[mpich-devel] ROMIO collective i/o memory use

Rob Latham robl at mcs.anl.gov
Tue May 7 16:38:25 CDT 2013

On Mon, May 06, 2013 at 03:05:10PM -0500, Bob Cernohous wrote:
> Ouch.  So you've seen the scaling failures at 8k... ranks? racks?
> Kevin is failing at... 16 racks x 16 ranks per node... I think ...
> so 256k ranks.

MOAB in particular has fairly straightforward workloads that will
exhaust memory (by attempting a 400 MiB memory allocation) at 2048 MPI


Rob Latham
Mathematics and Computer Science Division
Argonne National Lab, IL USA

More information about the devel mailing list