On Sun, Nov 2, 2014 at 8:57 AM, Fabian Glaser <fabian.glaser.gmail.com>
wrote:
> Hi Jason,
>
> Thanks for your quick response, the truth is I don't think it is a typo I
> have checked it in several occasions and it produces always the same
> result. I send you here the result of a test with -r in the last position
> and in other position, and the output file produced with the full content
> of the script. Is a short minimisation.
>
> I would like to mention that I am in the process of buying amber14, since
> amber12 I couldn't use in several nodes, it could be a bad compilation or a
> wrong version.....
>
> As said if -r min.rst is not the last flag, the default restrt file is
> produced. I know from a friend that he also thinks is a very weird
> behaviour, but it has always been like this in my cluster....
>
> Thanks a lot,
>
> Fabian
>
> Here it is:
>
> #!/bin/sh
> #
> # job name (default is the name of pbs script file)
> #---------------------------------------------------
> #PBS -N ETA_9_min
> # Submit the job to the queue "queue_name"
> #---------------------------------------------------
> #PBS -q short_l_p
>
> #-----------------------------------------------------------------
> #PBS -M fglaser.technion.ac.il
> #---------------------------------------------------
> #PBS -mbea
> #------------------------------------------------------------------
> #PBS -l select=1:ncpus=12:mpiprocs=12
> #
> #
> # specifying working directory
> #------------------------------------------------------
>
> cd $PBS_O_WORKDIR
>
> # running MPI executable with M*N processes
> #------------------------------------------------------
>
> source /usr/local/intel12/setup.sh
>
> #command
>
> mpirun sander -O -i min.in -o min.out -p
> ../clec2_system_ETA_ETA_9.prmtop -c ../clec2_system_ETA_ETA_9.prmcrd -ref
> ../clec2_system_ETA_ETA_9.prmcrd -r min.rst
>
"sander" is not a parallel executable, and so you should never run it with
mpirun. [1] If mpirun actually launches multiple threads, the behavior you
get will be unexpected. You should be using the ".MPI" versions of the
programs instead.
Amber uses standard intrinsic Fortran functions to loop through
command-line arguments, and the logic that parses them is _very_ simple and
not at all order-dependent. If the behavior you describe is actually
happening with correct command-lines, then it's more likely that mpirun or
the shell is misbehaving than sander or pmemd.
At the end of the day, if we can't reproduce this behavior, we can't even
begin to debug it.
All the best,
Jason
[1] In rare instances, like when you want to run multiple "sander"
simulations independently, you can have mpirun run it that way (i.e., MPMD
mode), as long as each sander instance is given exactly 1 thread.
--
Jason M. Swails
BioMaPS,
Rutgers University
Postdoctoral Researcher
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Sun Nov 02 2014 - 06:30:02 PST
