Re: [AMBER] cpptraj.MPI running in 'ensemble' mode.

From: Jonathan Gough <jonathan.d.gough.gmail.com>
Date: Sat, 9 Aug 2014 15:53:24 -0400

Thanks Jason!

As always, very helpful.


On Sat, Aug 9, 2014 at 2:55 PM, Jason Swails <jason.swails.gmail.com> wrote:

> On Sat, Aug 9, 2014 at 2:28 PM, Jonathan Gough <jonathan.d.gough.gmail.com
> >
> wrote:
>
> > Dear All,
> >
> > Is there any documentation on what it means to run cpptraj in parallel? I
> > got the error that it can only be run in ensemble mode. The Amber14.pdf
> > only has the following:
> >
> > "Note that currently this version requires that each thread be assigned
> to
> > a single member of the ensemble, i.e. the number of threads must equal
> the
> > ensemble size."
> >
> > I can't seem to locate a AmberTools14.pdf.
> >
>
> ​It's all in the same manual now.
>
> Best i can gather from the above is that you have to have the number of
> > trajectories = to the number of MPI threads, is that correct?
> >
>
> ​Not quite.
>
> ​
> > serial works fine, but would love to get things moving faster. Any
> insight
> > would be appreciated.
> >
>
> ​"ensemble" is only for analyzing REMD simulations. If you're not using
> REMD simulations, you should not be using ensemble. The number of threads
> must be == to the number of replicas you are running.
>
> MPI is fine for this task because each thread can take a different
> temperature and run the analyses on that one replica. For analyzing a
> single trajectory, though, MPI parallelization is not necessarily the best
> way to go. As a result, Dan used OpenMP to parallelize cpptraj
> calculations. The manual should have some details ​about which actions are
> OMP-parallelized as well as information about how to compile OpenMP
> programs.
>
> You can try that and see if it speeds up your analyses.
>
> HTH,
> Jason
>
> --
> Jason M. Swails
> BioMaPS,
> Rutgers University
> Postdoctoral Researcher
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Sat Aug 09 2014 - 13:00:01 PDT
Custom Search