Large-scale distributed computing for accelerated structure solution

Shankland, K. and Griffin, T.A.N. and van de Streek, J. and Cole, Jared H. and Shankland, N. and Florence, A.J. and David, W.I.F. (2009) Large-scale distributed computing for accelerated structure solution. Zeitschrift fur Kristallografie, 30. pp. 227-232. ISSN 0044-2968 (https://doi.org/10.1524/zksu.2009.0033)

Full text not available in this repository.Request a copy

Abstract

Improvements in SDPD methodology have meant that ever more complex structures are being tackled using global optimisation methods. As a very general rule of thumb, the more complex the structure, the more difficult it is to locate the global minimum in the real-space search. This difficulty can, to some extent, be circumvented by running many instances of the search; for stochastic search methods such as simulated annealing, each instance can be run independently of any other. Such search methods are therefore ideally suited to disposition on a distributed grid-type system that makes use of existing networked compute resources. At the Rutherford Appleton Laboratory, the DASH structure solution code has been adapted to run on a Univa UD GridMP system in order to distribute simulated annealing runs across hundreds of computers simultaneously with excellent scaling. The principles outlined are applicable to other structure solution codes and to other grid-type systems, such as the widely used and freely available CONDOR system.

ORCID iDs

Shankland, K., Griffin, T.A.N., van de Streek, J., Cole, Jared H., Shankland, N., Florence, A.J. ORCID logoORCID: https://orcid.org/0000-0002-9706-8364 and David, W.I.F.;
CORE (COnnecting REpositories)