Mining and improving composite web services recovery mechanisms

Ensuring composite services reliability is a challenging problem. Indeed, due to the inherent autonomy and heterogeneity of Web services, it is difficult to predict and reason about the behavior of the overall composite service (CS). Generally, previous approaches develop, using their modeling formalisms, a set of techniques to analyze the composition model and check "correctness" properties. Although powerful, these approaches may fail in some cases in order to ensure CS-reliable executions, even if they formally validate its composition model. This is because properties specified in the studied composition model remain assumptions that may not coincide with the reality (i.e., effective CS executions). Sharing the same issue, we present a reengineering approach that starts from CS executions log to improve its recovery mechanisms. Basically, we propose a set of mining techniques to discover CS transactional behavior from an event-based log. Then, based on this mining step, we use a set of rules in order to improve its reliability.