Maximizing the Performance of the Weather Research and Forecast Model over the Hawaiian Islands

The Hawaiian Islands consist of dramatic terrain changes over short distances, resulting in a variety of microclimates in close proximity. To handle these challenging conditions, weather models must be run at very fine vertical and horizontal resolutions to produce accurate forecasts. Computational demands require Weather Research and Forecasting (WRF) to be executed in parallel on the Maui High Performance Computing Center's Mana system, a Power Edge M610 Linux cluster. This machine has 1,152 compute nodes, each with two 2.8GHz quad-core Intel^® Nehalem processors and 24GB RAM. Realizing maximum performance on Mana relied on the determination of an optimal number of cores to use per socket, the efficiency of an Message Passing Interface (MPI)-only implementation, an optimal set of parameters for adaptive time-stepping, a way to meet the strict stability requirements necessary for Hawaii, effective choices for processor and memory affinity, and parallel automation techniques for producing forecast imagery.