On the performance of molecular dynamics applications on current high-end systems

Area of Science:

• Computational chemistry
• Molecular dynamics simulations
• High-performance computing (HPC)

Background:

• Effective utilization of modern high-performance computing (HPC) resources is crucial for molecular simulations.
• Current parallel molecular dynamics codes must efficiently use HPC components like CPUs and memory.

Purpose of the Study:

• Investigate the efficiency and scalability of popular molecular dynamics codes on UK national HPC resources.
• Compare performance across different HPC architectures (IBM p690+, SGI Altix 3700).

Main Methods:

• Performance and scalability analysis of AMBER, DL_POLY, and NAMD simulation codes.
• Testing on IBM p690+ cluster and SGI Altix 3700 systems.
• Comparison of distributed data versus replicated data approaches.

Main Results:

• Demonstrated significant performance advantages with a distributed data approach.
• Showcased major scalability benefits using distributed data strategies.
• Identified key differences in code efficiency on various HPC platforms.

Conclusions:

• A distributed data approach is superior for molecular dynamics simulations on HPC.
• Optimizing code for distributed data enhances performance and scalability.
• Findings are applicable to optimizing popular simulation packages like AMBER, DL_POLY, and NAMD.