Minutes of the QUEST-ESM NCAS FAMOUS meeting of Thursday, 3rd April 2008
Present: Manoj, Annette, Lois, Simon, Ros, Jeff, Robin, Jonathan
To complete deliverable D3 of “a documented and comprehensively evaluated version of FAMOUS-CI”, due Mar 2008, by the end of this month Robin will put a validation note and other notes on the website for his version including HadOCC with UTOPIA (thanks to Dave Storkey), which has run for 1000s of years on quest, and is in a near-steady state. The notes will comment on including TRIFFID and the carbon cycle. With the coupled C cycle, the model crashes at present after about 200 years, although there is no large drift or other peculiar symptoms. However there is at present no-one wanting to use this version, as far as we know. DESIRE will probably make further changes to the land-surface (based on the coastally tiled MOSES 2.2).
Annette and Robin have resolved the inconsistencies with Bristol’s list of modsets, by including some harmless diagnostic ones in our standard job xcps[ab]. Annette will switch off HadOCC in this version, and advertise it for adoption by Bristol. There will shortly be another standard version including HadOCC with UTOPIA and Robin’s spun-up state for those who want FAMOUS-C.
The problems with bit-reproducibility across PE configurations have been solved on all platforms (HPCx, quest and HECToR) by Simon. It requires a compile override for GCOM and is included in our standard FAMOUS job.
We decided not to do anything further about submission to quest from the UMUI since there is not at present a demand for it. Annette has documented how to submit jobs.
Jeff installed the modified small execs on HPCx required for post-processing and resubmission with Robin’s very long absolute time convention. They are not installed on quest, because post-processing and resubmission are not used.
FAMOUS on HPCx and quest are scientifically equivalent (no statistically significant difference in 100-year runs) but HECToR is different. Apparently HiGEM is also different. Perhaps there may be common problems. Annette will try a run without optimisation, and compare pgi and pathscale compilers.
Annette, Lois and Jonathan continue to consider the design of the program to modify dump contents using input from netCDF files.
Jeff has done some work to add a panel for making a “general” ancillary to xancil, in which the user specifies the stash codes. This facility can be used in cases where xancil does not yet have a specific panel for the ancillary the user wants to make.