Open mo-marqh opened 3 months ago
A couple of questions on this, what sort horizontal data size are we aiming for ~100x100 or smaller or larger? Would we be use using real data or analytic functions?
i think we would benefit from some pretty small data sizes first, i think smaller than 100x100 myself.
i think that using analytical functions is really useful for small scale testing, enabling result validation based on resolving the function.
for example
https://github.com/MetOffice/tcd-XIOS-demonstration/blob/main/xios_examples/read_domain_resample/domain_input_simple_linear.cdl
provides a 5x5 to 4x4 resample, using an x + y
data function with this in mind
it may well benefit us to produce some slightly larger demonstrations, but I prefer small numbers of x & y whilst we prove functionality
this demonstrates domain rank decomposition on a small array
produce demonstration code of conducting a horizontal domain resample on output, where there is rank decomposition of the domain
ensure that results are consistent, particularly at the boundaries between decomposed domain fragments
consider sharing a halo between ranks, consistent with the order of the interpolation