Designing Self-Adaptive Systems through Models at Run-time

Abstract

Adaptability is emerging as an underlying capability of software systems. There is an increasing demand for systems that dynamically adapt their behavior at run-time in response to changes in the surrounding physical environment, without or with minimal human intervention. These changes that occur while the system is in operation require the system adaptation to occur at run-time. Therefore, new challenges about how to specify, design, verify and realize these systems arise. Previous research has applied different engineering techniques to build dynamically adaptive systems. These approaches have obtained successful frameworks to support dynamic reconfiguration at run-time. However they lack dedicated design techniques or methodologies to systematically develop self-adaptive systems. The implications and repercussions of run-time reconfigurations are very difficult to foresee and control at design time. Thus, this master thesis provides a model-based method for the systematic design of trustworthy self-adaptive systems. The purpose of the method is (1) to analyze run-time reconfiguration effects at design time, and (2) to automatically refine reconfigurations to avoid faulty situations through execution. Potential design issues and guidelines to assist engineers in the design of self-adaptive systems have been highlighted. We have identified a series of design properties (Safe Reconfigurations and Reachability, Redundancy, Reversibility, Contextual Consistency and Contextual Determinism) that capture assertions about desirable behavior of the system. For each property, we have defined refinements to automatically incorporate interesting behavior issues in self-adaptive system specifications. We believe that dealing with these properties is essential for reliable systems as a next step in obtaining self-adaptive systems that fulfill many of the user's needs. We provide a set of tools that support the design method proposed. These tools ease the modeling of self-adaptive systems, and automate the analysis and refinement of reconfigurations. To evaluate the approach, we have applied the design method in a Smart Hotel self-adaptive system. This case study has demonstrated the difficulty to foresee the behavior of self-adaptive systems at design time. The proposed approach significantly enhances the ability to develop and maintain self-adaptive systems, by providing mechanisms to analyze run-time reconfiguration effects and automatically refine specifications before execution.