MDBenchmark: A toolkit to optimize the performance of molecular dynamics simulations

Area of Science:

• Computational chemistry and physics
• High-performance computing

Background:

• Molecular dynamics (MD) simulations are growing in size and complexity.
• Exascale computing presents challenges for efficient MD simulation execution.
• Hardware and software diversity necessitates easy benchmarking for optimal simulation setup.

Purpose of the Study:

• To develop a software tool, MDBenchmark, for streamlining MD simulation benchmarking.
• To demonstrate the necessity and benefits of performance benchmarking for MD simulations.
• To illustrate the capabilities of MDBenchmark in analyzing scaling studies.

Main Methods:

• Developed MDBenchmark, an open-source software for setting up, submitting, and analyzing MD benchmarks.
• Measured performance of 23 MD systems using GROMACS 2018 on CPU-only and CPU-GPU nodes.
• Optimized Message Passing Interface (MPI) ranks and Open Multi-Processing (OpenMP) threads for performance.

Main Results:

• MDBenchmark facilitates efficient setup and analysis of MD simulation benchmarks.
• Performance scaling was compared across different node types (CPU vs. CPU-GPU).
• Optimizing MPI and OpenMP parameters is crucial for maximizing simulation performance.

Conclusions:

• Benchmarking is essential for identifying optimal simulation parameters for specific hardware.
• Optimized MD simulations yield significant performance gains.
• Improved efficiency reduces monetary, energetic, and environmental costs of MD simulations.