Cluster-based parallel simulation for large scale molecular dynamics in microscale thermophysics

JW Shu and B Wang and WM Zheng, PARALLEL AND DISTRIBUTED PROCESSING AND APPLICATIONS, PROCEEDINGS, 3358, 200-211 (2004).

A cluster-based spatial decomposition algorithm for solving large-scale Molecular Dynamics simulation of thermophysics is proposed. Firstly, three kinds of domain division strategies are provided and their efficiency and scalability are analyzed. Secondly, a method called FLNP (Fast Location of Neighboring Particles) to accelerate the location of neighboring particles is proposed, which greatly reduces the cost of calculation and communication of interaction. Additionally, a new memory management technique called AMM (Adaptive Memory Management) is applied to meet the large memory requirement. The parallel algorithm based on these above technologies was implemented on a cluster of SMPs and tested on a system of 6,912,000 particles and achieved an efficiency of 77.0%.

Return to Publications page