GFDL Hybrid Model (version 5)
Quick-start guide for GFDL users

Get the files

• Create root, archive, and input paths. Recommended paths:
  • ROOT: /home/$USER/fms/run/hybrid/sfm1p5_180x96x25
  • ARCHIVE: /archive/$USER/fms/run/hybrid/sfm1p5_180x96x25
  • INPUT: /archive/$USER/fms/run/hybrid/sfm1p5_180x96x25/exp01/input
• Copy input_data.tgz to your INPUT directory.
  Unpack the data: gunzip input.tar.gz; tar -xvf input_data.tar
• Copy hybrid.xml to your ROOT directory.
• Copy and unpack preproc.tgz to your ROOT.
• Copy and unpack tables.tgz to your ROOT.

Setup your environment

• If necessary, change ROOT and ARCHIVE directories at top of hybrid.xml.
• On your Linux workstation: ensure that the RTS/FRE tools are in your path:
  setenv PATH /home/fms/bin:${PATH}
• I would also recommend copying any of my datasets listed in hybrid.xml (grep for atw) to your ARCHIVE, so that you will not be affected if I change or move them.  If you do this, change the paths in your hybrid.xml accordingly.

Compile

• In your ROOT directory, do: rtsmake -x hybrid.xml exp01
  This will check out all the source code, and generate a compile script. The output will indicate how to run the compile script. Submit this as indicated to the HPCS, to compile the code and generate the model executable.

Run a short test

• In your ROOT directory, do: rtsrun -x hybrid.xml -r basic exp01
  This will generate a run script. The output will indicate how to run this script. Submit this as indicated to the HPCS, to run ONE submission of ZERO MONTHS, ONE DAY, on 16 PROCESSORS (1x0m1d_16pe).
• Edit runscript so that the diag_table is set to output diagnostics every timestep. Do this by editing the lines below REGISTER NETCDF FILES to read:
  "ocean_month", 0, "months", 1, "hours", "Time"
"stat_month", 0, "months", 1, "hours", "Time"

• The job output will go to a subdirectory exp01/1x0m1d_16pe inside of ARCHIVE. If the run fails, the job output and error information will be found inside the directory exp01/1x0m1d_16pe/ascii; ensure that all paths listed in the run script are correct.
• Once the job succeeds, the diagnostic output will be found in exp01/1x0m1d_16pe/history and the restart files in exp01/1x0m1d_16pe/restart. Uncpio these files and analyze the NetCDF data. Compare your results against mine.

Run a longer test

• In your ROOT directory, do: rtsrun -x hybrid.xml exp01
  This will generate another run script. The output will indicate how to run this script. Submit this as indicated to the HPCS, to run TWO submissions of 12 MONTHS, ZERO DAYS, on 16 PROCESSORS. As this is a production run, the output will go to the directory exp01/history inside of ARCHIVE.
• Analyze the results to ensure that things look reasonable.
• To link history files together for easy analysis in Ferret, I've written a perl script (make_des) to automatically generate descriptor files. Run "make_des --help" for usage.

Modify the runtime options

• The XML file completely specifies the experiment, including all namelists. RTS inheritance makes it easy to define a new experiment with just a few changes in parameters -- see the RTS documentation for details. Then just do another rtsrun on the new experiment to generate a new runscript.

Modify the source code

• The source code is contained within the src subdirectory of ROOT. After modifying the source code, simply submit the compile script again and a new Makefile and executable will be generated.