Best way to set up UKESM1.1 Job

Unified Model
1

Hi CMS,

I’m looking to submit suite u-dd727 for a midHolocene experiment some time in the next week. The suite is UKESM1.1 AMIP and will need to run for a minimum of 200 years, maybe more. There is plenty of budget in n02-INSPIRE to do this but I’m not sure how to set up the suite cycling and wallclock time in the most effective way.

Previous 2-year test runs took ~ 12 hrs to complete so the model should be able to run at 4 years a day. However, I currently have another 2 yr test run active with a 1 yr cycling frequency, charged to n02-ncas. The first year submitted and ran immediately but the second year has been stuck in the queue now for 14 hours.

Ideally I’d like to set up for 250-300 years of runtime, cycling every 4 years, but avoiding my jobs getting stuck in a queue for potentially days at a time.

I have two questions. First, is the wallclock time requested in the suite conf the time expected to complete one cycle of the model? Second, can you recommend the most effective way for me to set up cycling and wallclock time for the run described above?

Thank you,
Alfred

2

Hi Alfred

Please don’t use n02-ncas.

First, is the wallclock time requested in the suite conf the time expected to complete one cycle of the model? Yes. It’s best to not inflate the time you request but allow for some variability in machine performance (say 10-20% increase of the best cycle time)

Second, can you recommend the most effective way for me to set up cycling and wallclock time for the run described above? There’s no hard rule - turn around will depend on what else is queued/running on the machine.

It might be best run so that in case of a failure, you don’t need to back up too far, so reasonably frequent dumps, but not so that you’re running out of space.

Grenville

3

This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.