A hybrid model for efficient decision-making in self-adaptive systems

Abstract

Engineering self-adaptive systems to guarantee the required quality properties is challenging and particularly in presence of uncertainties. Such uncertainties may occur in a variety of situations, ranging from variations in the system’s operating environment to ambiguity while selecting the appropriate adaptation option. Formal methods provide a rigorous means to specify and verify the behavior of self-adaptive systems. They are applied both during system design and at runtime to provide guarantees on the required properties of self-adaptive systems. However, existing approaches generally use exhaustive verification at runtime to pick adaptation options and achieve adaptation objectives, which is time and resource consuming. Aiming to tackle this shortcoming, we target a twofold objective. Firstly, we reduce the adaptation space, and then we predict the impact of each adaptation plan on the rest of system qualities, to assist the decision-making process in determining the most suitable adaptation plans and side effects. An Adaptation-space Reducer component is added to the analyser element; it uses deep learning to reduce the adaptation space. Furthermore, the planner element has been extended with a Decision Impact Predictor component, which employs quantitative analysis to forecast the impact of a decision. The DLA4EDM is defined as an approach for providing self-adaptive systems (SASs) with an efficient decision-making process. Our approach is applied on a self-adaptive Internet of Things application and the obtained results are compared to those of other approaches. Results show that the adaptation space is reduced by 97.57%, and the error rate in the decision-making is very low. Reducing the adaptation space and resolving uncertainties to be faced in the decision-making of self-adaptive systems contribute considerably to enhance the efficiency and quality of the adaptation process and hence ensure that the quality requirements are met. Evaluating the impact of the identified adaptation plans on the obtained guarantees ensures that the system is effectively operational.