MPI loses the error information

Report any suspected bugs and unexpected behavior here
Post Reply
Jukka Mettälä
Posts: 5
Joined: Tue Sep 20, 2011 10:37 am

MPI loses the error information

Post by Jukka Mettälä » Tue Jan 24, 2012 11:57 am

Hello !

I think this is not really a bug or even a Serpent issue but might be nice to know anyway.

I was doing some test calculations and increased neutron population to the limits and beyond.
At some point with -mpi option I got an error message (due to run out of memory?) witch was not very informative:
Starting the transport calculation cycle...

Sampling initial source...
--------------------------------------------------------------------------
mpirun noticed that process rank 1 with PID XXXXX on node NODENAMEX exited on signal 9 (Killed).
--------------------------------------------------------------------------
Running the same calculation without -mpi option ended up with a nice Segmentation fault error message at the same point of calculation.

And with still larger population I got:
Starting the transport calculation cycle...

***** Tue Jan 24 10:16:21 2012

Input error:

Insufficent memory for neutron population (pop = 135167552, nbuf = 0.0)

User avatar
Jaakko Leppänen
Site Admin
Posts: 2387
Joined: Thu Mar 18, 2010 10:43 pm
Security question 2: 0
Location: Espoo, Finland
Contact:

Re: MPI loses the error information

Post by Jaakko Leppänen » Tue Jan 24, 2012 5:25 pm

The first error message comes from mpirun, not Serpent. The calculation itself probably terminates in segmentation fault, similar to the non-mpi mode.

Have you tried running the calculation with Serpent compiled in the debug mode? This way the calculation could stop with an error message before segfaulting.
- Jaakko

Post Reply