Changes in ECLIS with V6.4

Eclis 6.4 prepares for CNRM-CM6 : it is adapted to coupling using Surfex, to the use of the MCT version of Oasis3, and to a new version of Trip.

Article mis en ligne le 25 juillet 2014
dernière modification le 17 juillet 2014

par senesi

Eclis 6.4 prepares for CNRM-CM6 : it is adapted to coupling using Surfex and the MCT version of Oasis3, and to a new version of Trip.

Warning : if you want to enforce backward compatibility for Oasis and Trip and coupling by Arpege, and you want to use version 6.4 or later, you must write LCOMPAT_64=1 in your experiment parameter file. If you have a more complex requirement (e.g. one component the old way and another one the new way) see below, and/or ask your local dealer.

  • Coupling :
    • shifting the coupling duty from Arpege to Surfex is implemented, and is now the default configuration. If you are defining a coupled experiment, and you use an Arpege binary which does handle the coupling (which is true at least up to Arpege Climat 6.1), you must set LSFXCPL =0 in your ’param’ file.
    • Eclis tries to be smart in diagnosing whether you run is coupled or not (from the list of binaries)
    • default number of coupler processes is now 0 (in line with Oasis-MCT design) ; but you may still set NPROC_CPL if you dare too.
  • Automating monitoring : Script verifrelan (in directory ’scripts’) can be used (with ’cron’) to monitor and re-lauch crashed experiments. See its auto-doc
  • Checking differences between experiments : script ’diffoce’, in directory ’scripts’, eases comparing all Nemo output files between two experiments for a given year+month
  • Various :
    • Eclis tries to diagnose which model is faulty (but, for the time being, Beaufix does not send send the diagnostic mail)
    • Surfex outputs tar files now also include the atmopsheric namelist
    • Parameter LSFXSPINOUT allows to toggle the saving of Surfex output files during the spinup period ; it defaults to 1 (= do save it) (for explanation about spinups with Eclis, see Running Surfex offline with a spinup)
  • Sea-ice cold start : Starting an experiment without any sea-ice restart file is possible, using LICEREST =0
    • MPI outputs are now ordered in stdout
    • For easy access to the ’steps’ directory of the previous run when necessary (i.e. using ARCHIVING=DURING), an additionnal symbolic link is available in the ’relance’ directory, named steps _previous
  • Advanced use  :
    • Parameter POSTOCE allows to customize an oceanic files post-processing script, which is called at step 3 of the simulation, and which defaults to script fix_nemo_time_counter from ECLIS toolbox directory
    • Parameter POSTRIV allows to provide an alternate script for post-processing river routing model outputs at step2. It defaults to (which now does ’tar’ the output files and name them with a date), except if LCOMPAT_64 is set, in which case its defaults to toolbox/ (for ’old’ trip)
    • Scripts for updating namelists for land surface, rivers and sea-ice models can now be explicitly provided using parameters UPDNAMSLM, UPDNAMRIV and UPDNAMICE (which default to standard Eclis scripts)
    • Charging an account other than the default one is possible, see Working with other accounts on HPC and on the storage system
    • Unix command ’ls’ is now replaced by $LS, which default to ’lfs ls’ on beaufix for taking advantage of Lustre
    • The install script will end immediately (without installing) if it is provided with argument ’-nogo’ ; this allows to source an ECLIS param file without installing, provided that it calls the installer as $INSTALLER $* ; this is useful when defining a new experiment with few changes w.r.t. the first one
    • Environment variable CM_TAR_OPTIONS (resp. CM_TAR_OPTIONS_OUTPUT) is introduced to provide options to ’tar’ when working on restarts (resp. diagnostic outputs) ; on beaufix, default value is --blocking-factor=2000 ; you may tune it in your .bash_profile or in the experiment’s ’.conf’ file.
  • Bugfixes :
    • Re-running or going on with old experiments will no longer crash on missing directories
    • On beaufix, Lustre buffers/locks are flushed before launching the single or coupled model, in order to avoid some issues when reading namelists
    • Updating the namcouple now accounts for the LSM spinup cases (correction of number of days in the month for the year of forcing, in order to account correctly for LEAP_YEAR)
    • for Nemo outputs, re-activate renaming of t_ave01month coordinate to time_counter (using a ’module load nco’)
    • Script ’setup’ now exits succesfully
    • Fix a bug with long files lists