Core Minutes 6/12/2012

• ScienceTools: (Jim) No new development news, but he did investigate behavior of the new astro with IGRF_11 after a report that the CRflux test program runs about 5 times slower than it did previously. The new performance behavior comes about because the code is reading in the spherical harmonic coefficients from a file each time the year changes by more than 0.001. Without the file I/O, i.e., for a fixed year, the new code execution time is only 10-20% slower. So if it is decided that the new code is too slow, this can be mitigated by caching the coefficients or reading them all in at the outset. (Joanne) is relieved that there is nothing more to it. Thanks for following up.

• FSSC: (Alex) All is quiet. They're waiting for the next release.

• Pass7 reprocessing: (Leon) It's done through March the target date of March 31st. We're sorting out some issues having to do with magnetic fields. At some point we'll embark on the backfill.

• Other Pass7 (Leon) Heather has made a new GR tag 17-35-24-lp30 with all the truncation code merged in. She will make systests, comparing this release to lp28.

• Pass8: (Tracy) Test data sets have been generated for GR 20-03-01 and people are looking at them. There is a new tag, 20-03-04, with all the latest little fixes. About ready to generate datasets for the summer; more or less back on track.

  (Tracy, Leon) A glitch has popped up using the skimmer with the new Pass8 datasets. They have an new tuple (PrimeTuple) that is a "friend" of the meritTuple. The immediate problem is that the skimmer cannot just copy the meritTuple, however it can copy both (meritTuple and PrimeTuple).

  (Tom G.) is concerned that skimmer support is limited to best-effort from David Chamont. What is the long term strategy? Wasn't there supposed to be a discussion of merit variables? (Tracy, Leon) Of course no one foresaw this glitch. The intent is that, ultimately, the PrimeTuple will replace the meritTuple. meritTuple has accumulated all sorts of obsolete junk, possibly in large part because, if someone wanted to test something, they might add a variable to the meritTuple (because there was no alternative), then forget to take it out. Ideally the new scheme would allow for an alternate tuple to be used for such experiments. For the indefinite future, we'll need both: meritTuple for Pass7 and PrimeTuple for Pass8. (Tom G.) This should all be sorted out before contacting David. (Tracy) It will be.

• Core talk:

  • GR validation (notes from Heather) As noted in mail.. the v20r3p4-scons versus v20r3p4 CMT rhel4 build had many reported failed plots. Others are looking into it. Heather is running off a re-test where the random numbers are event seeded...meaning seeds are applied for every event.

  • DB migration: (more notes from Heather) glastlnx01/02 migration to mysql-node01/node03 SCons RM will be unavailable from 10 AM Tuesday for about 6 hours during migration to MySQL 5.5
    calibration/mood outage is scheduled for Wednesday starting at 10 AM and should last about 1 hour keeping our fingers crossed that glastCalibDB alias will be transferred and the users/grants are appropriatedly configured.

  • End of glastlnx14: (still more notes from Heather) John Bartelt proposes the following:

    put glastlnx14 into IFZ (internet free zone)

    I think we can do that now..does anyone actually connect directly to glastlnx14 rather than logging into something else and ssh'ing into glastlnx14? I'll be sending a note directly to the list of users that still have glastlnx14 to check.

    remove all users other than batch — or at least cut down the list a bit

    once we've blessed a RHEL5 version of the P8 GR and L1proc GR, we'll remove all users other than batch.

    move to new regime where batch access is set up only by special request

    There might be such a request if there is a window where we've blessed RHEL5 for GR development and P8 but before it's been agreed that reprocessing could pick up a RHEL5 build of GR if needed. (Tom G. has put in a request to validate RHEL5 GR for reprocessing; hopefully this will get done in the same time frame as the other validations.)

    glastlnx14 disappears altogether sometime in August

  • SCons RM (Tom S.) There was a problem last week with the way certain paths were stored in the database which prevented the Installers from accessing Windows builds. It's now fixed. Otherwise, things have been smooth.

  • GR vc90 (Joanne) found and fixed a bug in mootCore affecting only Windows. (Tracy) needs to patch several places where an iterator is compared to the end of the sequence. vc90 doesn't like the syntax which has been commonly used. As was to be expected, there was a start-up penalty, but he's now able to do some debugging with vc90/SCons.

previous

minutes index

next

J. Bogart, Last Modified: