| Popis: |
We have parallelized GeoClaw using OpenMP to meet the urgent need of simulating tsunami waves at near- shore from Tohoku 2011 at a grid distance around 100 meters because of no available parallel version yet by then. It took us some efforts to obtain the correct results and speed-up in the computation. With this OpenMP parallelization, we achieved over 75% of the potential speed-up on an eight core Dell Precision T7500 workstation. Recently we obtained an unreleased OpenMP version of GeoClaw from David George, who developed GeoClaw as part of his Ph.D thesis. In this paper, we will show the tactics used for merging the two OpenMP codes and for speeding up the calculation with adaptive mesh refinement (AMR). We will also show the simulated inundation of the Tohoku 2011 Tsunami waves onto the Sendai airport and Fukushima Nuclear Power Plants, over which the finest grid distance of 20 meters is achieved through 4-level AMR. Our eventual goal of this work is to scale GeoClaw over at least several tens of cores so that it can meet the urgent need for solving real-world geophysical flow problems on many-core systems, which we believe is the most efficient and feasible means for AMR in the near future. Keywords-GeoClaw; OpenMP; Adaptive Mesh Refinement; Inundation of tsunami waves; Real-world Geophysical Flows. |
