Investigation of Nonlinear Dynamic Effects in Loaded Layer–Substrate Systems through Molecular Dynamics Simulation

Rentsch, R.; Brinksmeier, E.; Li, J.

doi:10.1002/3527602585.ch14

Cited by 1 publication

(1 citation statement)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…MD simulations are typically computationally intensive due to the long computation time set by its femtosecond step length and the large number of interactions that need to be evaluated. Millions of times steps (wall time can be hours, days, or months long depending on the problem size and computer capacity) are necessary to simulate phenomena, such as the movement of ATP, rapture of nano-scale materials, and shearing of confined micro-fluid which occur within several picoseconds [1,2]. In this context, parallel computation techniques have attracted much attention because of the significant reduction of wall time achieved by distributing massive computation among networked processors [3,4].…”

Section: Introductionmentioning

confidence: 99%

A Cyclic Force Decomposition Algorithm for Parallelising Three-Body Interactions in Molecular Dynamics Simulations

Zhou

Sadus

2006

First International Multi-Symposiums on Computer and Computational Sciences (IMSCCS'06)

View full text Add to dashboard Cite

A cyclic force decomposition algorithm is examined for parallelising three body interactions. The algorithm is based on the decomposition of a 3D force matrix into slices of 2D force matrixes in cyclic task assignments. The proposed decomposition algorithm is implemented using MPI and tested in computational experiments of MD simulations. The performance of the decomposition method is studied in terms of load balance, achieved speedup and parallel efficiency. Theoretical analysis of effective triplets is proposed and conducted by which load balance status can be predicted and compared with benchmark measurements. Reasonably good overall performance is achieved with the proposed algorithm. Both theoretical analysis and computation experiments demonstrate that the load balance is a key factor that impacts the parallel efficiency of the system examined in this study.

show abstract