Building dependable concurrent systems through probabilistic inference, predictive monitoring and self-adaptation

The infeasibility of statically verifying complex software is well established; in concurrent systems, the difficulty is compounded by nondeterminism and the possibility of ‘Heisenbugs’. Using runtime verification, one can not only monitor a concurrent system to check if it has violated a specification, but potentially predict future violations. However, a key challenge for runtime verification is that specifications are often incomplete. I will argue that the safety of concurrent systems could be improved by observing patterns of interaction and using probabilistic inference to capture intended coordination behavior. Actors reflecting on their choreography this way would enable deployed systems to continually improve their specifications. Mechanisms to dynamically add monitors and enforce coordination constraints during execution would then facilitate self-adaptation in concurrent systems. I will conclude by suggesting a program of research to extend runtime verification so systems an evolve robustness through such self-adaptation.