Publication

Nowadays, most of the approaches supporting self-adaptive systems (SASs) rely on static feedback control loops, for managing their adaptation process. One of the most popular feedback loops is the MAPE-K loop. In this loop, the Monitor element plays a crucial role since the quality of the monitoring data (e.g., timeliness, freshness, accuracy, availability, etc.) affects directly the performance of the rest of the elements of the loop, and in consequence the quality of the resulting adaptation decisions. Assuming static feedback loops implies that the structure and behavior of the elements of the loop should be determined at design-time and cannot change at runtime, i.e., in the case of the Monitor, systems’ owners should know everything to be monitored at design time. If that the case, current self-adaptive systems would not be able to react to unpredictable runtime events such as faults or changing requirements. Motivated by this fact, in this thesis, we address the automatic runtime adaptation of SASs’ feedback control loops, particularly the Monitor element, in order to respond to changes in the systems, the environment and the elements of the loop themselves. Concretely, we have presented HAFLoop, an architectural proposal for supporting the adaptation of the MAPE-K loop at runtime. We have identified open research challenges affecting SASs’ and feedback loops’ adaptation and analyzed whether and how existing approaches address those challenges. We have also studied how state-of-the-art approaches support adaptive monitoring in current monitoring systems. Although great efforts have been done for supporting the adaptation of feedback loops in SASs’, none of the state-of-the-art solutions satisfactorily addresses all the open research challenges. HAFLoop, in conjunction with its implementation in the form of a framework named HAFLoop4J, is a generic and reusable solution that easies the design and development of adaptive feedback loops, from higher to lower levels. Our solution enables loops to support different types of adaptation in a variety of settings. HAFLoop has been evaluated in different scenarios and in both simulation and real environments. The evaluations of HAFLoop have been conducted in the domain of smart vehicles with very promising results.