Blue Matter: Approaching the Limits of Concurrency for Classical Molecular Dynamics

Abstract

This paper describes a novel spatial-force decomposition for N-body simulations for which we observe O(sqrt(p)) communication scaling. This has enabled Blue Matter to approach the effective limits of concurrency for molecular dynamics using particle-mesh (FFT-based) methods for handling electrostatic interactions. Using this decomposition, Blue Matter running on Blue Gene/L has achieved simulation rates in excess of 1000 time steps per second and demonstrated significant speed-ups to O(1) atom per node. Blue Matter employs a Communicating Sequential Process (CSP) style model with application communication state machines compiled to hardware interfaces. The scalability achieved has enabled methodologically rigorous biomolecular simulations on biologically interesting systems, such as membranebound proteins, whose time scales dwarf previous work on those systems. Major scaling improvements will require exploration of alternative algorithms for treating the long range electrostatics.

Optimizing fine-grained communication in a biomolecular simulation application on Cray XK6

“…RELATED WORK Similar highly scalable MD codes, such as Blue Matter [3], Desmond [1], Amber [7] and GROMACS [9] have not yet demonstrated strong scaling results for molecule systems larger than 1x10 7 on over 200, 000 cores as in this paper, though the latter two have reported good GPGPU results on smaller systems on smaller machines. Impressive work has recently been done [22] optimizing FFT performance in a similar context on Anton, however unlike that work ours is focused on commodity hardware.…”

Enabling and scaling biomolecular simulations of 100 million atoms on petascale machines with a multicore-optimized message-driven runtime

“…Other scalable MD codes such as Blue Matter [8], Desmond [5], Amber [7] and GROMACS [9] have not shown strong scaling results for such a large molecule system as NAMD has done in this paper. Work by Schulz R etc.…”

Anton, a special-purpose machine for molecular dynamics simulation

“…• IBM's massively parallel, general-purpose Blue Gene/L machine, running its Blue Matter MD code, which was designed for maximal scalability [11,12,15].…”

“…Figure 6 shows measured or estimated performance for each MD platform using benchmark parameters chosen to be comparable to the parameters used by IBM in a set of performance results reported in 2005 for Blue Matter running the DHFR system on Blue Gene/L [11]. We also included a set of improved performance estimates for Blue Matter interpolated from performance results published recently for chemical systems other than DHFR [12]. We ran NAMD, GROMACS, and Desmond on our cluster using parameters identical to those used by IBM.…”

“…for smaller and larger systems [12], we estimated a simulation rate for DHFR of 40 ns/day on 8,192 processor cores (4,096 nodes) of Blue Gene/L and 50 ns/day on 16,384 processor cores (8,192 nodes). An even more recent report says that DHFR runs at somewhat more than 60 ns/day on 8,192 processor cores (4,096 nodes) of Blue Gene/L (R. S. Germain, personal communication).…”