Optimal Self-Healing of Service-Oriented Systems with Incomplete Information

Self-healing management of services aims to discover, diagnose, and react to disruptions as well as maintain the Quality of Service (QoS) at a desired level for a running service oriented system. Existing approaches assume that the state of a running service-oriented system can be fully monitored. However, the dynamic nature of the Internet environment coupled with the opaque internal status of third-party services makes such an assumption no longer hold. In this paper, we address the self-healing issue in service-oriented systems via a Partially Observed Markov Decision Process (POMDP). We determine the best action to minimize the operation cost caused by the QoS failure of particular component services by computing the optimal value of the POMDP. By relying on such a flexible technology, we are able to deal with the difficulties arising from the unpredictability of external partner services and the opaqueness of their internal status. We conduct a simulation to assess the effectiveness of the proposed approach.