Beginner Modeling

83 views
Skip to first unread message

James Polly

unread,
Jan 12, 2024, 10:07:26 AMJan 12
to HYCOM.org Forum
Hello, 

I've been downlaoding HYCOM  products for several years but would like to try to produce a run of my own.

I am aware of the following resources:
Since I'm ultimately interested in running a non-global region, I thought the GOMb0.08 example would be a good place to start. At present however I'm hoping to perform a small, proof-of-concept run to demonstrate that I have the workflow established. Nothing operational, nothing remotely resembling physical accuracy.

The following questions are mostly motivated from what I've found in the GOM example, and the natural assumption of a PBS scheduler:
  1. Are there other good resources beyond what I've listed above?
  2. Is a powerful workstation simply not going to work for even the simplest/coarsest of model configurations?
  3. Is there a model configuration that is best suited for understanding this workflow outside of an HPC environment? Even on cloud compute resources there will not be a PBS scheduler.
Any guidance is very much appreciated.
James Polly

Alan Wallcraft

unread,
Jan 15, 2024, 8:59:56 AMJan 15
to HYCOM.org Forum, James Polly
You should be able to run the GOMb0.08 example on a workstation.  It only requires about 2GB of memory and takes about 10 minutes to run one model month on 48 MPI tasks.

So if you have a 4 processor workstation and run on 3 MPI tasks (only use 4 tasks if the workstation is doing nothing else at all) it should take 2-3 hours per model month.  It might be easier to use 3 OpenMP threads (no MPI library required), which might take a little longer to run but is still manageable.

For gfortran with OpenMP use config/xc40-gnu-relo_omp,  your source directory name would end in _omp instead of _mpi.  If you want to use just a single processor use config/xc40-gnu-relo_one.

Use msub_csh in place of msub_pbs (which is linked to msub by default).

Alan.
Reply all
Reply to author
Forward
0 new messages