Improvements in Speed and Scalability of a DEM Code

Abstract

A number of near-optimal techniques were implemented to reduce computing times for the Discrete Element Method (DEM) code named DESOL. Among these, the following showed the largest improvements: multilevel bins, periodic rebuild, trimming and Symmetric Multiprocessor (SMP) parallelization. These improvements have led to Central Processing Unit (CPU) reduction of the order of 1:3-1:5 on scalar machines, while also showing excellent scalability up to the point of memory saturation, which on current Intel Xeon processors occurs at approximately 8 cores for double precision and 16 cores for single precision.