Parallel computers

Smith W 1991 Molecular dynamics on hypercube parallel computers Comput. Phys. Commun. 62 229-48... 

Beazley D M and Lomdahl P S 1993 Message-passing multi-cell molecular dynamics on the Connection Machine 5 Parallel Comput. 20 173-95... 

Hilbers P A J and Esselink K 1992 Parallel molecular dynamics Parallel Computing From Theory to Sound Praotloe ed W Joosen and E Mllgrom (Amsterdam lOS Press) pp 288-99... 

Hilbers P A J and Esselink K 1993 Parallel computing and molecular dynamics simulations Computer Simulation In Chemloal Physios /o 397 NATO ASI Series Ced M P Allen and D J Tlldesley (Dordrecht Kluwer) pp 473-95... 

This completes the outline of FAMUSAMM. The algorithm has been implemented in the MD simulation program EGO VIII [48] in a sequential and a parallelized version the latter has been implemented and tested on a number of distributed memory parallel computers, e.g., IBM SP2, Cray T3E, Parsytec CC and ethernet-linked workstation clusters running PVM or MPI. 

The new formalism is especially useful for parallel and distributed computers, since the communication intensity is exceptionally low and excellent load balancing is easy to achieve. In fact, we have used cluster of workstations (Silicon Graphics) and parallel computers - Terra 2000 and IBM SP/2 - to study dynamics of proteins. 

Trobec, R. Jerebic, L D.Janezic, D. Parallel Algorithm for Molecular Dynamics Integration. Parallel Computing 19 (1993) 1029-1039... 

The following sections cover the design goals, decisions, and outcomes of the first two major versions of NAMD and present directions for future development. It is assumed that the reader has been exposed to the basics of molecular dynamics [2, 3, 4] and parallel computing [5]. Additional information on NAMD is available electronically [6]. 

Almasi, G. S., Gottlieb, A. Highly Parallel Computing. 2nd edn. Ben-jamin/Cummings, Redwood City, California, 1994. 

Geist, A., Beguelin, A., Dongarra, J., Jiang, W., Manchek, R., Sunderam, V. PVM Parallel Virtual Machine A Users Guide and Tutorial for Networked Parallel Computing. MIT Press, Cambridge, Massachusetts, 1994. 

P. A. J Hilbers and K. Esselink, Parallel computing and molecular dynamics simulations , Computer Simulations in Chemical Physics, Proc. of the NATO advanced study institute on new perspectives in computer simulations in chemical physics, 473-95, 1993. 

T. G. Mattson and G. R. Shanker, Portable molecular dynamics software for parallel computing , ACS Symposium Series 592, 133-50. 

R. Murty and D. Okunbor, Efficient parallel algorithms for molecular dynamics simulations , submitted to Parallel Computing. 

W. Smith, Molecular dynamics on hypercube parallel computers , Comp. Phys. Comm., Vol 62, no 2 3, 229-48, 1991. 

R. Trobec, I. Jerebic and D. Janezic, Parallel algorithms for molecular dynamics integration , Parallel Computing, Vol 19, no 9, 1029-39, 1993. 

The program can be run either in a serial or parallel execution mode. Both execution modes were stable when tested on a multiprocessor Linux system. Parallel calculations can be run either on parallel computers or networked workstations. Benchmark information is available at the website listed below. 

OPW (orthogonalized plane wave) a band-structure computation method P89 (Perdew 1986) a gradient corrected DFT method parallel computer a computer with more than one CPU Pariser-Parr-Pople (PPP) a simple semiempirical method PCM (polarized continuum method) method for including solvation effects in ah initio calculations... 

The years since pubHcation of the third edition of the Eniyclopedia (1978—1984) have brought the rise and fall of the minicomputer, the worldwide ascendancy of microprocessor-based personal computers, the emergence of powerhil scientific work stations, the acceptance of scientific visualization, further advances with supercomputers, the rise and fall of the rninisupercomputer, and the realization that the future Hes in parallel computing. 

Supercomputers from vendors such as Cray, NEC, and Eujitsu typically consist of between one and eight processors in a shared memory architecture. Peak vector speeds of over 1 GELOP (1000 MELOPS) per processor are now available. Main memories of 1 gigabyte (1000 megabytes) and more are also available. If multiple processors can be tied together to simultaneously work on one problem, substantially greater peak speeds are available. This situation will be further examined in the section on parallel computers. 

The notion of an atomic operation is important for synchronization. An atomic operation is one that is indivisible. Once initiated, it will continue to completion. There are usually a large number of synchronization primitives in a parallel computer, most commonly test and set primitives, or semaphores implemented in hardware (10). A test and set operation tests the current value of a variable and optionally sets a new value, all in one indivisible operation. 

A recent victim of the killer micros was Evans and Sutherland s parallel computer development effort, halted ia 1990. Their architecture combiaed a small number of approximately 1-MFLOPS processors iato semi-iadependent functional units. Several of these units could, ia turn, be combiaed to form a processor hierarchy, building up to systems that were expected to cost between 1 and 8 million dollars. With the advent of lO-MFLOPS uniprocessor killer micros, such an architecture became irrelevant and the project was halted. The RISC killer micro could deUver the same level of performance as could the combiaed efforts of 10 of the 1-MFLOPS processors, evea with the unlikely assumptioa that the problem could be perfectiy parallelized across 10 processors. 

Speedup. The good performance merits discussioa. The ideal parallel computer has as many as an infinite number of processors, as much as... 

The obvious solution to the limitations imposed by shared bus communications is to fully connect each processor to all other processors via dedicated pathways. The problem is that the number of such pathways grows rapidly, N N — l)/2, where N is the number of processors. The inherent costs and complexity of such a system render it an impractical solution for large-scale parallel computing. 

Unfortunately, Flynn s classification, although commonly used, is quite restrictive when discussing parallel-architecture computers. There have been several attempts to formulate more detailed classification schemes for the great variety of parallel computers now available. None of these efforts have been entirely successful, and none appear to be in general use. A discussion of representative machines from some of the more common classes follows. 

The observation that certain kinds of parallel-computing architectures best support only certain kinds of problems seems to be general. The further observation that interprocessor communication can be the primary impediment to parallel performance is also general. As of this writing, any hope of a truly general purpose parallel computer seems to be remote. The best hope may He in software efforts that describe problems at higher levels of abstraction, which can then be ported and optimized for different parallel architectures (22). 

MIMD Multicomputers. Probably the most widely available parallel computers are the shared-memory multiprocessor MIMD machines. Examples include the multiprocessor vector supercomputers, IBM mainframes, VAX minicomputers. Convex and AUiant rninisupercomputers, and SiUcon... 

---