FAMOUS

Minutes of the QUEST-ESM NCAS FAMOUS meeting of Friday, 26th January 2007

Present: Robin, Annette, Ros, Lois, Simon, Jeff, Jonathan

Annette, Robin and Jonathan have overhauled and expanded the FAMOUS wiki website, which is now in place as the external website www.famous.ac.uk, with help from Andy and Katherine. This completes deliverable D1 of the QUEST-ESM contract. Julia thanked us for doing it.

The QUEST cluster is now installed at Bristol. Simon and Annette will ask for usernames. Gethin would like us to try it before the majority of users do. Simon says the PUM 4.5 already there is fine; the only difference from the standard PUM is a modset related to identifying the machine precision. Annette will run a standard FAMOUS job and compare with HPCx results. Once the QUEST cluster is in production, Robin and Jonathan can use it for Quaternary QUEST. (For QESM, it would be useful if Simon did a timing test for HadGAM? at N96 on the cluster.)

Robin discovered errors in the Mead diagnostics in FAMOUS on HPCx, but not in adtan (Met Office T3E). The bug might be related to other previously identified and fixed ones. Simon may be able to remember something if Robin shows him the details.

Ros, Andy and Lois are considering options for implementing a distributed UMUI. It would be good if all users found a central service hosted in Reading so attractive that they did not want to run the UMUI locally. However, local UMUIs will no doubt continue for the moment. The requirement is somehow to share jobs, not just the basis files, but also the many other files (stashmaster, compile overrides, modsets, ancillaries, start dumps) which define a job. This needs to be set up in such a way that by default users can obtain other user’s jobs, without the originator having to have taken special action to put their files in public places. Of course this should only be done within a limited and trusting community that we are supporting, such as QUEST, not with the whole world. For ad-hoc sharing of jobs, documentation is not required; for jobs which we make available as standard, it is important that we know what all the components are and where they came from.

Since we have decided to use MOSES 2.2 with FAMOUS, Ros knows which panels will have to added to the UMUI. MOSES 1 will still need to be supported i.e. FAMOUS as it is now. FAMOUS requires various modifications to job processing, including in the coupling macro. (As a digression, we discussed the future of the UMUI. There are no plans to replace it on QUEST timescales, so QESM changes will be implemented in the current UMUI framework, but fcm is being introduced.)

Annette is working on debugging problems with restartability and reproducibility in MOSES 2.2 in HadAM3. She will give the results of as a long a job as can be done to Robin, for a preliminary assessment of the climate. Michel Crucifix warned us it may be too warm, but apparently no-one else has tried. The next deliverable is that the carbon cycle should be technically working (i.e. not necessarily scientifically acceptable) by March, but that seems fine.

Robin will try running HadOCC, also required for March. He will ask Chris Jones for a dump and a HadCM3LC job for comparison.

Next month, we will give some thought to ancillary tools.

Ken Caldeira has written to Simon about FAMOUS technical problems and use for studying the C cycle. Simon will reply about the technical issues. Robin will ask Ken about scientific interests. Jonathan will also write to Ken and point out the website and mailing list.

NCAS doesn’t have enough tapes for storage of data on HPCx. For runs that don’t need to be kept, tape archiving should be switched off, as we have plenty of $DEVTDIR disk space (not backed up).

Page last modified on January 26, 2007, at 11:23 AM by JMG