businessorenterpriseapplicationinte-gration.Awebservicescompositiontoolcanhelpautomatetheprocess,fromcreatingbusinessprocessfunctionality,todevelopingexecutablework ows,todeployingthemonanexecutionenvironment.However,we ndthatthemainapproachestakenthusfartostandardizeandcomposewebservicesarepiecemealandinsu cient.Thebusinessworldhasadopteda(distributed)programmingapproachinwhichwebserviceinstancesaredescribedusingWSDL,composedinto owswithalanguagelikeBPELandinvokedwiththeSOAPprotocol.AcademiahaspropoundedtheAIapproachofformallyrepresentingwebservicecapabilitiesinontologies,andreasoningabouttheircompositionusinggoal-orientedinferencingtechniquesfromplanning.Wepresentthe rstintegratedworkincomposingwebservicesendtoendfromspeci cationtodeploymentbysynergisticallycombiningthestrengthsoftheaboveapproaches.Wedescribeaprototypeservicecreationenvironmentalongwithause-casescenario,anddemonstratehowitcansigni cantlyspeedupthetime-to-marketfornewservices.

Keywords

SemanticWeb,Planning,WebServicesComposition

1.INTRODUCTION

Thedemandforquicklydeliveringnewapplicationsisincreasinglybecomingabusinessimperativetoday.Forexample,giventheintensecompetitioninthetelecomsector,mobiletelephonyserviceprovidersneedtocon-tinuallydevelopcompellingapplicationstoattractandretainend-users,withquicktime-to-market.Often,ifacompetitorintroducesanewservice,theserviceprovidermusto erasimilarorbetterservicewithindays/weeks,toavoidlosingcustomers.Also,aserviceprovidercanattractenterprisecustomersbyo eringcustom-developedvalue-addedservicesthatleverageitstelecomandITin-frastructure.Enterprisecustomerstypicallyo ersigni -cantlyhighermarginsthanconsumers,andarethusmoreattractive.Serviceprovidersthereforeneedtoolsandstandards-basedruntimeplatformstoquicklydevelopanddeployinterestingapplicationsfortheirclients.Thiswouldassistintheirtransitiontowards“ondemand”,responsivebusinesses.

Muchofthisservice/applicationdevelopmentiscurrentlydoneinanadhocmanner,withoutstandardframeworksorlibraries,thusresultinginpoorreuseofsoftwareassets.Whenanewserviceisneeded,thedesiredcapabilityisinformallyspeci ed.Anapplicationdevelopermustthencreatethiscapabilityusingcomponentservicesavailablein-houseorfromknownvendors.Thisprocessisessentiallymanual.Forexample,ifamobileserviceproviderwishestoo erataxi-requestservicetoitsusers,thedevelopermustpickathird-partytaxiservice(withanadvertisednetworkinterface)apartfromin-houseserviceslikelocation-tracking,accounting,etc.anddesignawork owthatdeliverstherequiredfunctionality.Thedynamicnatureoftheenvironmentimpactsthedevelopmentprocessaswell.Forexample,newtaxiservicesmaybecomeavailable,o eringbetterand/orcheaperservices;physicalchangesinthenetworkorenvironmentmaynecessitatearedesignofthe ow,etc.

Webserviceshavereceivedmuchinterestinindustryduetotheirpotentialinfacilitatingseamlessbusiness-to-businessorenterpriseapplicationintegration[22,27].

CategoriesandSubjectDescriptors

H.3.5[InformationSystems]:InformationStorageandRetrieval—Web-basedservices;I.2.8[Arti cialIntelli-gence]:ProblemSolving,ControlMethods,andSearch—Planexecution,formation,andgeneration;D.2.m[SoftwareEngineering]:Miscellaneous

GeneralTerms

Design,Languages

CopyrightisheldbytheInternationalWorldWideWebConferenceCommittee(IW3C2).Distributionofthesepapersislimitedtoclassroomuse,andpersonalusebyothers.

WWW2005,May10-14,2005,Chiba,Japan.ACM1-59593-046-9/05/0005.

The demand for quickly delivering new applications is increasingly becoming a business imperative today. Application development is often done in an ad hoc manner, without standard frameworks or libraries, thus resulting in poor reuse of software assets. W

Webserviceso erstandardizedinterfacedescription,dis-coveryandmessagingmechanisms.Also,theprogrammingtoolsandruntimeenvironmentsforwebserviceshavenowmatured.Acomponent-orientedsoftwaredevelopmentapproachwhereeachsoftwareiswrappedasawebser-vicewouldo ersubstantialbene tsinthemobileserviceprovider’sscenario.Mobileuserapplicationsoftenrelyonseveral,relativelysimplebuildingblocks–userpro lelook-ups,addressbooks,location-trackingservices,accountingandbillingservices,etc.Manyofthesebuildingblocksarealreadyinplace,buttheyarenoteasytoreuseandintegrateintonewapplicationsbecausetheyarenotbuiltusingstandardizedframeworksorcomponentmodels.Thisleadstohighdevelopmentcosts,andsubstantialtime-to-marketfornewservices.Thiscouldbealleviatedbybuildingapplicationsusingtheservice-orientedarchitecture(SOA)paradigm,usingwebservicesastheunderlyingabstraction.We ndthattwodi erentapproacheshavebeentakentostandardizeandcomposewebservices.ThebusinessworldhasadoptedadistributedsystemsapproachinwhichwebserviceinstancesaredescribedusingWSDL,composedinto owswithalanguagelikeBPEL1,andinvokedwiththeSOAPprotocol.AcademiahaspropoundedtheAIapproachofformallyrepresentingwebservicecapabilitiesinontologies,andreasoningabouttheirfunctionalcompositionusinggoal-orientedinferencingtechniquesfromplanning[16].Theseapproachesbythemselvesarepiecemeal,andinsu cient.Theformerhasfocusedontheexecutionaspectsofcompositewebservices,withoutmuchconsiderationforrequirementscaptureandthedevelopmentprocess.Thelatterapproachhasstressedonthefeasibilityofservicecompositionbasedonsemanticdescriptionsofserviceca-pabilities,butitsoutputcannotbedirectlyhandedo toaruntimeenvironmentfordeployment.

Inthispaper,wedemonstratehowwebservicescompo-sitioncanbeleveragedforbusinessprocessintegration,bysynergisticallycombiningthestrengthsofboththeaboveapproaches.Themaincontributionsare:

Tothebestofourknowledge,wepresentthe rstendtoendwebservicescompositionmethodologythat,givenaformallyspeci edrequirementforanewservice,stitchestogethersemantically-annotatedwebservicecomponentsinaBPEL owthatdeliverstherequiredfunction. Weproposeaprincipledtwo-stagewebservicescompo-sitionapproachleveragingthedi erentiationbetweenwebservicetypesandinstances.Thishelpsinhandlingdi erentgoals,di erentdata,di erentratesofchangeofdataateachstage,anddi erentmeanstooptimizethem.Itallowsustoachievescalability. Wedescribeanendtoendworkingprototype:(a)Ontologymatching,compositionattypelevelwithservicematchmaking(b)Compositionatphysicallevelwithinstanceselection(c)Deploymentontoadecentralizedwork oworchestrationinfrastructure.Therestofthispaperisorganizedasfollows.Inthenextsection,wedescribeabusinessprocessintegrationscenarioandmotivatetheroleofwebservicescomposition.Wethen

The demand for quickly delivering new applications is increasingly becoming a business imperative today. Application development is often done in an ad hoc manner, without standard frameworks or libraries, thus resulting in poor reuse of software assets. W

Problem Ticket,

Problem Ticket (Customer Location)

Figure2:Location-basedAgentSelectorService.tovisitthecustomerand xherwashingmachine.ThebottomhalfofFig.2showshowthiscanbedonebycreatinga owlinkingtogetherseveralcomponentservices,feedingthemtherightinputs,etc.Doingthismanuallytakestime(andthedeveloperhastoknowwhichcomponentsexist,andhowtoconnectthemup).Instead,weprovideatoolthatdiscoverstherelevantservicesfromamongsttheavailableones,andcreatesthecontrol owbetweenthem.Theavailableservicesaresemanticallyannotated,providingmeta-informationabouttheirfunctionalityinthecontextofadomainmodel.Thedeveloperonlyneedsto(formally)specifytherequirementsoftheservicetobecreated.Thetoolcanthengeneratea ow,andwithsomedeveloperinputs,deploythe owontoaruntimeinfrastructure.Thisshouldleadtoquickerservicecreation,andthusfastertime-to-marketfornewservices.

Further,thenewlycreatedLocation-basedAgentSelector(LAS)serviceitselfbecomesavailableasacomponent.Itcannowbereusedincreatingother ows,suchastheoneinFig.1.Eachnewservicethusenrichestheinfrastructureandmakesthedeveloper’staskeasierinfuture.WewillusetheLASserviceasarunningexampletoexplainthephasesinthecompositionprocess.OurservicecreationenvironmenthoweverincludesadomainmodelandontologyfortheentireHelplineAutomationscenario,andwedemonstratetheautomatedcompositionofthecomplete owofFig.1inSec.6.

Figure3:SystemOverview.

compositionprocessrightfromspeci cationofthebusinessprocess,throughcreationofdesiredfunctionalityusingplanningtechniques,throughgenerationofadeployablework owbyselectionandbindingofappropriateserviceinstances,to nallydeployingandrunningthecompositeservice.Thisintegratedsolutionachievesthebestofbothworldsandprovidesscalabilitytothecompositionprocess.Wehavebuiltaservicecreationenvironmentthatrealizesthisapproachintermsofthefollowingphasesofcomposition:

1.LogicalComposition:Thisphaseprovidesfunc-tionalcompositionofservicetypestocreatenewfunctionalitythatiscurrentlynotavailable.2.PhysicalComposition:Thisphaseenablestheselectionofcomponentserviceinstancesbasedonnon-functional(e.g.qualityofservice)requirements,thatwouldthenbeboundtogetherfordeployingthenewlycreatedcompositeservice.ThisbasicapproachtoautomatingtheprocessofservicecreationisillustratedinFig.3.AServiceRegistrycon-tainsinformationaboutservicesavailablein-houseaswellaswithparticipating3rd-partyproviders.Thecapabilitiesofeachavailableservicetypearedescribedformally,usingdomain-speci cterminologythatisde nedinaDomainOntology.Whenanewserviceneedstobecreated,thedeveloperprovidesaServiceSpeci cationtotheLogicalComposermodule.Drivenbythespeci edrequirements,theLogicalComposerusesgenerativeplanning-basedauto-matedreasoningtechniquestocreateacompositionoftheavailableservicetypes.Itsgoalistoexplorequalitativelydi erentchoicesandproduceanabstractwork ow,i.e.aplan(assumingafeasibleplanexists)thatmeetsthespeci edrequirements.

Inordertoturntheplanintoaconcretework owthatcanbedeployedandexecuted,speci cinstancesmustbechosenforthecomponentservicesintheplan.ThePhysicalComposerusesschedulingandcompilationtechniquesinselectingthebestwebserviceinstancestoproduceanexecutablework ow.Thefocusisnowonquantitativelyexploringtheavailablewebserviceinstancesforwork owexecution.Itqueriestheregistryfordeployedwebserviceinstances,toaccomplishthistask.

3.SYSTEMOVERVIEW

Ourservicecreationmethodology,basedonwebservicecompositiontechniques,consistsofthefollowingsteps:1.ServiceRepresentation:Representingtheavailableservicesandtheircapabilities.2.RequirementsSpeci cation:position:positeServiceRepresentation:Representingthenewcompositeserviceanditscapabilitiessothatitcanbeprogrammaticallydeployed,discoveredandinvoked.Inpreviousworkasimilarprocesshasbeenappliedatdi erentlevelsofabstraction,noneofwhichindividuallyyieldsapracticalsolution.OursystemtakesanendtoendviewthatsynergisticallycombinestheAIapproachandthedistributedprogrammingapproachcurrentlyadoptedbyacademiaandtheindustryrespectively.Itdrivesthe

The demand for quickly delivering new applications is increasingly becoming a business imperative today. Application development is often done in an ad hoc manner, without standard frameworks or libraries, thus resulting in poor reuse of software assets. W

Thework owgeneratedbytheservicecreationenviron-mentmustthenbedeployedontoaruntimeinfrastructure,andexecutedinane cientandscalablemanner.Thisisespeciallyimportantinenvironmentslikethatofamobileserviceprovider,wherethenumberofend-usersislikelytobeveryhigh.Thestateoftheartistoexecutethework owusingawork owenginesuchasWebSphereProcessChoreographer2,withdata owingbackandforthfromthisenginetothecomponentwebservices.OurExecutionEnvironmentinsteadorchestratesthework owinadecentralizedfashion,withpartitionsofthe owexecutingconcurrently,innetwork-proximitywiththecomponentservicestheyinvoke.These owpartitionsaregeneratedautomaticallybyaDecentralizertool,usingstaticanalysisoftheinputBPEL ow.Thecommunicationamongstthesepartitionsisdesignedtominimizenetworkusage,whileretainingtheoriginal owsemantics.This,inconjunctionwiththeaddedconcurrency,resultsinbetterscalabilityandperformance.FormoredetailsonourExecutionEnvironmentpleasereferto[5,18].Inthispaper,wewillfocusontheLogicalandPhysicalcompositionstages.

4.LOGICALCOMPOSITION

Figure4depictsoursystemforimplementingthefourstepsofcompositionduringLogicalComposition.AvailableservicetypesandtheircapabilitiesarerepresentedinaServiceCapabilitiesRegistry.AnOntologycapturesthedo-mainmodel.WeuseIBM’sSNOBASE3asthemanagementsystemforourontologyandtheservicecapabilitiesregistry.Speci cationofthedesiredserviceissuppliedtoaLogicalComposermodulethat rstgetsitveri edforsyntacticcorrectnessusingaValidatormodule.TheMatchmakermoduleallowsqueryingtheserviceregistryforavailableser-vices.Baseduponthevalidatedspeci cation,Planner4Jretrievesthesetofcandidateservicesusingthematchmaker.TheFiltermodulehelpsinpruningthesetofcandidateservicesbeforePlanner4Jusesplanningtechniquestocreatethecompositeservice.Wenextdiscusstheissuesthatariseineachstepoflogicalcomposition.

ServicesServices

Figure4:LogicalComposition.

4.1RepresentationofServiceTypes

Toenableautomaticdiscoveryandcompositionofdesiredfunctionality,weneedalanguagetodescribetheavailablewebservices.Thiscantakeplaceattwolevels–webservicetypesandwebserviceinstances.Atthelogicalcomposition

4

The demand for quickly delivering new applications is increasingly becoming a business imperative today. Application development is often done in an ad hoc manner, without standard frameworks or libraries, thus resulting in poor reuse of software assets. W

Figure5:LogicalPlanfortheLASservice.andareusedduringplanninginlogicalcomposition5.Theinputsandoutputsareexpressionsinvolvinggeneraldatatypes(e.g.integers,strings,algebraicexpressions)whichareusedduringinstanceselectionand owconcretizationinthephysicalcompositionphase.Itispossibletoincorporatenumericinputsandoutputsduringplanningaswell–thisapproachtoplanningiscalledmetricplanning[6].Exploringthefeasibilityofmetricplanningforendtoendwebservicecompositionwouldbeaninterestingareaforfuturework.FortheLAScompositeservice,theprecondition(ortheinitialstateofthecompositionproblem)assertsthatCustomerLocationisknownandthee ect(orthegoalstate)isto ndtheagent(AgentID)nearesttothecustomerlocation.

4.3CompositionthroughPlanning

AIPlanningdealswith ndingacourseofactionsthatcantakeanagentfromtheinitialstatetoagoalstate,givenasetofactions(legalstatetransformationfunctions)inthedomain.Formally,aplanningproblem[30]Pisa3-tuple I,G,A whereIisthecompletedescriptionoftheinitialstate,Gisthepartialdescriptionofthegoalstate,andAisthesetofexecutable(primitive)actions.AnactionsequenceS(aplan)isasolutiontoPifScanbeexecutedfromIandtheresultingstateoftheworldcontainsG.Aplanner ndsplansbyevaluatingactionsandsearchinginthespaceofpossibleworldstatesorthespaceofpartialplans.Logicalcompositionofwebservicescanbecastasaplanningproblembyusingthedescriptionofwebservicesasactions,andforminginitialandgoalstatesfromthespeci cationoftheservicetobebuiltalongwiththedomainmodel[16].

Forourservicecreationtool,planningforwebserviceshassomeuniquecharacteristics(refertoFig.4):

Thenatureofplanningislimitedcontingencyplanning.Thevalueofalllogicaltermsmaynotbeknownin

The demand for quickly delivering new applications is increasingly becoming a business imperative today. Application development is often done in an ad hoc manner, without standard frameworks or libraries, thus resulting in poor reuse of software assets. W

brancheswhicharecontingentonspeci cconditions(calledbranchcontext)beingmet,actionsarelabeledwiththeircontext.Thedefaultcontextforanunconditionalactionistrue,alwaysvalid.

Theplanistranslatedtothework owrepresentationofBPEL,alanguageforexpressinginteractionsandmessageexchangesbetweenpartnerentities.Itcanbeautomaticallyinterpretedandexecutedbyawork owengine.ABPELspeci cationcanbeabstractorexecutabledependingonwhetherbindinginformationhasbeenexcludedorincluded.WerenderthegeneratedplanasanabstractBPELwork owsincewebserviceinstanceinformationisnotknownatthisstageofthecompositionprocess.TheactionsintheplanaremappedtocorrespondinginvokeactivitiesinBPELandorganizedintobranchesbyinsertingappropriateswitchandcaseactivities.

5.PHYSICALCOMPOSITION

Figure6:PhysicalComposition.

Inthisphase,theabstractwork owforthecompositeserviceisfedtothePhysicalComposer,whichbindseachserviceinthework owtoaconcreteserviceinstance.Theprocessofmatchingeachservicetypetoacorrespondinginstance,andthenorchestratingbetweenthesetofinstancesintheresultingwork owisanon-trivialmatchmakingproblem.Theproblemhasbeenaddressedextensivelyforwebservicesandinvolvesanumberofissuesrelatedtodata oworchestration,datatypeandinvocationprotocolmatching,QoSmatchingandSLAcomposition.Whilesomeoftheseissuescanberesolvedinanautomatedmanner,othersmightrequiremanualinterventionfromadevelopersupervisingthecompositionprocess.WenextdescribeeachofthestepsinvolvedinthePhysicalCompositionstage,illustratedinFig.6.

5.1RepresentationofServiceInstancesand

Requirements

AsinLogicalComposition,werequirearepresentationforserviceinstancesandcompositionrequirementstofacilitateComposition.Ithasbeenestablishedthatdirectoryservices,suchasUDDI,areimportantbutinsu cientforthispurpose[9]andneedtobecomplementedwithmatchmakingfacilitieslikesymmetryofinformationexchangebetweenservicesandtheirconsumers,theabilityofeachpartytodescriberequirementsfromtheotherparty,arichlanguagetodescribeservices’andconsumerdemands,andamethodologytochoosee cientlyamongcompetingserviceinstances.

Tothisend,weusetheWebServicesMatchmakingEngine(WSME)[9]–anenginecapableofmatchingcomplex

entities,andaDataDictionarytoolforde ningthelanguageforthematchingprocess.Matchingisperformedbetweenserviceinstancesandrequirementsspeci edbytheconsumer.Inourcase,therequirementscomefromtheabstractwork owandadditionalmatchingcriteriaspeci edbythedeveloperperformingtheservicecomposition.

TheengineisdeployedasaWebservicethatreceivesqueriesandadvertisementsfromthetwopartiesinvolvedinmatchmaking.Eachpartyessentiallysubmitsadescriptionofitselfandthedemandsfromtheotherparty.TheAdver-tisementissubmittedbytheprovidertoWSMEandislonglived,remaininginWSMEuntilitisexplicitlywithdrawnbytheprovideroruntiltheapplicationserverisstopped.Theadvertisementcontainsthefollowinginformation:(1)MyType-thisspeci estheadvertisementrecord-type.(2)YourType-thisspeci estherecord-typeexpectedtobesubmittedbytheconsumerquery.(3)Properties-alistofthepropertiesde nedasMyType.Someofthosepropertiesmaybede nedasdynamicpropertiesbytheproviderevaluatedatruntime.(4)Rules(optional)-whattheproviderrequiresfromtheconsumer.

AQuerysubmissionissentfromtheconsumertoWSMEandistransient,terminatingafterinitiatingthematch-makingprocessandbringingittoitsconclusion.Thequerycontainsthefollowinginformation:(1)MyType-thisspeci esthequeryrecord-type.(2)YourType-thisspeci estheprovider’sadvertisementrecord-typethatthequeryislookingfor.(3)Properties-alistofthepropertiesde nedasMyType.(4)Rules-whattheconsumerrequiresfromtheprovider.Thedescriptionsanddemandscanbedynamicallycreated,deletedandmodi edintheformofpropertiesandrulesrespectively,usingaDataDictionarytool.

TheWSMErulesallowbothsidestoselecttheotherpartytheywishtodealwithbyspecifyingtheireligibility.AruleisaWSMEscriptthatisevaluatedatmatchmakingtime,resultinginaBooleanvalue.Arulecanrefertothepropertiesofthetwopartieswhoseadvertisementandqueryareinvolvedinthematchmakingprocess.Exampleofapos-sibleconsumerruleisthefollowing:return(my.MaxCost≤your.cost).Aproblemarisesifarulereferstoapropertythatwasnotsupplied.Toavoidsuchasituation,theWSMETypesystemde nesthemandatorylistofpropertiesthatasubmissionmustprovide;thedatadictionarycontainsthosede nitions.

WeillustratetheDataDictionaryde nitionsusedforcomposingtheHelplineServicediscussedinFig.8.EachserviceinstanceneedstoadvertiseitselftotheWSMEserviceinstanceregistryusingtheadvertisementde nition.Eachadvertisementrecordcontainsbasicinformationliketheservicename,servicetype,methodnameandWSDLinformation,alongwithQoS-speci cmetricslike(expected)responsetime,throughputandcostofinvokingtheparticu-larinstance.Eachqueryrecordspeci esthemethodnameandservicetypethatneedstobeboundtoaninstancealongwithadditionalrulesthatarespeci edbythedevelopersupervisingthecompositionprocess.

5.2MatchmakingandInstanceSelection

Weemployatwo-stepapproachwhere,inthe rststep,weusetheWSMEMatchmakertoselectoneormorecan-didateinstancesthatmatchtherequirementspeci cations.Sinceaspeci cservicetypecanbematchedwithmorethanoneinstance,wenextadoptaheuristic-basedapproachto

The demand for quickly delivering new applications is increasingly becoming a business imperative today. Application development is often done in an ad hoc manner, without standard frameworks or libraries, thus resulting in poor reuse of software assets. W

Data Dictionary Tool

Service Provider

Service Consumer

Query Rules

Definition of advertisementDefinition of query

submission record type

submission record type

Figure7:AdvertisementandQueryFormatsforHelplineService.

selectoneamongthecon ictinginstances,tocompletethematchmakingprocess.

TheWSMEmatchmakingprocessisatwo-wayorsym-metricprocess-itbringstogethermatchingadvertisementsandqueriesbyapplyingtherulesofeachpartytothedescriptionoftheother,thusallowingbothpartiesto‘select’eachother.Amatchingadvertisementiscalledano er.Ifmorethanoneo erisavailable,theyarecollectedtogether.Zero,oneormorematchingo ersaresenttotheconsumer.Forfurtherdetailsonthematchmakingprocess,theinterestedreaderisreferredto[9].

Next,wedealwiththoseservicetypesthathavemorethanonematchingo ers(instances)fromWSME.Whiletherearemanywaystoperformtheinstanceselection,wechosetoemployagreedyheuristicapproachtosolvetheproblem.Inparticular,theInstanceSelector ndsinstancebindingassignmentsthatoptimizecertainqualityofservicemetrics.Forthecurrentprototype,wefocusoncommonlyusedQoSmetricslikecost,responsetimeandthroughput.Weassumethatvaluesofthesemetrics(asadvertisedinWSME)arestatisticallyguaranteed.Ifaservicetypehasmultiplematchinginstances,wechooseaninstancebasedontheoptimizationcriteriaspeci edbythedevelopersupervisingthecompositionprocess.

5.3BPELGenerationforCompositeService

Nowthateachservicetypeintheabstract owisboundtoaninstance,theBPELgeneratorproducesa(concrete)BPELwork owthatcanbedeployedontoaruntimeinfrastructure,torealizethecompositeservice.

We rstgeneratetheWSDLdescriptionforthecompositeservice.Itprovidesthenameandinterfaceofthecompositeserviceanddescribestheporttypesforstitchingtogetherthecomponentservices.OncetheWSDLhasbeengener-ated,partnerlinktypesarede ned,linkingthecomponentservices.ThenextstepisthegenerationoftheBPEL ponentsareinvokedinthemannerdescribedbytheabstractwork ow.Thecompositeserviceacceptsinputsfromtheuserthatisfedtothe rstcomponentserviceandsendsanoutputfromthelastcomponentservicebacktotheuser.Weintroducevariablesthatcapturetheoutputofoneserviceandprovideitasinputtothenext.

Speci c

Figure8:BPELcodefortheLASservice.

detailsforeachcomponentserviceareobtainedusingtheWSDLdescriptionforthecorrespondinginstance,presentintheWSMEserviceinstanceregistry.

ThoughBPELandWSDLareXML-basedstandards,wedonotmanipulateXMLdirectly.WeuseanEclipseMod-elingFramework(EMF)modelofBPEL(WSDL)thatisautomaticallycreatedfromaBPEL(WSDL)schema6.Themodelprovidesin-memoryrepresentationofconstructsandsupportforpersistenceto les(serialization)andloadingfrom les(de-serialization).BPELandWSDLmanipulationbecomesigni cantlysimpli edwiththecorrespondingEMFmodels.

NotethattheBPELgeneratedmightnotbereadilydeployableonawork owengine.Thisisduetothefactthatthecodeformessagingbetweencomponentservicesneedstohandleissueslike(input/output)typematchingandtransformation,mismatchininvocationprotocolsthatareused(synchronousvsasynchronous),orderingofparametersetc.WhiletheBPELwork owactsasthetemplateforthecompositeservice,itneedstobeexaminedandpossiblymodi edbythedevelopertoensurethatthedata owbetweencomponentservicesishandledproperly.Inthecurrentprototype,thisisdonebyallowingthedevelopertoedittheBPELwork owbeforeitisactuallydeployed.Wealsomaketheobservationthatthehandlingofsomeofthesematchingproblemscouldbedelegatedtothematchmakingengine(WSME),andweplantoinvestigatethisapproachinthefuture.

Figure8illustratespartoftheBPELcodegeneratedbythePhysicalComposerfortheLASservice.ItiscomposedofthefourcomponentservicesdescribedinSec.2.Further,oncephysicalcompositionisdone,theWSDLdescriptionofthisnewserviceisaddedtotheWSMEinstanceregistry,andcanbelaterusedinthecompositionofsomeotherservice.

The demand for quickly delivering new applications is increasingly becoming a business imperative today. Application development is often done in an ad hoc manner, without standard frameworks or libraries, thus resulting in poor reuse of software assets. W

(a)(b)

Figure9:(a)SpecifyinginputintheCompositionTool.(b)LogicalPlanforHelplineService.

POSINGTHEHELPLINESERVICE

WenowdiscusshowourservicecreationtoolcanbeusedforcomposingtheHelplineservicedescribedinSec.2.Re-callthattheHelplineserviceconsistsofmultiplecomponentsservicesliketheLASservice,MessageDeliveryandCallSetupservices.Bywayofrunningexample,weshowedhowanexecutablework owiscreatedfortheLASserviceusingthetool.Theusercanaddthecompositeservicetotheserviceregistrysothatitisavailableforreuse.

FordevelopingtheHelplineservice,theusermaychoosetousethetooltoexplorebasicservicesavailable,buildap-propriatecompositeservices,and nallybuildtheHelplineservice.Alternatively,theusercouldaskthetooltobuildtheHelplineserviceattheoutsetusingtheavailableservices,andletthetoolsearchthroughthesetofpossibleplans.Weexpecttheusertoprefertheformerapproach,whenthescenarioislargeandtheuserwantstocontrolthecomposition.

Wehaveapproximately100termsintheontologyand25servicetypes.AssumethatthepreviouslycreatedcompositeLASservicehasbeenaddedtotheregistry.NowthetoolisinvokedfortheoverallHelplineservicewithapreconditionofProblemHTMLForm,andthee ectofProblemResolutionStatusasshowninFig.9(a).TheLogicalComposerproducestheplanshowninFig.9(b).NotethattheLASserviceisreused.TheplancontainingLASserviceisselectedoveralternativeplanswithoutit,becausetheplan’sheuristiccostisless7.Finally,thePhysi-calComposertakestheabstractwork owandgeneratestheappropriateBPEL(similartoFig.9).

7.DISCUSSIONANDRELATEDWORK

Theliteratureonwebservicecompositionisextensive,

The demand for quickly delivering new applications is increasingly becoming a business imperative today. Application development is often done in an ad hoc manner, without standard frameworks or libraries, thus resulting in poor reuse of software assets. W

7.2LogicalComposition

Theliteratureoncomposingservicesbasedonannotations(semanticallyorganizedinontologiesorotherwise)hastakentwopaths.Onedirectionisdisambiguatingsimilarannotationsusingdomainmeta-data,rules,etc.Theotherdirectionisonmethodstocombineserviceswhoseannotationsmatchbasedonsomenotionofsimilarity.

In[20],matchingofwebservicesfromadirectoryisformalizedbasedonvariousinexactnessmeasures.In[12],theauthorshaveidenti edtheinformationthataSemanticWebServicemustexposeinordertoful lltheobjectiveofautomateddiscovery,composition,invocationandinteroperation.

SWORD[21]wasoneoftheinitialattemptstouseplanningtocomposewebservices.Itdoesnotmodelservicecapabilitiesinanontologybutusesrulechainingtocomposeswebservices.

Sirinetal.[24]usecontextualinformationto ndmatchingservicesateachstepofservicecomposition.Theyfurther lterthesetofmatchingservicesbyusingontologicalreasoningonthesemanticdescriptionoftheservicesaswellasbyusinguserinput.TheyattempttoovercomelackofsupportforservicetypesinOWL-SbycreatingaclasshierarchyofServicePro les.Anewsub-classiscreatedforeachvalueofanIOPEparameter.

Therearethreeproblemswiththeirapproach.First,alargesetofvaluesforanattributeofaservicewouldresultingenerationofthatmanyclasses.Second,torepresentafunctionalitywithmultipleattributesahugenumberofservices,oneeachforasetofpossiblevaluesofallattributes,wouldhavetoberepresentedasderivedclasses.Third,newclassesneedtobeaddedtotheontologyeverytimeanewtypeofserviceisintroduced.Acleanerapproachthatseparatesrepresentationoftheservicede nitionsfromserviceinstanceshasalreadybeendescribedinSec.4.

WebServicesModelingOntology8(WSMO)isarecente ortformodelingsemanticwebservicesinamarkuplanguage(WSML)andalsode ningawebserviceexecutionenvironment(WSMX)forit.Ourlogicalcompositionapproachisnotspeci ctoanyparticularmodelinglanguageandcanadapttonewerlanguages.

7.3PhysicalComposition

SeveralstandardizationproposalsaimedatprovidinginfrastructuresupporttoWebservicecompositionhavere-centlyemergedincludingSOAP,WSDL,UDDI,andBPEL.TherehasalsobeenalotofinterestintheareaofdynamicWebserviceandQoS-basedwork owmanagement.Pre-viouse ortsinthisarealikeeFlow[4]haveinvestigateddynamicserviceselectionbasedonuserrequirements.Zengetal.[31]proposethatthechoiceofcomponentservicesintheplanbemadeatruntimeforoptimality.Insteadofmakinglocalchoicesateachstepofthecomposition,thefocusisonoptimizationatacompositelevelbasedonagenericQoSmodel(basedonprice,duration,reliabilityetc.)andestablishedlinearprogrammingtechniques.OtherproposalssuchasMETEOR[3]andCrossFlow[10]haveconsideredQoSmodelsforwork owsalongfourdimensionsnamelytime,cost,reliabiltyand delity.Finally,therehasbeenaconsiderablee ortintheWebservicecommunityinidentifyingthechallengesinwork oworchestrationbetween

The demand for quickly delivering new applications is increasingly becoming a business imperative today. Application development is often done in an ad hoc manner, without standard frameworks or libraries, thus resulting in poor reuse of software assets. W

GoodwinandRamaAkkirajufortheircontributionstothedevelopmentoftheideaspresentedinthispaper.

10.

[1]

[2]

[3][4]

[5]

[6]

[7]

[8]

[9]

[10]

[11]

[12]

[13]

[14]

[15]M.MatskinandE.Tyugu.StrategiesofStructural

putingandInformatics,20:1–25,2001.

[16]S.McIlraith,T.C.Son,andH.Zeng.SemanticWebREFERENCES

Services.IEEEIntelligentSystems,SpecialIssueonV.Agarwal,N.Karnik,andA.Kumar.An

theSemanticWeb.,16(2):46–53,March/rmationModelforMeteringandAccounting.In

[17]CommonInformationModel(CIM)MetricsModel,ProceedingsoftheIEEE/IFIPNetworkOperations

Version2.7.DistributedManagementTaskForce,andManagementSymposium,April2004.

/standards/documents/CIM/J.Blytheetal.TheRoleofPlanninginGrid

DSP0141.pdf,puting.Proc.ofIntlConferenceonAIPlanning

[18]M.G.NandaandN.Karnik.SynchronizationandScheduling,2003.

AnalysisforDecentralizingCompositeWebServices.J.Cardoso.QualityofServiceandSemantic

InProceedingsoftheACMSymposiumonAppliedCompositionofWork ows.PhDthesis,Universityof

Computing,March2003.Georgia,2002.

[19]OWLServicesCoalition.OWL-S:SemanticMarkupF.Casati,S.Ilnicki,L.J.Jin,V.Krishnamoorthy,

forWebServices./services/andM.C.Shan.eFlow:APlatformforDeveloping

owl-s/1.0/owl-s.html,Nov.2003.andManagingCompositee-Services..Technical

[20]M.Paolucci,T.Kawmura,T.Payne,andK.Sycara.ReportHPL-2000-36,HPLaboratories,2000.

SemanticMatchingofWebServicesCapabilities.InG.B.Cha e,S.Chandra,V.Mann,andM.G.

FirstInt.SemanticWebConf.,2002.Nanda.DecentralizedOrchestrationofComposite

[21]S.PonnekantiandA.Fox.SWORD:ADeveloperWebServices.InProceedingsofthe13thInternational

ToolkitforWebServiceComposition.InProc.oftheWorldWideWebConference,2004.

11thInternationalWorldWideWebConference,2002.M.B.DoandS.Kambhampati.Sapa:AScalable

[22]S.Staabetal.Webservices:Beenthere,donethat?Multi-objectiveHeuristicMetricTemporalPlanner.

IEEEIntelligentSystems,pages72–85,Jan-Feb2003.JournalofAIResearch,20:155–194,2003.

[23]E.SirinandB.Parsia.PlanningforSemanticWebP.Doshi,R.Goodwin,R.Akkiraju,andS.Roeder.

Services.InSemanticWebServicesWorkshopat3rdParameterizedSemanticMatchmakingforWork ow

InternationalSemanticWebConference,position.TechnicalReportRC23133.Availableat

http://dali.ai.uic.edu/pdoshi/research/RC23133.html,[24]E.Sirin,B.Parsia,position-drivenFilteringandSelectionof

SemanticWebServices.InAAAISpringSymposiumK.Erol,J.Hendler,andD.S.Nau.HTNplanning:

onSemanticWebServices,plexityandexpressivity.InProceedingsofthe

TwelfthNationalConferenceonArti cialIntelligence,[25]B.Srivastava.ASoftwareFrameworkforBuildingSeattle,Washington,USA,1994.Planners.InProc.KnowledgeBasedComputer

Systems(KBCS2004),2004.C.Facciorusso,S.Field,R.Hauser,Y.Ho ner,

R.Humbel,R.Pawlitzek,W.Rjaibi,andC.Siminitz.[26]B.Srivastava,S.Kambhampati,andM.B.Do.AWebServicesMatchmakingEngineforWebPlanningtheprojectmanagementway:E cientServices.InProceedingsof4thIntl.Conf.onplanningbye ectiveintegrationofcausalande-CommerceandWebTechnologies,September2003.resourcereasoninginRealPlan.Artif.Intell.,

131(1-2):73–134,2001.J.Klingemann.Controlled exibilityinwork ow

management.InProc.ofthe12thInternational[27]B.SrivastavaandJ.Koehler.WebServiceConferenceonAdvancedInformationSystemsComposition-CurrentSolutionsandOpenProblems.(CAiSE),June2000.ICAPS2003WorkshoponPlanningforWebServices,

2003.A.Kumar,B.Srivastava,rmation

modelingforendtoendcompositionofwebservices.[28]X.SuandJ.Rao.ASurveyofAutomatedWebInIBMResearchReportRI05001.AvailablefromServiceCompositionMethods.InProceedingsofFirst/library/CyberDig.nsf/Home.,InternationalWorkshoponSemanticWebServices2005.andWebProcessComposition,ra,usen,S.Arroyo,J.deBruijn,and[29]P.TraversoandM.Pistore.AutomatedCompositionD.Fensel.SemanticWebServices:DescriptionofSemanticWebServicesintoExecutableProcesses.RequirementsandCurrentTechnologies.InIn3rdInt.SemanticWebConf.,November2004.InternationalWorkshoponElectronicCommerce,[30]D.S.Weld.RecentAdvancesinAIPlanning.AIAgents,andSemanticWebServices,September2003.Magazine,20(2):93–123,1999.D.J.MandellandS.A.McIlraith.Adapting[31]L.Zeng,B.Benatallah,M.Dumas,J.Kalagnanam,BPEL4WSfortheSemanticWeb:TheBottom-UpandQ.Z.Sheng.QualityDrivenWebServicesApproachtoWebServiceInteroperation.InComposition.InProceedingsofthe12thIntl.WorldProceedingsofInternationalSemanticWebWideWebConf.,May2003.Conference,Oct2003.

M.MatskinandJ.Rao.ValueAddedWebServicesCompositionusingAutomaticProgramSynthesis.InProceedingsofInternationalWorkshoponWebServices,E-business,andtheSemanticWeb,2002.

本文来源：https://www.bwwdw.com/article/3jsm.html