Core Minutes 1/15/2013

• ScienceTools: (Jim) made several updates, all tagged and tested, concerning handling time-dependent IRFs needed for trunc64. There are changes to the caldb_release package which must be exported to the FSSC.

  See Science Tools Development Notes for latest ST news.

  (Joanne) Made new tags of skymaps and pointlike omitting test programs since they fail on all platforms and tend to tie up Windows builds. We had agreed this was to be done some weeks ago.

• FSSC: (Tom S.) No news. We were at AAS.

• Reprocessing (Tom G.) New galactic diffuse model just released and undergoing validation. When complete, will rebuild all FITS files in the Pass7 full reprocessing (P202). Plan is to release fully-reprocessed Pass 7 data in late February or early March.

• Hardware and migration report (Tom G.) Unless he hears objections, 12 of the old sulky boxes will be turned off tomorrow. The remaining one is being used by CTA and will be left on.

  Wilko has sudo access to the 10 new lnx boxes. Three are being handed over to ISOC operations. Yemi is going to experiment with creating multiple VMs on one of the others. If this goes well, the plan is to install about as many VMs on the new machines as we used to have separate boxes. That way services which used to be isolated from each other will still be isolated, even if running on the same physical box. We'll avoid potential resource conflicts this way.

• Pass 7/8 (Tracy) Bill's recent work with GR-20-08-00 turned up several problems. Several of them were fixed in a new release on Friday, 20-08-02. It was built for linux only to start, but Heather will get out a Windows build shortly. See JIRA LPATE-68 for details. 20-08-02 should result in significantly faster reconstruction due to some work by Philippe, who demonstrated that step size for propagator may be increased without compromising resolution.

  Leon's reproducibility problem of last week turned out to have nothing to do with random numbers. It was occurring because of the use of a memory address as key in a map-key pair. There is no reason such addresses should come up the same for different runs on the same machine, and in general they don't. The fix is to sort the map. We should be on the alert for similar usage elsewhere.

  (Leon) meanwhile can get around the problem by using electrons rather than mouns for his work investigating uncertainty in calibrations.

  (Tracy) There are still a couple problems remaining, affecting only a small percentage of runs, which have not been fixed in 20-08-02. These will be addressed in another release.

• Core talk:

  • Root write failure (Heather) Starting with 5.26 a new feature was added: "optimize baskets". The idea was that ROOT would calculate for you what the optimum basket size would be. But in some cases this goes awry. This might have been a problem for us even before moving to 5.34. Heather is gathering as much information as possible about the failure and will send it to ROOT Central.

  • Windows accounts (Heather) She changed the password for the glast account on Windows. She is unable to log in to change passwords for glastrm or fermirm, but the right answer there is probably to eliminate the accounts on Windows. No on ever uses them.

  • RM and Mac (Heather) SCCS says, if we want to upgrade to Lion or Mountain Lion, we'll have to do without lsf. We could perhaps use Jenkins or even just a cron job to create the builds and make them available for distribution, which is the most important thing. (Joanne) It's not difficult to get Jenkins to run SCons. (Richard) What about notifications [of failures]? (Jim) would like to have them. (Joanne) is not opposed in principle, but who will do the work?

  • Continued CVS problems (Joanne) noticed, before making the new pointlike tag, that the previous tag, pointlike-08-00-00, was incomplete and failed to build. (Jim) could run the cron jobs for detecting and deleting locks more frequently. (Joanne) would like to gather more information about stag operations, perhaps writing to a central log. Another possibility would be to submit to batch as is now done for tagCollector, but that could be quite slow. (Jim) We could serialize CVS operations by submitting to a particular queue. (Joanne) is not eager to try to do the job CVS itself should be doing of handling contention, though admittedly it doesn't always do a very good job. If it were easy to fix, someone would have done it. She would rather start by gathering more information.

  • Windows and SCons (Joanne) has done nothing this last week about supersede problems. (Leon) Has been managing without supersede. It's a little clumsier but workable. (Heather) But we still need it for Bill.

  • GoGui (Joanne) GoGui 1.2.0, available for rhel5, rhel6 and Windows, dispenses with an explicit "init" step. Instead that work (checking for externals; computing dependencies) is done for every build and clean. SCons does some caching of information so the first build spends longer on these things than subsequent ones do. Both computers and SCons have gotten faster, so redoing init every time isn't so painful as it used to be. The new protocol will avoid a certain class of likely user errors (failing to do an init when one is required to pick up a new external, changes to SConscript, etc.). The new version of GoGui can be accessed from central Linux machines (assuming standard set up) with the command GoGui_new.

previous

minutes index

next

J. Bogart, Last Modified: