Exploiting Queuing Networks to Model and Assess the Performance of Self-Adaptive Software Systems: A Survey

Abstract

Abstract Self-adaptation has emerged as a primary concern in the context of modern software systems, due to the high dynamicity of the environments where they operate, which implies the need for such systems to properly face significant degrees of uncertainty. To this aim, much work has been done, mainly by coupling autonomic managers to the managed subsystem which perceives and affects the environment through its sensors and actuators, respectively. Such coupling often results into MAPE-K feedback loop(s), i.e. a Knowledge (K)-based architectural model that divides the adaptation process into four activities, namely Monitor (M), Analyze (A), Plan (P) and Execute (E). Performance modeling notations, analysis methods and tools, have been exploited and coupled to other kinds of techniques (e.g. control theory, machine learning) for modeling and assessing the performance of autonomic managers, possibly aimed at supporting the identification of more convenient architectural alternatives. Since moving in such a big arena is not trivial and it is easy to be overwhelmed, in this literature survey, we focus on a particular performance modeling paradigm, namely Queuing Networks, with the aim of clarifying the state-of-art in exploiting such a notation to model and assess performance of Self-Adaptive Software Systems. We conclude by bringing out some research opportunities that may be worth exploring in the near future.