Reputation-based Reliability Prediction of Service Compositions

Today, the concept of service oriented architectures provides a way of building integrated solutions out of existing services. To this end, services from different providers are composed using advanced orchestration and choreography techniques. However, while this principle allows for greater flexibility at a smaller cost, the use of third party services also includes a risk: Deployed services might not work as claimed by their providers. In this paper, we propose a technique for analyzing the expected reliability of service compositions based on ratings given by (previous) service users. Every service thereby comes with a reputation, and the analysis computes an overall reliability of a service composition from the reputations of its constituent services. The proposed model-driven approach proceeds by translating statechart models of service compositions into input for a probabilistic model checker (PRISM) using state-of-the-art model transformations techniques. The approach has been implemented as an Eclipse plug-in and is fully compliant with UML.