Performance Study and Capacity Planning of a Web Server

Using a Workload Generator

Bao Quoc Vu

Jin Fan

The objective of this project is to study the performance of a Web server, installed with Linux (Apache), in relation to the number of hits and the properties (such as CPUs and disks) of the system. The system is simulated by using workload generators installed in client sides. Workload generators are used to simulate a number of users making requests to the web server. Two kinds of users are supposed to have in the system: LAN and WAN users. The workload is partitioned to small, medium, and large components based on heavy-tail distributions of file sizes. The minimal spanning tree (MST) algorithm is employed to cluster the workload. The system is modeled as multi-classes open queuing networks (QN) using component-level modeling technique. The model is used to predict the server performance by changing the workload, CPU speed, number of CPUs, and number of disks.