An efficient Bayesian diagnosis for QoS management in service-oriented architecture

When running a business process in SOA, systems need an efficient mechanism to detect performance issues and identify root causes. In this paper, we study the Bayesian network diagnosis model to identify faulty services in a business process by monitoring a subset of services selected as evidence channels. Both local and global optimal evidence channel selection algorithms can be used to select the most informative services for runtime monitoring. In the local optimal algorithm, monitoring coverage on services is defined by the behavior similarity between individual services. In the global optimal algorithm, an iterative search is adopted to choose the service that reduces system entropy most in each round. We have implemented the Bayesian diagnosis capability in the Llama middleware. The system study shows that the new diagnosis approach can achieve a good diagnosis result for deployed business process by monitoring about 25% of the services.