View article

Autonomic control of a service system can take advantage of a performance model only if a way can be found to track the changes in the system. A Kalman Filter provides a framework for integrating various kinds of measured data, and for tracking changes in any time-varying system. This work evaluates the effectiveness of such a filter in tracking changes in performance parameters of a software system that occur at different rates and amplitudes. The time-varying system is a Web application deployed in a data centre with layered queuing resources, in which parameter variations happen at random instants. The tracking filter is based on a layered queuing model of this system, with parameters representing CPU demands and the user load intensity. Experiments were performed to evaluate the effectiveness of the filter in tracking the changes, and the requirements for the filter settings for fast and slow variations in the parameters. The target application is autonomic control of a service centre.