Using the Grid to run population dynamics simulations

Abstract

Analysis of population evolutionary dynamics using realistic models is a challenging task requiring access to huge resources. Estimates for simple models of population growth under different mutation and selection conditions yield running times of CPU years. As mutations are stochastic events, experiments can be split into many separate jobs reducing to a large Monte Carlo-like problem that is embarrassingly parallel and thus maps perfectly on the Grid. We have been able to run simulations with realistic population sizes (up to 1.000.000 of individuals) and growth cycles using the Grid with a ~190x efficiency gain, thus reducing execution time from years to a few days. This speedup allows us to accelerate the simulation cycle and work on data analysis and additional model refinements with minimal delays and effort. We have taken measures at various steps in the process to study the efficiency gains obtained. While our simple approach may arguably be far from achieving optimum efficiency, we were able to achieve significant gains. We conclude analyzing Grid efficiency and discussing which benefits can be realistically expected with the current technology and provide useful advice for future Grid developers.

All the tools described are available under GPL from http://ahriman.cnb.csic.es/sbg/tiki-download_file.php?fileId=16.