Analysis of web server performance

Abstract

Web server performance is a critical issue for sites which service a high volume of requests. This paper examines the performance of Web servers under high CPU loads. Simulations are used in conjunction with workloads which were obtained by analyzing Web logs and performance data from several real sites. Performance is significantly affected by the percentage of requests for dynamic HTML pages; dynamic HTML pages adversely affect server performance. In order to optimize performance, the number of dynamic pages should be kept as low as possible. When dynamic pages are required, techniques such as fast API's for invoking server programs and caching can be employed to keep the overhead of server programs generating the dynamic pages as low as possible. When the server is operating at or near peak capacity, there is a trade-off between average latencies and the percentage of requests rejected; performance is improved by rejecting a higher percentage of requests. For the real request distributions which we encountered, we found that Web servers should reject enough requests so that the average load on the system is 95% or less of the maximum capacity in order to prevent latencies from becoming too large.