How to increase the number of cores with 02_generate_openfast_model_for_dlcs example

[bluemiracle4862]

Hello.

I have another question with 02_generate_openfast_model_for_dlcs/iea15_monopile_modeling.yaml

How do I increase the cores ?

I try it with the following input but It didn's work.

---

openfast_configuration:
    use_exe: True
    cores: 8
    mpi_run: True

---

Thank you for your advice.

[dzalkind]

Hi,

Here is some documentation about running WEIS in parallel: https://weis.readthedocs.io/en/docs/running_weis.html#running-weis-in-parallel

Best, Dan

[dzalkind]

@ptrbortolotti, it seems like we're setting cores and mpi_run automatically now. Can you confirm that I'm understanding that correctly?

If so, I'll remove them from the input schema for clarity.

[ptrbortolotti]

I think you are good to remove them from the schema

[bluemiracle4862]

Thank you for your advice.

[briantempleton]

Hi all,

I have been encountering a similar issue with example #2 ("02_run_openfast_cases," e.g., weis_driver_loads.py). I am able to use the parallel instructions for running in parallel with MPI that are available at https://weis.readthedocs.io/en/latest/run_in_parallel.html.

I can clearly see that MPI fabric is being used. I see differences in execution with an mpiexec vs. an ordinary python run. For example,
-with mpiexec I see the following messages when running weis_driver_loads.py: "All MPI subranks closed."
-if I try a different example, 01_aeroelasticse/run_general.py, I can see the console output of multiple openfast simulations interleaved.

A few questions:

1. Do we expected the DLC-based OpenFAST simulations initiated by WEIS to run in parallel? (I am assuming so...)
2. If so, is there something else that I need to do to configure this example to realize WEIS's potential to run in parallel?
3. Bonus: Is there a way to configure WEIS to generate turbulent wind files once and then be able to reuse the generated wind files when run again?

Many thanks, I very much appreciate your insights and help!
-Brian

[dzalkind]

1. Yes, this should be the case.
2. Nothing I can think of, besides ensuring that mpi4py is installed and when MPI is imported in WEIS, it is not None.
3. If you re-run with the same output folder, WEIS will only generate the files that are not currently present. I don't think we have inputs configured to re-use the wind directory that WEIS creates previously, but that should be possible by creating a user input that maps to self.wind_directory here. We can add this to our list of requested features.

[briantempleton]

Thank you for your quick response! Answers for #3 make sense.

I am still thinking on #1/#2 related to MPI and not seeing evidence of parallel utilization during the OpenFAST run component. Upfront the only thing I could think of is that I am running on a low number of cores (currently have 2 allocated on this system)--but also tried running on 4 "cores" using --map-by :OVERSUBSCRIBE in case there are "primary" processes consuming some of the MPI workers. Didn't seem to help.

To answer your questions:
a) I do have mpi4py installed in my conda environment. (I followed the instructions in the main readme on this git site, taking note of specified Linux specific items, e.g., mpi4py.)
b) I do get an mpi object <module 'mpi4py.MPI' from '[home dir]/miniconda3/envs/weis-env3/lib/python3.12/site-packages/mpi4py/MPI.cpython-312-x86_64-linux-gnu.so'> for MPI when it loaded in WEIS/weis/main.py when using the following call:
mpiexec -np 2 python weis_driver_loads.py.
Additionally, I see related console outputs (e.g., "All MPI subranks closed.")
c) Also, as expected, when I run python weis_driver_loads.py, I get None.

Note that I edited the above and the attached text file to remove my home directory (has my work ID in it) and replaced it with [home dir].

I have attached some output from both running with mpi ("mpi") and without mpi ("nompi").

example02_output_mpi.txt
example02_output_nompi.txt

What strikes me is that if one diffs the files, the outputs with and without mpi are very similar for most of the run. There appears to be some evidence of multi-threading initially and then none when OpenFAST is executed. This is in contrast with if I run example 1, run_general.py I see interleaved OpenFAST output from both MPI workers, e.g.,

Generator speed: 2.3 RPM, Pitch angle: 3.4 deg, Power: 0.0 kW, Est. wind Speed: 2.9 m/s
Time: 10 of 200 seconds. Estimated final completion at 18:27:51 (in 0.002 days). Generator speed: 2.3 RPM, Pitch angle: 3.4 deg, Power: 0.0 kW, Est. wind Speed: 2.9 m/s
Time: 10 of 200 seconds. Estimated final completion at 18:27:46 (in 0.002 days). Generator speed: 2.7 RPM, Pitch angle: 3.4 deg, Power: 0.0 kW, Est. wind Speed: 2.8 m/s
Time: 20 of 200 seconds. Estimated final completion at 18:27:44 (in 0.002 days). Generator speed: 2.7 RPM, Pitch angle: 3.4 deg, Power: 0.0 kW, Est. wind Speed: 2.8 m/s
Time: 20 of 200 seconds. Estimated final completion at 18:27:43 (in 0.001 days). Generator speed: 2.9 RPM, Pitch angle: 3.4 deg, Power: 0.0 kW, Est. wind Speed: 2.9 m/s
Time: 30 of 200 seconds. Estimated final completion at 18:27:44 (in 0.001 days). Generator speed: 2.9 RPM, Pitch angle: 3.4 deg, Power: 0.0 kW, Est. wind Speed: 2.9 m/s
Time: 30 of 200 seconds. Estimated final completion at 18:27:45 (in 0.001 days). Generator speed: 3.2 RPM, Pitch angle: 3.4 deg, Power: 0.0 kW, Est. wind Speed: 2.9 m/s

Many thanks! Again much appreciate your insights!
-Brian