COSIMA TWG Meeting Minutes 2022

COSIMA TWG
,
1

Summary of my notes from the COSIMA TWG meeting today. Please add whatever I’ve missed/misrepresented

Date: 2022-11-09
Attendees: Micael Oliveira @micael, Andrew Kiss @aekiss , Russ Fiedler, Paul Leopardi @paulleopardi , Angus Gibson @angus-g, Harshula Jayasuriya @harshula, Rui Yang @rui.yang, Dougie Squire @dougiesquire (and perhaps others I’ve forgotten)

Regional domains

  • Discussion of regional modelling in breakout session at recent COSIMA meeting (see https://docs.google.com/document/d/1wW3Y1LhCcWkplSCCwf7-CLo2x3l00WNbMyI5Df0SmKE/edit#heading=h.51bba4etf8gv)
  • Identified technical work to be done:
    • Creation of tools to generate files/forcing for the model - Angus/Ashley have started developing some tools for this
    • Performance and benchmarking - work needs to be done benchmarking MOM6 generally/globally. This should in general be helpful for understanding scaling of regional configurations, but there are nuances, e.g. I/O cost of boundary forcing. Also need to think about load balancing in coupled configurations. Might NUOPC make this more straightforward than with past configurations? NCI would like a timeline for this work and idea of workload required.

Reproducible builds

  • Harshula working on making builds of existing configurations (ACCESS-OM2) as simple as possible and deterministic. Then will move onto looking at tools like Spack. Currently focussed on deterministic builds for the things we’re building ourselves - see ACCESS-NRI Github org.
  • Need to be sure that runtime environment is compatible with build. Can Spack help with this? EasyBuild makes this process easy

Topography

  • Updated topography workflow seems to be working well. Micael currently polishing the Fortran tools. Also, probably will add a Python script into workflow to ensure central longitude is correctly defined (this changed with the new gebco dataset). Also, the current tools create new file and add the topography to it, but this approach loses metadata. Probably better to modify in place, or to create a copy. New workflow still enables manual edits in the same way as the B-grid workflow. General agreement that it is better to create a modified version of the gebco dataset offline, rather than using available GFDL tools at runtime.

Action items

  • Dougie to schedule meeting about NUOPC work planning
2

Yes I believe spack covers that with it’s built in environments, but it is a bit difficult to say for sure without knowing the details of what (I presume) @micael was describing about EasyBuild.

3

@aekiss, I was wrong about the resolutions of the CESM MOM6-CICE6-DATM-SLND-SWAV configuration that I’m currently playing with. The ocean and ice components are actually ~1 deg, and the data atmosphere is ~2 deg. Still, a long way from ACCESS-OM2 resolutions…

4

The mentioned regional tools are at: GitHub - COSIMA/mom6-regional-scripts: Scripts for cutting out regional boundary forcing for MOM6. Ashley has wrapped up most of the process in a single notebook that I’m yet to review (oops!), but it has made me consider whether it would be a good use-case for nbdev

5

Yes I believe spack covers that with it’s built in environments, but it is a bit difficult to say for sure without knowing the details of what (I presume) @micael was describing about EasyBuild.

I simply pointed out that EasyBuild (and I assume also Spack) creates module files that ensure that one uses exactly the same dependencies at runtime that were used at compile time.