A requirements-based approach for the design of adaptive systems

Complexity is now one of the major challenges for the IT industry [1]. Systems might become too complex to be managed by humans and, thus, will have to be self-managed: Self-configure themselves for operation, self-protect from attacks, self-heal from errors and self-tune for optimal performance [2]. (Self-)Adaptive systems evaluate their own behavior and change it when the evaluation indicates that it is not accomplishing the software's purpose or when better functionality and performance are possible [3]. To that end, we need to monitor the behavior of the running system and compare it to an explicit formulation of requirements and domain assumptions [4]. Feedback loops (e.g., the MAPE loop [2]) constitute an architectural solution for this and, as proposed by past research [5], should be a first class citizen in the design of such systems. We advocate that adaptive systems should be designed this way from as early as Requirements Engineering and that reasoning over requirements is fundamental for run-time adaptation. We therefore propose an approach for the design of adaptive systems based on requirements and inspired in control theory [6]. Our proposal is goal-oriented and targets softwareintensive socio-technical systems [7], in an attempt to integrate control-loop approaches with decentralized agents inspired approaches [8]. Our final objective is a set of extensions to state-of-the-art goal-oriented modeling languages that allow practitioners to clearly specify the requirements of adaptive systems and a run-time framework that helps developers implement such requirements. In this 2-page abstract paper, we summarize this approach.