S. J. Plimpton. Fast parallel algorithms for short-range molecular dynamics. Journal Computation Physics, 117:1-- 19, 1995.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....time because each atom has few numbers of chemical bonds with other atoms, whereas the non bonded force calculations consume most of MD simulation time since each atom interacts with all other atoms in the system. In this paper, two approaches, data decomposition and force decomposition [6], have been employed to parallelize the non bonded force calculations. Although the regular force decomposition algorithm partitions atoms evenly over processors, the non bonded list is distributed unevenly and hence severe load imbalance occurs. This paper presents an irregular force ....

.... parallel algorithms have been proposed and implemented on MIMD systems [4] It has also been observed that spatial domain decomposition algorithms can be applied to achieve good load balance [8] More recently there have been scalable algorithms created for large MIMD multiprocessor systems [6, 9]. Several widely used MD programs have been parallelized on MIMD computers. Clark et al. 9] applied spatial decomposition algorithms to develop EULERGROMOS, a parallelization of the GROMOS MD program. Brooks and Hodo scek implemented a parallel version of CHARMM using the replicated data ....

[Article contains additional citation context not shown here]

Steve Plimpton. Fast parallel algorithms for short-range molecular dynamics. Technical Report SAND91-1144, Sandia National Laboratories, May 1993.

....is already very small for very low temperatures; e.g. for a Ca 40 Coulomb crystal at 1 K the wavelength is of order 10 m, a factor 50 100 smaller than the typical separation distance. sizes certain macroscopic properties of matter can be studied. There is a proliferation of programs for MD [4, 14, 30, 31, 32, 53, 77, 81, 86, 91, 98, 108, 110, 119, 124, 125] and several of these are robust production codes; some with scalable parallel implementations. They cover common MD problems and are excellent tools to perform simulations. However, many of the codes are legacy programs that are either poorly organized or extremely complex. One important factor ....

....dominate, since the integration is relatively cheap to carry out. Nevertheless, the terms of the sum of interactions U( x 1 ; x 2 ; x N ) are independent. Altogether, this makes MD to a certain extent inherently parallel.This has been exploited by several parallel MD programs [4, 14, 22, 23, 32, 53, 77, 81, 86, 91, 98, 110, 124, 125]. When parallelizing MD, there are two important considerations to make. First, the MD program must perform well for a small number of particles, i.e. less than 1000. There are several interests to carry out simulations with a few thousand particles over a long time scale, e.g. a protein ....

[Article contains additional citation context not shown here]

S. Plimpton. Fast parallel algorithms for short-range molecular dynamics. J. Chem. Phys., 117:1--19, 1995.

....Molecular Dynamics (MD) is a particle method that integrates Newtons equation of motion. The most time consuming part is the calculation of forces between the particles, which in principle is an N problem. For the common case of forces with short range several optimization techniques exist [6]. The linked cell algorithm used here, resembles the cell structure of the DSMC algorithm. However, the interaction is not limited to the same cell, but includes neighboring cells. Information about particles in the so called shadow or overlapping region has to be communicated in addition to ....

Steve Plimpton. Fast parallel algorithms for short-range molecular dynamics. Journal of Computational Physics, 117:1--19, 1995.

....to the communication to computation ratio of # P . While this is much better than RD, it is still not scalable according to our strict criteria: as we increase the number of processors, the proportion of communication cost increases even if we simulate a larger biomolecule. Research by Plimpton [38, 37] and Hwang et al. [19] shows that this method provides a better speedup than RD, and can be used with good speedups up to hundreds of processors. Results obtained by Hwang show that the proportion of communication cost goes from 6 percent of the computation cost on 32 processors to 36 percent ....

....of NAMD2, their patience during the development process, and their contributions to the Initial Applications section of this paper. This work was supported by the National Institutes of Health (NIH PHS 5 P41 RR05969 04) and the National Science Foundation (NSF GCAG BIR 93 18159 and NSF BIR 94 23827 EQ) JCP was supported by a Computational Science Graduate Fellowship from the United States Department of Energy. 24 ....

Steve J. Plimpton. Fast parallel algorithms for short-range molecular dynamics. Journal of Computational Physics, 117:1--19, March 1995.

.... scripting languages, visualization, parallel computing, SPaSM, SWIG Introduction With the development of massively parallel supercomputers, the materials science community has experienced an unprecedented explosion in both the size and complexity of short range molecular dynamics simulations [1,2,3,4,5]. The method of molecular dynamics (MD) has been used since the 1950 s for a variety of computational problems in physics, chemistry, and materials science [6] The idea is really quite simple given a collection of atoms, we solve Newton s equation of motion F=ma and track the atoms ....

.... simple pair potential to complicated many body potentials [7] Prior to 1992, MD simulations were usually performed on relatively small systems (often in 2D) involving fewer than 1 million atoms [8] However, in a span of only 3 years, MD simulation sizes grew to as many as 1 billion atoms in 3D [2,4,5]. Today, production simulations involving tens to hundreds of millions of atoms are possible, but analyzing the resulting data has proven to be extraordinarily difficult. As a result, most MD simulations remain small even though many researchers agree that large scale simulations are useful for ....

[Article contains additional citation context not shown here]

S.Plimpton, "Fast Parallel Algorithms for Short-range Molecular Dynamics," J Computational Physics, vol 117, (March 1995) p 1-19.

....communication algorithm are outlined in Section 4. In Section 5 we describe and extrapolate performance results to address the question regarding the data production rates that can be attained with the Intel Paragon and EulerGROMOS. 1 For a review of parallel molecular dynamics algorithms see [8]. The machine can of course be made to perform more efficiently by manipulating machine parameters such as cache size, instruction set, and so forth [9] 2 2 The AChE and tacrine study The enzyme acetylcholinesterase is responsible for degrading the neurotransmitter acetylcholine in species ....

Steve Plimpton. Fast parallel algorithms for short--range molecular dynamics. Journal of Computational Physics, 117:1--19, March 1995. 14

....A direct algorithm that solves the N body problem has a time complexity of O(N 2 ) where N is the number of particles involved. As N can be a very large number, a variety of parallelization techniques have been proposed to reduce the complexity down to O(N ) We refer the reader to the paper by [16, 25] for details. We present here a short discussion on the solution to the problem and discuss how the proposed UG ts to solve this problem. A Recursive Formulation of the N body Problem For N particles distributed in 2 or 3 dimensional space, the intent is to determine the forces the particles ....

S. Plimpton, \Fast Parallel Algorithms for Short Range Molecular Dynamics," Journal of Computational Physics, vol. 11, pp. 1-19, March 1995.

....three dimensional N b Theta N b Theta N b bins. The bin size is at least the cut off distance (r c ) used in the short range force evaluation. SOTON PAR exploits the symmetry of Newton s Third Law by examining atoms in only 13 (instead of 26) neighboring cells. It uses a spatial decomposition [8] and requires message exchanges with neighbor processors at each time step as atoms migrate across processor boundaries. Many of the MD algorithms used in SOTON PAR are described in the book Computer Simulation of Liquids [1] The code also uses a shifted force [9] Lennard Jones 6 12 potential ....

....processor. This yields an approximate overall rate (including message passing overhead) of 19.5MFlop sec for each node (with parallel threading enabled) Although we used single precision in our codes, our run times compare quite favorably with the times of 0. 24ms atom processor on the Intel Delta [8] and 0.26ms atom processor [2] performed in double precision) on a 1024 processor CM 5 2 both using a spatial decomposition and linked cell method. Acknowledgments The authors take this opportunity to express appreciation to Bob Marr, Ron Peierls and Joe Pasciak for the IPX package, which ....

[Article contains additional citation context not shown here]

S. Plimpton, Fast parallel algorithms for short-range molecular dynamics, Tech. Report SAND91-1144, UC-705, Sandia National Laboratories, Printed May 1993. -

....is the Inertial Method. This method produces partitions which are usually of a higher quality than those produced by ORB, at the expense of a slight increase in the computational effort required to produce the partitions. Alternate approaches to parallelization of particle methods can be found in [15]. Our scheme resembles ORB, but takes advantage of certain metric properties of the surface of the sphere to give good partitions. Before describing our algorithm, we note that one can consider stereographically projecting the points on the sphere onto the plane and then using an existing ....

Plimpton, S.: Fast Parallel Algorithms for Short-Range Molecular Dynamics. J. Comp. Phys. 117 (1995) 1--19

....Thus for fixed INB ave , the cost of communication will dominate for large P , independent of N . The global communication drawback of the replicated algorithm can be avoided with a domain decomposition of the simulation space, thereby improving scalability and reducing communication costs [2, 8, 14]. A general decomposition of the problem space is diagrammed in figure 1. The sixty four rectilinear process domains 1 X axis Z axis Y axis Figure 1: Spatial decomposition of the problem domain; we use P = p 1 p 2 p 3 = 4 4 4 processes and n tot = n 1 n 2 n 3 = 16 16 8 subboxes. Dotted ....

Steve Plimpton. Fast parallel algorithms for short-- range molecular dynamics. Journal of Computational Physics, 117:1--19, March 1995.

....(see Fig. 1) therefore every CPU is responsible for a part of the domain and has to exchange informations about the border of its domain with its adjacent neighbors. Instead of exchanging them directly with all 8 or 26 neighbors in two or three dimensions respectively the Plimpton scheme[1] is applied here (see Fig. 2) Fig. 1. Domain decomposition. The cells of the linked cell algorithm containing data that has to be shared with the neighbors (shadow cells) are in gray. Fig. 2. Plimpton scheme to exchange shadow cells. Molecular Dynamics with C 3 2 Design of the particle ....

Steve Plimpton. Fast parallel algorithms for short-range molecular dynamics. Journal of Computational Physics, 117:1-19, 1995.

.... architectures [18, 26, 32] languages and compilers [7, 13, 17, 25] and software systems [10] Efforts to improve molecular dynamics performance include sequential algorithms addressing the pairlist calculation [1, 22, 30] and numerous vectorization [3, 14, 15, 29] and parallelization efforts [4, 5, 6, 11, 12, 23]. A common application of a benchmark uses total time to ascertain the efficacy of an algorithm or computing system. The details about the benchmark that should be considered will vary according to the goals of the study and what is being measured. To compare two parallel molecular dynamics ....

....simulation details can be difficult to interpret and reproduce. Similarly, the picture one sees in evaluating a computing system with molecular dynamics may depend significantly on diversity in the simulation. In addition, because of variability in simulation parameters and algorithmic approaches [5, 9, 23, 29], one may benefit by considering attendant caveats when comparing different molecular dynamics algorithms, sequential and parallel alike. In the next section we look at features that are less dependent on the simulation input parameters, but, rather, are indicative of the molecular dynamics ....

Steve Plimpton. Fast parallel algorithms for short--range molecular dynamics. Journal of Computational Physics, 117:1--19, March 1995.

....and its rst derivative are continuous at the cut o distance. Particle positions are updated at each time step using a simple Verlet [20] leap frog scheme. Further details of the molecular dynamics algorithms can be found in the book Computer Simulation of Liquids [21] A spatial decomposition [22] is used which distributes the cells in blocks over a threedimensional mesh of processes so that each process is responsible for the particles in a rectangular sub domain. Point to point message passing is necessary to perform two tasks in the algorithm. First, particle information in cells lying ....

S. Plimpton, Fast Parallel Algorithms for Short-Range Molecular Dynamics, J. Comput. Phys., Vol. 117, pp. 1-19, March 1995. 12

....existing sequential codes, they can be shown to be theoretically non scalable: as the number of processors increases, the communication to computation ratio also increases, even if the problem size is arbitrarily increased. More sophisticated strategies, which are variants of force decomposition ([6, 12]) 4 are also non scalable in this sense, although in practice they may lead to reasonable speedups on medium size computers (up to 128 processors) Spatial decomposition schemes (used in [11, 13] where atoms are distributed into cubes depending on the spatial locations, and migrated among the ....

S. Plimpton. Fast parallel algorithms for short-range molecular dynamics. Journal of Computational Physics, 117:1--19, March 1995.

....lists in molecular dynamics) to preserve the set of possible interaction pairs for multiple timesteps. By greatly reducing the work required to identify interactions, this can significantly improve overall performance. More details can be found in the aforementioned references and in Plimpton [12]. 4 Crash Simulations Transient dynamics simulations are widely used to model low speed crash and impact phenomena. They are most commonly solved using explicit methods on Lagrangian grids. A prototypical example is the virtual crash worthiness analyses performed by car manufacturers. The car is ....

S. J. Plimpton, Fast parallel algorithms for short-range molecular dynamics, J. Comp. Phys., 117 (1995), pp. 1--19.

No context found.

S. J. Plimpton. Fast parallel algorithms for short-range molecular dynamics. Journal Computation Physics, 117:1-- 19, 1995.

No context found.

S. J. Plimpton. Fast parallel algorithms for short-range molecular dynamics. Journal Computation Physics, 117:1-- 19, 1995.

No context found.

S. J. Plimpton. Fast parallel algorithms for shortrange molecular dynamics. Journal Computation Physics, 117:1--19, 1995.

No context found.

S. J. Plimpton. Fast parallel algorithms for short-range molecular dynamics. Journal Computation Physics, 117:1-- 19, 1995.

No context found.

S. Plimpton. Fast parallel algorithms for short-range molecular dynamics. J. Comp. Phys., 117:1--19, 1995.

No context found.

S. J. Plimpton, "Fast Parallel Algorithms for Short-Range Molecular Dynamics", J. Comp. Phys.,117:1-19, 1995.

No context found.

S. J. Plimpton. Fast parallel algorithms for shortrange molecular dynamics. Journal of Computational Physics, 117:1 -- 19, 1995.

No context found.

S. Plimpton, Fast parallel algorithms for short range molecular dynamics, J. Comp. Phys., 117, 1, 1995.

No context found.

S. J. Plimpton. Fast parallel algorithms for short-range molecular dynamics. J. Comp. Phys., 117:1--19, 1995.

No context found.

S. J. Plimpton, Fast Parallel Algorithms for Short-Range Molecular Dynamics, J. Comp. Phys. Vol. 117 (1995) 1-19.

First 50 documents  Next 50

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC