[gmx-users] grompp -np option

[Berk Hess]

>From: "Nathan Moore" <nmoore at physics.umn.edu>
>Reply-To: Discussion list for GROMACS users <gmx-users at gromacs.org>
>To: "Discussion list for GROMACS users" <gmx-users at gromacs.org>
>Subject: Re: [gmx-users] grompp -np option
>Date: Fri, 10 Jun 2005 10:42:42 -0500 (CDT)
>
>So MAXNODES is defined in
>include/types/simple.h:
>with
>#define MAXNODES 256
>
>Can this definition be changed without code failure/corruption?

I think so.
Only the tpr files would become incompatible with code compiled
with MAXNODES 256.

>
>NT
>
>
> > Is the limit of -np to no more that 256 hard-coded into grompp for a
> > specific reason?  Is it a simple change to the code (I havn't looked)?
> >
> > More generally, is there much understanding of the "optimum" number of
> > atoms each node should have for peak performance in a parallel job?  eg 
>if
> > I have two systems, one with 5,000 atoms, and another with 500,000 
>atoms,
> > would they both be split up into blocks of 1000 atoms/compute node for
> > optimum (minimum real time) performace?

As the communication uses a ring structure, the number of MPI calls scales
with the number of processors. This means that the scaling will degrade
on a large numbers of processors. Summations are also done over the ring.
I think that (when you don't use PME) for the 'real work' there should be an 
optimum
number of atoms, but in practise the communication will become the 
bottleneck
at some point. This means the optimum number will grow as the number of
processors increases.

We are working on proper domain decomposition.

Berk.

_________________________________________________________________
FREE pop-up blocking with the new MSN Toolbar - get it now! 
http://toolbar.msn.click-url.com/go/onm00200415ave/direct/01/