-
Notifications
You must be signed in to change notification settings - Fork 1
Run experiments on Aaditya
Each Aaditya computing node has 16 cores, and 64 GB RAM (60GB usable). The directory /tmp/
for each computing node is 500MB so don't write out temporary files into /tmp/
during the experiments.
Please refer to the instructions from the mail sent by the Aaditya staff. You can find the username and password from the Eugenia's email "Re: VPN needed for us to access Aditya".
- After enabling VPN, connect Aaditya with your account and password provided by the Aaditya staff.
- Set the new password following the instructions given by Aaditya.
- Login into management node:
ssh mn1
with the password provided by the Aaditya staff. - Change the password, and set it as the same as your new password in Step 2.
- Wait a few minutes for the system to update this new password to all computing nodes.
- If changing the password in the future, log into
mn1
to make modifications and the change will be applied automatically to all nodes.
We can connect to GitHub through SSH. Please refer to https://docs.github.com/en/free-pro-team@latest/github/authenticating-to-github/connecting-to-github-with-ssh.
Note: When generating the SSH key, don't forget to set additional password!!!
- Download ESMF_3_1_0rp5 from https://earthsystemmodeling.org/static/releases.html.
- On Aaditya, load the Intel compiler 2014
module load intel/2014
. Make sure the version ofmpiifort
is consistent with yourifort
:which mpiifort
,which fort
- Set two environmental variables:
export ESMF_DIR=/backup2/your_usrname/your_preferred_directory
export ESMF_COMM="intelmpi"
make
make install
make installcheck
- get clone of CFS-LETKF from [email protected]:UMD-AOSC/CFSv2-LETKF.git
git clone [email protected]:UMD-AOSC/CFSv2-LETKF.git
- check out the branch
Aaditya
:
git checkout Aaditya
If switched successfully, you should see lsf.py
, lsf_cycle
under directory /run
.
- Modify var
ESMFDIR
in the fileconfig/makefile.Aaditya_intel2014.mk
. Make sureMACHINE=Aaditya
in the fileletkf-mom/config/machine.sh
. Then
ln -s makefile.Aaditya_intel2014.mk makefile.mk
ln -s setenv.Aaditya_intel2014.sh setenv.sh
Run the command source config/setenv.sh
and then start compiling by following the compiling instructions at https://github.com/UMD-AOSC/CFSv2-LETKF/wiki/Compiling with.
The branch Aaditya
has two main difference from the branch develop
:
- Aaditya uses LSF instead of SLURM.
- The computing nodes of Aaditya has small temporary space.
Based on these two situations, two solutions are developed:
-
lsf.py
, andlsf_cycle
are created. They are similar toslurm.py
, andslurm_cycle
. - Add option to disable
TMP_DIR_LOCAL
, and use shared directoryTMP_DIR_SHARED
only.
Note for lsf.py
, the function of banning nodes is not finished yet. So do not use it.
set properly the vars TMP_DIR_LOCAL
, TMP_DIR_SHARED
, USE_TMP_DIR_LOCAL=0
, FIX_DIR_AM
, FIX_DIR_OM
, OBS_ATM
Please copy the following files to your own directory:
- Fixed files for CFS:
/backup2/cheng/shared/CFSv2-LETKF.data
- Decoded PrepBUFR obs from
2006060100
to2006080118
:/backup2/cheng/shared/prepbufr
- Ocean profiles from:
- NOAA L4 SST from:
- Compressed sample 40-member initial condition of
2006060100
from:/backup2/cheng/shared/backup_compressed
- You can use the new script
init_copy
under/run
to generate the required file structure for CFS-LETKF running. For example,
./init_copy --source /backup2/cheng/CFSv2-LETKF/DATA/backup --ares 126 2006060100 /backup2/cheng/CFSv2-LETKF/DATA/mem5 5
- Start 1-day experiment: go to directory
/run
, then run the command
./lsf_cycle --aobs /backup2/cheng/CFSv2-LETKF/DATA/obs/atm_prepbufr --ares 126 --clear 24 --mem 40 /backup2/cheng/CFSv2-LETKF/DATA/mem40 2006060100 2006060200
- After finishing the experiment, check the controller log file to see if there is any error. Then check the ATM-LETKF log file for each cycle to see if observations are correctly assimilated.