A self-adaptive architecture to control the performance of multi-host web servers

In modern web applications, several web sites often inhabit one single web server, asking for some software mechanisms to allocate server resources among those virtual hosts. This paper tries to study this issue from the perspective of control theory. First, solid proofs are shown to demonstrate that web servers could hardly be modeled with fixed parameters, which cast doubt on foundations of existing parametric web QoS schemes. Furthermore, we present the self-adaptive architecture for web servers, whose objective is to perform finer control over the performance while suiting varying run-time conditions of web servers. A series of tests validate the efficiency of our strategy, showing that comparing with QoS schemes based on fixed parameters, self-adaptive architecture tracks demanding request latency more closely, even when run-time environments change dramatically. In addition, our architecture requires no superfluous off-line identification of system parameters. We also point out that the self-adaptive architecture for web servers could be implemented with other various QoS facilities.