Load balancing tool usage for CESM 2.3 alpha17b

[mike_scchen@gmail_com]

Hi,
We've ported the CESM to our machine, and would like to try the load balancing tool in it.
We've never tried this tool in previou CESM versions.
However, we have the following error message:

[u8935479@ilgn01 ~]$ export CIME=$HOME/cesm2_3_alpha17b/cime
[u8935479@ilgn01 ~]$ export CASE=$HOME/17b_2022_impi
[u8935479@ilgn01 ~]$ export PYTHONPATH=$CIME/CIME:$CIME/tools/load_balancing_tool
[u8935479@ilgn01 ~]$ ~/python-3.6.8/bin/python3 $CIME/tools/load_balancing_tool/load_balancing_solve.py --timing-dir=$CASE/timing --total-tasks=1024
found directory /home/u8935479/17b_2022_impi/timing
Reading timing file /home/u8935479/17b_2022_impi/timing/cesm.ESMF_Profile.summary.36837.240611-194217
Reading timing file /home/u8935479/17b_2022_impi/timing/cesm_timing.17b_2022_impi.36837.240611-194217
Using layout = IceLndAtmOcn
Traceback (most recent call last):
File "/home/u8935479/cesm2_3_alpha17b/cime/tools/load_balancing_tool/load_balancing_solve.py", line 432, in <module>
_main_func(__doc__)
File "/home/u8935479/cesm2_3_alpha17b/cime/tools/load_balancing_tool/load_balancing_solve.py", line 424, in _main_func
json_input,
File "/home/u8935479/cesm2_3_alpha17b/cime/tools/load_balancing_tool/load_balancing_solve.py", line 367, in load_balancing_solve
opt = optimize_model.solver_factory(data)
File "/home/u8935479/cesm2_3_alpha17b/cime/tools/load_balancing_tool/optimize_model.py", line 28, in solver_factory
expect(data.has_key("totaltasks"), "totaltasks not found in data")
AttributeError: 'dict' object has no attribute 'has_key'

Please check the attachments for the timing files.

Mike

 

[jedwards]

Hi Mike,

That load balancing tool has not kept up with model progress and is not suitable for use with the current model.
I'm sorry but the only option at this time is hand tuning. I will bring this up at our developers meeting but I suspect that
with all the things on our task list to prepare for the cesm3.0 release we will probably remove this tool rather than update
it to the current model structure.

Jim

 

[mike_scchen@gmail_com]

Hi Jim,
Thanks for the info!
I'll ask the instructor for manual PE arrangement guides then :)
I'm using 1280 PEs now and found that it's fastest to simply use 1280 PEs for all components (except the ESP) for B1850MOM compset so far.

Mike

 

[jedwards]

That should not be the case - use cesm/cime_config/config_pes.xml for example PE layouts that should perform better.
Generally ATM_PES=CPL_PES and LND_PES=RTM_PES and LND_PES+ICE_PES=ATM_PES with ROOTPE_ICE=LND_PES
and ROOTPE_OCN=ATM_PES. So that ocn runs concerently with everything else and lnd and cice also run concurrently.