Skip to Main Content
Summary form only given, as follows. Research in model checking is focused on increasing the size of the problems that tools can deal with. The ultimate wave has been the use of distributed computing, where a cluster of computers work together to solve the problem. In our work, we present a distributed model checker that is evolved from the Kronos tool and that can handle backwards computation of TCTL (timed computation tree logic) reachability formulae over timed automata. Our proposal, including the arguments of its correctness, is based on software architectures, using a notation adapted from C. Hofmeister et al. (1999). We find such an approach to be a natural and general way to address the development of complex tools that need to incorporate new features and optimizations as they evolve. We introduce some interesting features, such as a-priori graph partitioning (using METIS, a standard library for graph partitioning), sophisticated machinery to reach optimum performance (communication piggybacking and delayed messaging) and dead-time utilization, where every processor uses time intervals of inactivity to perform auxiliary, time-consuming tasks that will later speed up the rest of the computation. The correctness proof strategy combines an architecture evolution with the theoretical results about fix-point calculation developed by P. Cousot (1978).
Date of Conference: 25-25 May 2002