粒子法による大規模津波遡上解析に向けた拡張スライスグリッド法

抄録

When modeling the entire city with a highly detailed particle model, one must deal with an enormous number of degrees of freedom, and implementation of highly parallel calculation is indispensable. In this study, we developed a computational algorithm with high parallelization efficiency for the liquid spreading phenomenon over a flat plane problem, for later apply on tsunami run-up simulation. In particular, using a slice grid together with a dynamic load balancer as a method for dividing the domain, we propose an Expanding slice grid method to utilize maximum efficiency on memory utilization and computational speed. In this method, we gradually expand the analysis domain along with the fluid spreading. Finally, we measured the parallelization efficiency under highly parallel environment by calculation example. For the measurement, we used 256 and 4096 nodes of K computer. As a result, the calculation using 4096 nodes results in about 35 % efficiency compared with 256 nodes. It is necessary to further improve the efficiency by improving the program.