Help for CESM2.1.3 runs slowly

[Mengyao Xu]

Hello,

I noticed that the ported CESM2.1.3 version on my machine runs slowly. According to Katec, a fully coupled simulation for 5 years should be finished within 12 hours. However, I found the modeling efficiency(BSSP compsets) on my machine way too much lower than this, at around 7.3 h per model year using 6 nodes*32 cores. My resolution is f09_g17, and the output frequency is monthly(all output variables are set by default).
I wonder if this is related to any problem during the porting stage or in env_mach_pes.xml. So I attached the config_machines.xml, config_batch.xml, config_compilers.xml and env_mach_pes.xml together with env_run.xml. Could anyone help me with this problem?

Thanks a lot!

 

[erik]

Hello

Note that @katec said 5 simulated years will "likely" complete within 12 hours. But, this estimate would be based on usage for particular cases on Cheyenne and/or Derecho. And it completely depends on the compset and resolution as well as how many processors are used and specifics of the machine. So the above guess based on NCAR HPC machines is NOT a good benchmark to depend on. What you actually get out of your case and machine is what you should plan for.

What Kate was really getting at was how to set the STOP_N for your cases. She says here

You should choose a STOP_N number of years that will fit within the wall clock limits on your computer.

So nothing is really wrong for your setup...

 

[Mengyao Xu]

Hello

Note that @katec said 5 simulated years will "likely" complete within 12 hours. But, this estimate would be based on usage for particular cases on Cheyenne and/or Derecho. And it completely depends on the compset and resolution as well as how many processors are used and specifics of the machine. So the above guess based on NCAR HPC machines is NOT a good benchmark to depend on. What you actually get out of your case and machine is what you should plan for.

What Kate was really getting at was how to set the STOP_N for your cases. She says here


So nothing is really wrong for your setup...

Thanks Erik for your detailed explanation! I understand it now.