Minutes for the terameeting of 5th of August 2004.

Present: Emmanuel, Mireille, Yannick, Henry, Fred, Gerard (for part of the meeting).

SWarp (Emmanuel)

 Swarp still produces corrupted headers. Using the KDEtool 'khexedit' these can be easily corrected. If Emmanuel can fix this bug in a short period of time this would be good, because it is not possible to reformat headers produced in the fashion in an automatic way.

Scamp (Emmanuel, with testing by Mireille, Yannick, Henry)

 Mireille presents some comparisons between the FLUXSCALE computed by astrometrix and scamp. Three images in the 37 images of the pre-survey PreW3 cause problems: there are triple images in a part of the stack. The initial astrometric solution for these images is also bad. These images are heavily extincted. No solution for Prew1. Yannick attempted to process 360 ldacs but there was no solution at all (they included the PreW1, so that may be the same problem as Mireille with PreW1). This is now priority 0 for scamp

 Henry also highlighted the problem experienced with the astrometric solution found for the COSMOS i* data: more weight is given to internal overlapping sources than to the external sources, with the result that the reference solution isn't very good (residuals of a few arcseconds).

Ticket system (Fred):

 Fred will investigate available ticket-managing systems: this will be completed for the end of next week.

Spica (Laurent, in absentia):

 Scamp / Astrometrix will be used to test the pipeline, and verify that dbterapix can be used successfully in the pipeline environment.

dbterapix (Gerard):

 DBTERAPIX crashes if there is an image on input with a large (>60 character) filename. Not much progress wrt to yesterday. The segfault seems to come from a location which has nothing to do with the manipulation of the image name. Gerard will try a 'quick fix' which involves artificially shortening.

 The artificial shortening "_weight" into "_w" seems to work. However Laurent found there is no problem with Dbclient; the length of filename problem only occurs with Dbclient2... It is therefore attached to Dbclient2, not a fundamental limitation of filename length with the DB.

Hardware (Fred, Emmanuel):

Opterons (mix1, mix2 and the 4-Opteron)

With the replacement of the 3ware cards on the Opterons, all hardware stability (random crashes) problems appear to have been resolved. There may however still be some problems with the reliability of cables on the cards.

Fred has activated SMART self-testing on all the disks of the cluster and the Opterons which will also automatically send mails to Gimi, Magnard and Bertin if any disk fails a self-test.

 mix1 / mix2. Can we install mix2 on the cluster? Which operating system do we install? Do we stick with Gentoo?

 We should probably integrate mix2 in the cluster first, because both raid arrays are empty. When the new 4-processor Opteron arrives more tests will be made and comparisons made with mix1: once these tests are terminated both mix1 and mix2 will be integrated in the cluster. The 4-processor system will remain outside the cluster.

Gilles: mapix1 needs to be re-mounted. Gilles needs to compress his files. Note: It's better to use 'imcopy' rather than gzip to compress image files. It's much faster, and has a higher compression ratio.

Cosmix (Emmanuel). Emmanuel mentions that the nodes on cosmix still don't work. But this is relatively low priority for us right now.

Next tera-meeting: to be decided.