Data Flow Repair in Web Service Orchestration at Runtime

Business process modeling languages, such as BPMN, offer an abstract view to Web service composition, where business functionalities are decomposed into activities connected through high level control and possibly data flows. Closing the gap between abstract modeling and concrete business process execution requires to solve data mismatches between producer and consumer services with specific adapters. Automatically generating data adapters as well as selecting actual services such as to minimize mismatches are major barriers to this end. In this paper, we present some new pieces of framework providing an end-to-end solution to this problem, (1) first by evaluating the level of data mismatches between candidate producer and consumer services in a business process, (2) second by selecting services in such a way to globally minimize the overall data mismatches in a process, and (3) finally by generating transformations from output to input XML data everywhere needed in the data flow. Moreover, we extend this framework to cope with the failure of selected services at run-time by choosing a replacement service that locally minimizes the data mismatches with its providers and clients in the process. Our solution is implemented and performance evaluations show its feasibility since the cost of selecting a replacement service is typically less than the one of one call to a remote Web service.