MPI blow up for east-west periodic boundary conditions

[peisach]

Hi. I'm attempting some runs with east-west periodic boundary conditions. The model works fine in serial mode (on an Alpha) but blows up imediately in MPI mode (SGI).

The relevant (I believe) cpp options used are:

#undef EASTERN_WALL
#undef WESTERN_WALL
#define EW_PERIODIC
#define SOUTHERN_WALL
#define NORTHERN_WALL

I also read grid and initial fields from netCDF:

Any suggestions? Thanks,
Pål Erik Isachsen

[m.hadfield]

Hi. I'm attempting some runs with east-west periodic boundary conditions. The model works fine in serial mode (on an Alpha) but blows up imediately in MPI mode (SGI).

I'm not aware of any general problem with east-west periodic boundary conditions and MPI in ROMS 2.1. Some of the test cases use east-west periodic boundary conditions (UPWELLING, FLT_TEST) and I have run them successfully on a Cray T3E with MPI.

I also read grid and initial fields from netCDF:

There might be an issue there if the code fails to impose periodicity after it has read the fields.

Any suggestions?

Get out your favourite debugger. Turn on the compiler's bound checking option. Set aside a day and get to it. Report the solution back to the forum.