Hi,
I am following this post closely, actually Tom and I are working on same project and I am wondering because F2000climo does not have hourly outputs.
Is there any compost that allows to run tidally and get history (diagnose) files?
Additionally, are you running CESM on Cheyenne? It is because the time needed for those run models.
Thanks
You can manually change the frequency that the various components write history output - for instructions on how to do that, I would look in the various component forums on this bulletin board (e.g. questions about CAM should be asked
here).
I am running on cheyenne, and if you are as well I suggest you look at the
CESM 2 timing table to get a rough estimate for run time. I only see entries for the
f19_f19_mg17 resolution (I ran my test with
f09_f09_mg17; out of the box with that resolution, it looks like
F2000climo uses 360 MPI tasks with 3 OpenMP threads per task for a total of 1080 Total PEs) but if you are running at 2 degree resolution the default layout is 72 MPI tasks with a single thread per task. You can run
./pelayout from your case directory to see exactly what you are running on.
If you do look at the timing table, you should pay attention to the
ThruPut yrs/day number; that is the number of model years simulated in a 24 hour run. Since cheyenne limits jobs to a 12 hour wallclock, you should expect to be able to model half that number of years per submission; note that writing daily output will decrease throughput as the time it takes to write to disk will not be used to progress the model.
Lastly, after a successful run, you should notice a directory named
timing/ in your case directory; the
cesm_timing.${CASENAME}... files will give you information about your specific run. Here is the timing from when I ran for a single month to generate the monthly history files:
Code:
grid : a%0.9x1.25_l%0.9x1.25_oi%0.9x1.25_r%r05_g%gland4_w%null_z%null_m%gx1v7
compset : 2000_CAM60_CLM50%SP_CICE%PRES_DOCN%DOM_MOSART_CISM2%NOEVOLVE_SWAV_SIAC_SESP
run type : startup, continue_run = FALSE (inittype = TRUE)
stop option : nmonths, stop_n = 1
run length : 31 days (30.979166666666668 for ocean)
component comp_pes root_pe tasks x threads instances (stride)
--------- ------ ------- ------ ------ --------- ------
cpl = cpl 1080 0 360 x 3 1 (1 )
atm = cam 1080 0 360 x 3 1 (1 )
lnd = clm 1080 0 360 x 3 1 (1 )
ice = cice 1080 0 360 x 3 1 (1 )
ocn = docn 1080 0 360 x 3 1 (1 )
rof = mosart 1080 0 360 x 3 1 (1 )
glc = cism 1080 0 360 x 3 1 (1 )
wav = swav 1080 0 360 x 3 1 (1 )
iac = siac 1 0 1 x 1 1 (1 )
esp = sesp 1 0 1 x 1 1 (1 )
total pes active : 3240
mpi tasks per node : 36
pe count for cost estimate : 1080
Overall Metrics:
Model Cost: 1769.92 pe-hrs/simulated_year
Model Throughput: 14.64 simulated_years/day
The important line is
Code:
Model Throughput: 14.64 simulated_years/day
If I were trying to run 100 years, I would run
./xmlchange STOP_N=7,STOP_OPTION=nyears,RESUBMIT=13 so the model runs for 7 years at a time (14.64 / 2, rounded down), restarting itself 13 times for a total of 14*7=98 years. I would also run
./xmlchange --subgroup case.run JOB_WALLCLOCK_TIME=12:00:00 to request 12 hours per submission, and then
./case.submit will take care of the rest.
I'm getting a little ahead of myself, though. I think the first thing you should do is figure out the namelist settings to get the output you want at the frequency you want. Then run for a year or two (maybe with
./xmlchange --subgroup case.run JOB_WALLCLOCK_TIME=6:00:00) and look at the timing file to get a better sense of throughput before continuing the run in bigger chunks.
