[MITgcm-support] Adjoint tutorial problems

Marcello Magaldi Marcello.Magaldi at jhu.edu
Fri Feb 3 20:40:48 EST 2012


Hi Folks,

our MITgcm forward simulations (usually!!) are running fine with
different MPI compilers on our cluster.

I am observing a strange behavior when running the
"verification/tutorial_tracer_adjsens" experiment, however.
Specifically:

1) The adjoint tutorial runs fine when the processors involved are
within one node (e.g. PBS specification is like 1:ppn=8), regardless
of which compiler is used.

2) When the processors involved are NOT within one node  (e.g. 2:ppn=4, same
executable as before) and I am using OpenMPI (v1.4.3) compiled with
GNU compilers,
the tutorial still runs fine

BUT...

3) When the processors involved are NOT within one node and I am using
OpenMPI (v1.4.3) compiled with INTEL compilers, the tutorial quits
soon after printing the initial files with
a segmentation fault


Did someone else have similar experiences? Or can somebody try to
reproduce this problem?

Thanks in advance for any help/suggestion,
Marcello



More information about the MITgcm-support mailing list