FUSION: A Learning-Based Approach for Engineering Self-Adaptive Software Systems

Self-adaptive software systems are capable of adjusting their behavior at run-time to achieve certain functional or QoS goals. Such systems typically employ analytical models specified at design-time to assess their characteristics at run-time and make the appropriate adaptation decisions. However, prior to system’s deployment, engineers often cannot foresee the changes in the environment, requirements, and system’s operation profile. Therefore, any analytical model used in this setting relies on certain underlying assumptions that if not held at run-time make both the analysis and hence the adaptation decisions inaccurate. In this paper, we present and evaluate FeatUre-oriented SelfadaptatION (FUSION) framework, which aims to solve this problem by learning the impact of adaptation decisions on the system’s goals. The framework not only allows for automatic online fine-tuning of the adaptation logic to unanticipated conditions, but also reduces the upfront effort required for building such systems.