Reliability-driven dynamic binding via feedback control

Abstract

We are concerned with software that can self-adapt to satisfy certain reliability requirements, in spite of adverse changes affecting the environment in which it is embedded. Self-adapting software architectures are heavily based on dynamic binding. The bindings among components are dynamically set as the conditions that require a self-adaptation are discovered during the system's lifetime. By adopting a suitable modeling approach, the dynamic binding problem can be formulated as a discrete-time feedback control problem, and solved with very simple techniques based on linear blocks. Doing so, reliability objectives are in turn formulated as set point tracking ones in the presence of disturbances, and attained without the need for optimization. At design time, the proposed formulation has the advantage of naturally providing system sizing clues, while at operation time, the inherent computational simplicity of the obtained controllers results in a low overhead. Finally, the formulation allows for a rigorous assessment of the achieved results in both nominal and off-design conditions for any desired operation point.