Domain decomposition for grid with large masked regions

[Kosa]

I'm interested in defining an application-specific grid. Motivation: if we have a grid where the standard domain decomposition results in one or more tiles (MPI processes) that are composed entirely of masked out land cells, then this seems like an inefficient allocation of our computing resources. The application I have in mind is a section of the continental shelf with a narrow estuary, but I assume this is an issue relevant to global models where the continents are masked out.

I'm looking for information on other models which have dealt with this problem as well as any major foreseeable ROMS-specific roadblocks. Excuse me if this has already been discussed in the forum.

[kate]

I know that CICE won't compute on tiles which are completely land-covered. Also, HYCOM has done this for many years. I once asked Hernan about this and didn't get a satisfactory answer. It came up because in coupling to CICE, you have to make sure all the CICE tiles have some ocean in them or the coupling as I was using it failed to work.

[Kosa]

Thanks Kate

Also, HYCOM has done this for many years.

Page 56 of the HYCOM user's guide https://hycom.org/attachments/063_hycom_users_guide.pdf 20.3 Generating "Equal-Ocean" Tile Partitions, looks like the most promising lead so far.