Dual-processor parallelisation of symbolic probabilistic model checking

Abstract

We describe the dual-processor parallelisation of a symbolic (BDD-based) implementation of probabilistic model checking. We use multi-terminal BDDs (binary decision diagrams), which allow a compact representation of large, structured Markov chains. We show that they also provide a convenient block decomposition of the Markov chain which we use to implement a parallelised version of the Gauss-Seidel iterative method. We provide experimental results on a range of case studies to illustrate the effectiveness of the technique, observing an average speed-up of 1.8 with two processors. Furthermore, we present an optimisation for our method based on preconditioning.