Virtually there: speeding up the analysis of reflectivity data
New computational methods to aid model fitting.
Magnetic Contrast Reflectivity is an excellent method for studying the structure of membrane protein / phospholipid bilayers assembled on gold surfaces. Analysis of these complex layers is not simple and so a computational method has been developed to aid in model fitting and assessment. The approach ‘synthesises’ multiple (typically more than 1000) datasets based upon the physical data, which are then fitted. Each ‘dataset’ is produced from random variations of the physical dataset and thus can be analysed independently.
The analysis is ideally suited to execution on a distributed computing grid and we have created a Linux cluster of both real and virtual (Linux running under VMWare on Windows PCs) computers in order to do so. The (mainly) virtual cluster of some sixty CPU-cores is a powerful, scaleable, energy- and cost-efficient resource that has dramatically shortened the analysis timescale and which can, of course, be utilised to run many other Linux-based codes.
TAN Griffin, SA Holt (ISIS), JH Lakey (Newcastle University), F Heinrich (Carnegie Mellon University)