A Stochastic Petri-net-based Approach For Analysis Of BPEL-based Web Service Composition

Web service provides an effective way for sharing data and functions on the Internet. It provides a good solution for the interoperability and integration of heterogeneous applications in a loose-coupled environment.However, single function supplied by web service is limited. It is necessary to composite web service, in order to share web service sufficiently. At present, WS-BPEL (business process execution language) is widely accepted as the most prominent way of building service-oriented distributed business processes and orchestrating composite web service. There are some wicked characteristics of web service composition, such as autonomy, heterogeneity and dynamic. These characteristics may cause web service unsteady. Therefore, it is necessary to research in the feasibility of service composition, which will promote the development needs of web service technology.In this paper, we introduce a stochastic approach to evaluate the quality of BPEL service composition employing the general stochastic Petri net (GSPN) as the intermediate model.First, the abstract of web service composition. In order to get the general stochastic Petri net (GSPN) as the intermediate model from BPEL service composition, we need to extract the key factor of web service composition's service quality.Second, we introduce translation rules to map activities (send, reply and invoke), routing patterns (sequence, flow, pick, switch, while) and other constructs (link, scope, fault handling, event handling and compensation) into GSPN fragments. Also, we need to concern the factor which will influence the processes of internal activities evolution.Third, based on probabilistic transition matrix derived from GSPN, we present analytical methods to evaluate three metrics of quality (EPNCT, PNCP and EONC).Last, in order to verify the accuracy of theoretical results and the correctness of transformation rules, we employ ACTIVE BPEL execution engine to execute actual WS-BPEL samples and obtain experimental results. In the case study, we obtain experimental data and conduct a confidence interval analysis to validate the feasibility and good accuracy of the translation based approach.