Lattice-Boltzmann hydrodynamics on parallel systems

Realistic lattice-Boltzmann simulations often require large amounts of computational resources and are therefore executed on parallel systems. Generally, parallelization is based on one- and two-dimensional decomposition of the computational grid in equal subvolumes, and load balancing is completely ignored for simplicity. Besides reviewing the existing parallelization strategies we report here a new approach based on the Orthogonal Recursive Bisection (ORB) method. To illustrate the different decomposition methods, two realistic applications were simulated, namely fluid flow in random fibre networks and flow in a centrifugal elutriation chamber. For heterogeneously distributed workloads, the ORB method is found to be 12 to 60% more efficient compared to traditional parallelization strategies. It is shown that high parallel efficiencies can be obtained for both homogeneously and heterogeneously distributed workloads, thus supporting efficient simulations of a variety of realistic systems.