[phenixbb] Parallelization for optimization of wxu/wxc?

[Andreas Förster]

Hi Carsten,

I'm not much of an expert on phenix, but the latest newsletter states 
(on page 16)

The weight optimization procedure is easy to parallelize since the 
refinements with different trial weights are independent.  Starting with 
PHENIX version dev-­‐810, the refinement.main.nproc parameter is 
available to specify the number of CPUs the weight optimization 
procedure may use in parallel.  To give one example, the command 


phenix.refine 1av1.pdb 1av1.mtz \ optimize_xyz_weight=True \ 
optimize_adp_weight=True nproc=16

finishes in approximately 430 seconds on a 48­core 2.2GHz AMD Opteron 
system.  With nproc=1 the refinement requires more than 2000 seconds on 
the same machine.

Hope that help.


Andreas


On 26/08/2011 1:00, Schubert, Carsten [JRDUS] wrote:
> Hi,
>
> I waswonderingif the optimization of the xyz and adp target weight could
> be parallelizedand take advantage of multi-cpu systems?On the project I
> am working on now, the refinement time increased ~5.4 fold whenwith
> active optimization of theseweights.
>
> Are there any plans for that in the works?
>
> Cheers,
>
> Carsten
>

-- 
         Andreas Förster, Research Associate
         Paul Freemont & Xiaodong Zhang Labs
Department of Biochemistry, Imperial College London
             http://www.msf.bio.ic.ac.uk