Re: [AMBER] Atoms per processor

From: David A Case <david.case.rutgers.edu>
Date: Mon, 15 Apr 2019 10:37:15 -0400

On Mon, Apr 15, 2019, Charo del Genio wrote:

> I'm trying to understand how sander splits the workload of a simulation
> amongst processors, and I'm a bit at a loss. I thought that the idea was
> to assign approximately the same number of residues to
>each processor, but looking at my output, I read
>
> Number of triangulated 3-point waters found: 19880
> | Atom division among processors:
> | 0 12907 14831 16757 18683 20609 22535 24461
> | 26387 28313 30236 32159 34085 36008 37931 39857
> | 41780 43703 45629 47552 49475 51401 53324 55247
> | 57173 59096 61019 62945 64868 66791 68717 70640
> | 72563

I'm guessing(?) that you have ntp>0 (a constant pressure) simulation.
In that case, sander divides atoms on molecule boundaries; for constant
volume, it can divide molecules on residue boundaries. See the setpar()
subroutine in set.F90.

The pmemd program has a more balanced algorithm, so you should consider
that (if you can) for highly parallel jobs.

....regards...dac


_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Mon Apr 15 2019 - 08:00:02 PDT
Custom Search