Re: NAMD results depend on # of processors!!?? (was Re: Is clock skew a problem for charm++)

From: Jan Saam (
Date: Thu Jun 22 2006 - 06:45:47 CDT

Hi Jerome,

Jerome Henin wrote:
> Jan:
> In the present case, there might be a particular data corruption problem, but
> I just wanted to mention that NAMD runs are not necessarily reproducible when
> moved from an architecture to another one, or when run on clusters or similar
> parallel environments.
> The errors you mention, 'Periodic cell has become too small for original patch
> grid!' and 'Atoms moving too fast' often occur in the case of poorly
> equilibrated systems, but "bad things" leading to instability have a nonzero
> probability of happening anytime. The larger the timestep, the farther from
> zero the probability. Maybe your simulation that worked was just luck :)
This is what came to my mind first, but this is a restart of a well
equilibrated system. Also I tried this many times independently which
decreases the probability to just have good or bad luck.

I'm afraid I have to search the cause somewhere else. I really believe
it's a data corruption problem.
I'll ask Gengbin how to test this.

Best, Jan

> This is intended more as a general remark than as an answer to your particular
> problem, which, again, may have a specific cause.
> Jerome

Jan Saam
Institute of Biochemistry
Charite Berlin
Monbijoustr. 2
10117 Berlin
+49 30 450-528-446

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:42:15 CST