Improving log-based fault diagnosis by log classification

Abstract

In modern computer systems, system event logs have always been the primary source for checking the system status. As computer systems become more complex, such as cloud computing systems, the interaction among software and hardware is increasingly frequently. These components will generate enormous log information, including running reports and fault information. The massive data is a great challenge for analysis with manual method. In this paper, we implement a log management and analysis system, which can assist system administrators to understand the real-time status of the entire system, classify logs into different fault types, and determine the root cause of the faults. In addition, we improve the existing fault correlation analysis method based on the results of system log classification. We apply the log management and analysis system to cloud computing environment for evaluation. The results show that our system can classify fault logs effectively and automatically. By using the proposed system, administrators can easily detect the root cause of faults. © 2014 IFIP International Federation for Information Processing.