Thank you,
I thought that something like that'll be the problem. Now it's clear for polarizable simulations.
As I see the same problem for nonpolarizable simulations in SANDER.MPI, but not for PMEMD, does it mean that the work is divided among processors in different way in SANDER.MPI, and PMEMD?
Best regards,
Jan Heyda
> ------------ Původní zpráva ------------
> Od: case <case.biomaps.rutgers.edu>
> Předmět: Re: [AMBER] Polarizable simulation of the slab
> Datum: 27.11.2009 17:12:26
> ----------------------------------------
> On Fri, Nov 27, 2009, Jan Heyda wrote:
>
> > I started from equilibrated NPT simulation, which gave box with
> > dimensions approximately 32A x 32A x 32A.
> > System was then shifted by 75A in z-coordinate (by awk) and z-PBC
> > changed to 150, so the final system sizes were 32A x 32A x 150A, and all
> > atoms were in the z-region 59A-91A.
> >
> > The system was thou inhomogeneous in total, but in fact it's 60A of
> > vacuum, 32A of water, and 60A of vacuum. This should be stabile during
> > simulation.
>
> The nonbonded list code in sander assumes that the density of atoms in all
> regions will be approximately homogeneous. It uses this assumption to
> approximately divide work (and memory for data structures) among processors
> in multi-CPU simulations. So, the code will indeed work on a single CPU, but
> will find allocation failures in multiple-CPU runs.
>
> I don't know the code well enough to estimate how hard it would be to fix
> this. As it stands, what you want to do (parallel runs of polarizable
> potential in a system where part is in vacuum) is outside the capabilities
> of the code.
>
> ...regards...dac
>
>
> _______________________________________________
> 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
Received on Fri Nov 27 2009 - 08:30:03 PST
