[gmx-users] mdrun crashes with 'One or more interactions were multiple assigned in the domain decompostion'

[Mark Abraham]

On 12/08/2011 9:55 PM, Sebastian Breuers wrote:
> Dear all,
>
> searching for the mentioned error message I found a bug report for 
> mdrun. It seemed to be fixed, but in my setup it appears again and I 
> am not sure if I could do something about it.
>
> I did not attach the tpr file since it is bounced by the mailing list 
> and I'd like to get at least a hint without waitng for approval of the 
> rejected mail. :)
> The simulation crashes with 64 CPUs after step 11237000 with the 
> following entry in the log file:
>
> -------------------------------------------------------
> Program mdrun, VERSION 4.5.4
> Source code file: 
> /home/breuerss/local/src/gromacs-4.5.4/src/mdlib/domdec_top.c, line: 352
>
> Software inconsistency error:
> One or more interactions were multiple assigned in the domain 
> decompostion
> For more information and tips for troubleshooting, please check the 
> GROMACS
> website at http://www.gromacs.org/Documentation/Errors
> -------------------------------------------------------
>
> That means that the simulation already ran for some time. I could also 
> finish some runs successfully with the very same topology but 
> different simulation parameters.
>
> For any help or hints how I could fix it I would be grateful.

This one really can't be managed. If you have a checkpoint file from a 
time close before the crash, then it may be possible for you to 
reproduce the error, and so to produce a starting point just before the 
error. Now perhaps a developer could reproduce the error with a view to 
fixing it...

More useful from your point of view would be restarting from that 
checkpoint file on a different number of processors, to try to get past 
the problematic point.

A thorough description of your system, preparation and simulation 
protocol might help someone else suggest a cause, but I wouldn't hold my 
breath :)

Mark