SWAN coupling, with output

[nganju]

On separate occassions, at different times during the same simulation, ROMS/SWAN coupled runs halt in the middle of the run. There is no error in the ROMS output file, the SWAN PRINT file is usually in the middle of processing an output request, and the cluster issues an error file with the phrase "done."

When I turn output commands off in SWAN, all runs fine. When output commands are on, I have this problem. I need the output (spectral wave parameters) in order to drive a nested run, so I absolutely must have the output commands on. Any ideas are appreciated...

[arango]

That sounds like a parallel bug in SWAN. PRINT or WRITE statements in parallel codes require special care. I guess that we need to hunt for that problem. However, you do not give us enough information to track this problem.

[nganju]

The offending output command in the SWAN input file is SPECOUT (output 2D spectra); the TABLE command (output parameters in ascii format) has also caused the same problem. The ROMS output file stops after writing a typical line (time, kinetic energy, etc). The most recently written-to SWAN PRINT file (there is one for each processor) pauses in mid-output while processing a file request. I don't know what else is of use. Let me know if there are certain files that should be posted, thanks...

[jcwarner]

so i assume that when you have the output being written to spec2d files, the files have data in them for several output times, and the models haved synchronized and exchanged fields several times. So it does not sound like the coupling. If it worked once, then it should continue to be ok. I have run several cases for long durations and have not had problems, but there still could be issues. What are the file sizes of the spec2d files when the run stops? Can you try it with spec2d write out but just 1 processor to swan ( i know that takes a while longer, but it will help debug the issue)? Does the roms output look ok? have you updated to the latest version? I recently corrected an issue with spec2d out files (ticket # 154). I am aware of an issue with the table out, and am correcting that now.

[nganju]

SWAN spec2d files are about 1MB, ROMS output looks fine, I can try with one processor and report back. I will also update if necessary...

[mathieu]

Since we are at SWAN and ROMS, let me point out a problem others might have encountered.

It is necessary when running SWAN coupled with ROMS to output at least one field from the SWAN program, i.e. to have something like

Code: Select all

BLOCK 'COMPGRID' NOHEADER 'depth.mat'  LAY 4 DEPTH 1.    OUTPUT 20000101.000000 1 HR

in your swan.in file even if you do netcdf output of wave information from ROMS.

The reason is that the model coupling is done in the SWOUTP function, which is not called if there is no output and thus led to a forever blocked program.

[jcwarner]

As the coupling evolves, I am sure many things will change. Here is some more information on the coupling. Please comment back if this is clear/not clear.

https://www.myroms.org/wiki/index.php/swan.in