Efficient Web Service Composition : from Signature - Level to Behavioral Description - Level

Web services are software systems designed to support machine-to-machine interoperation over the Web. Many researches have been carried out for web service standards, and these efforts have significantly improved functionalities of Service Oriented Architecture (SOA) significantly. However, there still remain a number of research challenges. One of the remaining challenges is the web service composition (WSC) problem, i.e., when a single web service does not satisfy a given requirement, one wants to automatically combine web services to satisfy the requirement entirely. In this dissertation, we tackle this WSC problem in three levels, i.e., a signature level, a behavior description level and a QoS description level based on web service descriptions. First, for a signature-level approach where each web service is described by its signature in WSDL, we first analyze the topological landscape of a web service network formed by real-world web services. We then propose a SAT-based algorithm based on the analysis. Second, for web services that provide behavioral descriptions in addition to signatures, we first define a realistic model for the WSC problem, and investigate the computational complexities for the composition of web services on restricted (i.e., with full observation) and general cases (i.e., with partial observation). We then prove that the WSC problem with full observation is EXP-hard and the WSC problem with partial observation is 2-EXP-hard. To solve these high complexities, we also propose approximation-based algorithms using abstraction and refinement. Third, the previous two approaches consider only functional requirements specified in WSDL or BPEL. However, non-functional ones, such as Quality of Services (QoS) constraints, help clients select a service provider with good quality. In this case, the main aim of the WSC problem is to find a composite web service which satisfies a given complicated task with the optimal QoS value, which is called QoSaware WSC problem. We first propose to apply anytime algorithm based on beam stack search to the QoS-aware WSC problem. Moreover, to improve the basic anytime algorithm, we propose dynamic beam width with more heuristics, i.e., short backtracking and upper bound propagation.