Shared Memory Parallelization of the GROMOS96 Molecular Dynamics Code

This paper describes the parallelization of a commercial molecular dynamics simulation code, GROMOS96, on a SCI (Scalable Coherent Interface) interconnected PC cluster. The underlying programming model is that of shared data structures, exploiting SCI’s capabilities of enabling access to segments of remote memory in an entirely transparent way. Methodologies are elaborated that allow to obtain high performance in presence of the NUMA (Non-Uniform Memory Access) performance characteristic of the cluster platform. It is demonstrated that this type of cluster platform allows a step-by-step parallelization process in distinction to a message-passing parallelization with it’s partitioned and distributed data structures. Performance figures of the resulting parallel code are presented and discussed.

• Type: Conference
• Year: 1999
• Where: SCI
• Authors: Marcus Dormanns
• Keywords: Cluster Platform, Data Structures, Parallelization, SCI 1999, Software Engineering