Very short test runs of WACCM/CESM1.0.4 give segmentation fault

[andreasb]

Hi!
For testing a new module I want to make very short runs (e.g. 2 timesteps) to look at the output (output every timestep).

./create_newcase -res 10x15_10x15 -case test4 -compset F_2000_WACCM -mach absuseg95

However, this crashes with a segmentation fault before the wrapup of the netcdf files, so I cannot look at the output files (the data is in there, but the time axis is not updated and has 0 timesteps, so any reading programs cannot see the data).

What is the shortest possible model run length (e.g. with one day it works fine, but that takes 30 minutes on my laptop - too long for model development purposes.)

thanks for your help!
Andreas
NCAR-HAO, CG1

 

[andreasb]

Hi,
With more tests I managed to find a solution: The crash must be caused by the restart-component, because the following 2-step simulation works:
drv_in:

stop_option    ='nsteps'
  stop_n         = 2
  stop_ymd       = -999
  restart_option ='ndays'
  restart_n      = 1

regards,
Andreas

 

[santos]

I know that this has been here a while, but I thought I'd point something out. The CAM regression tests do restart and branch tests that run the model for 4-6 timesteps. So while I'm not sure what the minimum is for any given configuration, 4 is probably safe. With that few timesteps, most of the run time is usually in initialization and I/O, so 4 timesteps should not cost that much more than 2.