Optimizing Composite Web Services through Parallelization of Service Invocations

Current standardization efforts on web service interoperability provide ways to compose web services
distributed across heterogeneous network environments. Unfortunately, this has been attained at the
expense of efficiency, that is, substantial SOAP communication overheads have negative impacts on
performance of web services. One promising way to resolve this and improve performance is to hide
communication latency by means of \emph{parallelization of service invocations within a composite
web service}. However, this entails the difficulty of guaranteeing parallelization will not affect
the result of computation. Since components of a composite web service often have references to
each other and potentially interact with each other, elaborate dependency analyses could have to be
required to check validity of parallelization.
As a practical solution to this, we have developed a new mechanism to detect parallelizable service
invocations in a composite web service based on its \emph{behavioral analysis}. Our mechanism
exploits static analysis techniques to keep track of behavioral states of web service components
through execution steps and statically detects whether service invocations may have dependencies
with each other at corresponding states. The challenge here is that preside state-tracking would
still cost impractically high. To ameliorate this, our mechanism transforms behavioral
specifications of web services into simpler forms that still keep enough informations for analyzing
their dependencies. Evaluation shows strong potential of our mechanism: it substantially improves
performance of composite web services, including a real-world example such as portfolio optimization
in financial engineering.

By: N. Sato and S. Saito and K. Mitsui

Published in: RT0459 in 2007

This Research Report is not available electronically. Please request a copy from the contact listed below. IBM employees should contact ITIRC for a copy.

Questions about this service can be mailed to reports@us.ibm.com .