Partitioning of Complex Discrete Models for Highly Scalable Simulations

Abstract

The need for more and more accurate simulations of groups of autonomous beings directs the attention of researchers towards the ways of parallelizing simulation algorithms. Parallel execution of discrete simulation models update methods requires their division between workers. Existing methods used for grid division aim at providing equal areas of fragments and minimizing the length of created borders. However, in real life simulations, other factors also have to be considered. In this paper we present a method for grid partitioning, which also allows defining indivisible areas, considers complex shapes of real-life environments and supports division suitable for defined architecture of nodes and cores. The method is evaluated using several scenarios, providing satisfactory results.