#2675 closed help (answered)

WRF costs (and support) on ARCHER

Reported by: simon.tett Owned by: um_support
Component: WRF Keywords:
Cc: Platform:
UM Version:

Description

Dear Helpdesk,

I am planning a proposal for which, if successful, am considering doing some WRF simulations. I want to run a global model → regional model → 1km model. 1km model is roughly 1000x500 km domain and I want to run it for about 15 days. Would run the global model for about 6 weeks and the regional model for about 15 days. Suspect cost dominated by WRF at 1km. Roughly how many MAU would that need?

And what support, if any, do you provide for WRF?

If instead of WRF I was to use the UK 1km model what would the rough costs be and do you provide support for that model??

many thanks

Simon

Change History (10)

comment:1 Changed 11 months ago by grenville

Hi Simon

We do not support WRF - actually, we've seen diminished WRF usage on ARCHER from what used to be the main WRF users in Manchester and Leeds. They may be running on Tier 2 or other machines of course.

The UM has much improved nesting suite capability (see https://code.metoffice.gov.uk/trac/rmed RMED Suites) - we support that scheme in some sense, but I'm not really sure how to answer your support question - fiendishly complex nesting suites can be created through Rose/Cylc?, but ultimately they comprise UMs & UM utilities, which we do support.

Your proposed run will not cost much in AUs - I'm working an estimate.

Grenville

comment:2 Changed 11 months ago by simon.tett

Hi Grenville,

thanks very much. Need it reasonably quickly as doing this for CSSP project and if feasible will need to persuade Edinburgh to provide resources… I don't have access to rmed through met office link :-(

If running UM at high resolution I worry we might spend a lot of time setting model up — rumours are that WRF fairly easy to set up…

Simon

comment:3 Changed 11 months ago by grenville

Hi Simon

I was hoping to have access to a 1km model but that's been delayed.

For an N512 global model I estimate 250kAU/model month - that's on 66.8 million grid points, with a 600 sec time step =⇒ 0.0037AU/grid point/model month

For 1km we might use a 15 sec time step, so I'd estimate a cost of 0.148 AU/grid point/model month, so for a 1000x500 grid for 0.5 months

1000 x 500 x 0.5 x 0.148 = 37 000 kAU (37 MAU)

So, roughly 40MAU for a 15day global→1km run (more if you have an intermediate resolution nest). Thu global model cost is insignificant.

I can only guess at WRF numbers for the equivalent - but I think the WRF cost will be less than the UM cost.

Grenville

comment:4 Changed 11 months ago by simon.tett

Hi Grenville,

thanks so being more realistic. I think I want to run for 7 days so that is ¼ of a month, run at 2km x 2km resolution and with a domain size of 500x250 km. I think that reduces the cost for the high resoln model by a factor of 2(1→2km) x2(1-2km) x2 (timestep) x2 (½ domain) x2 (½ domain) x2 (7 rather than 15 days) = 64 do ~0.8 MAU…

Can we do N216 to set up the boundary conditions then go via 12km regional model? And how hard would this be to set up for South America?

Simon

comment:5 Changed 11 months ago by grenville

Simon

N216 is one of the global driving model resolutions - setting up a global→12lkm→2km model should be straightforward. The suite will automatically generate ancillaries for the nested regions and will create boundary conditions (automatically too).

I can't see why this would be significantly more challenging than other nesting suites - assuming you wish to run a common physics set up. Running over the Andes may be tricky.

Realistically, this feels like several months work.

Christophe should be able access to https://code.metoffice.gov.uk/trac/rmed - its well worth a visit

Grenville

comment:6 Changed 11 months ago by simon.tett

Hi Grenville,

thanks for that. Missed response — does all this run on ARCHER as we'll be doing this on CCSP brazil with I hope Edinburgh resources… Any chance of phone chat?

Simon

comment:7 Changed 11 months ago by grenville

Hi Simon

Yes - this runs on ARCHER. Tomorrow at 11am or 2pm OK for a call?

Grenville

comment:8 Changed 11 months ago by grenville

Simon

Double booked - tomorrow at 2pm or later?

Grenville

comment:9 Changed 11 months ago by simon.tett

3pm good for me — 0131 650 5341
Simon

comment:10 Changed 10 months ago by willie

  • Resolution set to answered
  • Status changed from new to closed
Note: See TracTickets for help on using tickets.