Assessing the effectiveness of automated service composition

Abstract

Background: Service Composition is an important feature of Service Oriented Computing, yet it remains mostly a manual process. Given the dynamic and decentralized nature of services, manual composition is a complex undertaking. Proposals to automate this process exist, but suffer from practical problems that hinder their implementation. Methods: In this paper, we introduce a pragmatic approach where we reverse engineer a service composition repository to obtain the necessary information for automated solutions to work. We then evaluate the quality of the automated compositions based on their similarity to the ones written manually. A classic planning algorithm was adapted in order to generate solutions closer to those expected by developers. Results: The use of classical planning tools is too time-consuming for agile development scenarios. A simplified, tailored implementation can be orders of magnitude faster than a generic planner, which suggests that expressive power may need to be sacrificed in favor of usability. Our evaluation showed that ensuring the adherence of the solution to the initial specification by enforcing the use of all input parameters was capable of significantly increasing the quality of the solutions. Conclusions: It is possible to increase the quality of automated composition by applying planning algorithms specially crafted for the service composition task. Comparisons with automated planning tools highlight the effectiveness of our proposal.