On the parallelization of molecular dynamics codes

1993

This paper is concerned with the implementation of the molecular dynamics code, CHARMM, on massively parallel distributed-memory computer architectures using a data-parallel approach. The implementation is carried out by creating a set of software tools, which provide an interface between the parallelization issues and the sequential code. Large practical MD problems is solved on the Intel iPSC 860 hypercube. The overall solution e ciency is compared with that obtained when implementation is done using data-replication.

Journal of Computational Physics, 1999

Molecular dynamics programs simulate the behavior of biomolecular systems, leading to understanding of their functions. However, the computational complexity of such simulations is enormous. Parallel machines provide the potential to meet this computational challenge. To harness this potential, it is necessary to develop a scalable program. It is also necessary that the program be easily modified by applicationdomain programmers. The NAMD2 program presented in this paper seeks to provide these desirable features. It uses spatial decomposition combined with force decomposition to enhance scalability. It uses intelligent periodic load balancing, so as to maximally utilize the available compute power. It is modularly organized, and implemented using Charm++, a parallel C++ dialect, so as to enhance its modifiability. It uses a combination of numerical techniques and algorithms to ensure that energy drifts are minimized, ensuring accuracy in long running calculations. NAMD2 uses a portable run-time framework called Converse that also supports interoperability among multiple parallel paradigms. As a result, different components of applications can be written in the most appropriate parallel paradigms. NAMD2 runs on most parallel machines including workstation clusters and has yielded speedups in excess of 180 on 220 processors. This paper also describes the performance obtained on some benchmark applications.

Journal of Computational Chemistry, 1998

In recent years several implementations of molecular dynamics Ž . Ž . MD codes have been reported on multiple instruction multiple data MIMD machines. However, very few implementations of MD codes on single instruction Ž . multiple data SIMD machines have been reported. The difficulty in using pair lists of nonbonded interactions is the major problem with MD codes for SIMD machines, such that, generally, the full connectivity computation has been used.

Codes simulating large particle systems present a high degree of spatial data locality and a signi cant amount of independent computations. However, neither stateof-the-art automatic parallelizers nor HPF compilers with standard data distribution directives are able to exploit e ciently such properties. Nowadays, most of the production applications in this area have been manually parallelized. The process of manual parallelization usually involves drastic rewriting of the original sequential code. This work presents a parallelization strategy that offers: high e ciency similar to that of manual parallelization; original program structure is preserved in resulting parallel code; global data structures are decomposed in smaller local structures with the same organization; initial data decomposition and further communications are handled by calls to an existing runtime support. The simplicity of the resulting code makes the technique very suitable for semi-automatic parallelization. At last, we analyze an intermediate approach in this direction: the introduction of HPF extensions to provide the compiler with information enough to guess the role of each data structure in particle codes.

40th Annual Simulation Symposium (ANSS'07), 2007

Molecular Dynamics (MD) involves solving Newton's equations of motion for a molecular system and propagating the system by time-dependent updates of atomic positions and velocities. As a severe limitation of molecular dynamics is the size of the timestep used for propagation, a key area of research is the development of efficient propagation algorithms which can maintain accuracy and stability with larger timesteps. We present MDL, an MD domain-specific language with the goals of allowing prototyping, testing and debugging of these algorithms. We illustrate the use of parallelism within MDL to implement the Finite Temperature String Method, and interfacing to visualization and graphical tools.

Computer Physics Communications, 1994

EGO is a parallel molecular dynamics program running on Transputers. We conducted a performance analysis of the EGO program in order to determine whether it was effectively using the computational resources of Transputers. Our first concern was whether communication was overlapped with computation, so that the overheads due to communication not overlapped with computation were less. With the assistance of performance tools such as UPSHOT, and with instrumentation of the EGO program itself, we were able to determine that only 8% of the execution time of the EGO program was spent in non-overlapping communication. Our next concern was that the MFLOPS rating of the EGO program was 0.25 MFLOPS, while the Transputers have a sustained rating of 1.5 MFLOPS. We measured MFLOPS ratings of small blocks of OCCAM code and determined that they matched the performance of the EGO code.

International Journal of High Performance Computing Applications, 1996

Page 1. NAMD: A PARALLEL, OBJECT-ORIENTED MOLECULAR DYNAMICS PROGRAM Mark T. Nelson William. Humphrey Attila Gursoy Andrew Dalke Laxmikant V. Kalé Robert D. Skeel Klaus Schulten Address reprint ...

IEEE Computational Science and Engineering, 1995

CHARMM (Chemistry at Harvard Macromolecular Mechanics) is a program that is widely used to model and simulate macromolecular systems. CHARMM has been parallelized by using the CHAOS runtime support library on distributed memory architectures. This implementation distributes both data and computations over processors. This data parallel strategy should make it possible to simulate very large molecules on large numbers of processors.

Journal of Optoelectronics and Advanced Materials

EGO is a parallel molecular dynamics program running on Transputers. We conducted a performance analysis of the EGO program in order to determine whether it was effectively using the computational resources of Transputers. Our first concern was whether communication was overlapped with computation, so that the overheads due to communication not overlapped with computation were less. With the assistance of performance tools such as UPSHOT, and with instrumentation of the EGO program itself, we were able to determine that only 8% of the execution time of the EGO program was spent in non-overlapping communication. Our next concern was that the MFLOPS rating of the EGO program was 0.25 MFLOPS, while the Transputers have a sustained rating of 1.5 MFLOPS. We measured MFLOPS ratings of small blocks of OCCAM code and determined that they matched the performance of the EGO code.

Journal of Parallel and Distributed Computing, 1997

To achieve scalable parallel performance in Molecular Dynamics Simulations, we have modeled and implemented several dynamic spatial domain decomposition algorithms. The modeling is based upon the Bulk Synchronous Parallel architecture model (BSP), which describes supersteps of computation, communication, and synchronization. Using this model, we have developed prototypes that explore the differing costs of several spatial decomposition algorithms, and then use this data to drive implementation of our Molecular Dynamics simulator, Sigma.