Analyzing logs of the university data repository

Abstract

Identification of execution anomalies is very important for the maintenance and performance refinement of computer systems. For this purpose we can use system logs. These logs contain vast amounts of data, hence there is a great demand for techniques targeted at log analysis. The paper presents our experience with monitoring event and performance logs related to data repository operation. Having collected representative data from the monitored systems we have developed original algorithms of log analysis and problem predictions, they are based on various data mining approaches. These algorithms have been included in the implemented tools: LogMiner, FEETS, ODM. Practical significance of the developed approaches has been illustrated with some examples of exploring data repository logs. To improve the accuracy of problem diagnostics we have developed supplementary log database which can be filled in by system administrators and users.