[J3] work on F202X at #217?

[Clune, Thomas L. (GSFC-6101)]

On Jul 26, 2018, at 2:15 PM, Keith Bierman <khbkhb at gmail.com<mailto:khbkhb at gmail.com>> wrote:



On Thu, Jul 26, 2018 at 7:20 AM Clune, Thomas L. (GSFC-6101) via J3 <j3 at mailman.j3-fortran.org<mailto:j3 at mailman.j3-fortran.org>> wrote:
Hi Vipul,

A full use case would be more writing than I have time for at the moment.   But a use case for coarray teams would be something like:

I have a model that is couples two semi-independent subcomponents.    Say a climate model with atmosphere and ocean components to be concrete.    Each of those components can be run as a model on their own.    When coupled they can

​When coupled, is there some periodic sharing of data? If not, what do we mean by "coupled”?

Yes.  As I said I did not have time to write down all aspects.

If so, is there some subset of the data that is specifically intended to be shared?

THe 0th order answer is that the ocean shares its skin temperature with the atmosphere (drives convection) and the atmosphere shares it surface winds with the ocean (drives ocean currents).   More complicated in modern coupled models though.


I'd have guessed that at some set of timesteps, there would be sharing of only some of the states ….

Correct.


If so, wouldn't we need not only "teams" but some what to indicate the subset of the teams that interact? And some way to limit the interaction to specific timesteps?​

This is all handled by the  coupler which spans both sets of images.   The ocean does not “know” that it is running the atmosphere but merely makes its state accessible to the coupler.  Likewise for the atmosphere.   The coupler is then in a position to drive both sub-components (tell them how many steps to run), allocate co-arrays that do span both sets of images,  exchange data at appropriate intervals, etc.      In theory one could substitute a different ocean model and make relatively modest changes to the coupler and have another functioning system.

Cheers,

- Tom





-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mailman.j3-fortran.org/pipermail/j3/attachments/20180726/879c6c7a/attachment-0001.html>