Considering Transient Effects of Self-Adaptations in Model-Driven Performance Analyses

Abstract

Self-adaptive systems reconfigure themselves to meet requirements under changing user load. Model-driven performance analyses for self-adaptive systems enable software architects to evaluate whether a self-adaptive system meets requirements under varying user load. It is essential to the efficiency of a self-adaptive system that it adapts its configuration at the right time. The effectiveness of adaptations depends not only on the time when an adaptation decision is made but also on its execution time. The execution of adaptations can cause additional stress on the system. This can further deteriorate system performance. Existing model-driven analyses do not consider these transient effects. We present an approach that enables systematic modeling and analysis of transient effects in software performance analyses. We apply our approach to a horizontally scalable media hosting application. By considering the transient effects of scale-outs we were able to increase prediction accuracy for response times of the applications services. Further experiments demonstrated that our approach enables detection and resolution of design deficiencies of self-adaptive systems.