Assessing Self-Adaptation Strategies Using Cost-Benefit Analysis

Self-adaptive software systems stand out from traditional ones in that they are able to autonomously change their behavior and structure during their execution using one or more self-adaptation strategies. The main objective of such a strategy is to maintain or even improve the functionalities/qualities of the system despite uncertainty in its operational environment. To date, a number of self-adaptation strategies have been proposed-following the architectural, control-theoretic, or online search paradigm-for several application domains. However, it is still unclear when a particular self-adaptation strategy needs to be developed and when it needs to be used at runtime (when a system can choose among several available strategies). In this paper, we aim to answer the above questions by relying on the assessment of a strategy's costs (e.g., development effort, performance decrease) and benefits (e.g., re-usability, performance improvement) at design time and runtime. The main novelty is that we provide a holistic view over the return on investment of a strategy and propose that the system itself uses cost-benefit analysis to decide on which strategy to apply at runtime.