Locking: Snr & Jnr waiting for one another
Solid lines show UMs working, while dotted lines shows them waiting.
Options to avoid/reduce locking
- Only pass dynamical core fields (red arrow) on hour
- Trusts Jnr to evolve fields completely alone for one hour
- Early call to UKCA_MAIN1, say at +0 or, for some locking, +15 or
+30 mins
- Chemistry is +30-60 mins old, instead of 15 mins old as before
- Share chemistry workload across timesteps for Jnr, so DO_CHEMISTRY
every timestep and
- reduce NTZS and something similar for SPLINSLV2 & SPFULJAC
routines, or
- introduce some convergence criteria