Lattice Gas: An Efficient and Reusable Parallel Library Based on a Graph Partitioning Technique

We present a parallel library which can be used for any lattice gas (LG) application. A highly reusable implementation, as well as a general parallelization scheme based on graph partitioning techniques are developed. We show that the performance we obtain with our approach compares favorably with the plain, classical implementation of LG models on regular domains whereas on irregular domains, it can even be better. We propose a theoretical expression for the execution time and we validate our analysis in the case of the problem of wave propagation in urban areas.