Core Minutes 4/19/2011

• ScienceTools: (Jim) The new tag 09-23-01 incorporates changes since April 11 included a patch concerning MapCubes and the addition of a new isotropic source facility. See Science Tools Development Notes for details.

• FSSC: (Eric W.) has imported the new ST tag (see above); looks routine so far.

• Documentation: (Chuck) sends the following:

  The newsletter currently has top priority as it's targeted to be released this Thursday. In between articles, I'm implementing inputs received from Gino, which primarily affect the Science Analysis section and also address several navigation issues. Last week, I made significant changes to the "splash" pages for the Science Analysis and Science Tools sections and also updated the FSSC links on the Data Access green navigation bar.

• astroserver proposal: (Richard) Gino is asking to put standard ST analysis on the astroserver as an afterburner: after it gets the ft1/ft2 files, it would run the standard chain of science tools (gtmktime, gtltcube etc) and autobuild the source xml file from the catalogue, all the way through binned likelihood. We are starting by using the expert pipeline customisation the astroserver already allows to prototype it. We'll try to find a postdoc to do it.

• Disks: (Richard) Currently we're 6 servers in the red. After cancelling a contract some months ago because it ostensibly didn't have the servers we had been promised, Oracle recently notified us it has found some refurbished ones. We came back with a couple conditions: 1) price must be very good 2) we want 10 of them. They offered us 10 at $270/TByte so the purchase is likely to go through. For future purchases, LSI seems to be the best option. We're in the process of getting a quote. (Tom G.) The refurbished units from Oracle would buy us some time while we figure out what to do for next time.

  (Richard) Next year's plans are for 2 Pbytes of tape, 1 Pbyte of disk. Pass7 reprocessing is coming up, to be followed next year by Pass8 reprocessing.

• GR/Onboard filter: (Tracy) On Windows he has everything needed from JJ. It builds all right. He and has made requisite changes — nothing too extensive was required — to get GR up and running with it. Indications so far are that it produces the same results as our production obf, as it should.

  JJ and Tony thought they had given us everything needed on Linux, but they haven't quite. The remainder will show up in due time. Then it will need some repackaging into a standard SCons external.

  Tracy now has to switch gears and start getting ready for Rome, so further obf work will have to wait.

• Pass8: (Tracy) There is a lot of activity in a lot of different areas, evidence that the code has gotten to a sufficiently mature state to be of interest to a wider crowd.

  (Leon) Everyone is busy with posters now, so don't expect to see much activity until after Rome.

  In MC, it looks like we could improve the overall PSF at high energy by rearranging the way the TKR hits are read out to allow for more hits. The timing of the tree-based patrec, as currently implemented, is very sensitive to the number of hits in large events; Tracy has been working on this issue.

  (Richard) What can be said about processing time? (Tracy) We can do better by recognizing which events are bad candidates for pattern recognition because of large number of hits and giving up. Sometimes in such cases Cal recon can do a reasonable job. We haven't yet made serious attempts to cut off processing this way. However Philipe has been looking at recon on events abouve 1 TeV and has found he can do a better job by spending more time on analyis; the extra time sometimes buys you something. (Richard) It also means we have to wait longer for data to be reprocessed.

• Core talk:

  • Externals to be promoted (Heather) thought she already had promoted packages needed for use of the new data-type externals diffuseModels and catalogProducts as well as the newly-named cfitsio library, but apparently not. [Promoted later this morning in HEAD 891.]

  • Gaudi, G4 and so forth (Heather) Francesco reports he continues work on the Geant4 upgrade commit to CVS for GlastRelease HEAD. He plans to work over the Easter break to hopefully complete it. Meanwhile Heather successfully built Geant 4.9.4.p01 with VS 2003 after modifying about 20 source files. The new source has been set aside and the binaries are being installed on the V: drive. We have managed to build most, if not all, of the GlastReleasev18r7p1 branch using this new version of G4 and VS2003.

  • Upgrading MySQL (Heather) Kim is looking into ramifications and procedures for upgrading the CMT RM database from MySQL 4 to 5. Navid has pointed out some issues to consider in an email which Heather will forward to interested parties. (Joanne) There are databases calib_test and mood_test, identical in structure to the production calibration and moot databases, which may be used to test migration procedures.

  • Tagging debacle (Heather, Joanne) Josh Lande tried to make a new ST HEAD with the tagCollector script and it hung up. After 3 hours he finally killed it and tried again. This time the operation went through, but the new HEAD tag was missing some packages. Since tagCollector makes new HEAD tags by starting with the previous one, successive attempts also made incomplete HEAD tags. Joanne was able to fix the problem by adding the new HEAD tag to the packages which didn't already have it, but we still have no idea what went wrong in the first place. It could have to do with CVS — CVS is known to have some protocol holes so that, under certain circumstances, it leaves old lock files lying around which can interfere with future operations — but we don't yet have a good guess for what the exact sequence of events might have been in this case. Heather asked SCS to look into possible disk or server problems on a couple dates when we had trouble with CVS tagging; they didn't turn up anything suspicous. There has been some talk of using a different primary source (packageList.txt file or contents of SCons RM database rather than CVS tags) to keep track of exactly what is in any given build.

  • SCons miscellany (Joanne) Johann, whose development environment normally consists of a tagged ST release plus a "supersede" development area, ran into a problem when he changed ST releases to 09-23-00. An environment variable normally set in the routine facilities::commonUtilities::setupEnvironment( ) was not getting a good value. The soruce of the problem was that the file _setup.sh in the supersede area had not been rebuilt since Johann starting using the newer ST release. There is no good way for SCons to know when to rebuild this file, so I implemented Johann's suggestion to always rebuild it when the --supersede option is used — committed but not yet tagged.

    While attempting to duplicate Johann's problem, using an RM build at SLAC as the base release I inadvertantly wrote some files in the RM build directory (I have permission to write there; typical users do not), files like sconsTools/astroLib.pyc. What would happen for users who don't have permission? Should we think again about generating .pyc files for release builds (for local use at SLAC only; should not be part of Installer tarballs).

    (Joanne) There is a new version of stag. 0.2.3. The only difference is the addition of a new option, -S, which should be used when tagging SConsFiles (or any other package with an SConstruct rather than SConscript, but there aren't any others).

    (Jim) Johann's problem concerned the environment variable CALDB. Conceivably one might want to set this to something other than the usual value (derived either from root of the base installation or from location of the user's supersede area) but there is no way to do it. Any value set by the user will be overridden at run-time. (Joanne) If the current scheme doesn't have enough flexibility it could perhaps be changed to respect existing values for some environment variables.

    (Tom S.) added the -h option to externals tarball generation so that sym links will be followed.

    (Tom S.) RM is now sending out emails when there are Windows build failures.

  • SCons meeting There will be a short one tomorrow (Wednesday), 10:30 AM PDT

• Announcements Heather will be at SLAC for a few days starting June 20th.

previous

minutes index

next

J. Bogart, Last Modified: