Abstract |
Service-Oriented Architecture (SOA) has emerged as a prominent design style that enables an IT infrastructure to allow different applications to participate in business processes, regardless of their underlying features, by encapsulating
them as platform-independent entities that become available via a certain network,
primarily the Internet. In order to effectively discover and use the most
suitable services, service description should provide a complete behavior model,
describing the inputs and preconditions that are required before execution, as
well as the outputs and effects of a successful execution. Such service specifications
are indispensable in a variety of activities, such as conformance and verification
checks, adaptation evaluation and deducing composability of services.
Service specifications rely on the expression of conditions that should hold
before and after service execution. Such specifications are prone to a family of
problems, known in the AI literature as the frame, ramification and qualification
problems. These problems deal with the succinct and flexible representation of
non-effects, indirect effects and preconditions, respectively. Research in services
has largely ignored these problems, at the same time ignoring their effects, such
as compromising the integrity and correctness of services and service compositions
and the inability to provide justification for unexpected execution results.
To address these issues, this thesis proposes the Web Service Specification
Language (WSSL), a novel, semantics-aware language for the specification and
composition of services, independent of service design models. WSSL’s foundation
is the fluent calculus, a specification language for robots that offers solutions
to the frame, ramification and qualification problems. Further language extensions
achieve three major goals: realize service composition via planning, supporting
non-deterministic constructs, such as conditionals and loops; include specification
of QoS profiles; and support partially observable service states.
To investigate WSSL’s applicability and demonstrate its benefits, we analyze
correctness of the composition extension, decidability and complexity of the underlying
theory, as well as compatibility with other related languages in service
science. Moreover, an innovative service composition and verification framework
is implemented, that advances state-of-the-art by satisfying several desirable requirements
simultaneously: ramifications and partial observability in service and
goal modeling; non-determinism in composition schemas; dynamic binding of
tasks to concrete services; explanations for unexpected behavior; QoS-awareness
through pruning and ranking techniques based on heuristics and task-specific
goals and an all-encompassing QoS aggregation method for global goals.
Experimental evaluation is performed using synthetically generated specifications
and composition goals, investigating performance scalability in terms of
execution time, as well as optimality with regard to the produced composite process.
The results show that, even in the presence of ramifications in some specifications,
functional planning is efficient for repositories up to 500 specifications.
Also, the cost of functional discovery per single service is insignificant, hence
achieving good performance even when executed for multiple candidate plans.
Finally, optimality relies mainly on defining suitable problem-specific heuristics;
thus, its success depends mostly on the expertise of the composition designer.
|