Using a Progressive Model for Performance Diagnosis of Distributed System

Abstract

It is important to keep an information system work properly with efficient performance in enterprise environment. The heterogeneous compositions and complex architectures of distributed systems make performance diagnosis difficult. When the system performance goes down, how to find out the root cause of inefficiency is a challenging task. Basically, the performance diagnosis is the iterative processes which should be repeatedly performed until the problem is solved. In this paper, we propose the three-phase progressive model for constructing the performance diagnosis system. The main purpose of our model is to acquire experts\' knowledge in performance diagnosis, and transform it into rule bases to perform performance diagnosis and locate the possible issue. By using this model, we further construct a prototype system to assist IT department employees to solve the problem progressively in trouble shooting. Two online systems in enterprise environment are adopted to evaluate the performance of the progressive model. The experimental results show that our prototype system can find out the root cause efficiently when system goes wrong.