Hybrid

The 2UMs model is being know as UKESM1.0-N216ORCA025hybrid. I don't think we're ready to give estimates for the runtime of this, but we are being asked so here's my best shot.

My work on profiling Snr on IBM suggested that compared to a CLASSIC run (atmosOnlyN216) for 5 day which ran for 1,209s.

  • RADAER add 190s
  • Scatter and gather of UKCA fields adds 45s
  • If we add UKCA_ACTIVATE
    • Looking at the full chemistry profiling, it looks like UKCA_ACTIVATE is about 95/3072 = 0.0309 or 3.09% of UKCA_MAIN1
    • If the estimate of UKCA_MAIN1 is +4,878s, this suggests that UKCA_ACTIVATE will be 4,878 * 0.0309 = 151s.
  • For JNR at 16x24, OASIS3_GET_SNR adds 334s

The total is +720s or 59.6% extra time and using 2.5 times the resources (+1.5 resources for Jnr). Hence the factor to increase the node hours is 1.596*2.5 = 3.99 (basically *4).

This isn't much better than full chemistry at low speeds - about 4.5 times more resources than atmosOnlyN216. However, this model should be able to go a lot faster than fullChemN216.