Abstract: The linear time invariant (LTI) model for Apache Web server was identified experimentally. This model was used to describe the relationship between connection delay ratio and service thread ratio for two classes of Web client connections. Based on this model,a controller was designed to implement proportional delay guarantee by adjusting the number of service thread for different class of connections. An extension for multiple connection classes was implemented by using multiple controllers. Simulation results indicated that, Apache Web server in the closed-loop system gua ranteeds the proportional delay for different connection classes quite well, even if the number of concurrent client connections changes abruptly under heavy load conditions.

Key words: Web Server, Queuing theory, Feedback control, Proportional delay guarantee