Dynamic adaptation with distributed control in Paradigm

Adaptation of a component-based system can be achieved in the coordination modeling language Paradigm through the special component McPal. McPal regulates the propagation of new behavior and guides the changes in the components and in their coordination. Here we show how McPal may delegate part of its control to local adaptation managers, created on-the-fly, allowing for distribution of the adaptation indeed. We illustrate the approach for the well-known example of the dining philosophers problem, by modeling migration from a deadlock-prone solution to a deadlock-free and starvation-free solution without any system quiescence. The system migration goes through various stages, exhibiting a shift of control among McPal and its helpers, and changing degrees of orchestrated and choreographic collaboration. The distributed system adaptation is formally verified using the mCRL2 model checker.