Application Performance Monitoring and Analyzing Based on Bayesian Network

Abstract

Application performance monitoring and analyzing system has been widely used in online applications and services, to obtain high performance and reliability. However, existing performance monitoring systems only raise an alarm when certain parameter exceeds its threshold and do not provide further analysis. This paper presents an application performance monitoring and analyzing model based on Bayesian Network, in which we discover the implied causality between performance parameters and user experience. We also design a feedback correction algorithm that can improve the validity of our model. A series of experiments and tests have demonstrated that our model provides proper analysis. With these desired properties, the performance monitoring systems with our model and algorithms can provide high performance and reliability to users.