Modeling Recovery Strategies in Service-Oriented Architecture Using a Markov Decision Process

Service-oriented architecture (SOA) requires fault-tolerant implementation because the heterogeneous nature of services is likely to cause faults and failures. Focusing on the runtime re-composition and exception handling strategies for execution faults, we propose a recovery model in SOA using a Markov decision process (MDP). Various 'quality of service' (QoS) criteria and possible recovery strategies can be incorporated into our model to determine the optimal policy, which entails cost optimization in service selection. We show how a typical SOA scenario can be translated into our model and how an optimal policy can be determined. Analytical results reveal the usefulness of our approach as compared to sole consideration of service cost. We also analyze the rationale for the selection of the optimal policy.