General methodology for characterizing access patterns and analyzing Web server performance

Abstract

We develop a general methodology for characterizing Web server access patterns based on a spectral analysis of finite collections of observed data from real systems. Our approach is used together with the access logs from the IBM Web site for the 1996 Olympic Games to demonstrate some of its advantages over previous methods and to analyze certain aspects of large-scale Web server performance.