Tools for Composite Web Services: A Short Overview

10 Pages·0.262 MB·English

by  Hull R., Su J.

Tools for Composite Web Services: A Short Overview(cid:3)(cid:3)
(cid:3)
Richard Hull Jianwen Su
BellLabsResearch DepartmentofComputerScience
LucentTechnologies UCSantaBarbara
[email protected] [email protected]
ABSTRACT describingemergingframeworksforstudyingcompositeser-
vices,andidentifyingemergingtoolsandtechniquesforboth
Webservicestechnologiesenable(cid:13)exibleanddynamicinter-
automated design and analysis of composite web services.
operationofautonomoussoftwareandinformationsystems.
The focus of this short survey is on the foundations of
A central challenge is the development of modeling tech-
composition and related issues. From this perspective, it is
niquesandtoolsforenablingthe(semi-)automaticcomposi-
worthwhile to understand the overall process of designing
tionandanalysisoftheseservices,takingintoaccounttheir
composite services. Fig. 1 shows the key elements in the
semanticandbehavioralproperties. Thispaperpresentsan
overview of the fundamental assumptions and concepts un-
Goals
derlying current work on service composition, and provides
asamplingofkeyresultsinthearea. Italsoprovidesabrief
tour of several composition models including semantic web
Activities
services, the \Roman" model, and the Mealy/conversation
model.
1. INTRODUCTION Discovery, Orchestration, Information Sharing
Self-description Monitoring (Messaging)
Theweb services paradigm promises to enablerich, (cid:13)ex-
ible, and dynamic interoperation of highly distributed and
Figure 1: Anatomy of Web Service Composition
heterogeneousweb-hostedservices. Substantialprogresshas
already been made towards this goal (e.g., emerging stan-
typical design process. It is not surprising that the design
dards such as SOAP, WSDL, BPEL) and industrial tech-
processissimilartoplanningstudiedinarti(cid:12)cialintelligence
nology (e.g., IBM’s WebSphere Toolkit, Sun’s Open Net
(e.g., [42]). In this case, activities that may contribute to
Environment and JiniTM Network technology, Microsoft’s
accomplish the overall goal need be discovered and orches-
.Net and Novell’s One Net initiatives, HP’s e-speak, BEA’s
trated. Theorchestration hastobeconstrainedbythe(cid:13)ow
WebLogic Integration). Several research e(cid:11)orts are already
of available information, which is generally agreed to be in
underwaythatbuildonor takeadvantageof the paradigm,
theformofmessages. Notethatmonitoringiscloselyrelated
including the DAML-S/OWL-S program [14, 36, 24], and
toorchestration;speci(cid:12)cmodelsofthelattermaydetermine
automata-basedmodelsforwebservices[8,26,7]. Butthere
howandtosomedegreewhattomonitor. Interestingly,the
isstillalongwaytogo,especiallygiventheostensiblelong-
design process embodies elements from all four task groups
termgoalofenablingtheautomated discovery, composition,
listed above.
enactment, and monitoring of collections of web services
Fig. 2 illustrates three dimensions that \measure" Web
working to achieve a speci(cid:12)ed objective. A fundamental
service description/composition models. The component
challenge concerning the design and analysis of composite
service complexity dimension indicates the information ca-
webservicesistodevelopnecessarytechniquesandtools to
pacityof thelanguagesandmodelinrepresentingaservice.
handle the novelaspects of the web services paradigm.
In this dimension, OWL-S is low since it describes only the
The challenge raises a variety of questions, several of
input and output. On the other hand, WSDL captures
which are relevant for the database research community.
richer structural information with XML Schema and uses
Some of the questions are: What is the right way to model
messagesforinputandoutputs,andothermodelsincluding
webservicesandtheircompositions? Whatistherightway
BPEL, CSP [29] and (cid:25)-Calculus [34], the Mealy model [8],
to query them in order to support automated composition
the Roman model [7], and BPML also model service states
and analysis algorithms? And how can the data manage-
and message or activity sequences. Clearly, the ability of
mentaspectsofcompositewebservicesbeincorporatedinto
assembling individual services together in this \cluster" of
currentwebservicesstandards? Thispaperattemptstopro-
models make them high along the glue language complexity
vide the groundwork needed to address these questions, by
dimension. The third dimension is the ability to describe
(cid:3)
Supportedin partbyNSFgrant IIS-0101134. \semantics". OWL-Scan describe the properties on the in-
(cid:3)(cid:3)
DatabasePrinciples Column. put and output of an operation, and also specify how the
Column editor: Leonid Libkin, Department of Computer Science, Uni› serviceinteractswithanabstractmodelofthe\realworld",
versity of Toronto, Toronto, Ontario M5S 3H5, Canada. E›mail: which distinguishes it from the other models.
[email protected].
OWL-S will most likely be guided by already established standards
andpractices. Someofthecurrentstandardsareillustrated
by the layered structure shown in Figure 3. Brie(cid:13)y, web
CTR-S services interactbypassing XMLdata, withtypesspeci(cid:12)ed
Commitment using XML Schema. SOAP can be used as the communi-
protocols p-Calc
Mealy BPML cation protocol, and the i/o signatures for web services are
“semantics” Complexity of glue language RomanWSCL CSP BPEL sBwgeiePvrbveEnicLseeb,rsyvWciWacSenCsSbDDteoLLd,.eeaBA(cid:12)cPlnhleMoodftLhut,ehsDrien.sAgeBMhceaiLhgn-ahSveb,ireoerltdeacvel.e(cid:12)dlnesestdcarnbipdetafiorordnessbsouinfcdhwineabgs
Complexity of component services WSDL ChorDeiosgcroavpehryy UDDI WS-Choreorgaphy
Composition BPEL4WS OWL-S ServiceModel
Figure 2: Anatomy of Web Service Composition (Individual) WSCL
Service
Thispaperisorganizedasfollows. Section2givesashort Description WSDL OWL-S ServiceProfice
overview of several standards related to Web services and
XML
composition. Section 3 is intended to provide some high SOAP
Messaging
levelofkeyaspectswebservicecompositionandadiscussion
Network HTTP, SMTP, FTP, etc.
on models thathavebeen studied in the context. Section4
brie(cid:13)ysummarizestechniquesandapproachesforanalyzing
Figure 3: Web Service Standards Stack
Webservices. Section5 concludes the paper.
In Figure 3, XML messaging and Network layers pro-
2. WEBSERVICES ANDSTANDARDS vide the foundation for interoperations or interactions be-
tween services. The starting point for service descriptions
Agoodstartingpointforunderstandingthewebservices
is the W3Cs Web Service Description Language (WSDL)
paradigm is to consider the stated goals, as found in the
[50]. WSDL describes a Web service via its set of visible
literature and the standards communities. The basic moti-
operations. They can be thought of as message endpoints,
vation of standards such as SOAPand WSDL is to allow a
or the set of messages that it can send and receive. WSDL
high degree of (cid:13)exibility in combining web services to cre-
speci(cid:12)es on the one hand \reactive" operations in which a
ate more complex ones, often in a dynamic fashion. The
message is received by the service. If the reactive opera-
current dream behind UDDI is to enable both manual and
tion is declared as \one-way", then it does not return a re-
automated discovery of web services, and to facilitate the
sponse, otherwise it is a \request-response" operation, and
construction of composite web services. Building on these,
the return type is also declared. It also describes proactive
theBPELstandardprovidesthebasisformanuallyspecify-
operations that send out messages from the service. \Noti-
ingcompositewebservicesusingaprocedurallanguagethat
(cid:12)cation" operations send out messages without waiting for
coordinates the activities of other webservices.
a response, while \solicit-response" operations block wait-
Much more ambitious goals are espoused by the OWL-
ing for a response, with the response type being speci(cid:12)ed
S coalition [14] and more broadly the semantic web ser-
with the operation. The receive and response types of the
vices community (e.g., [15]). These goals are to provide
operationsaremappedontoconcreteXMLSchematypesto
machine-readable descriptions of web services, which will
be used in messages. WSDL can thus be seen as an exten-
enable automated discovery,negotiation with, composition,
sion of traditional input/outputsignatures inprogramming
enactment,andmonitoringofwebservices. OWL-Sisanon-
languages and distributed computing to a peer-to-peer set-
tologylanguagefordescribingwebservices,intermsoftheir
ting. A service is viewed both as a server (via its reactive
inputs,outputs,preconditionsand(possiblyconditional)ef-
operations) a client (via its proactive operations).
fects, andoftheirprocessmodel. Importantly,OWL-Spro-
WSDL models services that are essentially stateless.
vides a formal mechanism for modeling the notion of the
(WSDL2.0incorporatesalimitednotionofstate,sinceiten-
state of the "real world", and describing how atomic web
ablesspeci(cid:12)cationofcertainmessagepatternsthataservice
services impact that state over time. (This is described in
should satisfy.) While it is clear that the complete internal
more detail in Subsection 3.1 below.) OWL-Salso provides
state of a running service is not expected to be known, it
a grounding, which provides mechanisms for mapping an
is also commonly accepted that some executionstates need
OWL-S speci(cid:12)cation into a WSDL speci(cid:12)cation (e.g., see
be visible either due to the fact that theyare observable or
[40]).
thenecessity(e.g.,tointeractcorrectlywithotherservices).
Akindofmiddlegroundisalsoemerging,whichprovides
TheWebServiceConversation Language (WSCL)[49] pro-
abstract \signatures" of web services that are richer than
posalisaninterestingapproachinde(cid:12)ningtheoverallinput
WSDL but retain a declarative (cid:13)avor. Most popular here
and output message sequences for one service using essen-
is the use of automata-based descriptions of permitted se-
tially a (cid:12)nite state automaton (FSA) over the alphabet of
quencingpatternsofthewebservices,withafocusoneither
message types. The FSAis called a conversation.
activities performed [7] or messages passed [26].
Note that WSCL complements extremely well with
The underlying structure for the web services paradigm
WSDL. A WSDL description along with a WSCL conver-
sationofaserviceprovidearatherrichsemanticsaboutthe for process (cid:13)ow.
servicewhilekeepingtheinternalimplementation/execution The work on the di(cid:11)erent models has centered around
encapsulated. Forthisreason,theyarecombinedintoalayer three di(cid:11)erent aspects of composition. To describe these,
named \individual service description" in the stack in Fig. we (cid:12)rst establish some vocabulary. When a family of web
3. services interact, the overall topology may be, speaking in-
WhileWSDLandWSCLcande(cid:12)neWebservices,clearly formally,mediator-basedorpeer-to-peer. Bymediator-based
there is a need for languages in the corresponding level of we mean that there is one web service, called the media-
abstraction to compose services. BPEL [12] was developed tor, which has the specialized role of controlling the opera-
as a language both for programming Web services and for tionandinteractionof theotherservices; theotherservices
speci(cid:12)cation of Web services (e.g., in legacy systems). For arecalledcomponent services. Inapeer-to-peerframework,
example, [26] identi(cid:12)ed two topologies for service composi- each participating service is called a component service.
tion: \peer-to-peer",and\hub-and-spoke"withamediator. Itisalsousefultodistinguishbetweencompositions that
Mediatorsplaytheroleofcoordinatingtheactivitiesofother are intended for single or multiple use. In the former case,
webservices. AprimarygoalofBPELisprovidealanguage a composition algorithm is invoked each time that a client
forspecifyingthebehaviorofmediatorservices,ratherthan identi(cid:12)es a desired goal service, similar to typical scenarios
for general-purpose programming. on AI planning. In the latter case, the goal established for
In contrast to specifying individual services that BPEL the composition algorithm is to produce a (composite) ser-
provides, WS-Choreography (WSCDL) [48] attempts to vicethat can berunmultipletimes, insupportof the same
specifygloballyhowdi(cid:11)erentcomponentservicesshouldbe- or di(cid:11)erentclients.
have. WSCDL emphasizes on the \choreography" aspects: The three families of initial results are now described;
the roles of the participating services, information that be- more details are given below.
ing passed between services, and channels that enable the
Synthesis ofmediator fromatomiccomponent services. The
information (cid:13)ow.
(cid:12)rstcategoryofresults,whichhasbeenusedsuccessfully[38]
At the top of the standards stock is the UDDI (Version
withOWL-S,isfocusedonbuildingawork(cid:13)owschema(e.g.,
3 was recently adopted as an OASIS standard). UDDI al-
(cid:13)owchart, Petri net, composite OWL-S service, ...) that
lows services to be registered with a registry. Services in a
invokes atomic services in order to achieve a speci(cid:12)ed goal
registrycanbesearchedwithqueryingabilitiesprovidedby
within speci(cid:12)ed constraints. This work(cid:13)ow schema would
the standard.
typically be enforced by a mediator, and in practice might
be speci(cid:12)ed using BPEL. This work tends to focus on the
3. FOUNDATIONS FOR SERVICE COM›
single-use case.
POSITION:ASAMPLING
Selection of multi-step components and synthesis of medi-
Web service composition addresses the situation when a ator. In this work, initiated with the Roman model, the
client request cannot be satis(cid:12)ed by a single pre-existing componentservicesaregenerallynon-atomic,andhavepro-
service, but can be satis(cid:12)ed by suitably combining some cess (cid:13)ow speci(cid:12)ed using a transition system or (cid:12)nite-state
available, pre-existing services. Reference [25, 1] identi(cid:12)es automata. Unlike the OWL-S work, however, the atomic
two aspects of composition: composition synthesis is con- servicesinsidethecomponentservicesareveryabstract. As
cerned with synthesizing a speci(cid:12)cation of how to coordi- describedbelow,seminalresultsherefocus ontheconstruc-
natethecomponentservicestoful(cid:12)lltheclientrequest;and tion(ifoneexists)ofaspecializedkindofmulti-usemediator
orchestration, isconcernedwithhowtoactuallyachievethe thatachieves a desired goal behavior.
coordination among services, byexecuting the speci(cid:12)cation
produced by the composition synthesis and by suitably su- Synthesis ofcomponent services inpeer-to-peer setting. The
pervising and monitoring that execution. The focus here is results here focus on message-passing Mealy machines. In
on composition synthesis andrelated issues. this context,a composition schema is de(cid:12)nedas a template
The theoretical study of (automatic) composition syn- thatcomponentwebservices canbepluggedinto. Thegoal
thesis for web services is still in its infancy. The models behavior is speci(cid:12)ed as a family of permitted message ex-
underlying this work on automatic composition have roots change sequences, or conversations, that should be realized
in AI, logic, situation calculi, transition systems, and au- bythesystem. Keyresultsherecharacterizewhenaconver-
tomata theory, and thus rely on a variety of philosophical sation language can be realized, and synthesize component
bases. The discussion here is not intended to be compre- services in that case. As detailed below, these results can
hensive, but rather highlights some of the most important be used to help with practical composition, and they pro-
foundations and results that might be most interesting to videanapproachtomap\global" choreographyconstraints
the database community. onto\local" constraints concerning the message sequencing
The initial results on automatic composition considered behaviors of the componentservices.
herearegroupedaroundthreedi(cid:11)erentmodelsforwebser-
In Subsection 3.1 below we describe the OWL-S model
vices,whichweidentifyhereas(a)OWL-S(e.g.,[14]),which
andresultsinmoredetail. Specialemphasisisplacedonthe
includes a rich model of atomic services and how they in-
model, and a natural bridge from the AI foundations that
teract with an abstraction of the \real world", (b) Roman
OWL-S relies on to a formal basis that essentially builds
(e.g., [7]), which uses a very abstract notion of atomic ser-
on a relational database framework. In Subsection 3.2 we
viceina(cid:12)nite-stateautomataframeworkfordescribingpro-
describetheRomanmodelandresults,andinSubsection3.3
cess(cid:13)ows,and(c)(message-based)Mealymachine(e.g.,[8]),
we describe the model and results on message passing and
whichfocuseson(message-based)\behavioralsignatures"of
Mealy machines. Finally, Subsection 3.4 presents a brief
services, andagainusinga(cid:12)nite-stateautomataframework
overview of recent work on web service models and results
thatcombinethefundamentalaspectsof theabovemodels. core aspects of the Situation Calculi. PSL is a (cid:12)rst-order
logic ontology for describing the core elements of processes.
3.1 OWL›S: Modelandcomposition
PSL is layered, with PSL-Core at the base and many ex-
We now consider salient aspects of the OWL-S model tensions. PSL-Outercore is a natural family of extensions
and some of the key composition results obtained. In our above PSL-Core, useful for modeling processes. FLOWS is
presentation we do not follow exactly the original formu- a family of extensions on topof PSL-OuterCore.
lation [14], but instead adopt the approach recently intro- PSL-OuterCore provides (cid:12)rst-order predicates that can
duced by work on the First-order Logic Ontology for Web be used to describe the basic building blocks of processes
Services (FLOWS) [9]. FLOWS provides the basis for the and there execution. To give the (cid:13)avor, we mention a few
(cid:12)rst-orderlogiccomponentoftherecentlyreleasedSemantic highlights. Fundamental is the class of activity; an activ-
Web Services Ontology (SWSO) [15]. We use the FLOWS ity can be viewed as a process template, and there may be
framework here, as it is somewhat closer to a relational zero or more occurrences of an activity (corresponding to
databaseformulationthantheunderpinningsoriginallyused instancesofaprocessorenactmentsofawork(cid:13)ow). Thebi-
byOWL-S. narypredicateoccurrence of(occ;a)isusedtospecifythat
Akeycontributionof OWL-Sis the explicit modeling of occisanoccurrenceof activitya. Therearecomplexactivi-
how web services in OWL-Sinteract with the \real world". ties,andthebinarysubactivitypredicateisusedtospecify
ThebasicbuildingblockoftheOWL-Sprocessmodelisthe sub-activities of an activity. There are also atomic activi-
notion of \process"; thisincludes bothatomic andcompos- ties, such that, intuitively, the execution of an occurrence
iteprocesses. OWL-Sprocessesarespeci(cid:12)edtohaveinputs, of an atomic activity is \atomic" in the typical database
outputs, pre-conditions, and conditional e(cid:11)ects (IOPEs). sense. A notion of time points in a discrete linear ordering
The IOPEs of two example OWL-S atomic processes taken isincluded;occurrenceshaveabegintimeandanendtime.
from a stock brokerdomain are informally sketchednow. In a typical usage of PSL (FLOWS), an application do-
mainiscreatedbycombiningthePSL-OuterCore(FLOWS-
select_stock
Core)axiomswithdomain-speci(cid:12)cpredicatesandsentences
input: stock_name, quantity
toforma((cid:12)rstorderlogic)theory. Themodelsofthistheory
output: price, reservation_id
pre-condition: the quantity of stock is canbethoughtofasatreeorforest,whosenodescorrespond
available for sale to occurrences of (atomic) activities, and with edges from
conditional effects: one occurrence to another if theyfollow each other with no
if true, then modify world state to intervening occurrence. A path in this tree (forest) can be
reflect the fact that this stock
viewedasone possible sequenceof stepsintheexecutionof
is being held for sale
the overall system.
purchase_stock Speakingloosely,eachsentenceinanapplicationdomain
input: reservation_id, billable_account theory can be viewed as a constraint or restriction on the
output: confirmation_id models (inthesense of mathematical logic) thatsatisfy the
pre-condition: reservation_id is valid theory. In particular, and following the spirit of Golog [41]
and has not expired
andsimilarlanguages,evenprocessmodelconstructssuchas
conditional effects:
while or if-then-else correspond formally to constraints
if enough money in billable_account then
transfer of $$ to stock owner and rather than procedural elements. A mapping of OWL-S to
transfer of stock to buyer PSL is providedin[24].
if not enough money, then Wenowgetbacktoourstockpurchaseexample,andpro-
make reservation_id void videadatabase-orientedcastonPSLandOWL-S.ThePSL-
anddomain-speci(cid:12)cpredicatesassociated withourexample
A client would use select stock in order to check the
can be viewed essentially as relations in some (potentially
availability of a given quantity of some stock, and if it
vast)relational database. Someof the domain-speci(cid:12)crela-
is available, then reserve that for purchase. Speaking in-
tionsmightbe\accessible"byjustoneservice,othersacces-
tuitively, successful execution of the service would result
siblebyseveralservices,andstill othersmightbeaccessible
a commitment that this stock will be held for purchase
byallservices. Intheexample, wemighthavethefollowing
(e.g., for 10 minutes). After a successful reservation of
domain-speci(cid:12)c relations
that sort, the client might invoke purchase stock to ac-
tually make the purchase, using some bank account to pay
available_stock(stock_name, qty, price)
for it. purchase stock will execute as long as the reserva- reserved_stock(stock_name, qty, price, res_id)
tion is valid, but the outcome of this execution depends on purchase_reservations(res_id, start_time)
whethertherearesu(cid:14)cientfundsintheaccount. Acompos- billable_accounts(acct_id, current_balance)
iteOWL-Sservicemightbecreatedbycombiningthesetwo
Usingtheaboverelations itis nowrelativelystraightfor-
atomic processes using the sequence construct; in this case
wardtospecifythebehaviorsoftheatomicprocesses(which
theIOPEofthiscompositeprocesscanbeinferredfromthe
can be viewed as atomic activities in the parlance of PSL).
IOPEs of the atomic processes.
Forexample,ifselect stockiscalledwithinputss;q,then
To provide a formal basis for this interaction with the
the pre-conditioncan bespeci(cid:12)ed as a testagainst the cur-
realworld,OWL-StakestheapproachoftheSituationCal-
rentcontentsofrelationavailable stock,toseeifthereisa
culus [41], a logic-based formalism which explicitly models
tuple(s;q0;p)withq0 (cid:21)q. Thee(cid:11)ect(whichinthiscasewill
thefactthatovertimedi(cid:11)erent\situations"orworldstates
will arise. To brie(cid:13)y illustrate this here we follow FLOWS, always occurifthepre-conditionholds)mightbetoreplace
andusetheProcessSpeci(cid:12)cationLanguage(PSL)[43,23],a that tuple with (s;q0(cid:0)q;p), create a \new" reservation id
r, and insert (s;q;p;r;t) into reserved stock, where t is a
recentISOstandardwhichamongotherthingsincorporates
time-stamp. Alloftheseoperationsshouldbeconsideredas Delegator select_stock
anatomic database transaction. (Are(cid:12)nementwouldbeto Target select_stock
enable the case where the shares of stock ful(cid:12)lling a single S1 S1
pay
request would havemultiple prices.) pay
Now, thereis one step remaining before we havethe full pay
pictureonhowthebehavioroftheOWL-Satomicprocesses select_bond S2 S2
can be formally speci(cid:12)ed in PSL. Two general approaches
select_bond
haveariseninthecontextofthesituationcalculi(see[39]for
a detailed discussion). Under one approach, relations such
as available stock are transformed byaddinganew (cid:12)eld, “UDDI++”
S1 select_stock S2 pay
which holds a time value. Under a second approach, taken available services
byPSL,thisrelation wouldbetransformed intoafunction,
pay
say, f available stock that takes three arguments. Two select_bond
PSLpredicates,holdsandpriorareusedtoassociate(intu-
itively)truthvaluestotermscreatedusingthisfunction. For Figure 4: Illustration of composition with Roman
example, if the predicate holds(f available stock(s,q,p), model
occ)istrueinamodel,thiscorrespondstotheintuitionthat Horn clauses). If the overall family of constraints is satis(cid:12)-
the term holds(f available stock(s, q, p) is true \about able,thenthereisacomposition,whichcanbeequatedwith
the world" immediately after the occurrence occ was exe- a branch of the execution tree of the generic program that
cuted, or in terms of the relational perspective, that tuple satis(cid:12)es certain properties. Reference [35] describes how a
(s, q, p) was in the relation available stock at that time. ConGolog interpretercan be usedto (cid:12)ndsuchbranches.
Thepredicatepriorworksanalogously,referringtothetime
immediately prior to the start of an occurrence. 3.2 Roman Model perspective: Actions with
OWL-Sprovidesavarietyof featuresbeyonditsrichno- Automata›based ProcessModel
tion of IOPEs. We mention one here, which is a family
Wenowconsiderafamilyofresults involvingautomated
of constructs for building composite processes from atomic
composition of multi-stepservices. This workwaslaunched
ones. TheseconstructsareinspiredbyGOLOGandarerem-
by the seminal work [7, 5], which is of interest for at least
iniscent of (cid:13)owchart-style constructs, but in OWL-S they
two reasons: (a) the paper develops a novel, general frame-
are viewed as declarative constraints in the speci(cid:12)cation of
workforstudyingcomposition of human-machinestyleweb
a web service, ratherthan procedural imperatives.
services,and(b)thetheoreticaltechniquesdevelopedinthe
Wenowbrie(cid:13)ydescribetwoofthekeyresultsconcerning
paper can begeneralized to a muchbroader context.
automaticcompositionforOWL-SandOWL-S-likeservices.
The paper starts with a very abstract model of web ser-
The (cid:12)rst is developed in [38], where salient aspects of the
vices, based on an abstract notion of activities. Basically,
OWL-S model are mapped (cid:12)rst into a Situation Calculus,
there is a ((cid:12)nite) alphabet of activity names, but no inter-
and from there into the Petri net domain. In this compo-
nal structure is modeled (no input, output, or interaction
sition result, it is assumed that all relevant aspects of the
with the world). To specify the internal process (cid:13)ow of a
applicationdomainarerepresentedusinga(cid:12)nitenumberof
web service, [7, 5] starts with transition systems. In the
propositional(cid:13)uents,andthefamilyofpermittedinputand
mostgeneralcase,thesearepotentiallyin(cid:12)nitetrees,where
output argument values is (cid:12)nite. (In a relational database
each branch corresponds to a permitted sequencing of exe-
setting, this can be achieved by restricting attention to a
cutionsof the activities. Forthe theoretical results theyre-
(cid:12)nite domain.) The goal of the (single-use) composition is
strict attention to (cid:12)nitely speci(cid:12)able transition systems; in
then speci(cid:12)ed in terms of an overall e(cid:11)ect to be achieved
particular onsystemsthatcan bespeci(cid:12)ed as deterministic
(on all of the relevant propositional (cid:13)uents) starting from
(cid:12)nite-state automata, where the activities act as the input
some initial state (which in essence includes relevant input
alphabet (i.e., where the edges are labeled byactivities). It
argumentvalues). Anotherelementoftheresultisthetrans-
is convenientto refer to this as the \Roman" model.
lation of OWL-Satomic processes intoPetri netfragments;
References [7, 5] focus on human-machine style web ser-
these are used to assemble a Petri net that corresponds to
vices. Asasimpleillustrationofhowserviceswork,consider
all possible sequencing of the processes. (For simplicity, it
the service labeled Target in Figure 4. When this service
is assumed that no atomic service is invoked twice in these
is being used bya client (could be human or automated) it
sequences.) The overall complexity of determining the ex-
performs the following sequence of steps:
istence of a composition is expspace. Of course, various
heuristics can be applied to make this tractable. Also, the (1) Give the client a set S of activities to choose from
techniquecanbegeneralizedtosupportcompositionofcom- (possibly including the special \activity" terminate).
posite processes, as well as atomic ones.
(2) Waitfor the client tochoose exactly oneelementof S
Another approach to composition, also developed in the
context of OWL-S, is developed in [35]. This work pro- (3) Executethe chosen activityandreturnto step (1), or
poses a two-tier framework, based ongeneric programs and terminate if that was chosen.
customization via constraints. Beginning with a family of
Ineachiteration,thesetSischosenaccordingtothecurrent
atomic OWL-Sservices, a generic program can be speci(cid:12)ed
state of the service { in a state p the set S includes each
usingtheGOLOGlanguage. Notethatthisgenericprogram
activity that labels an out-going edge from p, and includes
may not be completely speci(cid:12)ed. As a second phase, addi-
\terminate" if p is a (cid:12)nalstate.
tional constraints can be written to capture customizations
To illustrate, when the Target is launched it gives the
neededbyagivenclient. (Itistypicaltoexpresstheseusing
clientachoiceoffselect stock,select bond,terminateg.
Target Credit_card the client, but also the choice that the client will make in
Commit_
. . . to_buy . . . the next move. Reference [22] shows that there is a non-
collapsing hierarchy of k look-ahead delegation problems,
Mail_order and there are examples requiring \in(cid:12)nite" or arbitrarily
long look-ahead. Also, the problemof checkingexistence of
a k look-ahead delegator can be transformed into a prob-
“UDDI++”
lem of checking existence of a conventional (0 look-ahead)
Commit_ Commit_ available services
to_buy Credit_card to_buy Mail_order delegator, and has exptimecomplexity.
3.3 Conversationsand Mealyservices
Figure 5: A composition problem with a 1 look- The OWL-S and Roman models focus on what a service
ahead solution or composition does, either in terms of the input/outputof
services (activities) and their impact on the world, or the
Supposing the client chooses select stock, in the next it-
sequencing of the activities in a service. Atthe operational
erationtheservicegivesthesetfpayg. Executioncontinues
level, neitheraddressesthe issueof howthe componentser-
until the client chooses terminate.
vices in composition should interact with each other. The
Wecannowdescribethebasiccompositionproblemstud-
notion of \conversations" was naturally formulated in ad-
ied with the Roman model. Consider now the two services
dressingthisneed[27,28,49]. Preliminarytheoreticalwork
S1 and S2 shown in Figure 4. Suppose that these are the
onconversationswas reportedin[8, 18]. Inthissubsection,
onlypre-existingservicesavailable. Thequestioniswhether
wehighlightkeytechnicalnotionsandresults. Muchofthis
itispossible tousesome orallof thoseto\build"aservice
work assumes a peer-to-peerframework.
thatactsthewayTarget does. Onewaytoaccomplish this
Tobeginwith,weassumetheexistenceofanin(cid:12)nitesetof
is with the \service" shown as Delegator in Figure 4. The
servicenames. Thesewillactessentiallyasplace-holdersfor
operationofthisissimilartotheoperationoftheotherRo-
service spec(i(cid:12)cation)s, which can be viewed as completely
manservices,exceptthatwitheachtransitionofDelegator
orpartiallyspeci(cid:12)edimplementationsfortheservicenames.
it is also assumed that one or more of the pre-existing ser-
Wealsoassumeanin(cid:12)nitesetofmessageclasses,whereeach
vices performs a transition. In particular, the pre-existing
class mc has a service name as source and a service name
service must perform the same activity as the Delegator.
astarget. Acomposition schemaisaapair(P;M)whereP
Intuitively, one can think of Delegator as not executing
is a (cid:12)nite set of service names and M is a (cid:12)niteset of mes-
the services, but rather, as delegating the execution to the
sage classes. whose sources and targets are in P. Figure 6
pre-existing services. Furthermore, in a valid execution of
shows a composition schema for a \stock analysis service"
Delegator, eachof thepre-existingservicesmustperforma
(SAS)[20]. TheSAScompositionschemausesthreeservice
valid (andterminating) execution.
names: Investor, Stock Broker, and Research Department,
Thecentralresultof[7]isthattheexistenceofadelegator
canbedeterminedinexptime,andthatthereisaconstruc- anduses the 9message classes indicated inthe (cid:12)gure. This
composition schema models abstractly a domain in which
tiveapproachforbuildingsuchdelegators. Importantly,the
the Investor service can request reports on di(cid:11)erent stocks
form or skeleton of the delegator may be di(cid:11)erent than the
from the Research Department. Therequestsare controlled
form or skeleton of the target service (e.g., in the example
by the Stock Broker service, which handles issues such as
the delegator has more states thanthe target service).
permissions and billing.
Thisresultcanbeprovedbyusingatransformationofthe
problem into Propositional Dynamic Logic (PDL), a well-
register, ack, cancel
knownlogicforreasoningaboutprograms[31]. Asdetailed Investor StockBroker
in[5],thereisanatural,polynomial-sizetranslationofaRo- accept, reject, bill
man composition problem into deterministic PDL. The re- request,
report
sultthenfollowsbecausesatis(cid:12)abilityfordeterministicPDL terminate
is exptime, and a constructive technique is available. Fur- ResearchDepartment
ther,[6]describeshowadescriptionlogicreasonerhasbeen
adapted to perform this test andconstruction. Figure 6: A Composition Schema
We brie(cid:13)y mention two extensions of the results in [7].
A composite service for composition schema (P;M) is
Reference [10] extendsthe basic modelbyconsidering (pre-
an association of actual web services to each service name
existing)services thatcan themselvesdelegate tootherser-
in P, where the actual web services are capable of send-
vices. In the model there, if an edge in a service S1 has
label>>athisindicatesthatS1canperformactivitya,and ing/receiving messages coming from the relevant message
classes in M. Given a composite service, a basic approach
ifanedgeinS2haslabel a>> thisindicatesthatS2isable
fortheunderlyingoperationalmodelisthateachservicehas
todelegateactivityatosomeotherservice. Thedelegation
a FIFOqueuefor all incoming messages (possibly from dif-
feature adds a level of non-determinism, but checking the
ferentservices). Thiscorrespondsabstractlytothemessage
existence of, and constructing, a (generalized) delegator is
orientedmodelinSOA[25,1]. Queuingofmessagesprovides
still possible in exptime.
a model asynchronous communication, which is realistic in
ConsidernowthecompositionproblemillustratedinFig-
the web setting.
ure 5. There is no delegator that can simulate the target
Given a composite service S over composition schema
service in this case. However, as discussed in [22], there is
(P;M), aconversation of S isthe (global) sequenceof mes-
a \1 look-ahead" delegator, that is, a delegator which can
sagessentduringonesuccessfulexecutionofS. Severalstan-
make the correct delegations, as long as the delegator is
dards and research works are concerned with properties of
given information not only about the immediate choice of
conversation languages. WS-Choreography [48], and refer- a ?a
ences [27,28,48]focus on\global" constraintsonconversa- ?b !a
b
tions. In contrast, the standards WSCL [49] andWSDL2.0 !b
[51] focus on \local" constraints on conversations, in the
sensethattheyfocusonconstraintsaboutmessagepatterns Figure 9: A Non-regular Conversation Language
of the component services of a composite service.
can simulate Turing machines [13]. When all queues are
Reference[8]initiatedtheformalstudyoftherelationship
bounded,the conversation languages are always regular.
between such global and local constraints. A conversation Turningtoquestion(Q1),a(cid:12)neranalysisrevealstwoim-
protocol for a composition schema (P;M) is a (cid:12)nite state portant aspects of conversations languages: (a) What indi-
machineoverthesetofmessageclassesinM. Figure7shows vidual Mealy service sees vs. the (global) conversation, and
aconversationprotocolfortheSASservice. Inthisexample,
(b)WhentwoMealyservicesarebothreadytosend,itisn’t
the register message from INV can include a list of stock
possibletoforceaparticularordering. Thesetwofactorsare
symbols. Upon acceptance of a message of this type, each
formulated intotwo operators in [8]:
symbolwillresultinaseparaterequestfromtheStockBroker
service to the Research Department service (corresponding (cid:15) Projectingaconversationtothemessagealphabetofa
to the request-report-ack loop inthe protocol). single service name and joining message sequences of
Mealy services into a conversation, and
cancel
(cid:15) Preponing (i.e., swapping) messages from di(cid:11)erent
report
senders (ina local perspective).
register acceptrequest ack requaesctk report uItndtuerrn(s1)ouptrotjheacttioMn-ejaoliyncaonndve(2rs)aptiroenpolnaninggu.ages are closed
reject bill cancel colTishencoltosculoreserdesuunltdseirnediitchaetre othpaetraitfoar,cothnvenersitatcioannnportotboe-
terminate terminate bill realized byanycomposite service. Intuitively,this suggests
that during design it may not be a good candidate as a
Figure 7: A Conversation Protocol for SAS global constraint onthecomposition duringdesign. Onthe
other hand, it is interesting to know if the projection-join
Three natural questions are (Q1) Given a conversation and prepone closure of a regular language is realizable as
protocolovercompositionschema(P;M),canitberealized theconversationlanguageofsomeMealycompositeservice.
bysome composite service S over(P;M)? (Q2) If so, what The answer turns out to be positive [8], and, relevant to
can be inferred about the individual services in S? (Q3) question (Q2),a construction is provided.
If the individual services in a composition have a message- Itisnotclearwhetheronecanalwaysdetermineifareg-
passing behavior corresponding to a regular language, then ularlanguageisclosedunderprojection-joinandpreponing,
is the resulting conversation also characterized bya regular thougha su(cid:14)cient condition was givenin [18].
language (i.e., bya conversation protocol)? Weconcludethisdiscussionbydescribingtwobroadways
In [8] these questions are studied using the notion of in which the above results can be applied in practical set-
Mealy services. Theseareservicespecsbasedon(cid:12)nitestate tings. Oneistohelpintheprocessofpracticalcompositions.
automata whose transitions correspond to sending a mes- Suppose that one seeks to create a (peer-to-peer) composi-
sage m(denoted as !m)or receiving a message m(denoted tion, based on a pre-determined composition schema, and
as?m). Fig. 8showsaMealyserviceforINV.AMealycom- satisfying a given conversation protocol. Suppose further
posite service is a composite service over Mealy services. thatafamilyof componentMealyservices issynthesizedto
WenotethataMealyservicecan beviewedinthreedif- satisfy this protocol. This family of Mealy services can be
ferentways: (a)asanimplementationofaservicename;(b) usedtoselectafamilyofcandidate(real-world)webservices
as a \behavioral signature" that abstractly describes some that satisfy the \local" constraints speci(cid:12)ed by the Mealy
properties of a service behavior; and (c) as a constraint on services. Thenperformfurtheranalysisofthecandidateser-
the local messaging behavior that a service shouldsatisfy. vices,onaspectsthatarenotcapturedbytheMealyservices
With regards to question (Q3), in [8], it was shown that alone.
theconversationlanguagesofsomeMealycompositeservices Thesecondpracticalapplicationistoviewaconversation
are not regular (see Fig. 9), nor even context-free in the language as a global constraint that should be satis(cid:12)ed by
language hierarchy, but they are always context-sensitive. afamily of webservices, essentially inthespiritof theWS-
The key factor here is the use of unbounded queues. This Choreography standard. The results above can be used to
is not surprising, since (cid:12)nite state automata with queues determine whether this global constraint can be realized at
all. Itcanalsobeusedtoinferlocalconstraints,thatspecify
properties of the behavioral signatures of each component
!register
service.
?accept
?reject
!ack 3.4 Selected AdditionalDevelopments
Wecloseoursamplingofcompositionmodelsandresults
?report
with a brief overview of two very recent e(cid:11)orts, which hold
!cancel ?bill
thepromiseofprovidingaunifyingframeworkforthisarea.
?bill
The (cid:12)rst, already touchedupon in Subsection 3.1, is the
Figure 8: A Mealy Service for Investor work on FLOWS [9, 15], which provides a (cid:12)rst-order logic
ontology for web services. We provide here a little more Using the Colombo model, [4] develops a framework for
detail onFLOWS. posing composition problems,thatclosely parallels the way
FLOWS includes objects of type service. These can be compositionwillhavetobedoneusingstandards-basedweb
related to several types of object, including non-functional services. Speci(cid:12)cally,thebasiccompositionproblemstudied
properties (e.g., name, author, URL), and also to its activ- is how to builda mediator that simulates the behavior of a
ity. This is a PSL activity, and may have specialized kinds target web service, where the mediator can only use mes-
of subactivities. This includes (FLOWS) atomic processes, sage passingtogetthepre-existingweb servicestoperform
which are based on OWL-S atomic processes with inputs, atomic activities (which in turn impact the \real world").
outputs, pre-conditions and conditional e(cid:11)ects. Some of Undercertainrestrictions, [4]demonstratesthedecidability
these atomic processes, as in OWL-S, are domain-speci(cid:12)c, of the existence of a mediator and develops a method for
in that they interact with the \real world" as represented constructing them. This result is based on (a) a technique
intheapplicationdomain,andothersareservice-speci(cid:12)c, in for reducingpotentially in(cid:12)nitedomainsof datavaluesinto
that they focus on the standardized mechanics of services, a(cid:12)nitesetof symbolicdata values,and(b)inageneraliza-
e.g., message-passing andchannelmanagement(seebelow). tionof[5],amappingofthecompositionproblemintoPDL.
The (cid:13)ow of information between services can occur in The results reported in [4] rely ona numberof restrictions;
essentially two ways: (a) via message passing and (b) via a broad open problem concerns how these restrictions can
shared access to the same "real world" (cid:13)uent (e.g., an in- be relaxed while still retaining decidability.
ventorydatabase,areservationsdatabase). Withregardsto
message passing, FLOWS models messages as (conceptual) 4. ANALYSIS COMPOSITESERVICES
objects that are created, read, and (possibly) destroyed by
The need for analysis is particularly acute for composite
web services. A message life-span has a non-zero duration.
services,especiallyiftheyaretobecreatedfrompre-existing
Messages have types, which indicate the kind of informa-
services using automatic algorithms. The ultimate goal is
tionthattheycantransmit. FLOWSalsoincludesa(cid:13)exible
toensurethattheeventualexecutionofacompositeservice
channel construct.
produces the desired behavior. Ideally, one would be able
To represent the acquisition and dissemination of infor-
to statically verify properties (e.g., in temporal logic) for
mation inside a Web service, FLOWS follows the spirit of
composite services. There have been various attempts at
the ontology for "knowledge-producing actions" developed
developingsuchstaticanalysismethodsforwebservicesand
in [45]; this also formed the basis for the situation calculus
work(cid:13)ow systems.
semantics of OWL-Sinputsande(cid:11)ects [38].
Based on work(cid:13)ows represented in concurrent transac-
FLOWS is very open-ended concerning the process or
tion logic, [16] studied the problem of whether a work(cid:13)ow
data (cid:13)ow between the atomic processes inside a services.
speci(cid:12)cation satis(cid:12)es some given constraints similar to the
Thisisintentional,asthereareseveralmodelsfortheinter-
EventAlgebraof [44]. Algorithmsweregivenforsuchanal-
nal processing in the standards and literature (e.g., BPEL,
ysis. An alternative approach is developed in [47], which
OWL-S Process Model, Roman model, Mealy model) and
mapsconventionalwork(cid:13)owstoPetrinets,andthenapplies
many other alternatives besides (e.g., rooted in Petri nets,
standard results to analyze properties such as termination
in process algebras, in work(cid:13)ow models, in telecommunica-
and reachability for work(cid:13)ows. Similar results have been
tions). TheFLOWSworkisstillquitepreliminary,butholds
obtained for OWL-Scompositions [38].
thepromiseofprovidingaunifyingfoundationforthestudy
There are two recent projects that use model checking
and comparison of these many variations on the notion of
techniques to verify BPEL composite web services. In [21],
web service.
mediatedcompositeservicesspeci(cid:12)edinBPELwereveri(cid:12)ed
A second very recent work is reported in [4, 3]. That
against the design speci(cid:12)ed using Message Sequence Chart
work develops Colombo, a formal model for web services
andFiniteStateProcessnotations,withafocusonthecon-
that combines
trol (cid:13)ow logic. In [19], both conversation protocols and
(a) Aworldstate,representingthe\realworld",viewedas Mealy services were extended to guarded automata which
adatabaseinstanceoverarelational databaseschema incorporate (i) XML messages and (ii) XPath expressions
as the basis for verifyingtemporal properties of theconver-
(b) Atomic processes in the spirit of OWL-S,
sations of composite web services. The following shows an
(c) Message passing, including a simple notion of ports example of a transition ina protocol for the SASservice:
and links, as found in web services standards (e.g.,
WSDL,BPEL) t14 {
s8 -> s12 : bill,
(d) An automata-based model of the internal behavior of Guard {
web services, where the individual transitions corre- $request//stockID =
$reginfo//stockID [position() = last()]
spond to atomic processes, message writes, and mes-
=>
sage reads.
$bill[//orderID := $reginfo//orderID]
The (cid:12)rst three elements parallel in several aspects the core }
},
elements of FLOWS.Colombo also includes
(e) a \local store" for each web service, used manage the Roughly, the transition is de(cid:12)ned from state \s8" to
dataread/writtenwithmessagesandinput/outputby \s12" on message \bill". The part of guard prior to
atomic processes; and \=>"de(cid:12)nesanadditionalequalityconditionwithbothside
XPathexpressions. Thepartafter\=>" speci(cid:12)edanassign-
(f) a simple form of integrity constraints on the world
ment.
state
The extended model makes it possible to verify designs I/O signatures? What is an appropriate way to model the
ata moredetailedlevel andtocheckproperties aboutmes- data transformations occurring in a web service, which will
sage content. A framework is presented where BPEL spec- enablereasoningaboutthebehaviorofdatabeingpassedor
i(cid:12)cations of web services are translated to an intermediate writtentodatabasesbyacompositewebservice? Arethere
representation, followed by the translation of the interme- specialized models of Web service composition that will be
diate representation to the veri(cid:12)cation language Promela, moresuitableforapplicationsthataretargetedprimarilyat
input language of the model checker SPIN [30]. Since the data processing? Astarting pointhere mightbe [2, 17].
SPIN model checker is a (cid:12)nite-state veri(cid:12)cation tool, the
Acknowledgements: Theauthorswishtothanktheircol-
tool can only achieve partial veri(cid:12)cation by (cid:12)xing the sizes
laboratorsformanyfruitfuldiscussionsandinspirations,in-
of the inputqueuesin the translation. Su(cid:14)cientconditions
cluding Michael Benedikt, Daniela Berardi, Tev(cid:12)k Bultan,
for complete veri(cid:12)cation are also obtained.
Diego Calvanese, Vassilis Christophides, Guiseppe De Gia-
An important step in statically analyzing Web services
como, XiangFu,Cagdas Gerede, Michael Gruninger,Oscar
de(cid:12)ned in BPEL (or other languages) is to translate them
Ibarra, Grigorios Karvounarakis, Maurizio Lenzerini, Mas-
into formalisms that are suitable for analysis. Such for-
simo Marcella, andSheila McIlraith.
malisms include (cid:12)nite state machines [37, 21], extended
Mealymachines[19], processalgebra[32]. E(cid:11)ectively,these
translations provide formal semantics to BPEL. Most of 6. REFERENCES
thetranslationsmentionedabovefocusonlyoncontrol(cid:13)ow
structures, with an exception of [19] (where XML message [1] G.Alonso,F.Casati,H.Kuno,andV.Machiraju.Web
Services.Concepts, Architectures and Applications.
types,localdata,andXPathexpressionsarealsotakeninto
Springer-Verlag,2004.
consideration. More recently, an extensive translation of
[2] S.Abiteboul,V.Vianu,B.Fordham,andY.Yesha.
BPELintoaversionofPetrinetswasdeveloped[33], which
Relationaltransducersforelectroniccommerce.InProc.
deals with scoping rules, variable assignments, correlation ACM Symp. on Principles of Database Systems,1998.
sets, among other things.
[3] D.Berardi,D.Calvanese,G.DeGiacomo,R.Hull,and
The presence of databases makes the static analysis ex- M.Mecella.Automaticcomposition ofwebservicesin
tremely hard but not entirely impossible. In [17], a rule Colombo.InProc. of13th Itallian Symp. on Advanced
based language was developed for specifying interactive Database Systems,June2005.
Web services. The rules may access associated (relational) [4] D.Berardi,D.Calvanese,G.DeGiacomo,R.Hull,and
databases. Web service properties are de(cid:12)ned using a lin- M.Mecella.AutomaticCompositionofTransition-based
SemanticWebServiceswithMessaging.Technicalreport,
ear temporal (cid:12)rst-order logic. Based on the Abstract State
UniversityofRome,\LaSapienza",Roma,Italy,April,
Machinetechniques[46], itwas shownthatsomeproperties
2005.
are veri(cid:12)ablefor a subclass of Webservices.
[5] D.Berardi,D.Calvanese,G.DeGiacomo,M.Lenzerini,
Serviceorientedarchitecturemay\fundamentallychange andM.Mecella.AutomaticServicesCompositionbased on
the way software is made and used" [11]. The current de- BehavioralDescriptions.International Journal of
velopment platforms may no longer be suitable. While this CooperativeInformationSystems(IJCIS),2004.Toappear.
is a rare chance to revisit important software development [6] D.Berardi,D.Calvanese,G.DeGiacomo,M.Lenzerini,
environment,thereisnodoubtthatanalysistools(staticor andM.Mecella.ESC: AToolforAutomaticComposition
ofe-ServicesbasedonLogicsofPrograms.InProc.
dynamic),testinganddebuggingtools,monitoringtoolsare
Workshop on Technologies for E-Services,2004.
becoming critically importance.
[7] D.Berardi,D.Calvanese,G.DeGiacomo,M.Lenzerini,
andM.Mecella.Automaticcompositionofe-servicesthat
5. CONCLUSIONS exporttheirbehavior.InProc. 1st Int. Conf. on Service
The web services paradigm raises a vast array of ques- OrientedComputing (ICSOC),volume2910ofLNCS,
pages 43{58,2003.
tions, including many that will be of interest to, and can
[8] T.Bultan,X.Fu,R.Hull,andJ.Su.Conversation
bene(cid:12)tfrom, thedatabase researchcommunity. Firstisthe
speci(cid:12)cation: Anewapproachtodesignandanalysisof
questionof (cid:12)ndingappropriate models andabstractions for
e-servicecomposition.InProc. Int.World WideWebConf.
representing Web services and their \behaviors", which are (WWW),May2003.
suitable to the web services paradigm, and can support ef- [9] D.Berardi,M.Gruninger,R.Hull,and S.McIlraith.
(cid:12)cientqueryingandmanipulationas neededbywebservice Towardsa(cid:12)rst-orderontology forwebservices.InW3C
compositionandanalysisalgorithms. Morebroadly,towhat Workshop on Constraints and Capabilities for Web
extentcantheproblemofautomatedcompositionbere-cast Services,October 2004.
to be a problem in writing and answering one or several [10] D.Berardi,G.DeGiacomo,M.Lenzerini,M.Mecella,and
D.Calvanese. Synthesisofunderspeci(cid:12)ed composite
queriesagainstbehavioraldescriptionsofservices? Another
e-servicesbased onautomatedreasoning.InProc. Second
aspectconcernstheapplicationofXMLconstraint-checking
International Conference on Service-OrientedComputing,
techniquesto performcompile-time orrun-timecheckingof
pages 105{114,2004.
Web service speci(cid:12)cations (e.g., in WSDLandBPEL, or in
[11] J.Bloomberg.Thesevenprinciplesofservice-oriented
emerging behavioral speci(cid:12)cation languages). development. XML& WebServices,3(5):32{33, 2002.
Asecond category of questionsis how tobring data ma- [12] BusinessProcessExecutionLanguage forWebServices
nipulation more clearly into the web services paradigm and (BPEL),Version1.1.
their associated standards. The standards and most re- http://www.ibm.com/developerworks/library/ws-bpel,
search at present are focused primarily on process model May2003.
and I/O signatures, but not on the data (cid:13)ow and the ma- [13] D.BrandandP.Za(cid:12)ropulo.Oncommunicating(cid:12)nite-state
nipulationofthedataasitpassesthroughthis(cid:13)ow. Isthere machines.Journal of the ACM,30(2):323{342, 1983.
value in associating integrity constraints with web service [14] OWLServicesCoalition.OWL-S:Semanticmarkupforweb
services,November2003. and Reasoning (KR2002),pages 482{493,April2002.
[15] SWSLCommittee.SemanticWebServiceOntology [36] S.A.McIlraith,T.C.Son,andH.Zeng.Semanticweb
(SWSO), 2005.Availablein services.InIEEE IntelligentSystems,March/April2001.
http://www.daml.org/services/swsl/report/. [37] S.Nakajiama.Veri(cid:12)cationofwebserviceswithmodel
[16] H.Davulcu,M.Kifer,C.R.Ramakrishnan,andI.V. checking techniques.InProc. First Int. Symposium on
Ramakrishnan.Logicbased modelingandanalysisof Cyber Worlds.IEEEComputerSocietyPress,2002.
work(cid:13)ows.InProc. ACM Symp. on Principles of Database [38] S.NarayananandS.McIlraith.Simulation,veri(cid:12)cation and
Systems,pages 25{33,1998. automatedcompositionofwebservices.InProc. Int.World
[17] A.Deutsch,L.Sui,andV.Vianu.Speci(cid:12)cationand Wide WebConf. (WWW),2002.
veri(cid:12)cation ofdata-drivenwebservices.InProc. ACM [39] J.A.Pinto.Temporal Reasoning inthe Situation Calculus.
Symp. on Principles of Database Systems,2004. PhDthesis,UniversityofToronto,1994.
[18] X.Fu,T.Bultan,andJ.Su.Conversation protocols: A [40] M.Paolucci,T.Kawamura,T.R.Payne, andK.Sycara.
formalismforspeci(cid:12)cationandveri(cid:12)cationofreactive ImportingthesemanticwebinUDDI.InProc. Workshop
electronicservices.InProc. Int. Conf. on Implementation on WebServices, E-businessand Semantic Web(at
and Application of Automata (CIAA),2003. CAISE conference),2002.
[19] X.Fu,T.Bultan,andJ.Su.AnalysisofinteractingBPEL [41] R.Reiter.Knowledge inAction: Logical Foundations for
webservices.InProc. Int.World Wide WebConf. Specifying and Implementing Dynamical Systems.MIT
(WWW),May2004. Press,Cambridge,MA,2001.
[20] X.Fu,T.Bultan,andJ.Su.Modelchecking XML [42] S.RussellandP.Norvig.Arti(cid:12)cialIntelligence: AModern
manipulating software.InProc. Int. Symposium on Approach. PrenticeHall,1995.
Software Testingand Analysis (ISSTA),July2004.
[43] PSLstandardsgroup.PSLhomepage.
[21] H.Foster,S.Uchitel,J.Magee,andJ.Kramer. http://ats.nist.gov/psl/.
Model-basedveri(cid:12)cation ofwebservicecompositions.In
[44] M.Singh.Semanticalconsiderationsonwork(cid:13)ows: An
Proc. the 18th IEEE Int.Conf. on Automated Software
algebraforintertaskdependencies. InProc. Workshop on
Engineering Conference (ASE 2003),2003.
Database Programming Languages (DBPL),1995.
[22] C.E.Gerede,R.Hull,OscarH.Ibarra,andJ.Su.
[45] R.B.ScherlandH.J.Levesque.Knowledge,action,and
Automated compositionofe-services: lookahead.InProc.
theframeproblem.Arti(cid:12)cial Intelligence,144:1{39,2003.
Intl. Conf. on Services-OrientedComputing,2004.
[46] M.Spielmann.Veri(cid:12)cationofrelationaltransducersfor
[23] M.Gru(cid:127)ningerandC.Menzel.Processspeci(cid:12)cation
electroniccommerce.Journal of Computer and System
language: Principlesandapplications.AI Magazine,
Sciences,66(1):40{65, 2003.
24:63{74,2003.
[47] W.M.P.vanderAalst.Ontheautomaticgenerationof
[24] M.Gru(cid:127)ninger.ApplicationsofPSLtosemanticweb
work(cid:13)owprocessesbased onproductstructures.Computer
services.InProceedings ofSWDB’03, The(cid:12)rst
in Industry,39(2):97{111, 1999.
International Workshop on Semantic Weband Databases,
[48] WebServicesChoreographyDescriptionLanguageVersion
2003.
1.0(W3CWorkingDraft).
[25] H.Haas.Architectureandfutureofwebservices: from
http://www.w3.org/TR/2004/WD-ws-cdl-10-20041217/,
SOAPtosemanticwebservices.
December 2004.
http://www.w3.org/2004/Talks/0520-hh-ws/, May2004.
[49] WebServicesConversationLanguage(WSCL)1.0.
[26] R.Hull,M.Benedikt,V.Christophides,andJ.Su.
http://www.w3.org/TR/2002/NOTE-wscl10-20020314/,
E-services: Alookbehindthecurtain.InProc. ACM
March2002.
Symp. on Principles of Database Systems,2003.
[50] WebServicesDescriptionLanguage(WSDL) 1.1.
[27] J.E.Hanson,P.Nandi,andS.Kumaran.Conversation
http://www.w3.org/TR/wsdl, March2001.
supportforbusinessprocessintegration.InProc. 6th Int.
[51] WebServicesDescriptionLanguage(WSDL) Version2.0
Enterprise DistributedObjectComputing (EDOC),Ecole
Part2: Prede(cid:12)nedExtensions(W3CWorkingDraft).http:
Polytechnic,Switzerland,2002.
//www.w3.org/TR/2004/WD-wsdl20-extensions-20040803/,
[28] J.E.Hanson,P.Nandi,andD.W.Levine.
August2004.
Conversation-enabled webservicesforagentsand
e-business.InProc. International Conference on Internet
Computing (IC),pages791{796.CSREAPress,2002.
[29] C.A.R.Hoare.Communicatingsequentialprocesses.
Communications of theACM,21(8):666{677, 1978.
[30] G.J.Holzmann.The SPIN Model Checker: Primer and
Reference Manual.Addison-Wesley,Boston,Massachusetts,
2003.
[31] D.KozenandJ.Tiuryn.Logicsofprograms.InJanvan
Leeuwen,editor,Handbook of Theoretical Computer
Science |Formal Models and Semantics,pages 789{840.
Elsiever,1990.
[32] M.KoshkinaandF.vanBreugel.Vericationofbusiness
processesforwebservices.TechnicalReportCS-2003-11,
Department ofComputerScience,YorkUniversity,2003.
[33] A.Martens.Analysisandre-engineeringofwebservices.In
Proc. 6th Int. Conf. on Enterprise Information Systems,
2004.
[34] R.Milner.Communicating and MobileSystems: The
(cid:25)-calculus.CambridgeUniversityPress,1999.
[35] S.McIlraithandT.Son.AdaptingGologforcomposition
ofsemanticwebservices.InProc. of the Eighth
International Conference on Knowledge Representation