In a few days time (from the time of writing) I need to give a estimates for the runtime and resources required to run some low resolution versions of UKESM1.0. For some reason I don't know, it isn't currently possible to couple the UM to ORCA1, but we can couple to ORCA025. Hence, it isn't going to be possible to run the actually configurations, and I'll need to estimate run times based on the run speeds of the stand alone components.
My key assumption is that - based on comparisons in the IBM - coupling a component seems to slow it by about 10%.
The only coupled job I have is u-aa677, which is a copy of Yongming's u-aa369 and puma-aa127, and is GC2 + UKCA (StratTrop) + TRIFFID + Multilayer Snow + iBVOC for UM10.0 and NEMO VN3.4, and I think this is approximately
This models uses PEs for OASIS3, whereas we'll be using OASIS3-MCT, so I'll ignore the resources required for OASIS3 here.
Early testing suggests that fullChemN96-dt30 is about the same speed on Cray as on IBM, while nemoCiceOrca025 is 1.5 times slower on Cray than IBM.
Based on my assumptions
| Speed (model years/day) | Speed/0.9 (model years/day) | Speed on IBM | Predicted components | Recommended components | Total nodes | |||
|---|---|---|---|---|---|---|---|---|
| fullChemN96 -dt30 | nemoCice Orca025 | fullChemN96 -dt30 | nemoCice Orca025 | fullChemN96 -dt30 (nodes) | nemoCice Orca025 (nodes) | |||
| 1 | 1.11 | 1.11 | 1.67 | 184 (8x23) | 160 (16x10) | 192 (16x12) (6) | 160 (16x10) (5) | 11 |
