Statistical Model Checking of Response Times for Different System Deployments

Performance testing is becoming increasingly important for interactive systems. Evaluating their performance with respect to user expectations is complex, especially for different system deployments. Various load-testing approaches and performance-simulation methods aim at such analyses.However, these techniques have certain disadvantages, like a high testing effort for load testing, and a questionable model accuracy for simulation methods. Hence, we propose a combination of both techniques. We apply statistical model checking with a learned timed model and evaluate the results on the real system with hypothesis testing. Moreover, we check the established hypotheses of a reference system on various system deployments (configurations), like different hardware or network settings, and analyse the influence on the performance. Our method is realised with a property-based testing tool that is extended with algorithms from statistical model checking. We illustrate the feasibility of our technique with an industrial case study of a web application.