High performance examples

[glwagner]

Once we can run examples on Caltech's cluster via #104, we would like to put together some computationally expensive examples that showcase ClimaOcean's capabilities for realistic simulations.

Here are a few examples that we would like to set up first:

• An idealized hydrostatic example (perhaps Abernathy channel with a slope?) (@glwagner lead)
  • this example should use diagnostics or tools from ClimaOcean to motivate its inclusion in the ClimaOcean docs
• A realistic regional example with ECCO-derived restoring at the boundaries (@simone-silvestri lead)
• A global 1/4 degree example on 1 GPU (@simone-silvestri lead)
• A global 1/4 degree example with biogeochemistry on 4 GPUs (@seamanticscience lead?)
  • we have to figure out how to run Documenter / Literate via MPI (will require additional work on top of Run Examples on GPU for Documentation #104)

@simone-silvestri @navidcy @siddharthabishnu @seamanticscience @sandreza @rafferrari

[francispoulin]

I was planning to put together a regional model of the Gulf Stream, or something along those lines. This could be in connection with the second example. @simone-silvestri , if you wanted to chat sometime, please let me konw. I'd be happy to discuss whether I could help with your example.

[glwagner]

Nice @francispoulin !

We discussed a few regions for the example. I think we want to start with one that doesn't have such a significant boundary inflow as the Gulf Stream region (because boundary inflows are not yet supported). Possibly you can work with @simone-silvestri on the regional example, and then use that case as a template for the case you are interested.

Ultimately I would like to get as close to possible as having a tool that allows us to input a lat-lon box (or geometry) and spits out a Simulation.

[francispoulin]

Good idea @glwagner !

I agree that the Gulf Stream has very strong inflow and a tough starting example. I am happy to consider any regional model. Maybe the arctic? Whatever @simone-silvestri has in mind I'm sure is a good starting point and happy to contribute.

[simone-silvestri]

Hi @francispoulin, I have started with the mediterranean, but we can consider other regions maybe a little more "open" that do not require an open boundary condition for momentum. I you want to take a look at how restoring is done you can look at this simulation in the meantime.

I would stay away from the arctic for the moment until we have all the bits and pieces of the sea ice model in place. What other region were you thinking?

[francispoulin]

Thanaks @simone-silvestri . I think the mediterranean is a great idea! I have other ideas but yours seems better. The ACC is one thought for a next step. We could go around the globe but cut off at the drake passage.

Thanks for sharing the code . That looks very advanced and exciting. I am tempted to try and run it to see what happens. I presume this is an expensive run, even though it only uses on GPU?

Also, what could I do that would help you at this stage?

[vtamsitt]

I'll just chime in here to say that we have an upcoming project to develop a regional setup of the Salish Sea region with ECCO restoring at the boundaries (and compare to the ROMS LiveOcean model). So I'd be interested to contribute to the example development with the Med as we'll be working on a similar setup ourselves, and could potentially contribute the Salish Sea regional configuration as an additional example down the line.

[glwagner]

Salish Sea seems like a nice region to try... it's way smaller than the Mediterranean too so would be a much cheaper example at the same resolution.