B. Wang, J. Shu, W. Zheng, J. Wang, and M. Chen (PRC)
Molecular dynamics, Hierarchical algorithm, Clustercomputing,The rest of this paper is organized as follows. Section 2gives a simple description about the decompositionstrategies of FD and SD,
This paper provides a new parallel algorithm named Hierarchical algorithm, which combines the advantages of FD and SD and limits their disadvantages to some extent. HA algorithm has two levels. Spatial decomposition strategy is used in the outer level to reduce the communication cost and improve the scalability; force decomposition strategy is used in the inner level to give a better load balance performance. Especially when dealing with a large scale MD simulation with non-uniform density, HA performs better than SD and FD. For example, the performance of HA on 72 processors are 44% and 36% better than that of SD and FD respectively when used to simulate nucleation process of a system with 4,000,000 particles. This paper presents a new parallel algorithm in Molecular Dynamics simulation--HA (Hierarchical Algorithm). HA algorithm has two levels. Spatial decomposition strategy is used in the outer level to reduce the communication cost and improve the scalability; force decomposition strategy is used in the inner level to give a better load balance performance. When used to simulate a large nucleation system with non-uniform density, HA performs better than pure SD and FD algorithms.
Important Links:
Go Back
