Optimized Scaling for GROMACS

Simulations of big molecular dynamics systems can easily occupy a compute cluster for weeks or even months. Continuous efforts are being made to ensure that our computing power is used most efficiently. This includes network fine-tuning and code optimizations to reach the best possible parallel scaling.

Contributions that enhance the parallel scaling include:

• Parallelization of the Essential Dynamics + Flooding module, making use of Gromacs 4 new domain decomposition features
• A patch [GPL license] for Gromacs 3.3.1 optimizes the all-to-all communication for better PME performance on ethernet clusters [1, 2]
• Multiple-Process, Multiple-Data PME: This type of PME treatment is available in Gromacs from version 4 on. PME efficiency is enhanced by splitting up a part of the processors for the calculation of the reciprocal part of the Ewald sum [3]
• For Gromacs 4.0.7 there is a GPL tool that finds the optimal performance with PME on a given number of processors [download g_tune_pme] (Unpack with tar -xvzf). From version 4.5 on, g_tune_pme is part of the official Gromacs package. There is also a poster describing g_tune_pme. [PDF]

References

3. B. Hess, C. Kutzner, D. van der Spoel, E. Lindahl (2008). GROMACS 4: Algorithms for Highly Efficient, Load-Balanced, and Scalable Molecular Simulation. J. Chem. Theory Comput. 4: 435-447 [pdf]
2. C. Kutzner, D. van der Spoel, M. Fechner, E. Lindahl, U. W. Schmitt, B. L. De Groot, and H. Grubmüller (2007). Speeding up parallel GROMACS on high-latency networks. J. Comp. Chem. 28: 2075-2084 [pdf]
1. C. Kutzner, D. van der Spoel, M. Fechner, E. Lindahl, U. W. Schmitt, B. L. de Groot, and H. Grubmüller (2006). Improved GROMACS scaling on ethernet switched clusters. Recent advances in parallel virtual machine and message passing interface, 13th European PVM/MPI User's Group Meeting, Bonn, Germany, September 17-20, 2006. Proceedings Series: Lecture Notes in Computer Science, B. Mohr, J. L. Träff, J. Worringen, J. Dongarra (Eds.) 404-405 [pdf]