Weighted decomposition in high-performance lattice-Boltzmann simulations: are some lattice sites more equal than others?

摘要

Obtaining a good load balance is a significant challenge in scaling uplattice-Boltzmann simulations of realistic sparse problems to the exascale.Here we analyze the effect of weighted decomposition on the performance of theHemeLB lattice-Boltzmann simulation environment, when applied to sparsedomains. Prior to domain decomposition, we assign wall and in/outlet sites withincreased weights which reflect their increased computational cost. We combineour weighted decomposition with a second optimization, which is to sort thelattice sites according to a space filling curve. We tested these strategies ona sparse bifurcation and very sparse aneurysm geometry, and find that usingweights reduces calculation load imbalance by up to 85%, although the overallcommunication overhead is higher than some of our runs.