Closed nikizadehgfdl closed 9 years ago
I can provide evidence that I already have 25+ GB of CM4i analysis in /nbhome and I've hit my quota.
Dr. John Krasting Physical Scientist (NOAA Federal) NOAA/Geophysical Fluid Dynamics Laboratory Biogeochemistry, Ecosystems, and Climate Group Princeton University Forrestal Campus 201 Forrestal Road Princeton, NJ 08540
P. (609) 452-5359 F. (609) 987-5063
On Mon, Oct 27, 2014 at 6:41 PM, Niki Zadeh notifications@github.com wrote:
/nbhome/$USER is limited to 10GB and can easily fill up and when that happens the analysis scripts will fail. I already went over quota on /nbhome because of the figures and the following items even though I do not have too many files there. I imagine this is going to be a problem for real model runners.
-
Some analysis tool try to build (MIDAS) libraries in that space.
Some analysis scripts (TS_depth_integrals.py) save larger .nc files in analysis figure destination in addition to small .png.
Can we try /ptmp instead?
— Reply to this email directly or view it on GitHub https://github.com/CommerceGov/NOAA-GFDL-MOM6-examples/issues/4.
Should we add a cleanup of the .nc files to the TS_depth_integrals.py script?
Yes. Also, one way to save space on /nbhome is to put mom6 clone (or anything not needed "permanently") in /ptmp and link it to /nbhome for any given experiment (I have tested that out of necessity since my /nbhome kept filling up).
Is this still an issue?
This also relates to #25 (+x permissions of gcp'd scripts).
The reason we need particular versions of the analysis scripts is they can/do/will evolve with the code and diag_tables. I agree, however, that it is very wasteful to be installing MOM6-examples for each run. Here's some break downs:
We can reduce the footprint by an order of magnitude by doing a sparse clone but this is a little involved. :-1:
Here's a different suggestion: IF we knew the commit id used in the remote repository (think tag) then instead of cloning/copying to the experiment we could clone to /nbhome/$user/mom/<id>
so that there is only one copy per user for all their experiments using the same tag/source. Bonnie runs dozens of runs per tag so the above saves an order of magnitude. :+1:
I think the question then becomes how to get the commit id or tag back to GFDL from the remote platform. I believe you said that FRE does not know it at the frepp stage? Which is why you have to gcp the entire directory? If so, then here's an alternative way to get the tag, still using gcp but this time only transferring 41bytes: gcp <path_to_mom>/.git/HEAD gfdl:/<nbhome_path_to_expt>/HEAD
and then use the contents of HEAD to check for the existence of /nbhome/$user/mom/dev/master/2015.09.21a/
. If missing, clone MOM6-examples into it.
In the .frepp files, we have
set script_dir = ${out_dir}/mom6/tools/analysis
which would need to become
set mom_dir = `cat ${out_dir}/HEAD`
set script_dir = /nbhome/$user/mom/${mom_dir}/mom6/tools/analysis
:cake: (As in "piece of" - I can't find an emoji for :Genius!:)
Most of the space is used up by the analysis figures, mom6 clone takes only a minute part. This is really a workflow issue and not a MOM6 issue. It is more about finding a permanent home for the analysis figures. E.g., a 60 year run takes almost 10G of space (even after .nc files are removed) with mom6 clone being only 178M.
I think we can close this ticket.
/nbhome/Niki.Zadeh/ulm_201505_awg_v20150702_mom6sis2_2015.08.06b_ncrc2.intel15/CM4_c96L32_am4g5r2_2000_sis2_low_mixing3 %
216M atmos_0001_0005
194M atmos_0001_0010
199M atmos_0001_0015
1.1G atmos_0001_0020
209M atmos_0001_0025
214M atmos_0001_0030
219M atmos_0001_0035
223M atmos_0001_0040
227M atmos_0001_0045
232M atmos_0001_0050
236M atmos_0001_0055
237M atmos_0001_0060
25M atmos_0006_0010
25M atmos_0011_0015
25M atmos_0016_0020
25M atmos_0021_0025
849M atmos_0021_0040
25M atmos_0026_0030
17M atmos_0031_0035
17M atmos_0036_0040
17M atmos_0041_0045
1.1G atmos_0041_0060
17M atmos_0046_0050
17M atmos_0051_0055
17M atmos_0056_0060
3.7M atw_atmos_av_annual
2.4G atw_atmos_ts_monthly
439M cvdp
804K db
532K ice_0001_0005
528K ice_0001_0010
528K ice_0001_0015
524K ice_0001_0040
524K ice_0001_0045
524K ice_0001_0050
524K ice_0001_0055
524K ice_0001_0060
528K ice_0021_0040
364K kap_land_0001_0005
364K kap_land_0001_0010
368K kap_land_0001_0015
196M kap_land_0001_0020
380K kap_land_0001_0025
376K kap_land_0001_0030
380K kap_land_0001_0035
380K kap_land_0001_0040
388K kap_land_0001_0045
388K kap_land_0001_0050
388K kap_land_0001_0055
388K kap_land_0001_0060
196M kap_land_0021_0040
196M kap_land_0041_0060
2.3M mdt_xml
178M mom6
6.9M ocean_0001-0005
6.8M ocean_0001-0020
7.0M ocean_0006-0010
6.5M ocean_0011-0015
7.5M ocean_0016-0020
7.5M ocean_0021-0025
6.9M ocean_0021-0040
7.5M ocean_0026-0030
7.5M ocean_0031-0035
7.5M ocean_0036-0040
7.5M ocean_0041-0045
6.9M ocean_0041-0060
7.5M ocean_0046-0050
7.5M ocean_0051-0055
7.5M ocean_0056-0060
1.2M ocean_monthly_0001_0005
1.2M ocean_monthly_0001_0020
1.2M ocean_monthly_0006_0010
1.2M ocean_monthly_0011_0015
1.2M ocean_monthly_0016_0020
1.2M ocean_monthly_0021_0025
1.2M ocean_monthly_0021_0040
1.2M ocean_monthly_0026_0030
12K ocean_monthly_0031_0035
1.2M ocean_monthly_0036_0040
1.2M ocean_monthly_0041_0045
1.2M ocean_monthly_0041_0060
1.2M ocean_monthly_0046_0050
1.2M ocean_monthly_0051_0055
1.2M ocean_monthly_0056_0060
8.9M refineDiag_ocean_annual
Done
/nbhome/$USER is limited to 10GB and can easily fill up and when that happens the analysis scripts will fail. I already went over quota on /nbhome because of the figures and the following items even though I do not have too many files there. I imagine this is going to be a problem for real model runners.
Can we try /ptmp instead?