Re: [AMBER] MMPBSA Job Parallel Running

From: Jason Swails <jason.swails.gmail.com>
Date: Wed, 13 Jan 2016 11:40:12 -0500

On Wed, Jan 13, 2016 at 10:46 AM, Guqin Shi <shi.293.osu.edu> wrote:

> Hi Pallavi,
>
> I'm 100% sure all .dat files are out and whole job is done.
> I had finished around ren mmpbsa.MPI jobs on Super Computer, about half of
> them didn't terminate by themselves after all .dat files are produced. I
> requested #PBS -j oe so I had all xxx.o12345678 file when job is done or
> aborted due to wall time limit. It showed that mmpbsa job is done.
>
> I also got reply back from Super Computer Tech Support. They said according
> to their log, mmpbsa.MPI was running after all .dat files have been done.
> They claimed that the pbs platform worked all right and they thought it's
> because something is stuck in program.
>
> What I can do right now is to restrict the wall time limit to minimum so
> that I can save resource units (charged by super computer center) if the
> program has to be aborted.
> According to Kenneth's post, it happened to her, too. It might because some
> minor issues in mmpbsa.MPI itself...
>
> Thank you for the help! Overall, I think it's not a big problem. It will be
> better if any developer could have a look and fix that.
>

I made a change that I *think* should fix this problem, but since I haven't
observed this behavior before specifically with MMPBSA.py.MPI, I have no
way of testing that my change really does resolve this. I've attached a
patch with my changes. You can apply this patch by going to $AMBERHOME and
running the command

patch -p0 -N < mmpbsa_mpi.patch

Then rebuild AmberTools 15 in parallel.

Basically what I did was add a call to MPI.Finalize() before the
sys.exit(0) call. This *should* clean up the MPI stuff so that it knows to
properly terminate. At least that's what the MPI spec says should happen...

Let me know if this patch works for you. It will be part of the upcoming
AmberTools 16 release in April.

Thanks,
Jason
‚Äč
-- 
Jason M. Swails
BioMaPS,
Rutgers University
Postdoctoral Researcher
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed Jan 13 2016 - 09:00:03 PST
Custom Search