Regional grid run computing time issue

[adrienD]

Hello everyone,

I recently had to reinstall CTSM on a new version of the supercomputer I use (Levante from DKRZ). I took the opportunity to reinstall a more up-to-date version of CTSM (from dev023 to dev086).

So far, with your help, I have managed to build and run a default global grid with a much improved computation time compared to my previous setup.

However, when I try to reuse a regional grid that I used in my previous setup, the computation time is quite slow. I only managed to run 31 days with 594 nodes/hr. I was able to run a full year or more with this calculation in my previous setup.

Due to the nature of my problem, I cannot work on logs issues and I wonder if you have any idea what I could do.

I have attached my case file and the logs from a run it produces.

Thanks!

 

[jedwards]

I see that you are setting PNETCDF_PATH, but the model is using netcdf not pnetcdf.
This is probably because you are only using 1 node for your run and so only have 1
iotask. I don't think that this performance is entirely due to io performance, but try changing
PIO_REARRANGER=1.

 

[adrienD]

I see that you are setting PNETCDF_PATH, but the model is using netcdf not pnetcdf.
This is probably because you are only using 1 node for your run and so only have 1
iotask. I don't think that this performance is entirely due to io performance, but try changing
PIO_REARRANGER=1.

Thanks @jedwards. Sorry for my late answer, the forum was down for a few days and I didn't have time to see your answer.

I have tried to run without PNETCDF LND PIO_REARRANGER=1, and it's now working like before! Maybe better, need to do more tests.

Thank you so much