While looking over the CLEX and NRI tutorials for AUS2200. I’m wondering if 21stCW would be well served by creating a very small UM test-suite (i.e. a minimum working example).
refers to suite u-aa003 which runs a very simple N48 Endgame GA6 standard job (build, reconfigure, atmospheric). Of course, all the paths are hardcoded to the UKMO filesystem.
I’ve come across Paul Leopardi’s minimum-working-example for rose/cylc on gadi (u-cq161) which is a great introduction to how rose/cylc runs basic tasks on gadi.
A good workflow for new PhDs/Post-Docs could be:
start with u-cq161
progress to an ACCESS/gadi equivalent of u-aa003
then move onto AUS2200 (e.g. Chermelle’s u-dg767 and u-dg768)
Does ACCESS-NRI currently have any UM minimum-working-example rose suites?
If not, is there any appetite within NRI to help build and support one?
I agree with Scott on on this. Also @cbengel u-dg767 and u-dg768 are not AUS2200, they are ACCESS-NRI copies of the Regional Ancillary Suite and Regional Nesting Suite.
If you look through the Run a Model section is that what you meant?
Please note, we have not finished all the steps required at ACCESS-NRI to officially support the regional nesting suite as yet. Once we have, we will add documentation for running the limited area (regional) model there.
In terms of providing support for a global hind-cast (as opposed to a global climate model set-up), if that is what you mean, I think it would be best to start talking with the Atmosphere Working Group. It may be non-trivial due to the initial condition data required, etc.
For those types of reasons the regional model example may be a better start despite the learning required.
Yes, I’ve seen the hive docs on how to run ACCESS GCM’s. What I was after was a smaller suite (the smallest possible!) to just run the UM, so it could be used to demonstrate how rose/cylc uses rose-app.conf and suite.rc to create a task graph to run UM executables and co-ordinates UM namelists and ancillary files etc.
I think it would be a good step before a user proceeds to running an ACCESS regional nesting suite.
I’ll follow Scott’s suggestion and build one based on GAL9.
I think what you describe can be very useful.
Something similar has been also voiced by other people in the community, so I think there are good reasons to start working on that.
I also think there might be very good space for collaboration both on the creation of the suite and for its documentation.
Does NRI (or anyone else) have an existing N96 global atmosphere-only suite running on gadi?
I’ll use u-aa003 as a starting point for a minimum working example. From first glance, some of the information I’ll need to run it includes:
relevant modules for fortran compilation
At atmospheric start file
Some global ancillaries (the suite uses N48 ancils)
The suite can run with a pre-built UM, it assumes the existence of vn13.6 pre built executables from the UKMO computer at /home/h01/frum/cylc-run/vn13.6_prebuilds/share
The link @Scott shared earlier suggests u-dh838 could also be used with vn13.6, however is that only available with cylc8.
So I’ve checked out u-df541 which uses UM vn 13.4 and runs in cylc7. That appears to have all the required information to run on gadi.
Please let me know if you’d prefer another suite or configuration to use as a template.
Hi Paul,
Please note that u-aa003 is not a portable suite. It was built to work in the UK Met Office environment at Exeter. In particular, it does not use PBS at all, does not use NCI modules, and uses hardcoded path names. In my opinion, you might be better off to reference a suite that has already run at NCI.
For example, you may want to contact Jin Lee and Anderson Murray at the Bureau of Meteorology about using some of the work done on u-cr043 to assist you with porting u-aa003 to NCI Gadi. See also roses-u ticket #521 and its Ticket Details.
I don’t think u-cr043 is a good comparison for portability specifically, since it’s very complex and you’re hoping for something simple. I have u-cx707 which is normally used in GC5 config but can run GAL9 fine, in which case it is already fairly close to the minimum practical suite. I haven’t run it at N96 however so some of the ancils/etc. would be missing from the suite config I expect.
I would be happy to help you bounce ideas around as far as porting u-aa003 (it doesn’t look like there’s much to that suite, less complexity does make porting easier, though Paul L. is correct it’d involve starting from scratch) or modifying u-cx707 to make it more of a toy suite for demonstration purposes.
