An Active Method to Building Dynamic Dependency Model for Distributed Components

Currently many research show that, in a sophisticated application system, the faults which are impossible to occur theoretically, may take place in practice. J2EE (Java 2 Platform, Enterprise Edition) distributed environment has been popularly applied to EAI (enterprise application integration). With the growth of the numbers of Jsp, Servlet and EJB components, for a specific J2EE application, it is difficult for administrators to locate the fundamental position of the faults, and delay recovering the faults. Dependency models provide the effective method to trace all possible sources of the faults from the problem vertices against the relationship edges. Bayesian network was presented in 1981 by R.Howard and J.Matheson. It has been successfully applied to fault diagnosis field. Bayesian networks provide a method to describe consequence information naturally. In this paper, we construct the dependency models of software components with the construction algorithm of Bayesian networks. Dependency models can be represented with Bayesian networks. The vertices in Bayesian networks corresponds to the vertices in dependency models, and the conditional probability expressed with the edges in Bayesian networks corresponds to the relative strength expressed with the edges in dependency models. Consequently, it is possible to develop a tool to analyze and recover the faults automatically, and be helpful to find fundamental reasons for various faults, based on dependent relations between the components in dependency models