Meta-Adaptation Goals: Leveraging Feedback Loop Requirements for Effective Self-Adaptation

Self-* systems focus on optimizing systems regarding adaptation goals, e.g. availability, that relate to properties of the managed system. While these adaptation goals are important, the feedback loop must also be considered. It is problematic if systems adapt too often, make too time-consuming changes that involve insecure transition states, or react too slowly when planning adaptations. Current approaches do not allow to enforce requirements on the adaptation loop—including sensitivity, sta-bility, and transition states. When planning adaptations, those meta-adaptation goals need to be considered and potentially traded off against traditional adaptation goals. Such a tradeoff might entail, for instance, that it may be worth it to adapt a system less frequently, although the overall availability of the system could be higher with more adaptations. We propose to explicitly consider feedback loop requirements for self-adaptive systems. Concretely, we propose using machine learning to predict quality measures based on the state of the feedback loop and the managed system. This information is used to plan for better adaptations that consider the adaptation goals and the meta-adaptation goals of the feedback loop. The explicit representation of meta-adaptation goals and adaptation goals enables the optimization of adaptation strategies, tradeoff analysis, and evaluation of adaptations.