NAMD: Biomolecular Simulation on Thousands of Processors
James C. Phillips Gengbin Zhengy Sameer Kumary Laxmikant V. Kaley
Abstract
NAMD is a fully featured, production molecular dynamics program for high performance
simulation of large biomolecular systems. We have previously, at SC2000, presented scaling
results for simulations with cuto electrostatics on up to 2048 processors of the ASCI Red
machine, achieved with an object-based hybrid force and spatial decomposition scheme and an
aggressive measurement-based predictive load balancing framework. We extend this work by
demonstrating similar scaling on the much faster processors of the PSC Lemieux Alpha cluster,
and for simulations employing ecient (order N log N) particle mesh Ewald full electrostatics.
This unprecedented scalability in a biomolecular simulation code has been attained through
latency tolerance, adaptation to multiprocessor nodes, and the direct use of the Quadrics Elan
library in place of MPI by the Charm++/Converse parallel runtime system.
1 Introduction
NAMD is a parallel, object-oriented molecular dynamics program designed for high performance
simulation of large biomolecular systems [6]. NAMD employs the prioritized message-driven exe-
cution capabilities of the Charm++/Converse parallel runtime system,1 allowing excellent parallel
scaling on both massively parallel supercomputers and commodity workstation clusters. NAMD is
distributed free of charge via the web2 to over 4000 registered users as both source code and conve-
nient precompiled binaries. NAMD development and support is a service of the National Institutes
of Health Resource for Macromolecular Modeling and Bioinformatics, located at the University of
Illinois at Urbana-Champaign.3
In a molecular dynamics (MD) simulation, full atomic coordinates of the proteins, nucleic acids,
and/or lipids of interest, as well as explicit water and ions, are obtained from known crystallographic
or other structures. An empirical energy function, which consists of approximations of covalent in-
teractions in addition to long-range Lennard-Jones and electrostatic terms, is applied. The resulting
Newtonian equations of motion are typically integrated by symplectic and reversible methods using
a timestep of 1 fs. Modications are made to the equations of motion to control temperature and
pressure during the simulation.
With continuing increases in high performance computing technology, the domain of biomolecu-
lar simulation has rapidly expanded from isolated proteins in solvent to include complex aggregates,
often in a lipid environment. Such simulations can easily exceed 100,000 atoms (see Fig. 1). Sim-
ilarly, studying the function of even the simplest of biomolecular machines requires simulations
Beckman Institute, University of Illinois at Urbana-Champaign.
yDepartment of Computer Science and Beckman Institute, University of Illinois at Urbana-Champaign.
1http://charm.cs.uiuc.edu/
2http://www.ks.uiuc.edu/Research/namd/
3http://www.ks.uiuc.edu/
0-7695-1524-X/02 $17.00 c
2002 IEEE
1