Network Fault Management System Based On Knowledge Discovery And Realization

As modern computer networks increase in size and complexity, effective network fault management and maintenance become more important and more difficult. In order to make use of the large amount of raw management information to meet future challenges for network fault diagnosis and prediction, intelligent information processing techniques based on knowledge discovery are required urgently. In addition, autonomic computing is an exciting research direction in IT industry currently, which aims to provide self-configuration, self-optimization, self-healing and self-protection capabilities to computer systems, thus computer systems can manage themselves according to high-level objectives given by administrators.This paper focus on a new generation network fault management system, which is using the knowledge discovery techniques as basic implementation techniques, using autonomic computing model as primary architecture, and achieved following results:(1)Based on the generic control theory and the concept model of autonomic computing provided by IBM, proposed an autonomic network fault management framework, which combines the reactive autonomic approach based on feedback control with the predictive autonomic approach based on feedforward control.(2)Based on the PATRICIA theory, designed and implemented a common alarm analysis algorithm, which is able to analyze alarms and provide explanations and recommended actions under multi-vendors network environment. In addition, it is able to support any network devices added dynamically.(3)Based on ARIMA model of time series analysis, create an ARIMA model by analyzing the time series data of numeric network history data, then using the model to analyze and predict the trend of network status.(4)Discover potential correlation rules from history alarm series using correlation analysis firstly. Then, discuss and define the presentation, the creation algorithm of correlation rules. At last, discuss the method and process to predict network fault using correlation rules.