We present a tool-supported approach on creating load models that can be used for performance testing. The models are created automatically from historical web access log data. The result is a stochastic model represented by Probabilistic Timed Automata (PTA) and describes how users interact with a system. Stochastic models have more advantages over traditional models that simply play back scripted or pre-recorded traces: they are easier to create and maintain and achieve higher coverage of the tested application. The purpose of these models is to mimic real user behavior as closely as possible when generating load.
Fredrik Abbors, Dragos Truscan, Tanwir Ahmad: Tool Support for Automated Workload Model Creation from Web Server Logs. TUCS Technical Reports 1112, TUCS, 2014.