Core Minutes 5/14/2013
Core Minutes 5/14/2013ScienceTools: (Jim) is looking into adding duffuse models to caldb..has to be FITSified.
Joe sent a bug report of a failure in an optimized build, unfortunately on a platform that Jim does not have access to.
FSSC: (Joe) All quiet here. (Tom S.) We had a bit of a security incident on the web servers again and we've had all our PHP scripts disabled. That doesn't affect the dataserver but does affect the back end of the mobile app and a few other minor portions of our website.
Pass7 Reprocessing (Tom G.) No news.
Pass8 (Tracy) Concerning Recon output status: recall that last week it looked like the compression achieved of tracking tree output was going to be sufficient to put us in an acceptable range. But there was a bug causing us to lose ACD output. In fact, it turned out we had been missing a lot of ACD output, enough to put us well out of the desired range. So far efforts to involve ACD people in pruning or compression have not been successful.
We plan to look into re-instating double32 (rather than double) in merit to see
(Heather) Stephan reports seeing savings 30% with compression techniques, but so far he has just done comparisons using a single source file.
(Tracy) Alignment in pattern recognition: Leon's code has been committed and tagged.
New code in GR 20-08-12 put in to address (large event) tails is also connected with a loss in resolution. Tracy is working on getting the resolution back, hoping for a new tag in a day or so.
A memory leak peculiar to Pass8 was discovered. Heather did some investigation with Valgrind. She discovered a leak in GCRCalib but it has been around since Pass7. This leak has now been fixed in Pass8; (Heather) will see about backporting to Pass7. (Tracy) Subsequently another leak, which probably is the leak was found in AnalysisNtuple.
Current ETA for reprocessing readiness: about a week. (Tom G.) has set up the reprocessing task and pushed 10 runs through. He encountered some issues but will ready to go when the time comes, just needing to substitute in correct GR release and so forth. See also this comparison between Pass7 and Pass8 reconstruction.
Core talk:
RM move to fermilnx (Heather) Believes all issues with SCons RM have been successfully addressed. (Tom S.) confirms. (Heather) However, CMT RM is now defunct. Can we also get rid of CMT-style tags? [no audible complaints].
MOOD moved (Joanne) Tony Waite's carefully crafted procedure to move MOOD to fermilnx01/fermilnx02 went off without a hitch on Monday; L1 processing did not so much as raise an eyebrow. The old production database on glastlnx06 has been shut down permanently. The only thing left to do — if it seems worthwhile — is to reinstate the daily dump to mysql-node03. Since this hasn't been working for quite some time there is no urgency, but I'm still inclined to do it if it can be accomplished without much investment of time and effort.
Dodging bullets (Tom G.) Various people have run into problems with their jobs now that the rhel6 bullet machines have been added to the public batch queues. We might not be done yet. You can avoid the bullets by specifying the resource rhel50. Tom is happy to supply details if you need them.
previous
|
minutes index
|
J. Bogart, Last Modified: