[MITgcm-support] MITgcm flt package
Jordan Dawe
jdawe at eos.ubc.ca
Wed Apr 4 15:19:38 EDT 2007
Ok, let's try this again...
I am trying to get the MITgcm flt package to work in a multiprocessor
MPI setup. I have had success running the flt package on a single
processor without MPI, and on MPI multiprocessors if flt is turned off,
but both of them together causes the model to run but not do anything;
CPU usage shows 0% on all CPUs. The model does not halt or output any
error messages when this occurs.
I am running with the MNC package turned on as well for output; however,
I am initializing everything from binary files. On the successful
single-processor runs, flt has output a binary file. Turning off the
MNC package does not alleviate the problem.
I have found this bug on a 32-processor AMD Opteron cluster and running
a 2 process MPI run on a single-processor Athlon. The bug occurs with
both gfortran and pgf90.
I know very little about MPI coding, and even less about how MITgcm
wraps its MPI calls. I have spent a while trying to debug this problem,
but I have had no success. Does anyone have any suggestions?
Jordan
More information about the MITgcm-support
mailing list