Difference between revisions of "ACCESS1.0 CMIP"

Line 105: Line 105:
# Change '/short' to '/scratch'
# Change '/short' to '/scratch'
export SHORTDIR="/scratch/$PROJECT/$USER"
export SHORTDIR="/scratch/$PROJECT/$USER"
Also in the .fin file (e.g. 'exp/a10_hist/a10_hist.fin') change 'mppnccombine.RAIJIN' to 'mppnccombine.VAYU' in the 'archjob' line
<syntaxhighlight lang=bash>
archjob=`qsub -v PROJECT,sdir=$sdir,tdir=$tdir,idate=${enddate},mppncombine_exec=$ACCESS_BINDIR/mppnccombine.VAYU $ACCESS_BINDIR/do_mppncombine-cmip5.ksh`

Revision as of 20:46, 6 February 2020

ACCESS 1.0 Coupled Model as used in CMIP5. Includes UM 7.3 atmosphere, MOSES land surface, CICE sea ice, MOM 4 ocean coupled using Oasis3

The model is not available in the UMUI, instead it is run from a set of shell scripts.

A guide to setting up and running the model on the old supercomputer Raijin can be found at https://accessdev.nci.org.au/trac/wiki/access/ACCESS1.0_raijin

Running on Gadi

NOTE: ACCESS-CM1 is only maintained for historical reference and is no longer supported (existing projects excepted). New work should use ACCESS-ESM 1.5, it is much simpler to use and supported.

The component models of ACCESS-CM 1.0 need to be rebuilt in order to run on Gadi, and there are few modifications required to the run script. The instructions here are for the historical experiment mentioned at https://accessdev.nci.org.au/trac/wiki/access/ACCESS1.0_raijin, you may need to modify them if running a different experiment.

On Gadi the model should take approximately 1,400 SU per model year, and run at a speed of about 5 model years per day.

1. Build the models

Follow the instructions at https://github.com/coecms/access-cm10-build-gadi to download the source code and build the model

2. Change the executable paths in your init file to point to the newly built model executables (e.g. exp/a10_hist/a10_hist.init)

Note the newly built executables have a timestamp in the file name that will need to be changed, e.g. edit the lines

cp -f ~access/apps/oasis3/3/bin/oasis3.MPI1.x $cpl_exe
cp -f $ACCESS_BINDIR/mom4_MPI1.exe.20131017.RAIJIN    $ocn_exe
cp -f $ACCESS_BINDIR/cice_MPI1.RAIJIN_6p_20131028    $ice_exe
cp -f $ACCESS_BINDIR/um_hg2-r1.1.exe-20131017        $atm_exe

to instead say (this happens two times in the script):

cp -f $ACCESS_BINDIR/oasis3_MPI1.${EXE_TIMESTAMP}.RJ         $cpl_exe
cp -f $ACCESS_BINDIR/mom4_MPI1.exe.${EXE_TIMESTAMP}.VAYU     $ocn_exe
cp -f $ACCESS_BINDIR/cice_MPI1.VAYU_6p_${EXE_TIMESTAMP}      $ice_exe
cp -f $ACCESS_BINDIR/um_hg2-r1.1.exe-${EXE_TIMESTAMP}        $atm_exe

3. Set the recommended PBS headers for Gadi in your main run script (e.g. exp/a10_hist/a10_hist)

#PBS -q normal
#PBS -l ncpus=144
#PBS -l mem=192gb
#PBS -l walltime=00:30:00
#PBS -l wd
#PBS -W umask=0022
#PBS -l storage=scratch/access+gdata/access

4. Set the recommended Decomposition (for a total of 144 cpus, 192 gb memory) by modifying variables in the main and init scripts (e.g. exp/a10_hist/a10_hist, exp/a10_hist/a10_hist.init)

Component Variable Value
MOM oce_nx 8
MOM oce_ny 5
MOM nproc_ocn 40 (oce_nx * oce_ny)
CICE nproc_ice 6

5. Fix the main script for Gadi (e.g. 'exp/a10_hist/a10_hist')

A few edits are required throughout the script:


# Should point to the 'bin' directory from building the model

# Also load openmpi here
module load openmpi/4.0.2

# ...

# Replace the square brackets and their contents with 'true'
# if [[ "$PBS_O_HOST" == "raijin*" || "$PBS_O_HOST" == "r"*{0,2}(\d) ]] ; then
if true; then
         export exec_site="NCI"
         export exec_system="gadi"

# ...

# Change '/short' to '/scratch'
export SHORTDIR="/scratch/$PROJECT/$USER"

Also in the .fin file (e.g. 'exp/a10_hist/a10_hist.fin') change 'mppnccombine.RAIJIN' to 'mppnccombine.VAYU' in the 'archjob' line

archjob=`qsub -v PROJECT,sdir=$sdir,tdir=$tdir,idate=${enddate},mppncombine_exec=$ACCESS_BINDIR/mppnccombine.VAYU $ACCESS_BINDIR/do_mppncombine-cmip5.ksh`

6. Perform a test run

Change the run length settings in the init script (e.g. 'exp/a10_hist/a10_hist.init') to perform a 1 model month run:

#-- Initial and Final Date of the Experiment
iniyear=1951;   finalyear=1951;         typeset -Z4 iniyear  finalyear
inimonth=1;     finalmonth=1;           typeset -Z2 inimonth finalmonth
iniday=1;       finalday=31;            typeset -Z2 iniday   finalday

inidate=${iniyear}${inimonth}${iniday}                  #YYYYMMDD
finaldate=${finalyear}${finalmonth}${finalday}          #YYYYMMDD

#-- Duration of a run/chunk (HERE is most often visited for test/short runs!)
nyear=0                 # number of years
nmonth=1                # number of months
nday=0                  # number of days

Then qsub the main script (e.g. 'exp/a10_hist/a10_hist')

Raw model output gets sent to $SHORTDIR/archive/history/$EXPT (e.g. /short/w35/saw562/archive/history/a10_hist)

MOM, CICE & Oasis Model logs and namelists are in $SHORTDIR/work/um_coupled/$EXPT

UM logs are in $SHORTDIR/data/um_coupled/$EXPT

This should take approx 30 minutes to complete

Model Configurations