Machine Learning Techniques for Predicting Web Server Anomalies

This paper describes an approach for identification of internal web server system anomalies affecting the quality of service. We assume that the problems are due to system resource starvation. We observe the response time of a web server while under various artificial workload. Simultaneously we collect data on several system resource parameters. Supervised machine learning based on regularization is done to correlate the high response time with observed system data. The research described is done with artificial workload, but we argue that the approach is applicable for any running web server. This type of analysis could be useful in web server, operating system or virtual machine rejuvenation.