We have an instance where we are running a model (UKESM and then also an atmos only model) for say 10 years in a CRUN with jobs each of 1 year. So I’d envisage setting the total run length to 10, the cycling frequency to 1 year, the dump frequency to 6 months.
Here are my questions:
- I want to perturb a restart dump before the model next executes (i.e. at the start of each CRUN job). So, can I do a CRUN but with something like CONTINUE=FALSE set? If so then am I right to say its “rose suite-restart” I would use to prompt the next job to run – and before I do this I can tweak the restart dumps.
- I cant find the documentation for the relevant variables and changing an NRUN to CRUN… I think I saw it once but not when just looking for it!
We don’t generally talk about NRUN and/or CRUN in cylc, so I may not have fully understood what you want to do - I think all you need do is add a task that perturbs the dump written at the end of each 2nd cycle - no need to stop/restart the suite.
The idea of stop/restart is that the perturbations are determined by analysis of the last year’s data in a number of concurrent runs, not just one run:
So I had in mind the loop is:
- set of runs does the next year
- review data and determine perturbations to apply
- Restart all
As suggested by Ros, you could set the run length for 1 year, run the model(s?), perturb the dump(s), then extend the run to 2 years and rose suite-run --restart. The run will continue as if it was originally set to run for 2 yrs (ie as a CRUN) - then at 2yrs, perturb, set run length to 3 yrs, restart …
A bit ugly but it should work. There is probably a cleverer way to do this with interacting suites.