Performance of Parallel Molecular Dynamics Algorithm with Ewald Coulomb

B.B. Karki, H. Kikuchi, and S. Saini (USA)

Keywords

Molecular dynamics, Parallel algorithms, Largescale simulations, Polycrystalline materials

Abstract

We have reported important benchmark tests of the parallel molecular dynamics algorithm, which treats the long-range electrostatic interactions with the Ewald summation method. The algorithm has recently been implemented to perform large-scale atomistic simulations of polycrystalline materials. It uses the well-known spatial decomposition approach and dynamic management of distributed linked cell list. Performance tests conducted using up to 512 processors have shown that the algorithm scales well. With increasing the number of atoms, the execution time initially increases rapidly and then more gradually showing a linear-scaling behavior when the number of atoms becomes large (two millions or more). Several millions of atoms can thus be simulated with the execution time per MD step remaining in the range of several seconds even with moderate resources.

Important Links:



Go Back