An Approach to Predictive Analysis of Self-Adaptive Systems in Design Time

Predictive analysis methods offer the possibility of estimating the impact of design decisions, which may help in the accomplishment of operational optimal results, before the deployment of the system is made, and therefore minimizing the required maintenance effort and cost. However, current predictive methods are not effective when used on self-adaptive systems, and specifically when used on cloud environments, because of its complexity and dynamic nature. The main goal of this thesis project is to investigate different methods for the specification of adaptive systems, and to propose techniques and tools for the modeling of self-adaptive systems and environments, considering adaptation mechanisms, and approaches for the estimation of different Quality of Service (QoS) metrics that help in the analysis of the systems to be developed. Specifically, we will provide generic mechanisms for the modelling of adaptive systems and environments, the definition of metrics as transformation rules, and tools using such system specifications for their analysis. We will focus on the kind of systems and adaptation mechanisms we find in the cloud, and will evaluate our proposal on state-of-the-art cloud applications. We will present an approach of predictive analysis, based on graph transformation, which provides the capability of taking decisions about elasticity-related QoS from the definition of an adaptive model of the system and the specification of adaptation mechanisms described in a formal language to control elasticity in cloud applications. The proposed approach will enable the simulation of cloud environments and their elasticity features at design time, which allows the prediction of different QoS metrics in cloud scenarios, and provides the capability of specifying and tuning elasticity monitoring, constraints, and strategies at different levels.