Parallel Performance and Optimization of the Lattice Boltzmann Method Software Palabos Using CUDA

Abstract

The open-source fluid dynamics software Palabos based on the Lattice Boltzmann Method (LBM) has been widely used in porous media, biological fluids, free interfaces and other physical problems. Palabos has excellent MPI parallel performance and can complete large scale computation of computational fluid dynamics. To realize the heterogeneous parallelism of Palabos, we test the performance of Palabos on a large-scale simulation on a general-purpose cluster at first. The experimental results show that Palabos has a good performance in 16000 MPI processes. Then, we designed a CUDA parallel optimization algorithm for the case of cavity flow according to address mapping and shared memory optimization. Numerical experiments results show that the speedup ratio can achieve about a 1.5x acceleration ratio when the number of the grid is 512×512×512.