Efficiency Testing of Self-adapting Systems by Learning of Event Sequences

Adding self-adaptation as a property to systems aims at improving the efficiency of this system. But there is always the possibility for adaptations going too far, which is very detrimental to the trust of users into a self-adapting system. In this paper, we present a method for testing the efficiency of a self-adapting system, more precisely the potential for inefficiencies after adaptation has taken place. Our approach is based on learning sequences of events that set the system up so that a second following learned sequence of events is reacted to very inefficiently by the system. We used this approach to evaluate a self-adapting system for solving dynamic pickup and delivery problems and our experiments show that the potential inefficiencies due to self-adaptation are smaller than the inefficiencies that the non-adapting base variant of the system is creating. Keywords-testing; learning; dynamic optimization