A team of researchers from the Department of Energy's (DOE) Oak Ridge National Laboratory (ORNL), Brookhaven National Laboratory (BNL) and the State University of New York (SUNY) at Stony Brook have set a new world record for system size in molecular dynamics, using Intel's 1,024-node Paragon supercomputer.
Molecular dynamics, which models the interactions between the atoms in a chemical, biological or solid state system, is a cornerstone of applications ranging from the study of DNA-protein interactions to the design of new materials.
In two record-breaking runs, members of the Partnership in Computational Sciences (PICS) consortium performed molecular dynamics simulations for systems of 200 million and 400 million particles, with each simulation step taking 80 and 160 seconds, respectively. These results far exceed recently reported simulations of 10 million particles on the 512-node Intel Touchstone Delta system at the California Institute of Technology at Pasadena and 180 million particles on a 1,024-node Thinking Machine CM-5.
"This breakthrough is yet another example of the High Performance Computing and Communications program bearing fruit for U.S. science and industry," said Ken Kliewer, director of the Center for Computational Sciences (CCS) at Oak Ridge. "It demonstrates the ability of key molecular dynamics algorithms to scale up to the large sizes needed for biologically interesting problems. This should lead to major breakthroughs in molecular biology by permitting the simulation of realistic systems over significant time scales."
The MD code was developed by Osman Yasar of ORNL's CCS, Robert B. Marr and Ronald F. Peierls of BNL, and Yuefan Deng and R. Alan McCoy of SUNY. All three institutions are members of the PICS consortium, which uses the large Intel Paragon systems at Oak Ridge National Laboratory.
The algorithm used by the PICS team is based on classical, short-range force models and uses a link-cell method. Particle interaction is modeled with a pair-wise 6-12 Lennard-Jones potential. Innovations in the team's approach include a new methodology for a synchronous concurrent execution to handle message passing and synchronization, the preservation of geometric locality to avoid communication costs, and a dynamic load balancing mechanism for increased efficiency.
One of the applications using this MD package, modeling of thin film depositions, will be described at the Materials Research Society annual meeting in Boston on Nov. 30.
The record-setting runs were made on ORNL's 1,024-node Paragon XP\S 150 MP system. The system is currently undergoing testing at the Intel manufacturing facility in Beaverton, Ore. Each node possesses 64 MB of memory. Intel, the world's largest chip maker, is a leading international manufacturer of personal computer, networking and communications products. Paragon is a trademark of Intel Corp.
The work at ORNL and BNL was supported by the U.S. Department of Energy. At SUNY, the work was partially supported by the Applied Mathematics Subprogram of the U.S. Department of Energy's Office of Scientific Computing.
ORNL, one of the Department of Energy's multiprogram national research and development facilities, is managed by Martin Marietta Energy Systems, which also manages the Oak Ridge K-25 Site and the Oak Ridge Y-12 Plant.