An Implementation and Optimization of Lattice Boltzmann Method Based on the Multi-Node CPU+MIC Heterogeneous Architecture

摘要

The lattice Boltzmann method (LBM) is a computational method based on kinetic description of fluid mechanics. The Intel Many Integrated Core (MIC) architecture designed for highly parallel computing has not been fully evaluated for LBM. This paper proposes a hybrid MPI/OpenMP parallel model of LBM based on CPU+MIC heterogeneous architecture, using the two-dimensional partitioning method for data division and offload, MPI, OpenMP. Furthermore, we use various methods to optimize the algorithm. By the numerical experiments in the cluster of Shanghai University, the presented hybrid parallel models has the better performance and scalability on the multi-node CPU+MIC heterogeneous architecture.