Modeling Bayesian Networks for Autonomous Diagnosis of Web Services

We took an innovative approach to service level man-agement for network enterprise systems by using inte-grated monitoring, diagnostics, and adaptation services in a service-oriented architecture. The autonomous di-agnosis for trouble-shooting of web service interrup-tions is based on Bayesian network models. In this pa-per, we present our methods for building the diagnostic models. We focus on two types of Bayesian network models of different structure complexity. Our result shows that the two-layer model outperforms the three-layer model in the applied domain. This challenges the common belief that adding unnecessary nodes in a Bayesian network and growing its structural complexity does not deteriorate performance. Hence such practice of building more complex models than necessary should be approached cautiously within the context of the ap-plied domain