UM - Minimum working example?

Atmosphere Unified Model
, , ,
21

Are you able to use the ancillaries under /g/data/access/TIDS/UM/ancil/atmos/GC5/n320e? This is the main mirror of ancil data from the Met Office.

1 Like
22

I support @Scott’s assertion/suggestion. (If it works that is, and I can’t see why it wouldn’t.)

23

Ok the ancils are no problem. The issue is the start_dumps.

There are three directories in /cylc-run/u-dl058/share/data/etc that link to /g/data/dx2/gc4/2013-2014_ecmwf_initial_conditions/:

ocean_startdumps -> /g/data/dx2/gc4/2013-2014_ecmwf_initial_conditions/FOAMv14d
startdumps -> /g/data/dx2/gc4/2013-2014_ecmwf_initial_conditions/ecmwf_raw
surf -> /g/data/dx2/gc4/2013-2014_ecmwf_initial_conditions/surf

But I can remove those links from that directory and um_recon works fine. Evidently you don’t need those files when running in uncoupled mode.

However, the UM atmospheric dump comes from the dp9 project:
/g/data/dp9/ukmo_mass/opfc/atm/mogreps-g/rerun/202001.file/20200116T0000Z_engl_t+3_000

So @griff it looks like you’re off the hook and I can remove all dx2 dependencies.

But we are left with a single dp9 dependency.

Can we move some of the UM start dumps into ~access?

The entire directory /g/data/dp9/ukmo_mass/opfc/atm/mogreps-g/rerun/202001.file is 398 Gb, but a single start dump is only 17 Gb.

A ‘minimum working example’ would only require one file?

24

Happy to, can you and the NRI folks give me a good path to put it?

25

This path already exists and seems to be appropriate for the task:

/g/data/access/GC5_ancil/gc5/startdumps/
26

@MartinDix the directory /g/data/access/GC5_ancil/gc5/startdumps is only writable by you, if this is a good location can you please change the permissions and I’ll add this file

27

Was thinking it might be good to put a README.txt in that directory so in the future, anyone wandering into that directory knows the UM restart file is used by suite u-dl058

EDIT : Was also thinking it might be good to store the outputs so users can compare against the baseline. The output netCDF files for a one day’s worth of forecasts (output at 6 hourly intervals) are 644 Mb.

See

/scratch/gb02/pag548/GC5_runs/u-dl058/atmos/

for an example.

28

@Paul.Gregory What is the current status of this suite? Has your question been answered?

29

Hi Paul

I’m just getting back from leave today. Currently the suite still has a dp9 dependence due to the start dump file

/g/data/dp9/ukmo_mass/opfc/atm/mogreps-g/rerun/202001.file

Once this file has been copied to

/g/data/access/GC5_ancil/gc5/startdumps/

the suite should be available to use by the general community.

30

I have marked this topic as solved. Please reply if this is not the case.

31

Hi Paul

I’d like to wait until the start dump in question has been copied to

/g/data/access/GC5_ancil/gc5/startdumps

before the topic is solved.

Once we remove the dp9 dependency the suite will be available for general users, requiring no additional NCI project memberships outside of the standard access/rose projects.

Thanks

Paul

32

/g/data/access/GC5_ancil was some some ancillary data downloaded from JASMIN before it became part of the standard ancillary collection. Now superseded by ~access/umdir/ancil/atmos/GC5. No idea why there was ever a startdumps subdirectory. I don’t think it’s an appropriate location for a MOGREPS restart becasuse I’d been intending to delete it.

Any other ideas?

33

Here are some other existing directories.

~access/data/input_dumps/
~access/data/ACCESS_initial/atm/
~access/data/UM_Input/
34

This does raise the larger question of when to use /g/data/access for start dumps vs when to use other projects. I will pose that question as a separate topic.

35

I noticed that I am no longer a member of dp9. Perhaps I got purged over the summer, as I’m not currently working at the Bureau?

In any case, I’ve just run the suite again and the task retrieve_atmos still works fine, without dp9 membership.

The task pulls the start dump via the Moose archive with the following commands

  ATMDUMP_MOOSE=moose:opfc/atm/mogreps-g/rerun/${YYYYMM}.file/${DATE_restart}00Z_engl_t+3_$( printf '%03d' $en_instance )

  echo "Retrieving ATM ENGL DUMP " $en_instance " for " $YYYYMMDDTHH " from MASS" 
  echo $ATMDUMP_MOOSE
  moo get $ATMDUMP_MOOSE $ROSE_DATAC/

I’m guess you can use the moo command without the dp9 membership, as the client works around the project membership? From the job.out file

2025-02-20T03:58:51Z INFO - started
Retrieving ATM DUMP for  1  ensemble members starting at 0
Retrieving ATM ENGL DUMP  0  for  20200116T00  from MASS
moose:opfc/atm/mogreps-g/rerun/202001.file/20200116T0000Z_engl_t+3_000
/g/data/access/projects/access/apps/moose-client-wrapper/1.0.4/moo/moo: STORAGE_TYPE=gdata
/g/data/access/projects/access/apps/moose-client-wrapper/1.0.4/moo/moo: STORAGE_TOP_DIR=/g/data/dp9/ukmo_mass
/g/data/access/projects/access/apps/moose-client-wrapper/1.0.4/moo/moo: Copying /g/data/dp9/ukmo_mass/opfc/atm/mogreps-g/rerun/202001.file/20200116T0000Z_engl_t+3_000 to /home/548/pag548/cylc-run/u-dl058/share/cycle/20200116T0000Z/
2025-02-20T03:59:09Z INFO - succeeded

In which case - can we mark this as solved and the suite can be disseminated amongst the community?

36

Sure. I will now close this topic.