Adaptable Configuration of Decentralized Monitors

Prominent challenges in runtime verification of a distributed system are the correct placement, configuration, and coordination of the monitoring nodes. This work considers state-of-the-art decentralized monitoring practices and proposes a framework to recommend efficient configurations of the monitoring system depending on the target specification. Our approach aims to optimize communication over several features (e.g., minimizing the number of messages exchanged, the number of computations happening overall, etc.) in contexts where finding an efficient communication strategy requires slow simulations. We optimize by training multiple machine learning models from simulations combining traces, formulae, and systems of different sizes. The experimental results show that the developed model can reliably suggest the best configuration strategy in a few nanoseconds, contrary to the minutes or possibly hours required by direct simulations that would be impractical at runtime.