Minutes for the Tera-meeting of Friday, 12th June 2004

Present: everyone, except Gerard, Mireille.

Scamp (Emmanuel) Perhaps the major remaining bug has been found. It turns out that the most robust mode in SCAMP is one in which we assume that the instrument (camera) is the same from one run to the other. But we need properly account for anamorphosis. It seems that the problem (perhaps) comes from the fact the matrix solutions calculated by the ATLAS libraries are returned from the library functions in the wrong order -- changing the order of matrix parameters produces a very good fit!

 Need to implement USNOB-B with flag information. Seems to improve accuracy of astrometric solution.

 There is still much work to be done: this new version of Scamp must be tested by Mireille, Yannick and Henry, and there is still work to done in the code, namely the addition of rejection of outliers.

 Additionally, some work has still to be done in SWARP to correctly handle the astrometric solution produced by SCAMP.

Spica (Laurent): New version of spica-web finished, which uses db-spica; much faster than the old version that used db-terapix. dbclient compiled successfully on the Opteron. Spica interactive mode not working. Problem with the "panier". Gerard should go through this urgently when back.

Skywatcher (Jean-Christophe): memory bugs fixed and tables created. This should mean Skywatcher should be very much faster when there are many images to display. Important to implement single FITS image plot, like a big stacked image. Will make the tool more useful for most users.

Processing (Henry, Yannick) Some discussion followed about astrometric accuracy: what reference system are we in (is it fk5)? ICRS? 0.3''-0.2''. Can we invite an 'astrometric expert' to terapix to discuss these questions (perhaps someone from Bordeaux?)

 All of the wide data <0.85'' has been reduced. Still the deep data to reduce. Steady progress is being made on the d1.

Hardware (Fred, Yannick, Emmanuel)

We will ask for a devis from Landat for cables, USB box, and 2gb of RAM for cosmix, as well as external disk of 250GB and backup RAID card for the Opterons.

 We still don't know what the origin of the instability on mix1/mix2. Fred tried again to allocate very large amounts of memory, but it didn't cause any crashes. Yannick reports that this morning he started several DS9 processes without crashes [it now seems Tuesday 15th that the instability comes from writing/reading from the system disk].

 mix2: perhaps the problem with the RAID card comes from the bad cables, we will see if replacing them helps or not.

New hardware: (Fred, Emmanuel)

 Should we buy a 4-Opteron or a 2-Opteron system? The problem is that for the 4-Opteron systems, we are limited to 4.4TB/3.3TB of disk per box. For the bi-Opteron we could have 8.8TB disk. The problem with buying two bi-Opteron is that after that there is no money left for other hardware needs (replacing workstations, screens and so forth). A bi-Opteron costs 4000 eur less than a quad-Opteron for the same configuration. Yannick and Emmanuel favour the quad-Opteron solution.

 Emmanuel benchmark SWarp on Stephane Colombi's 4-Opteron system to provide 'real-life' comparisons with mix1/mix2 (our bi-Opteron system).

 The other idea is to use the money from EFIGI to buy the 4-Opteron system and then terapix money to buy the disks.

 We also need to buy a new fibre-channel disk for incoming/outgoing disks.

Next Tera-meeting:

Monday 21st of June 2004, maybe?