Block Explicit implementation for unstructured meshes

[aliabdolali]

Is your feature request related to a problem? Please describe.
The explicit scheme in WW3 can be improved in term of efficiency by the implementation of the block explicit scheme o on unstructured meshes and on a global scale. In principle, exchanges are only called for each freq in this solver which reduces latency and therefore improves the performance on runs with a large core count.

[aliabdolali]

this is a part of wise PR#888 and the highest priority to be integrated before we continue other issues.

[aliabdolali]

@AvichalMehra-NOAA we need to check block explicit on the target mesh (1.8M) and see if the same efficiency improvement I have seen in the regtest is gained or not.

[aronroland]

For defining the block we can even sacrifice the dispersivity of the waves in terms of CFL to reduce latency. This would then basically be equivalent to the implicit schemes (in terms of communication within one sub-step of the explicit scheme), however exchanging would happen for all components on the lowest time step. This can be still of benefit when using large core counts e.g. > 10000

[aliabdolali]

@aronroland I totally agree with you. We have different applications, for coastal, nobody will use explicit but for global cases like our GFS and GEFS, we still would like to take advantage of the great performance of unstructured meshes, at the same time make sure it scales well. So, if block explicit is going to be used for only large global meshes, running on a large number of CPU cores, we can go in that direction. However, other users who do not have access to large cpu counts will suffer. We need to test more Aron.

[aronroland]

Well Ali, u brought it to the point. The parallelization strategy and the decision, which parallelization or scheme to implement for the certain problem and the given environment, grid resolution, physical processes is to be chose by the experienced modeller, nevertheless it i on us to provide some guidance or as a different strategy to keep just one scheme, which is good in the mean ...

[JessicaMeixner-NOAA]

Close with merge of PR #888