WRF model efficiency is unusually low

[MSdueon12]

Hello,
I am currently running the WRF model on the Derecho HPC system under {{/glade/work/zhouxu/wrfv4.3}}. However, I found that the model efficiency is unusually low.

I submitted the job using the {{qsub.wrf}} script with 10 CPU cores, which should be sufficient for this simulation. However, after approximately 6 hours of wall-clock runtime, the model only simulated about 13 minutes of model time. This performance is far lower than expected.

I have previously run the same model configuration and settings, and under normal conditions I would expect approximately 5–6 hours of simulated model time within 6 hours of runtime using 10 cores. I tested using the default version, which suggests that no custom modifications were made to the original configuration.

Therefore, I suspect there may be an improper setting or configuration issue causing the low simulation efficiency. I would greatly appreciate it if someone could help me identify and solve the problem.

The {{qsub}} submission script, and namelist.input have been attached for reference. The relative files are under the directory of /glade/work/zhouxu/wrfv4.3/run

 

[Ming Chen]

Hi,
I experienced a long waiting queue a few days ago. But the issue seems to be fixed and my MPAS jobs run normally today. I suppose it should be the same for WRF. However, If the situation continues, please send email to help@ucar.edu. I believe this is a machine issue.

In your namelist.input, you set "mp_physics = 10,10,32, ", ---- is there any special reason that you want different microphysics in D03? Note that it is recommended to use same physics over all domains (cumulus is an exception).

 

[MSdueon12]

Hi Dr. Chen,
The experiment was deliberately designed in this manner because the activation processes of CCN and IN needed to be considered. Given the associated computational costs, the full SBM scheme was applied only within the innermost domain.

 

[Ming Chen]

Thanks for the explanation. I suppose you also turn on feedback, is this correct? Please let me know whether the case can run successfully and whether the results are reasonable.