Periodic hidden Markov model-based workload clustering and characterization

Workload of a Web server is a complicated stochastic process with non-stationary properties. Userspsila access to Websites is governed by the activities of their daily life. Workload of servers has significant periodicities that reflect daily, weekly and seasonally effect of userspsila access. In this paper, we present a periodic hidden Markov model to characterize the stochastic behavior and the periodicity of workload. Based on this model, we present a method for clustering and classification of workload patterns. Our approach is validated against workloads collected from tens commercial Websites. This approach provides a new solution for traffic modeling and characterization, workload and performance prediction, capacity planning, and statistical anomaly detection of network intrusions.