Charles Pelletier a5c609c527 nemo rs před 3 roky
..
readme.md a5c609c527 nemo rs před 3 roky
redisplay_restart.py a5c609c527 nemo rs před 3 roky

readme.md

Redisplaying restart files for "tricking" NEMO into starting from a custom restart

At initialization, NEMO either starts from:

  • rest, with temperatures and salinities taken from namtsd in namelist_cfg, and ice fields potentially initialized in namelist_ice_cfg:

  • a full restart file, which is a snapshot of a run's RAM at a given time. NEMO restarts are specific to the domain decomposition of your setting: configuration, number of cores, potential use of "land suppression" (check nammpp in NEMO namelist).

Hence, for starting NEMO from a "custom" restart file (e.g. spin up), you'll need to redistribute the restart according to the domain decomposition you want to run with. This is what this tool does. It requires:

  • A set of processor-specific NEMO restart files defined with a certain domain decomposition, that you'll want to run at (2*n_proc, n_proc for oce and n_proc for ice). You can get this by doing a dummy run with the number of processors you'll eventually want.

  • A pair of rebuilt (processor-wise) NEMO restart files, which you'll want to get the actual restart data from. Check the REBUILD_NEMO tool for rebuilding NEMO outputs.

Step 1: on the ELIC servers

Prepare the restart files with the help redisplay_restart.py (the code is commented), and the ingredients above. Upload the resulting restart file sets to the cluster you want to run on.

Step 2: on the cluster where NEMO is running

In experiment.cfg, set:

    special_rs_nemo=true
    special_rs_nemo_stencil="<path_to_restart_files>"

Coral expects restart files to be named <path_to_restart_files><oce,ice>_XXXX.nc, where XXXX are the 4 digits distinguishing processor numbers.

In build_namelist.cfg.sh, set ln_rstart = .TRUE., and nn_rstctl = 0.

Build and submit.