Performance Evaluation of Hybrid MPI/OpenMP Implementation of a Lattice Boltzmann Application on Multicore Systems

The Lattice Boltzmann method is a novel method used in lieu of the traditionally used Navier-Strokes equation for the simulation of fluid flows. In our paper, we discuss how we implemented a hybrid (OpenMP/MPI) parallel Lattice Boltzmann application based on an existing MPI-only version. Then we use two multicore systems – Hydra IBM p5575 from Texas A&M and Dori from Virginia Tech – to collect the execution times and power consumptions of the MPI version of the Lattice Boltzmann application and the hybrid version that we evaluate by increasing the number of cores for each problem size. This will help to determine which one of the two parallel programming paradigms performs better. The results show that the original MPI-only application performs better than our hybrid application on smaller numbers of cores, and the hybrid application performs better on larger numbers of cores.