ACCESS-OM2: Release Information

12/04/2024

ACCESS-OM2 Release Update

New configurations

ACCESS-NRI has released global 1°, 0.25° and 0.1° physics only and biogeochemical (ocean-bgc) configurations of ACCESS-OM2 with both repeat-year and interannual JRA55-do forcing.

There are twelve configurations in total. Experiments configurations are stored as separate specially named branches beginning with release- in the access-om2-configs repository. where you can also find instructions on how to access the configurations.

The ACCESS-Hive instructions for using the ACCESS-OM2 model have also been updated to use ACCESS-NRI infrastructure.

Details of the releases are here:

All the configurations that are being released today use the ACCESS-OM2 model version 2024.03.0 that has restart reproducibility enabled and so are not bitwise reproducible with the equivalent COSIMA configurations on which they are based.

For more information on restart reproducibility and ACCESS-OM2:

Configuration versioning

Model configurations use a MAJOR.MINOR versioning scheme (a modified semantic versioning (semver) scheme).

In this scheme the two components of the version have the following semantic meaning:

  • MAJOR : indicates changes in bitwise reproducibility
  • MINOR : any other change that does not change bitwise reproducibility

What does this mean?

Any configuration that has the same MAJOR version as another version of the same configuration is bitwise comparable, i.e. those experiments are directly comparable within scientific limitations.

Configurations are bitwise reproducible within MAJOR versions. For example, if a configuration is versioned as 2.3, it is bitwise reproducible with any other 2.X configuration.

What version are these released configurations?

All the configurations that are part of this release are versioned as 2.0 as they are not bitwise reproducible with the equivalent COSIMA configurations on which they are based.

Automated Testing

Automated continuous integration (CI) testing has been added to the access-om2-configs repository.

What does this mean?

Every time there is an update to a release-* configuration branch Quality Assurance (QA) checks ensure the model configuration doesn’t have any easily detectable errors. This helps ensure released configurations are of high quality and helps configuration developers avoid common errors.

If QA checks pass then reproducibility checks are done: the configuration is run on gadi for a very short time (typically a day) and a comparison made between the model state and the same model state from the previous version of the configuration. Short runs are also done to test for restart reproducibility.

What if it is bitwise reproducible?

If the model reproduces previous results it is bitwise reproducible. If the proposed changes are merged into the release- configuration the configuration MINOR version is incremented, e.g. if the current version is 2.5 the new version would be 2.6.

What if it isn’t bitwise reproducible?

If the model doesn’t reproduce previous results it is not bitwise reproducible. If the proposed changes are merged into the release- configuration the configuration MAJOR version is incremented and the MINOR version set to zero, e.g. if the current version is 2.5 the new version would be 3.0. In this case the reproducibility checksums that represent the model state would also be updated to the values for the new MAJOR version.

Release

When a change to release- branch is merged the version is always updated. This also triggers the merge commit to be automatically tagged. The tag contains the branch name with the version appended, and this also automatically creates a release that is visible in the release page

The releases page contains details of the changes for each release, and is a useful source of information about configurations.

Why is this testing important?

Automatic reproducibility testing means configurations will not be altered and change bitwise reproducibility without that being obvious (change in MAJOR version). It makes it clear what experiments are comparable and provides certainty.

The testing is also part of the automated release process, so you can be certain that the configuration used for your experiments has been tested and tagged appropriately.

Where do I ask questions?

Replies to this topic are disabled.

If you have specific questions about this release follow the guidelines for requesting help from ACCESS-NRI .

If you have questions about access-om2 create a topic in COSIMA and tag it with access-om2. If you require assistance follow the guidelines for requesting help from ACCESS-NRI .

19/04/2024

Updated to include released 0.25° configurations

22/04/2024

Updated to include link to ACCESS-OM2 model version release.

4 Likes