Abstract

Performance of Web servers is critical to the success of many corporations and organizations. However, very few results have been published that quantitatively study the server behavior and identify the performance bottlenecks. In this paper we measure and analyze the behavior of the popular Apache Web server on a uniprocessor system and a 4-CPU SMP (Symmetric Multi-Processor) system running the IBM AIX operating system. Using the AIX built-in tracing facility and a trace-analysis tool, we obtained detailed information on OS kernel events and overall system activities while running Apache driven by the SPECweb96 and the WebStone benchmarks. We found that on average, Apache spends about 20-25% of the total CPU time on user code, 35-50 % on kernel system calls and 25-40 % on interrupt handling. For systems with small RAM sizes, the Web server performance is limited by the disk bandwidth. For systems with reasonably large RAM sizes, the TCP/IP stack and the network interrupt handler are the major performance bottlenecks. We notice that Apache shows similar behavior on both the uniprocessor and the SMP systems. After quantitatively identifying the performance bottlenecks, we proposed 8 techniques to improve the performance of Apache. We implemented all but one of these techniques. Our experimental results show that these techniques improve the throughput of Apache by 61%. These techniques are general purpose and can be applied to other Web servers as well. Finally, our results suggest that operating system support for directly sending data from the file system cache to the TCP/IP

Citations