There's no problem, this is the expected behavior. MMPBSA.py.MPI is
parallelized on the script level -- that is, when you do something like
this:
mpirun -np 2 MMPBSA.py.MPI -O -i ....etc.
what it does is starts 2 MMPBSA.py.MPI threads. Each of these threads runs
every other program in serial. I will provide (too many) more details in
the postscript below.
HTH,
Jason
P.S. The alternative (and original) approach was to use sander.MPI inside
MMPBSA.py (so MMPBSA.py was not parallel itself, it just called parallel
executables). There are many downsides to this approach. First of all,
there's no 'general' way of launching an MPI program within a Python script
appropriately that is guaranteed to work for all MPIs. Secondly,
sander.MPI only parallelizes within each snapshot -- that is, every
processor you allocate will work on the same frame at the same time (and
adjusting this behavior for imin=5 is too difficult to be worthwhile).
Therefore, you're limited to the scaling you can achieve within a single
snapshot. That limits parallel scaling to essentially *just* MM/GBSA
calculations. The way we're doing it now, we can split up a trajectory
into smaller chunks and balance out the workload pretty evenly among all
processors, allowing MMPBSA.py to scale with the number of frames you're
analyzing rather than how well your particular solvent model of choice is
parallelized.
On Thu, Apr 12, 2012 at 9:12 AM, Alfredo Quevedo <alfredoq.fcq.unc.edu.ar>wrote:
> Good morning. I followed the instructions in the AmberTools12 manual to
> install the serial and parallel versions of Amber and AmberTools. All
> serial and parallel test passed OK, but when I try to use the
> MMPBSA.py.MPI module I find that for the corresponding calculations
> sander is called instead of sander.MPI. I compiled openmpi using the
> provided script and everything runs ok. Also sander.MPI is present in
> $AMBERHOME/bin. Can anyone point me out which may be the problem,
> thank you very much in advance for the help,
> best regards
> Alfredo
>
>
> --
> ************************************************************************
> Dr. Mario Alfredo Quevedo.-
> Departamento de Farmacia.-
> Facultad de Ciencias Químicas.-
> Universidad Nacional Córdoba.-
> 5000 Córdoba, Argentina
> TE: 54351 4334163 Ext: 106
>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
--
Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Candidate
352-392-4032
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Thu Apr 12 2012 - 06:30:06 PDT
