MASSIVE PARALLELISM: THE HARDWARE FOR
COMPUTATIONAL CHEMISTRY?
M.F. GUEST AND P. SHERWOOD
Department for Computation and Information,
CCLRC Daresbury Laboratory, Daresbury,
Warrington WA4 4AD, Cheshire, UK
AND
J.A. NICHOLS
High Performance Computational Chemistry Group,
Env ironmental Molecular Sciences Laboratory,
Pacific Northwest National Laboratory,
PO Box 999, Mail Stop Kl-90, Richland,' WA . 99352, USA
1. Introduction
Computational chemistry covers a wide spectrum of activities ranging from
quantum mechanical calculations of the electronic structure of molecules,
to classical mechanical simulations of the dynamical properties of many-
atom systems, to the mapping of both structure-activity relationships and
reaction synthesis steps. Although chemical theory and insight play im-
portant roles in this work, the prediction of physical observables is almost
invariably bounded by the available computer capacity.
The potential of massively parallel computers , with hundreds to thou-
sands of processors (MPPs) , to significantly outpace conventional super-
computers in both capacity and price-performance has long been recog-
nised. While increases in raw computing power alone will greatly expand
the range of problems that can be treated by theoretical chemistry meth-
ods, it is now apparent that a significant investment in new algorithms
is needed to fully exploit this potential. Merely porting presently avail-
able software to these parallel computers does not provide the efficiency
required, with many existing parallel applications showing a deterioration
in performance as greater numbers of processors are used . New algorithms
T. Ebisuzaki et al. (eds.), New Horizons of Computational Science
© Springer Science+Business Media Dordrecht 2001