computational resources question

[renfrew]

Good afternoon

I have a question about running (wallclock) time for some climate simulations that I'm planning to conduct with the MetUM on ARCHER and wondered if you might be able to help please.

I am considering regional atmosphere-only MetUM simulations (driven by CMIP5 data at the boundaries) with a largeish domain (~3,000 x 3,000 km) at 30 km horizontal grid spacing centred near Iceland and extending across Greenland, Iceland and the surrounding region then nested within that a domain with 12-4 km variable spacing. The variable spacing domain would be approximately 1,500 x 1,500 km with 4km region occupying about a third to a quarter of this area.

I plan to use a recent version of the MetUM, although don't know exactly which since this still at the stage of planning for a grant proposal. Therefore only a rough estimate is required at this stage.

Can you estimate the wallclock time for such a setup run on ARCHER (maybe based on previous similar set-ups)? If this isn't possible, can you recommend someone else that I could ask?

Regards,

[willie]

Hi Ian,

I think the best thing here is to pick a version of the UM that has the right science for your project, ensure that the STASH setup is close to what you require and then do a test run of a reasonable length. This will allow you to estimate the data volumes produced, the run time and the cost in AUs. The resulting model would then be a good starting point for your project.

I am running a UM8.2 job which is 1380x1020 at 4km (regular grid) covering North Africa. This has quite a lot of STASH and we've had to use netCDF to keep the data volume down. Using 48x40 processors with 12 IO servers, the job runs at 10 model days per day and produces 3.5TB data per real day. This may or may not be relevant.

Regards

Willie