Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Experiment Videos

Speeding up parallel GROMACS on high-latency networks.

Carsten Kutzner1, David van der Spoel, Martin Fechner

  • 1Department of Theoretical and Computational Biophysics, Max-Planck-Institute for Biophysical Chemistry, Am Fassberg 11, 37077 Göttingen, Germany. ckutzne@gwdg.de

Journal of Computational Chemistry
|April 4, 2007
PubMed
Summary
This summary is machine-generated.

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same author

How proteins fold.

Nature reviews. Molecular cell biology·2026
Same author

Enhancing Relative Binding Free Energy Calculation with Grand Canonical Monte Carlo, Water-swap Monte Carlo, Terminal-flip Monte Carlo and Replica Exchange Solute Tempering.

Journal of chemical theory and computation·2026
Same author

Structural basis for activation and potentiation in a human α5β3 GABA<sub>A</sub> receptor.

Nature communications·2026
Same author

An atomic interaction conserved for over 600 million years gates inhibitory neurotransmission.

bioRxiv : the preprint server for biology·2026
Same author

Bayesian Modeling of Polarizable Water: Lessons for Force Field Development.

Journal of chemical theory and computation·2026
Same author

Molecular Dynamics Workflows to Compute Large-Scale Sets of Absolute Binding Free Energies Aiding Drug Candidate and Binding Pose Selection.

Journal of chemical theory and computation·2026

Optimizing the GROMACS molecular dynamics code on Ethernet clusters significantly improves parallel scaling. Implementing an ordered all-to-all communication routine prevents packet loss, boosting performance on any cluster size.

Area of Science:

  • Computational chemistry
  • High-performance computing

Background:

  • GROMACS 3.3 demonstrates excellent performance on specialized supercomputers and clusters with high-speed interconnects.
  • On standard Ethernet clusters, GROMACS scaling is limited by network bottlenecks, particularly with more than two nodes.

Purpose of the Study:

  • To investigate the parallel scaling of GROMACS on Ethernet Beowulf clusters.
  • To identify prerequisites for effective scaling on limited-bandwidth, high-latency networks.
  • To overcome communication bottlenecks in GROMACS for improved performance.

Main Methods:

  • Analysis of GROMACS parallel scaling on Ethernet Beowulf clusters.
  • Identification of communication bottlenecks using LAM MPI.
  • Implementation and testing of an optimized all-to-all communication routine.

Related Experiment Videos

  • Evaluation of network flow control and node connection strategies.

    • Main Results:

    • The all-to-all communication step was identified as the primary scaling bottleneck due to message flooding and TCP packet loss.
    • Ethernet flow control provided substantial scaling improvements for up to 16 CPUs.
    • An optimized, ordered all-to-all routine completely prevented packet loss, dramatically improving GROMACS scaling on any number of nodes.
    • Optimal performance on HP ProCurve 2848 switches depends on specific node-to-port connections.

    Conclusions:

    • Optimized communication strategies are crucial for achieving high parallel scaling of GROMACS on commodity Ethernet clusters.
    • The developed all-to-all routine significantly enhances GROMACS performance, making it viable on a wider range of hardware.
    • Network topology and switch configuration play a vital role in maximizing performance for molecular dynamics codes.