Fractal Characterization of Web Workloads

Understanding the workload of Web and e-commerce sites is a fundamental step in sizing the IT infrastructure that supports these sites and in planning for their evolution so that Quality of Service (QoS) goals are met within cost constraints. This paper discusses the use of fractal-based methods to simplify and characterize Web and e-commerce workloads. Fractal clustering is quite appropriate to find sets of points that are somehow “similar” with respect to a fractal dimension. This implies that clusters do not have to be shaped as hyperspheres, as is the case with traditional clustering techniques. We apply the fractal techniques to an actual e-commerce workload and use the results to understand what customers do, what navigational patterns they follow, and to identify groups of users that have similar behavior. A comparison with results obtained with k-means analysis is also discussed. The main contributions of this work are techniques that improve the process of workload characterization.