Version 5 (modified by flavoni, 5 years ago) (diff)

List of possible errors

How to perform a second simulation after a run with PeriodState=Fatal

rm Job_name/run.card; rm Script_Output* rm everything created during the simulation in IGCM_OUT

but do not cancel run.card. init file, if not run.card file is not re-created.

( File run.card.init is created with ins_job script )

Submit the main job —- or you can use clean_month.job script: cd EXPERIENCE
../../../lib IGCM/ clean_month.job

How to continue a simulation

Update config.card with the new DateEnd . Do not change DateBegin . Change PeriodState in run.card to OnQueue (or let clean_month job do it for you) Submit the main job

How to continue a simulation that crashed previously

If you want to relaunch a crashed simulation you need first to remove all files produced for the crashed period. Use the script modipsl/libIGCM/clean_month.job and answer yes to delete files and change run.card Submit the main job

RebuildFromArchive=NONE

In MPI it is good to use Rebuild in "archive" machine, to not ovrerload calcul machine.
You have to put in config.card in POST Session:

RebuildFrequency =1Y
RebuildFromArchive =true

NOTA BENE: If you put RebuildFrequency =1Y and you forget to set RebuildFromArchive =true, it is very dangerous. NEVER DO IT!

How to do a long job without exit from queue

You can play with Duration Experience, Period Lenght? and Period Nb? , elapstim_req in this way :
in config.card Duration Experience is = DateBegin - DateEnd
in config.card PeriodLenght is how many time you do restart files
in Job_Name PeriodNb is in Job_Name elapstim_req is elapsed time limit

Your job is out from queue every time that is finished a Period Nb? , so you can do :

PeriodNb = ( Duration Experience ) / PeriodLenght

but be careful with elapstim_req, because this is a limit, you've to be sure that it exists a queue that can hold you.

Simona Flavoni last update: Timestamp?