Steve Plimpton

A New Parallel Method for Molecular Dynamics Simulation of Macromolecular Systems

Short‐range molecular dynamics simulations of molecular systems are commonly parallelized by replicated‐data methods, in which each processor stores a copy of all atom positions. This enables computation of bonded 2‐, 3‐, and 4‐body forces within the molecular topology to be partitioned among processors straightforwardly. A drawback to such methods is that the interprocessor communication scales as N (the number of atoms) independent of P (the number of processors). Thus, their parallel efficiency falls off rapidly when large numbers of processors are used. In this article a new parallel method for simulating macromolecular or small‐molecule systems is presented, called force‐decomposition. Its memory and communication costs scale as N/√P, allowing larger problems to be run faster on greater numbers of processors. Like replicated‐data techniques, and in contrast to spatial‐decomposition approaches, the new method can be simply load balanced and performs well even for irregular simulation geometries. The implementation of the algorithm in a prototypical macromolecular simulation code ParBond is also discussed. On a 1024‐processor Intel Paragon, ParBond runs a standard benchmark simulation of solvated myoglobin with a parallel efficiency of 61% and at 40 times the speed of a vectorized version of CHARMM running on a single Cray Y‐MP processor.

Parallel strategies for crash and impact simulations

We describe a general strategy we have found effective for parallelizing solid mechanics simulations. Such simulations often have several computationally intensive parts, including finite element integration, detection of material contacts, and particle interaction if smoothed particle hydrodynamics is used to model highly deforming materials. The need to balance all of these computations simultaneously is a difficult challenge that has kept many commercial and government codes from being used effectively on parallel supercomputers with hundreds or thousands of processors. Our strategy is to load-balance each of the significant computations independently with whatever balancing technique is most appropriate. The chief benefit is that each computation can be scalably parallelized. The drawback is the data exchange between processors and extra coding that must be written to maintain multiple decompositions in a single code. We discuss these trade-offs and give performance results showing this strategy has led to a parallel implementation of a widely used solid mechanics code that can now be run efficiently on thousands of processors of the Pentium-based Sandia/Intel TFLOPS machine. We illustrate with several examples the kinds of high-resolution, million-element models that can now be simulated routinely. We also look to the future and discuss what possibilities this new capability promises, as well as the new set of challenges it poses in material models, computational techniques, and computing infrastructure.

AN EFFICIENT PARALLEL ALGORITHM FOR MATRIX-VECTOR MULTIPLICATION

The multiplication of a vector by a matrix is the kernel operation in many algorithms used in scientific computation. A fast and efficient parallel algorithm for this calculation is therefore desirable. This paper describes a parallel matrix-vector multiplication algorithm which is particularly well suited to dense matrices or matrices with an irregular sparsity pattern. Such matrices can arise from discretizing partial differential equations on irregular grids or from problems exhibiting nearly random connectivity between data structures. The communication cost of the algorithm is independent of the matrix sparsity pattern and is shown to scale as for an n×n matrix on p processors. The algorithm’s performance is demonstrated by using it within the well known NAS conjugate gradient benchmark. This resulted in the fastest run times achieved to date on both the 1024 node nCUBE 2 and the 128 node Intel iPSC/860. Additional improvements to the algorithm which are possible when integrating it with the conjugate gradient algorithm are also discussed.

Massively parallel methods for engineering and science problems

Scientific research and engineering development are relying increasingly on computational simulation to augment theoretical analysis, experimentation, and testing. Many of todays problems are far too complex to yield to mathematical analyses. Likewise, large-scale experimental testing is often infeasible for a variety of economic, political, or environmental reasons. At the very least, testing adds to the time and expense of product development

Transient dynamics simulations: parallel algorithms for contact detection and smoothed particle hydrodynamics

Transient dynamics simulations are commonly used to model phenomena such as car crashes, underwater explosions, and the response of shipping containers to high-speed impacts. Physical objects in such a simulation are typically represented by Lagrangian meshes because the meshes can move and deform with the objects as they undergo stress. Fluids (gasoline, water) or fluid-like materials (earth) in the simulation can be modeled using the techniques of smoothed particle hydrodynamics. Implementing a hybrid mesh/particle model on a massively parallel computer poses several difficult challenges. One challenge is to simultaneously parallelize and load-balance both the mesh and particle portions of the computation. A second challenge is to efficiently detect the contacts that occur within the deforming mesh and between mesh elements and particles as the simulation proceeds. These contacts impart forces to the mesh elements and particles which must be computed at each timestep to accurately capture the physics of interest. In this paper we describe new parallel algorithms for smoothed particle hydrodynamics and contact detection which turn out to have several key features in common. Additionally, we describe how to join the new algorithms with traditional parallel finite element techniques to create an integrated particle/mesh transient dynamics simulation. Our approach to this problem differs from previous work in that we use three different parallel decompositions, a static one for the finite element analysis and dynamic ones for particles and for contact detection. We have implemented our ideas in a parallel version of the transient dynamics code PRONTO-3D and present results for the code running on a large Intel Paragon.

Parallel transient dynamics simulations

Transient dynamics simulations are commonly used to model phenomena such as car crashes, underwater explosions, and the response of shipping containers to high-speed impacts. Physical objects in such a simulation are typically represented by Lagrangian meshes because the meshes can move and deform with the objects as they undergo stress. Fluids (gasoline, water) or fluid-like materials (soil) in the simulation can be modeled using the techniques of smoothed particle hydrodynamics. Implementing a hybrid mesh/particle model on a massively parallel computer poses several difficult challenges. One challenge is to simultaneously parallelize and load-balance both the mesh and particle portions of the computation. A second challenge is to efficiently detect the contacts that occur within the deforming mesh and between mesh elements and particles as the simulation proceeds. These contacts impart forces to the mesh elements and particles which must be computed at each timestep to accurately capture the physics of interest. In this paper we describe new parallel algorithms for smoothed particle hydrodynamics and contact detection which turn out to have several key features in common. Additionally, we describe how to join the new algorithms with traditional parallel finite element techniques to create an integrated particle/mesh transient dynamics simulation. Our approach to this problem differs from previous work in that we use three different parallel decompositions, a static one for the finite element analysis and dynamic ones for particles and for contact detection. We have implemented our ideas in a parallel version of the transient dynamics code PRONTO-3D and present results for the code running on the Pentium-based Intel Teraflop machine at Sandia.

Parallel many-body simulations without all-to-all communication

Abstract Simulations of interacting particles are common in science and engineering, appearing in such diverse disciplines as astrophysics, fluid dynamics, molecular physics, and materials science. These simulations are often computationally intensive and so are natural candidates for massively parallel computing. Many-body simulations that directly compute interactions between pairs of particles, be they short-range or long-range interactions, have been parallelized in several standard ways. The simplest approaches require all-to-all communication, an expensive communication step. The fastest methods assign a group of nearby particles to a processor, which can lead to load imbalance and be difficult to implement efficiently. We present a new approach, suitable for direct simulations, that avoids all-to-all communication without requiring any geometric clustering. We demonstrate its utility in several parallel molecular dynamics simulations and compare performance against other parallel approaches. The new algorithm proves to be fastest for simulations of up to several thousand particles.

Computational limits of classical molecular dynamics simulations

The system sizes and time scales accessible by classical molecular dynamics techniques on current-generation parallel supercomputers are briefly discussed. The implications for simulation of glasses and glass-forming materials now and in the near future are highlighted.

Molecular dynamics simulations of athermal polymer blends: Finite system size considerations

Molecular dynamics simulations of binary, athermal blends of chains consisting of 50 tangent sites were carried out over a range of compositions at liquidlike packing fractions. The sites interact with repulsive Lennard‐Jones potentials and have effective hard sphere diameters of approximately 1.0 and 1.2 for the two site types. The intrachain and interchain correlation functions were found and, except for the single component systems, the interchain correlations were seen to be strongly dependent on system size. Trivial, long range correlations due to the finite system size can be approximated from simple physical arguments, and the coarse, interchain, radial distribution functions can be ‘‘corrected’’ for these effects. The resulting correlation functions are seen to behave at large separation as would be expected of interchain radial distribution functions in an infinite sized system, permitting meaningful comparisons with the predictions of liquid state theory.

Scalable parallel molecular dynamics on MIMD supercomputers

Presents two parallel algorithms suitable for molecular dynamics simulations over a wide range of sizes, from a few hundred to millions of atoms. One of the algorithms is optimally scalable, offering performance proportional to N/P where N is the number of atoms (or molecules) and P is the number of processors. Their implementation on three MIMD parallel computers (nCUBE2, Intel Gamma, and Intel Delta) and performance on a standard benchmark problem as compared to vector and SIMD implementations is discussed. The authors also briefly describe the integration of one of the algorithms into a widely-used code appropriate for modeling defect dynamics in metals via the embedded atom method.<<ETX>>