Abstract
Understanding the user behavior is a need to analyze the performance and the scalability of Web servers. This knowledge is used, for instance, to build workload generators that help evaluating the performance of those servers. Current workload generators are typically memory-less, being unable to mimic actual user interaction with the system. In this work we propose a hierarchical characterization and simulation model focused on the user behavior, named USAR. We use the latency and inter-arrival time of the requests to model user actions, which are the basis of our model. We validate this model through a proxy-cache server case study, where we perform the characterization and construct a user behavior simulator. We foresee from the results the possibility to generate more realistic workloads.