On overload control through queue length for web servers

We investigate the performance of overload control through queue length for two different web server architectures. Both of them use finite queue lengths to prevent servers from being overloaded. One architecture prioritizes requests from established sessions while the other treats all requests equally. First, we introduce queueing models for these two systems. Then, we define and explain a new web server performance metric that is a function of session throughput, error rate for connection within sessions and average request response time. Finally, we use simulation to evaluate the performance of these two types of web servers. The result suggests that the benefit of request prioritization is noticeable only when the capacities of the sub-systems match each other