Core Minutes 11/20/2007

• Collaboration meeting: (Richard) It was a good meeting; a lot is going on.
  • Bill asked for a task force to look into improving AcdRecon. Minimally it will use errors to cut on track extrapolation to the ACD instead of hard cuts. He expects this will take a month, during which we can generate backgrounds, then reprocess with the new code. Then generate the sky, while completing Pass6 (current) then 7 with new AcdRecon, BT tuning etc. with the new code.
  • C & A wants to incorporate Beamtest results
  • IRF wants many mono-energetic MC sources to calibrate with
  • GRB wants a thousand GRBs in a few orbit days for systematics
  • L1proc needs to be stress tested
• GlastRelease/Big run: (Heather) About 1-1/2 weeks ago Tom requested a RHEL4 opt build, which Navid provided in time for v13r4, built over the weekend. This incorporated fixes for some would-be show-stoppers:
  • Large numbers of bad values in merit tuple. Leon tracked this down to several uses of abs in Tracker where the floating point version, fabs was intended.
  • Memory problems (Joanne) The long-standing double free error occuring at the very end of certain test programs running on RHEL4 was eliminated by replacing the declaration of a static std::vector with a pointer to same, initialized to zero, and modifying code to allocate (once) with new at an appropriate time. The memory corruption in AcdUtil::AcdCalib was fixed by a similar technique. To keep gcc 3.4 (and perhaps other, less articulate compilers) happy,

    Do not declare potentially variable length objects, such as vectors, to be static.

    The amount of static memory required should be knowable at compile time, without having to run executable code.

  (Richard) The new release was used for a muon run and sample day test, demonstrating that there were still outstanding issues, chief among them the efficiency of ACD, which was only 35%! Another unpleasant characteristic was bifurcated values per run for latitude and longitude. It was looking bleak for the home team. However, late yesterday Eric C. determined the ACD problems were due to inconsistent calibrations. Job options need to change for the new ACD code. He will put together an appropriate set of calibrations, we will adjust job options and try again. The latitude/longitude strangeness is probably something we can live with for now, perhaps attributable to the short runs. The first event or three have one pair of values, the rest a different pair. (Toby) How does one know what the right job options should be? (Richard) We need to clean up job options again, rationalizing use of includes [hearty agreement from all sides].

  RandomSvc Default is still to set seed once per event, rather than once per run as agreed at a previous meeting. (Toby) Let's change the default. (Heather) Yes, let's. (Richard) will make the change in job options for the upcoming test runs to use the new scheme.

  CAL status Due to issues with EbfWriter, we're not yet fully using Zach's Cal upgrades; in particular we're still running CalDigiAlg before TriggerAlg.

  EBF data direct (Heather) [sent me the following, for which much thanks!]

  ..concerning putting the EBF data from the evt files onto the TDS, rather than having EbfWriter recompute the EBF data blob and putting it on the TDS for the OBF to use, as it does now. This issue came out of the OktoberTest, where it was determined that the GCR events were getting dropped due to some filtering done by EbfWriter - not really a deficiency in EbfWriter, as it was never written to handle processing of "real" data (even in this case, where the real data was simulated..but that's where my head starts to spin and I best not think too deeply about that). So the solution first suggested by Bryson was that we take the EBF data that already exists in the evt files, and just put that on the TDS. So I've implemented that. Unfortunately, I haven't quite got the exact blob I need. There must be extra bytes or something, since when OBF gets my blob, it fails and complains somewhere in the FSW. I'm looking into it.

  Composite Event Lists (Heather) CEL will go into HEAD soon, after David Chamont has had a chance to respond to her query for a good set of tags to use.

  Interleave(Tracy) He's restructuring, inventing new GaudiTools, to use CEL and to support possible Level 0 interleave. It's all checked into CVS, awaiting the promotion of CEL to HEAD. (Toby) suggests Interleave should keep statistics on how many times each group of events is selected. (Tracy) has been using Leon's new job options arrangement; it seems fine. He would like to move the Interleave-specific job options file out of the Gleam package to Interleave.

  See the blog for late-breaking news.

• OnBoardFilter: (Tracy) There have been discussions recently of wants and needs w.r.t. OBF for TVAC. It boils down to 3 things:
  1. Upgrade to the latest version (B1-0-5) of fsw. We're currently at B1-0-1. Most of the work is straightforward, but there is one substantial chunk.
  2. Need mode-switching capability
  3. Up till now we'be had two parallel sets of TDS output. We need to change over completely to the new scheme which allows output for mulitple filters, abandoning the old scheme for reporting gamma filter details for something better.

  (Richard) Expect Rob to call a meeting next week to discuss MOOT aspects of Filter Configuration and related issues.

• (Joanne) RM provides invaluable information, but we need to eliminate clutter where possible so the useful stuff isn't buried, then pay attention to what's left. gcc 3.4 issue warnings for suspicious occurrences of abs which look something like this:

  warning: passing `double' for converting 1 of `int abs(int)' 

  but some packages have so many warnings (in v13r4, rhel 4 build, CRflux is the leader, with 254) that it is impossibly tedious to read through them all. Most of them are benign and can be eliminated with proper recoding. It's just as important to keep test programs up to date so that failures are meaningful. (Jim) Warnings can come from include files belonging to codebases not under our control. This was the case for Likelihood and Xerces. (Joanne) Many warnings from Xerces code disappeared when we moved from 2.6 to 2.7 [in fact currently the Likelihood build has no Xerces-related warnings], but this kind of thing certainly can be a problem. (Toby) This can be handled by a filter configured to look for certain known patterns.
• RM, SCons and so forth
  • u09 and RM (Navid) The u09 disk, where Installer versions are normally kept, has become badly fragmented, making it impossible to copy new releases. (Windows releases may appear to be there even when they're not because of the intermediary Samba.) In the short term he has been erasing old RM stuff which is no longer needed but we may need some other strategy to get us to the time when we can switch to SCons. At that point he's hopeful we won't have to use u09 for this purpose any more.
  • ST v9r3 and Installer (Richard, Navid) Emmanuel has made a mac Installer version of ScienceTools v9r32p2.
  • SCons (Navid) Pure SCons work is on hold until Jim finishes adapting python code to use commonUtilities. (Jim) was distracted by Collaboration Meeting; he expects to commit everything to CVS this week. (Navid) in the meantime is working on the new RM. New batch submission is almost done.
• AOB (Chuck) Who owns user.cshrc? Right now some people are running into a problems with it because of the new ssh. user.cshrc thinks it needs an afs token in some situations when it actually doesn't, forcing the user to type in a password an extra time. [I've changed user.cshrc as recommended by Pat Nolan so that the test should now always do the right thing. Anyone who has previously copied it to their own .cshrc should make a similar change. ed.]

previous

minutes index

next

J. Bogart, Last Modified: 01-Jun-2010 15:47:21 -0700