A statistical approach to predictive detection

Service providers typically define quality of service problems using thresh old tests, such as "Are HTTP operations greater than 12 per second on serve r XYZ?" Herein, we estimate the probability of threshold violations for spe cific times in the future. We model the threshold metric (e.g., HTTP operat ions per second) at two levels: (1) non-stationary behavior las is done in workload forecasting for capacity planning) and (2) stationary, time-serial dependencies. Our approach is assessed using simulation experiments and me asurements of a production Web server. For both assessments, the probabilit ies of threshold violations produced by our approach lie well within two st andard deviations of the measured fraction of threshold violations. (C) 200 1 Elsevier Science B.V. All rights reserved.