Core Minutes 8/10/2010

• ScienceTools: (Jim) will be working on a new PSF implementation motivated by this weeks C&A discussion (see Riccardo's presentation). A new ST tag is expected by the end of the week.

• FSSC: (John) is finishing up documentation. (Eric) is currently working on resolving issues from user problem reports.

• Documentation: (Chuck) sends the following:

  After restoring the WB's search functionality last week, I created a new tag, and Berrie has since updated the mirror site. (The URL to the SLAC search engine used in the WB had been changed without notification; fortunately, however, I think we the change was discovered shortly after the it was made.) I should probably remind everyone that there's a link to the WB's Change Log on the main "splash" page, and entries in the log date back to December 17, 2008.

  On another front, work is progressing on the new Instrument Analysis section, which is designed to incorporate all of the changes being made to the SAS Software infrastructure. The volume of the work needed continues to increase as I find more "linked to" pages that require updating, and the biggest challenge is how to incorporate these changes so that updates appropriate to inclusion in other sections of the WB can be made and referenced without undue duplication.

• Reprocessing: (Tom) Pass 7.3 reprocessing has two parts:

  • Rebuilding merit files This is in progress, now about 1/4 done. There was a problem with xrootd overload with one particular server. Diagnosis is difficult since Ganglia does not always report back accurately.
  • Regenerating fits files. He has a working recipe. 10 runs have been processed. C&A needs to look them over. Tom will also wait for the new ScienceTools tag before starting the production run.

  Tom has also heard rumors of another (last?) Pass6 reprocessing.

  Wilko and Tom have come up with an algorithm which will do a better job of spreading the load among xrootd servers. Wilko will be implementing it over the next few weeks.

• Storage model: (Richard) There have been recent discussions. Throughput to/from tape is significantly better than previously thought, so it would make sense to keep only a few months-worth of data on disk and push the rest to tape.

• Pass8: (Tracy) made a vow to concentrate on tree-based pattern recognition for a while and it payed off. He found and fixed a memory leak and made everything shipshape overall.

  There has been a perennial problem with crash-on-exit due to a double-free of memory in G4HadronSim. With the planned upgrade to G4, rumor has it we won't need this package; we'll be able to use the G4 native code instead. Tracy needs to find out from Francesco exactly how we detach from G4HadronSim.

  There is a meeting this Thursday, expected to be largely devoted for planning the next Face-to-face, coming up in 3 weeks.

  (Leon) Geometry and new alignment constants are in place, but not used by default.They may be selected via job options.

• CAL Xtal status word: Now that there is a CAL hardware failure the need for keeping track of CAL hardware and configuration status is no longer hypothetical. Eric Grove sent around a proposal for per-crystal 32-bit status words with the bits subdivided into 3 categories: hardware status, configuration status and recon status. The hardware and configuration status will come from a new calibration file; recon status will be filled in by recon code. (Tracy) added a status word to the CalXtalRecData object and modified existing clustering routines to look at the bits to decide whether or not to include a crystal. Chul has the big job of actually fixing the problem. (Joanne) Ideally the configuration status should come from Moot and the hardware status from a new calibration. I had suggested that, if it would be too hard in the short term to get the configuration status from Moot we start out by putting it in a separate calibration file, very similar in format to the hardware status file. However the consensus among CAL people is that in practice hardware status and configuration status are closely linked, tend to change at the same time (hardly at all), etc., so there is no harm in sourcing them from the same file.

• WIRED: (Leon) The mechanism to automatically start WIRED from within Gleam has been tested and works, but the pieces are not actually in place yet.

  A remaining question is whether WIRED gets started from the web-site or from a locally downloaded version. The former has advantages for Tony as the maintainer, and for the user in not having to know anything, but violates the runable-without-the-network rule.

  A compromise would be to allow both methods with the choice made through the jobOptions. Then the only decision to make would be which to make the default.

• Missing GCR (Richard) The GCR files stopped being generated about a year ago after an FSW upgrade - to do with checking FSW vs OBF copies of bits. A workaround is now in place and in L1, so we need to remake those missing GCR files. We think it can be done without a full reprocessing, but need Johann to confirm. Then Tom would do it with some priority in the queue with all the other reprocessings.

• Core talk:

  • SCons progress (Joanne) There has been some:

    • Avoid copying source into variant directories: This is no problem on Linux; one can just call SCons with an option which says to prefer symbolic links. Another strategy is to turn off the copies altogether. This works fine for most packages, but may not work in cases where part of the build creates additional source, as is the case for Rootcint (and also Swig, but Swig still runs correctly without copies; Rootcint does not). I've implemented a scheme which makes copies only when absolutely necessary: on Windows, only for packages using Rootcint, and only copying include files, not other source. There are small, backwards-compatible changes to the SConscripts for packages using Rootcint: two in ScienceTools and about 8 in GR. These have been committed and tagged. The remaining step, once these packages have been promoted to HEAD, will be to commit and tag changes to a couple global files implementing our SCons policy: SConstruct and processExternals.scons.
    • New GoGui release, 0.9.9. It uses the SCons option for symbolic links, mentioned above. There are a couple other minor enhancements, made some time ago but never included in a release.
    • A solution to another outstanding Windows issue — how to embed manifest files in .exe or .dll files — was recently posted to the SCons users list. It should be easy for us to implement in our system.

previous

minutes index

next

J. Bogart, Last Modified: