Hello,
On Thu, Oct 28, 2010 at 4:48 PM, zhihong ke <ebelleke.gmail.com> wrote:
> Hello.
>
> I have installed the MMPBSA.py.MPI and have finished the test in
> mmpbsa_py.test.tar.gz downloaded from the tutorial A3.
> During the test, when I tried to used 8 processes, it said that the number
> of the processes should not be larger than the number of the frames. (there
> are only two frames in the test). That means one process was used for
> either
> frame.
>
MMPBSA.py.MPI is parallelized across frames, as this was the easiest (and
most naive) load-balancing that we attempted. I didn't consider the speedup
introduced by allowing each frame to be split up by multiple processes worth
the added complexity. It does allow you to use a thread count that's not an
even multiple of the number of frames you want to analyze, as it just gives
an extra frame to as many processors as it has *leftover* frames.
> However, I found that there is an option "nproc", which could be 3 or any
> multiple of 2.
This was in the old version of the script that was not actually parallel,
but called sander.MPI utilizing multisander to run the complex, receptor,
(and ligand if nproc=3) simulations simultaneously. This scaled far worse,
and couldn't be used past 3 processors for PB (and couldn't be parallelized
at all for nmode), so it was discarded in favor of the new parallelization
scheme.
> My question is :if I set the nproc = 4, then could I use 8
> processes for 2 frame? And is this option available for the Nmode
> calculation? Is it possible that I use 8 processes for one-single-frame 's
> Nmode calculation? And does it help?
>
This can no longer be used. In fact, nproc and mpi_cmd will no longer be
recognized as valid input options since they're not used (they're supplied
directly on the command line when you call MMPBSA.py.MPI now).
>
> On the other hand, it is known that the Nmode is very expensive. In some
> literature, a truncated model was used, such as a sphere of 8 Angstrom was
> truncated. Is it possible that I just use a task or some function like
> IBELLY to provide a residue list, which is allowed to move in the Nmode
> minimization and is activated for the Nmode calculation?
>
This is not implemented. You would need to adjust source code to get this
to work.
HTH,
Jason
> Thank you very much!
>
> Zhihong
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
--
Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Graduate Student
352-392-4032
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Thu Oct 28 2010 - 14:30:03 PDT
