Generating Environmental Models for Testing Self-adaptive Systems

Self-adaptive systems (a.k.a. SASs) are useful but error-prone. This stems from the complexity of the interaction between a self-adaptive system and its running environment. Therefore, a testing approach of self-adaptive system has to consider the system's running environment. However, due to their poor controllability and observability, neither the real environment nor the environmental simulators could support SAS-testing effectively and efficiently. In this paper, we propose a novel approach AutoModel to generate environmental models for testing self-adaptive systems effectively. Our key insight is that a self-adaptive system's execution traces naturally encode the behavior of its running environment, especially for the logic of how the environment interacts with the system. Based on the collected execution traces, our AutoModel approach synthesizes an environmental model and learns the model's reaction logic. The derived environmental model is able to imitate the real environment's behavior in program-environment iteration. Our primitive evaluation on real-world self-adaptive systems validates the effectiveness of our AutoModel approach. The average predictive R-squared value of the generated environmental model's prediction results is 55.0%.