Towards Extracting Realistic User Behavior Models

Abstract

Workloads can be characterized by intensity and user behavior. Combining multiple intensities and behaviors can be used to create workload profiles to evaluate software design and support the prediction of system utilization. The central challenge for workload profiles is their fit to real workloads and in particular the match to specific behaviors. This is especially relevant for understanding and identifying specific user groups and support workload composition by operators. In this paper, we address the identification of such realistic user behaviors utilizing domain specific attributes, evaluate the fitness of potential behavior clustering approaches, and discuss our setup to evaluate further clustering approaches.