Performance evaluation of a consensus algorithm with Petri nets

This paper presents an application of Hierarchical Coloured Timed Petri Nets to the modeling, the evaluation and the improvement of the performance of a distributed consensus algorithm. The asynchronous system model is augmented with Failure Suspectors, in order to overcome the impossibility of reaching consensus in the presence of crash failures. We consider here two different implementations of the Failure Suspectors: a general implementation that shows poor performances, and an implementation optimized for the execution of the consensus. The termination time of the consensus algorithm is determined by the communications cost, i.e., by the timing characteristics of the network used for exchanging messages. In other words, the network latency is taken into account in the system model. By simulating the consensus Petri net model it is possible to provide probabilistic guarantees on the termination time of the consensus and to compare the performances of the different implementations of the Failure Suspectors.