Dynamic load balancing of Lattice Boltzmann free-surface fluid animations

Abstract

We investigate the use of dynamic load balancing for more efficient parallel Lattice Boltzmann Method (LBM) Free Surface simulations. Our aim is to produce highly detailed fluid simulations with large grid sizes and without the use of optimisation techniques, such as adaptive grids, which may impact on simulation quality. We divide the problem into separate simulation chunks, which can then be distributed over multiple parallel processors. Due to the purely local grid interaction of the LBM, the algorithm parallelises well. However, the highly dynamic nature of typical scenes means that there is an unbalanced distribution of the fluid across the processors. Our proposed Dynamic Load Balancing strategy seeks to improve the efficiency of the simulation by measuring computation and communication times and adjusting the fluid distribution accordingly.