Notes from meeting of Annette, Robin and Jonathan on Tuesday, 23rd October 2007
Without the optimisation mods, FAMOUS is bit-reproducible across processor configurations on HPCx but not on quest or HECToR. We need Simon to help with this. Robin notes that Ron Kahana said it was not reproducible even on a single processor; that could be a more serious problem, and we should enquire about it.
FAMOUS does not run on 1×16, because a condition fails on one of the PEs to do with it not having enough rows. Annette can make it run by assigning the odd row to the N Pole instead of the S Pole, but it is slightly slower than 2×8, which is therefore recommended as being the fastest configuration.
Robin has run the new-physics version (MOSES 1 still) for 200 yr on both FAMOUS and quest and will arrange for there to be a parallel section (of 100 yr) for statistical comparison. On quest it did 100 yr/day with dumps every 15 days. If the statistical comparison is OK, this version can be released, with a caveat about bit-reproducibility. The documentation is ready.
We decided to put all the MOSES 2.2 + coastal tiling into one mod. A problem of subroutines with the same names has been avoided by renaming the wanted version in the mod, to avoid the unwanted precompiled version being picked up. Reconfiguration runs harmlessly if all the required fields are already present in the dump.
The climate difference between MOSES 1 and 2.2 in HadAM3 will be compared with the difference in FAMOUS. Annette will check transa2o conservation. Robin will look for any consistency checks that can be done with tiled diagnostics. Jonathan thinks there may be a new treatment needed for global energy conservation considering TOA.
Some attention may be needed to the O→A CO2 flux, probably weighting it by the sea fraction, and combining it with CO2 flux from land in coastal boxes. Robin has managed to run TRIFFID for a couple of days.