Re: [AMBER] MMPBSA.py.MPI with MVAPICH2

From: Jason Swails <jason.swails.gmail.com>
Date: Wed, 29 Sep 2010 12:37:42 -0400

Hello,

Try this: fire up the Python interpreter (just type "python") and try the
commands

import mpi4py

and

from mpi4py import MPI

If either of those fails, then the installation has problems. If the first
one fails but you did install mpi4py, then it's likely that you need to add
a directory to PYTHONPATH before python can find it. Look in the
mpi4py/build/ directory and find the directory containing the "mpi4py"
directory. The directory name should correspond to your operating system
(i.e. linux or something). Then, add this directory to PYTHONPATH by the
command

export PYTHONPATH=/path/to/mpi4py/build/folder\:$PYTHONPATH

in bash, or the equivalent in csh using setenv based on your shell.

Hope this helps,
Jason

On Wed, Sep 29, 2010 at 12:31 PM, Senthil Natesan <sen.natesan.yahoo.com>wrote:

> Hi Jason,
>
> Thanks for the reply.
>
> yes, MMPBSA.py works fine. I will be running this calculation on large
> number
> of complexes and it seems that for a single calculation, it takes more than
> an
> hour or so,
> (still my first calculation is not completed yet). This particular protein
> has ~
> 299 residues.
> But I gave instruction to read every 1 ps frame of 1 ns trajectory.
> Probably I
> should
> increase this number to 10 ps.
>
> By the way, I am working on Craycx1 machine which has 4 computing nodes
> (each
> with 8 processors).
> I haven't compiled mpi4py with MVAPICH yet. I did it with MPICH2 only as I
> didn't see MVAPICH
> given as one of the options.
>
> I do have working mpicc compiler with MVAPICH2 and I will try now.
>
> The error message was to install mpi4py, as if it was not able to find it.
> Though it is installed.
>
> are there any other tips to increase the speed of calculations without
> loosing
> the quality of the output?
>
> thanks,
> Senthil Natesan
>
>
>
> Jason wrote:
> Hello,
>
> Here are some words of warning: MMPBSA.py.MPI is an advanced option, and
> you
> should only try and use this if you really require the benefit of running
> on
> multiple processors. Next, if the available MPI is mvapich, then I'm
> guessing you're running this on a cluster with infiniband, correct? It's
> tricky to run multiple MPIs on a cluster, and care has to be taken to keep
> them separate and not to have one stepping on the toes of the other
> (different MPI implementations are generally incompatible with each other).
>
> There should not be any issues in building mpi4py with mvapich2 to my
> knowledge. I installed it just fine with mvapich. You just have to make
> sure that it has a working mpicc compiler. You can run "python setup.py
> build" in the mpi4py source directory to build the binaries in the build/
> directory inside mpi4py-1.2.1b/ . Then you have to add the corresponding
> library path to PYTHONPATH. (python setup install will likely not work
> without root privileges unless you're using a python that you've built)
> This will all hopefully be streamlined in the future.
>
> Finally, what error messages do you get when you try to run MMPBSA.py.MPI?
> Can you run MMPBSA.py?
>
> Good luck!
> Jason
>
> On Wed, Sep 29, 2010 at 11:45 AM, Senthil Natesan <sen.natesan.yahoo.com
> >wrote:
>
> > Hi Amber Users,
> >
> > Greetings. I compiled Amber11 and Ambertools1.4 with MVAPICH2 and
> > everything
> > works fine.
> > I am wondering if MMPBSA.py. MPI can be run using MVAPICH too ?
> > I don't see any option for compiling mpi4py with MVAPICH, so I compiled
> > with
> > MPICH2.
> > Now I am not able to run MMPBSA.py.MPI ?!!.
> >
> > Please suggest me how to go about ?
> > thanks,
> >
> > Senthil Natesan
> >
> >
> >
> >
> > _______________________________________________
> > 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
>
>
>
>
> _______________________________________________
> 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 Wed Sep 29 2010 - 10:00:13 PDT
Custom Search