Core Minutes 10/29/2013

• ScienceTools: (Jim) No code updates this week. Matthew finished notes on Pass8 event types. He and Jim will meet to decide on next steps.

  Jean reported two issues with the MINOS function in Minuit:

  • For sources with best-fit fluxes near zero, the MINOS would try to evaluate the log-likelihood at a negative value of the flux that resulted in the log of a negative value being taken. He proposed a way of dealing with this by thresholding the probability densities for sources to be positive, but it was not clear that what he proposed satisfied the requirements (continuity, differentiability) for the optimizers, like Minuit, to work.
  • MINOS would sometimes encounter an error condition, but rather than producing an exception or returning an error code, it would simply print to stderr and stop.

  (Richard) Does anyone use the ST Windows builds. (Jim) No. And he has no platform on which to debug problems there. (Heather) There have been build failures but no complaints about them. (Richard) So it should be ok to turn off these builds? (Universal response) YES!

• Pass7 reprocessing: (Tom G.) Should be done next week. Target date is Nov. 5.

• Pass8 reprocessing: (Tom G.) About 87–88% done; looking good. Soon we will reach Sept. 9, 2012, at which time new ACD calibrations will be picked up. We should examine output to be sure this happens corectly. (Richard) Will we pause at this point? (Tracy) Probably we'll keep going.

  (Tom) Note there is substantial overhead (about 3 days) for any new task, no matter what period it covers. If there will be many such tasks it might be worthwhile to design a better scheme (keeping in mind that that in itself could take significant effort).

• Pass8 development (Tracy) GR-20-09-05 is nearly ready to roll, now that it's getting some attention; just need to develop standard job options for all-gamma and backgrounds simulations.

  With the new CAL calibrations the issue of noise scaling has arisen. A scaling parameter was set years ago, but at that time we were using ideal calibrations. We need to work out a new strategy. He is hopeful the new GR build will happen this week. Decision about noise scaling should be made on a similar timescale.

• Hardware: (Tom G.) The new 350 Tbyte xroot server is up and configured. It's been handed over to Wilko for testing. He may also be considering redistributing data to optimize reading. We still have about 200 Tbyte available on old servers so there is no rush.

  (Richard) New file servers are working well. We expect to get about two new ones a year but generally hardware expenses will be down. Most servers (except for astroserver) have been replaced fairly recently.

• glastlnx to fermilnx migration: (Tom G.) It's proceeding apace. We are not confident that all services, cron jobs, etc. running on the old machines have been accounted for in plans for the new machines. Those who might be affected should see a summary page in Confluence and follow links to details as appropriate to make sure everything they care about has been covered. (Jim) has a cron job on glastlnx10 which will need to be moved. (Tom G.) This one has already been accounted for.

• Core talk:

  • GR tags (Heather) 17-35-24-lp61 is out. It includes Tracy's G4Propagator fix and Leon's updates to TkrUtil among other goodies. It has undergone systests and looks ok. During the process it was discovered that systests have been using ideal calibrations for a long time. Whether this is the right thing to do in all cases should be discussed.

  • Compression (Heather) In the ongoing work to fix up job options for the upcoming GR 20-09-05, Tracy raised questions concerning the compression scheme to be used by default; e.g. LZMA versus the ROOT default of gzip. However the default in RootIo is already set to LZMA for all ROOT files (except merit) so it's not necessary to put anything in the jo. There is no reason not to use LZMA, so we'll stick with that. And, at some point, for consistency, perhaps change over merit as well.

  • Mountain Lion macbook (Heather for Tom S.) Not much news since last week; Tom left on vacation and won't return until next week. He gave up on building QT natively for the time being; it depends on xcode which has not been installed. He will use the snow leopard build. He hopes to get RM builds working by the end of the year.

  • Windows and G4 (Heather) Bill encountered a known, Windows-only bug in the version of G4 we're using. The supposed solution — rebuild including an extra specified link flag or two — has not fixed it. One tactic which would take care of it (we're told) would be to advance to a somewhat newer version of G4, but that hardly seems worthwhile.

  • Calibration selection (Joanne) At Leon's request, after experiences with sys tests, she changed the behavior of the calibration service ever so slightly so that the criteria for finding an appropriate calibration now include a test for event timestamp greater than or equal to the start of the validity interval instead of strictly greater than. This update should go in the new GR tag along the main branch and also sometime in the L1 branch.

• Announcements

  Far, far away Richard leaves for New Zealand next Wednesday and won't return until the 25th. Chances are he'll be out of touch most if not all that time.

previous

minutes index

next

J. Bogart, Last Modified: