dmcG : A Distributed Symbolic Model Checker Based on GreatSPN

We encountered some limits when using the GreatSPN model checker on life-size models, both in time and space complexity. Even when the exponential blow-up of state space size is adequately handled by the tool thanks to the use of a canonization function that allows to exploit system symmetries, time complexity becomes critical. Indeed the canonization procedure is computationally expensive, and verification time for a single property may exceed 2 days (without exhausting memory). Using the GreatSPN model-checking core, we have built a distributed modelchecker, dmcG, to benefit from the aggregated resources of a cluster. We built this distributed version using a flexible software architecture dedicated to parallel and distributed model-checking, thus allowing full reuse of GreatSPN source code at a low development cost. We report performances on several specifications that show we reach the theoretical linear speedup w.r.t. the number of nodes. Furthermore, through intensive use of multi-threading, performances on multiprocessors architectures reach a speedup linear to the number of processors.