A SUCCINCT ANALYSIS OF WEB SERVICE COMPOSITION
Wassam Zahreddine, Qusay H. Mahmoud
Department of Computing and Information Science
University of Gueph, Guelph, ON, N1G 2W, Canada
Keywords: Web service composition, BPEL4WS, OWL-S, WSCI, WSCL, software agents, P2P.
Abstract: Numerous standards are being proposed by industry and academia to find ways to best compose web
services together. Such standards have produced semi-automatic compositions that can only be applied in a
limited number of scenarios. Indeed, the future is moving towards a semantic web and fully automatic
compositions will only occur when semantics are involved with web services. This paper presents brief
notes on the state of the art in the field of service composition. The paper classifies service composition into
two streams: semi-automatic and fully automatic, then compares and contrasts the available composition
techniques.
1 INTRODUCTION
Web services are becoming an attractive solution for
businesses and consumers alike because of their
simplicity and reusability. Presently, web services
are designed to be modular and loosely coupled to
perform a specific set of operations such as
retrieving a stock quote. However, what if a client
requires a service that no one web service can
satisfy? The modularity of web services has created
a composition problem that is still in need of an
optimal solution. Web service composition involves
an amalgamation of two or more web services to
fulfil a request that no one web service is able to
provide. The endless possibilities of a composite
web service will bring forth a new wave of online
applications.
Web services are designed to perform a specific
set of operations and ideally perform them well.
Presently, the user would pick and choose the
organization that offers a web service that performs
optimally, or for the best price, or any other required
criteria. If the user requires more than what any one
particular web service has to offer, they can
manually invoke other web services and organize the
results on their own. However as the need for
various web services grow so does the complexity of
choosing the appropriate service and managing the
results. The time involved in searching, selecting,
and invoking individual web services could be
immense depending on the number of web services
needed. Having the ability for an application to
perform these tasks would be ideal. Therefore, there
needs to be some tools to manage and ensure that
services perform correctly together.
2 SEMI-AUTOMATIC METHODS
There have been various attempts by industry to
build composite web services. This section will
focus on the many important technologies developed
by industrial efforts, namely: BPML, WSCI, WSCL,
BPEL4WS, XLANG, and WSFL.
2.1 BPML & WSCI
The Business Process Management Language
(BPML) provides an abstract model and grammar
for describing business processes and was developed
by Business Process Management Initiative
(BPMI.org). Initially it was designed for the BPMS
system to support business processes. However, the
first draft used the Web Services Choreography
Interface (WSCI), this is an extension of WSDL to
describe the behaviour of a web service and the flow
of messages. BPML handles the orchestration of
web services while WSCI handles the choreography
between web services. Hence, WSCI only describes
the visible behaviour and not the definition of
executable business processes as BPEL4WS does
(Peltz, 2003) but BPML makes up for that with its
own orchestration techniques. The BPML model is
a composition of activities that represent a business
process. Each activity performs a specific function
(a unit of work) and the process directs how and
193
Zahreddine W. and H. Mahmoud Q. (2006).
A SUCCINCT ANALYSIS OF WEB SERVICE COMPOSITION.
In Proceedings of the Eighth International Conference on Enterprise Information Systems - SAIC, pages 193-196
DOI: 10.5220/0002495301930196
Copyright
c
 SciTePress

when these activities are executed. This approach is
very similar to UML activity diagrams.
Moreover, during a message exchange, only one
partner's participation is described in a single WSCI
interface and is therefore designed from the
perspective of one side of the partnership (Peltz,
2003). Realistically, you need to map the actions of
both sides during collaboration. WSFL had already
accounted for this scenario with the concept of a
global model. Later on in the development of WSCI
global models were included. The formation of a
global model using WSCI would be assuming the
two parties know a lot about each other’s process. A
bad approach considering the global model is
assumed immutable. This made BPML difficult to
use for B2B (Business to Business) scenarios and
was likely one of the main reasons why the WSCI
protocol was removed from the BPML 1.0
specification (but it is still supported).
2.2 WSCL
Web Services Conversation Language (WSCL) is a
service description language using a simple state-
transition model for organizing the sequence of
WSDL operations. WSCL can be used to describe
service interactions and to specify a web service
interface. Also this protocol orchestrates the
message exchanges that occur at each stage of the
conversation. WSCL portrays the conversation
pattern that a web service will be engaged in by
describing the order is which WSDL operations
should be invoked. WSCL has the following basic
concepts that describe conversations:
DocumentTypes to reference XML Schemas,
Interactions for one or two way message exchanges,
and lastly Transitions to describe how to move from
one interaction to another. It is expected that WSCL
will be extended to describe more complex scenarios
such as multi-party conversations and composition,
service attributes, and transactions. Unlike other
modelling languages it does not support contexts,
exception handling, time-outs, and contexts.
2.3 BPEL4WS, XLANG & WSFL
One modelling language with major industrial
backing is BPEL4WS (Business Process Execution
Language for Web Services) which evolved from the
amalgamation of Microsoft's XLANG and IBM's
WSFL (Web Services Flow Language). This new
language combines the best features of WSFL and
XLANG. The merge involves taking the graph
oriented, transitioned-based process representation
of WSFL and the block structured processes of
XLANG. Together they build a new orchestration
language defining interactions between web
services. WSFL supports two model types namely
flow and global models. The flow model describes
business processes that use predetermined set of web
services. The global model describes how web
services will interact with each other. And XLANG
orchestrates how individual web services become a
business process and composite web service.
Furthermore, BPEL4WS was also released with
two other specifications, WS-Coordination and WS-
Transaction. WS-Coordination defines a framework
that allows different coordination protocols to
manage operations between participants. WS-
Transaction allows businesses to monitor
coordinated activities in a business process. These
will ensure that all the transactions complete
successfully or fail as a group. In addition, the model
has scopes and handlers to manage exceptions and
allow alternative actions to be taken or reverse work
in a previously completed scope (Curbera et al,
2003). This language scales well and offers
primitive constructs of the language such as
‘sequence’ and ‘while’. Even more interesting is
when the composition is advertised as a web service,
other compositions may include it their workflow.
Furthermore, BPEL4WS also gives the composer
flexibility in developing a workflow since two
different styles of modelling are supported: the
graph-oriented style of WSFL and the block
structured algebraic style of XLANG (Curbera et al,
2003). The merging of these two languages also
means that BPEL4WS supports all the patterns that
WSFL and XLANG, however this flexibility comes
at the price of a language with greater complexity.
Especially when considering the overlapping
constructs and the compromises that had to be made
between Microsoft and IBM to create the language.
Moreover, BPEL4WS is still in its infancy and is
the first step towards building a technology that may
become the industry's first choice in web service
composition. Furthermore, BPEL4WS is semi-
automatic although it does support "runtime
binding", where the parties involved in the
composition is not known until they are needed.
Realistically, it can dynamically assign a partner not
dynamically discover new partners. Binding to
service partners depends on the descriptions of what
the services do and how they work, this is done by
references to the <portTypes> in WSDL.
Unfortunately, since WSDL lacks semantics it fails
to describe how the web service works or what it
does, and only provides the methods, parameters,
and return values. In this manner, service behaviour
is restricted by XML making it a difficult decision
on whether to bind with a service partner or not.
Many models have been presented here however
they all fall in relatively the same scope. These
languages focus on a syntactical approach and can
only produce semi-automatic composition. What it
ICEIS 2006 - SOFTWARE AGENTS AND INTERNET COMPUTING
194

really comes down to is having a person in the loop
taking care of the workflows and business processes.
This may suffice for services that rarely change or
knowing a priori the relationships between business
partners. However the growing number of web
services will create new options that may entice new
business partnerships, hence complicating the
building of workflows.
3 AUTOMATIC COMPOSITION
The automation of web services is difficult with
standard WSDL descriptions because they do not
provide any machine comprehendible metadata to
assure that the service is being used correctly. More
importantly a machine cannot understand whether a
web service will perform the needed service or not.
The liability is on the user to ensure the appropriate
web service is selected.
3.1 Semantic Web
Presently, web sites are designed to describe their
appearance. However, as automation becomes more
important, web pages will have to describe their
meanings as well.
Searches based on syntactical means fall victim
to false positives when the word is a homograph or
false negatives when the word is a synonym.
Researchers are looking for solutions to solve the
semantically deprived Internet. This has led to the
creation of semantic languages that are meant for
computers, and not for humans. RDF (Resource
Description Framework) is one of the first languages
for representing information about web resources.
However, RDF lacked the expressiveness needed
and the semantics themselves were underspecified.
Then OWL (Web Ontology Language) was
developed on top of RDF to support processing
information on the web. Currently, web services are
defined by RDF using RDFS (RDF Schema). Other
languages such as DAML-ONT, OIL, and
DAML+OIL have emerged as a way to improve the
semantic richness of web resources. These languages
are written in XML and describe UDDI and WSDL
documents.
Moreover, it is important to ensure that the
ontology being used is standardized so that all web
services follow the same ontology. One unpleasant
scenario would involve invoking a web service with
non-standardized ontology that delivers unintended
material or even spam. Therefore, not only does the
web service need to follow a standardized ontology
but it must also have the means to verify that the
service is indeed following a standard. Security
issues with ontology-based systems still need to be
addressed. And what were to happen if a standard
ontology changes? The affected web service would
have to redefine its ontology and ensure it complies
with the updated standard.
3.2 OWL-S/DAML-S
A major player in web service automation is the
DAML (DARPA Agent Markup Language)
organization. Their initiatives in web services have
produced DAML-S, and now its successor OWL-S.
This language is built upon OWL and its
predecessor ontology, DAML+OIL. These
languages are based on the DARPA knowledge
sharing format and works by restricting services to
use a common syntax, ontology, and protocols.
OWL-S has three properties to describe a service.
The ServiceProfile and ServiceModel are the
abstract representations of a service and
ServiceGrounding deals with the concrete level of
specification. The ServiceProfile describes the
capabilities and parameters of the service. The
service model describes what happens when the
service is carried out. More formally, it describes
how the service works by specifying the workflow
and possible execution paths. The service grounding
explains how the service can be accessed and used.
Grounding will specify the protocol to use, message
types and other details specific to the service.
Orchestration by OWL-S is performed similarly to
other choreography languages such as BPEL4WS. It
has control constructs such as ‘sequence’, and ‘if-
then-else’ which form the composition plan.
3.3 Software Agents
A software agent is a program that performs a
specific task on behalf of the user or another agent.
Agents can autonomously solve problems, and are
goal-oriented. The agent is more commonly created
from an agent platform, where it then moves from
one platform to another invoking the methods
accessible to the agent. Unlike remote method
invocation, the state of the agent can be stored and
can continue where it left off on the new platform. A
mobile agent is extremely useful in the field of
wireless computing because they improve latency
and the use of bandwidth since the invocations are
handled locally on the hosts' platform. Agents can
also communicate with other agents and act as
autonomous communicative middleware. For
instance, an agent that acts as a yellow page to find
other agents.
The use of agents with web services is
interesting because of the benefits agents bring to
web services (Zahreddine et al, 2005). For instance,
web services are stateless and know only of
A SUCCINCT ANALYSIS OF WEB SERVICE COMPOSITION
195

themselves; conversely agents can be personalized,
adaptable, and context-aware. Also, agents can act
as a middleware for composition, manage results
and exception handling. However, it is important to
note that not every agent-based approach supports
fully automatic composition. In order to have
automatic composition, an intelligent discovery
service is required to find the right web services and
understand how to piece them together.
Work in (McIlraith et al, 2001) is one of the
initial discoveries that involve using agents and
semantic web services together. Their work has
taken the ordinary web services and wrapped them
in a DAML-S ontology and shown how agents can
successfully perform automatic web service
composition. A Plan Domain Description Language
(PDDL) is used as the artificial intelligence planner
for web service choreography. Using the control
constructs provided by DAML-S the planner decides
how and when to invoke web services. Agents
written in ConGolog perform the actual invocations.
Furthermore, using agents in composition does
have its disadvantages. Building and testing an agent
society is difficult. There are also security risks
when using agents. Firstly, protecting the host from
malicious agents is imperative. Many techniques
have been documented to solve this problem such
as: authentication and authorization (Berkovitz et al,
1998) or sandbox approaches that limit privileges.
Secondly, protecting agents from malicious hosts;
this facet of security has not been solved by software
means. Some believe the answer is in a mutually
trusted third party (Algesheimer et al, 2000). These
security fears are important set backs in software
design and are one of the reasons why agents do not
have a wide spread employment over the net.
3.4 P2P
Another approach involves the field of P2P (peer to
peer) computing such as in (Benatallah et al, 2003;
Pitoura et al, 2003; Abiteboul et al, 2002). In
particular, the P2P orchestration model Self-Serv
(Benatallah et al, 2003) is an interesting approach
towards a middleware infrastructure for web service
composition. The Self-Serv model proposes that
composition of web services through a decentralized
dynamic environment. Self-Serv brings together
elementary and composite services. An elementary
service is an individual web service that does not
rely on other web services. A composite service is
referred to as a component and aggregates more than
one web service together; business logic is expressed
as a state chart.
4 CONCLUSION
Semi-automatic compositions can only be applied in
a limited number of scenarios. The future is moving
towards a semantic web and fully automatic
compositions will only occur when semantics are
involved with web services. Automatic web service
composition does have its costs, such as in service
discovery. Searching a UDDI takes time, especially
when dealing with multiple registries. Although
efforts in the P2P field have helped, and knowing
the parties you are dealing with at design time rather
then runtime is quicker. However, this form of
composition is only appropriate if you know what
web services you need, know how to use them, and
that they rarely change. Realistically, there will
always be circumstances when services need to be
discovered at runtime especially when automatic
composition is needed, and as the number of web
services grows.
REFERENCES
Abiteboul, S., Benjelloun, O., Manolescu, I., Milo, T.,
Weber, R., 2002. Active XML: Peer-to-Peer Data and
Web Services Integration. VLDB, pp. 1087-1090.
Algesheimer, J., Cachin, C., Camenisch, J., and Karjoth,
G., 2000. Cryptographic Security for Mobile Code.
Technical Report RZ 3302 (# 93348), IBM Research.
Berkovitz, S., Guttman, J.D., and Swarup, V., 1998.
Authentication for Mobile Agents. Lecture Notes in
Computer Science, Vol. 1419.
Benatallah, B., Sheng, Q.Z., Dumas, M., 2003. The Self-
Serv Environment for Web Services Composition,
IEEE Internet Computing, Vol 7 No 1, pp. 40-48.
Curbera, F.C., Khalaf, R.Y. and Leymann, F., 2003.
Composing Web Services Using BPEL4WS, OMG
Web Services Europe 2003: Web Services for
Integrated Enterprise.
McIlraith, S., Son, T.C., and Zeng, H., 2001. Semantic
Web Services, IEEE Intelligent Systems. Special Issue
on the Semantic Web. 16(2):46-53.
Pitoura, E., Abiteboul, S., Pfoser, D., Samaras, G.,
Vazirgiannis M., 2003. DBGlobe: A Service-Oriented
P2P System for Global Computing, Sigmod Record,
SIGMOD Record 32(3): 77-82.
Peltz, C., 2003. "Web Services Orchestration and
Choreography", IEEE Computer, Vol. 36, No. 10, pp
46-52.
Zahreddine, W., and Mahmoud, Q.H., 2005. Blending
Web Services and Agents for Mobile Users. Proc. of
the 7
th
ISADS Workshop on Cooperative Computing,
Networking, and Assurance, Chengdu, China, pp. 585-
590.
ICEIS 2006 - SOFTWARE AGENTS AND INTERNET COMPUTING
196