The paper describes how to model and control performance of a trusted web server using Feedback Error Learning and H¡Þ control theory. Web server load may experience enormous variations and unpredictability. Existing web servers are difficult to achieve performance guaranties in such environment. One of challenges in web server research is how to model non-linearity computing systems. We demonstrate that Web server model can be approximated with a two tank system that includes many uncertainly components. In contrast with web server delay control that only control delay, our architecture based on H¡Þ robust control and queuing model predictor enforce server utilization and delay guarantees for different classes on virtual servers. A simulation evaluation demonstrates that H¡Þ robust control can provide robust request utilization and request response time for Internet servers.