.p000 files when running metgrid.exe and real.exe not working after

[sstrey]

I am trying to run a WRF with a very large domain at 4 km. Running metgrid.exe on Derecho produced *.p000 files that I hadn't seen before.

It appears that metgrid.exe runs correctly and auto-deletes the .p000 files. I then go to run real.exe, and I get the attached error before real.exe quits. I've checked other met_em files that have produced viable runs and compared them with my met_em files, but I can't see a difference in the number of variables.

Any tips or insight as to why this is doing this to me are welcome. My instinct is saying that something is wrong with running the large domain in WPS.

Thanks in advance!
Sara

 

[Ming Chen]

Sara,
Your namelist.input looks fine to me.
Can you run a test case with smaller grid numbers using the same forcing data? We would like to know whether the large grid numbers caused the trouble.
Thanks.

 

[sstrey]

Sara,
Your namelist.input looks fine to me.
Can you run a test case with smaller grid numbers using the same forcing data? We would like to know whether the large grid numbers caused the trouble.
Thanks.

Running the same north pole centered grid at 4km and 500X500. I will keep you posted.

 

[sstrey]

I only changed the domain from 2500X2500 to 500X500, keeping the 4 km resolution and polar projection with the North Pole at the center. For, the smaller domain ran both metgrid.exe and real.exe successfully with no *.p000 files or killing errors. It seems to be related to the domain size or processing. Any ideas?
Thanks,
Sara

 

[sstrey]

Sara,
Your namelist.input looks fine to me.
Can you run a test case with smaller grid numbers using the same forcing data? We would like to know whether the large grid numbers caused the trouble.
Thanks.

Do you think the problem will be solved if I used WPS compiled for dmpar instead of serial? If yes, do you know of a successful compile strategy for Derecho? I've been trying to compile it to match the precompiled WRF 4.5.2 (glade/work/wrfhelp/derecho_pre_compiled_code/) with no success getting all three executables. The problem there is with ugrib.exe and may have been addressed in another thread. Thanks!
Sara

 

[Ming Chen]

Sara,
Thanks for the update. Again, I think this is a memory issue because your case is way too large.
In derecho, I would suggest that you compile WPS and WRF using gfortran. Please follow the steps below:
(1) module load gcc
(2) go to WRF directory, compile WRF with the option 34
(3) go to WPS directory, choose the option 3, and compile WPS

Then you can run metgrid in parallel mode. Note that ungrib.exe only works with a single processor.

Hope this is helpful for you. Let me know if you have any issues during the process.