Design Of Parallel Algorithms For Lattice Boltzmann Method And Numerical Simulations

Large scale computing is well recognized as the third significant method besides experimental and theoretical approaches. However, because of its hardware limitations, a single computer does not have the capability to satisfy the requirements of the large scale computing. High Performance Computing (HPC) becomes the only approach for it. As we know, simulation of fluid flows is always one of the most important and challenging branches of HPC. Lattice Boltzmann method, emerging as a new numerical technique based on kinetic theory, is adapted to simulate fluid flows in recent years. This method is simple, and easy implementation of boundary conditions, intrinsically parallel. Significant progress has been achieved in employing the lattice Boltzmann equation for fluid dynamics and modeling complex physics in fluids recently, there still exist a lack of studies on High Performance Computing using LB method. In order to apply LBM more efficiently, the author compares its parallel capacity under different communication models and different region partitions.We discuss the parallel implement using LBM method under RedHat Linux environment. Based on ANSIC and MPI, we workout the parallel process of blocking communication mode and non-blocking communication mode about cavity flow problem, and see about its parallel ability under different region partitions. A new phenomenon is found while the grid is refined. With parallel algorithms, we observe the periodic phenomena of 2-D cavity flow at Re=10000. A new phenomenon in 3-D cavity flow is discovered, and the forecast in our previous work is proved true.