Information Technology Reference
In-Depth Information
1.3 WORKFLOWDESIGN: STATE OF THE ART
This topic focuses on the design (i.e., build-time) aspect of service-
based workflow systems. This section surveys the related work and
categorizes them into five areas, that is, automatic service composition,
mediation-aided service composition, verification of service-based
workflows, support for decentralized execution, and scientific work-
flows. This categorization does not imply that the first four areas are
related to business workflows and the last one to scientific ones. Instead,
it reflects the fact that the research on business workflows has a longer
history and more comprehensive literature coverage. Many of the
studies in the first four categories are applicable to both business
and scientific workflows, although most of them were originally
targeted at business workflows. On the other hand, we have a separate
subsection for scientific workflows to discuss their specific topics.
1.3.1 Automatic Service Composition
Composability and reusability are among the eight principles of SOA, as
proposed by Erl [3]. Services need to be reusable to make sense for their
presence, and need to be composable so as to be reusable. Methods to
compose services in a full or semiautomaticway play an important role in
SOA, due to the large number of candidate services and the complexity
that is required to perform such a composition. Manual composition can
be tedious, error-prone, and more importantly, not able to yield satisfac-
tory solutions in a timely manner. Automatic service composition means
that given a goal or an abstract process, a desired composite service that
consists of the existing services is constructed in an algorithmic way.
Automatic service composition methods can be classified into two
categories, that is, planning based [95] and optimization based [43,44].
The former transforms the service composition problem into an
AI-planning method, in which the goal state of the composition is given
and a chain of service invocations is constructed from the initial state to
reach it. They mainly concern the behavior, that is, the sequence of
service invocations, of a composite service.
The optimization-based methods mainly concern the nonfunctional
aspect, that is, QoS of a composite service [44] after the desired function
is fulfilled. QoS constraints include cost, time, availability, and reli-
ability. Optimization methods such as linear programming [43] and
