Identifying worst-case user scenarios for performance testing of web applications using Markov-chain workload models

Abstract The poor performance of web-based systems can negatively impact the profitability and reputation of the companies that rely on them. Finding those user scenarios which can significantly degrade the performance of a web application is very important in order to take necessary countermeasures, for instance, allocating additional resources. Furthermore, one would like to understand how the system under test performs under increased workload triggered by the worst-case user scenarios. In our previous work, we have formalized the expected behavior of the users of web applications by using probabilistic workload models and we have shown how to use such models to generate load against the system under test. As an extension, in this article, we suggest a performance space exploration approach for inferring the worst-case user scenario in a given workload model which has the potential to create the highest resource utilization on the system under test with respect to a given resource. We propose two alternative methods: one which identifies the exact worst-case user scenario of the given workload model, but it does not scale up for models with a large number of loops, and one which provides an approximate solution which, in turn, is more suitable for models with a large number of loops. We conduct several experiments to show that the identified user scenarios do provide in practice an increased resource utilization on the system under test when compared to the original models.