Abstract: A flexible and efficient method of service monitoring and prediction are presented. The time correlation is used to model the service behavior, the discrete Fourier transform is used to determine the reasonable monitoring period and the Bayesian theory is used to predict the abnormal behavior. In the end, a set of measurement data is used to show the validity of the method.

Key words: service measurement, service model, performance prediction