Scaling Benchmark of ESyS-Particle for Elastic Wave Propagation Simulations

Abstract

The Discrete Element Method (DEM) is a popular particle-based numerical method for simulating geophysical processes including earthquakes, rock breakage and granular flow. Often simulations consisting of thousands of particles have insufficient resolution to reproduce the micromechanics of many geophysical processes, requiring millions of particles in some instances. The high computational expense of the DEM precludes execution of such problem sizes on desktop PCs. ESyS-Particle is a parallel implementation of the DEM, designed for execution on cluster supercomputers. Three-dimensional spatial domain decomposition is implemented using the MPI for interprocess communications. We present results of scaling benchmarks in which problem size per worker remains constant. As the number of workers increases from 27 to 1000, execution time remains near-constant, permitting simulations of 8.7M particles in approximately the same real time as simulations comprising 240K particles.