On Oct 3, 2009, at 9:39 AM, Joe Krahn wrote:
A parallel version of PHENIX was discussed before. I know that parallelization of refinement is difficult, and probably not worth the effort when there is still a lot of function/methods work. However, running parallel independent tasks could pay off without a lot of work. For example, optimization of WXU and WXC can spawn multiple tasks with different test weights. Annealing tasks could also run multiple jobs, and pick the best result, or even merge results based on real-space fit analysis.
The AutoBuild wizard already does the last bit - it will use as many processors as you want, you can turn on simulated annealing, and you can pass it parameter files for phenix.refine. -------------------- Nathaniel Echols Lawrence Berkeley Lab 510-486-5136 [email protected]