Towards stochastic performance models for web 2.0 applications

Abstract

System performance is one of the most critical quality characteristics of Web applications which is typically expressed in response time, throughput, and utilization. These performance indicators, as well as the workload of a system, may be evaluated and analyzed by (i) modelbased or (ii) measurement-based techniques. Given the complementary benefits offered by both techniques, it seems beneficial to combine them. For this purpose we introduce a combined performance engineering approach by presenting a concise way of describing user behavior by Markov models and derive from them workloads on resources. By means of an empirical user test, we evaluate the Markov assumption for a given Web 2.0 application which is an important prerequisite for our approach.