基于日志信息的故障诊断分析word论文.docx

Abstract
In the puter system, log is usually to be the first source for obtaining the system situation and diagnosing the failure of system. However, with the development of puting and cluster environment, the architecture of system and software is plicated. munication between different layer of hardware and software in the system is ing more frequency. It enhances the coupling of system and makes fault diagnosis using system logs difficult. The traditional system fault diagnosis based on log relies on artificial analysis. It is difficult to analyze system in real-time prehensive, when dealing with massive log generated by ponents in modern system. Therefore, it is a significant meaning for current management puter system that to design a log management and analysis system.
The unified management and analysis system for log (UiLog) collect fault logs gener- ated by ponent of cloud system and provide management and analysis functions. When a failure occurs, UiLog will determine the fault catalog and sort the entire fault logs in causality to help administrator for fault diagnosing. Firstly, UiLog manage all the logs in the system to ensure UiLog can obtain the information of fault. Secondly, UiLog take ad- vantage of new fault classification measure to reduce the dependency of human knowledge library. Through the Fault Keyword Matrix, UiLog can classify fault log efficiency. On the other hand, UiLog improves the traditional fault correlation analysis based on time. Using the result of fault classification to determine the size of time windows, it can improve ac- curacy of fault classification and help administrator for finding the root cause of fault.
The result of evaluation shows that UiLog can manage logs of entire system and clas- sify logs by fault type. Meanwhile, UiLog also can find root-cause of fault. For log classi- fication, the new approach improves the accuracy of fault classification to 95%. For log correlation, the truncation fault rate and collisio