Core Minutes 9/27/2005  

Present:    Joanne Bogart, Toby Burnett, Jim Chiang, Seth Digel, Richard Dubois, Warren Focke, Berrie Giebels, Tom Glanzman, Navid Golpayegani, Michael Kuss, Matt Langston, Chuck Patterson, Igor Pavlin, Leon Rochester

• DC2 news: (Richard) There is one remaining issue to be resolved: misbehaving ACD ribbons. They are less efficient than they should be, which shows up in plots of Active Distance (distance of each track to nearest ACD tile edge of a tile which has fired). One thought is to try a different definition of "fired": determine by readout threshold rather than trigger threshold. We will try a 100 million event run using this strategy. Berrie will make a similar run in Lyon.

  Toby will put together a new GR head

• Reproducibility: (Michael) He described his comprehensive exploration of the topic. He can regenerate complete and correct ntuple files from mc files as long as he sticks to full runs. When he just asked for individual events by run id, event id, they were either generated correctly or not at all. If he asked for an interval of events, depending on the interval he got
  • all events reconstructed correctly
  • one event reconstructed incorrectly, others ok
  • one event reconstructed incorrectly, one not reconstructed at all, others ok

  He has no explanation at the moment. Toby has seen similar things on Windows which, now that it's clear it's not just a Windows problem, he suspects might be caused by use of random numbers in some rarely-invoked part of Geant. However this is at the moment just a hypothesis; he knows of no candidate parts of Geant which might be doing this.

• Release Manager: (Navid) VC8 builds should now happen automatically; the next time we have a new release needing building this will get exercised.
• Data handling: Tony couldn't be at the meeting, but he provided this report. Main points were:
  • Linux version of GUI Installer is up. See instructions on the GUI Installer home page.
  • Data Catalog is now basically working, though there are a couple minor problems to be worked out before it's released.
  • Work on interface to pruner/peeler is ongoing.
  Igor spoke to the last point and has kindly summarized what he said then as follows:

  There will be a new web application coming in one/two weeks for pruner/peeler jobs that is a modification of the one that Tony wrote earlier. It will include database persistence for user batch submissions, which means that a user will be presented with a set of the batch parameters from his last batch submission whether it was a pruner or a peeler job. This way he can modify only a few batch job parameters and resubmit the batch job.

  There will be also a history of all user batch jobs, so the user can relate to an earlier pruner/peeler parameters. Richard also asked if there will be a smarter way to estimate the job duration, so that we do not lower unnecessarily batch job priority when we submit the job to the farm. I answered that we can expose some hidden (so far) batch parameters, which will make it easier to enter batch duration time, based on the history of the previous runs.

  I also plan to add some other hidden so far batch job parameters later, depending on the user's input in the trial run. In particular, Tom has some new ideas that Tony and I will collect to make batch submission a better experience.

• CMT for Science Tools: (Jim) He would like to revisit this issue. Originally it was felt that users (as opposed to developers) should not have to deal with CMT, but, at this time, the work-arounds needed to provide necessary parts of CMT functionality are clumsier and more difficult to maintain than CMT itself would be, particularly since the Installer knows how to fetch CMT. Several people agreed with Jim and no one objected. Navid mentioned that there is a certain amount of annoying set-up one has to do for CMT.

  Speaking of CMT, its maintainers have decided to switch from CVS to Subversion; will this have any impact on support?

• Python, Gaudi and ROOT: (Tom) He and summer student Andrey Goder looked into getting these things to work together. There are version problems: the Python interfaces are much superior in newer versions of ROOT (5 rather than our current 4.02) and Gaudi (16 rather than 12), but the Gaudi move, especially, is unlikely to happen any time soon.

  Issues to address include

  • need to convert job options to a Python script. They were able to do this successfully.
  • When do you or should you get a Python prompt? Which program gets to be Main?

previous

minutes index

next

J. Bogart, Last Modified: 01-Jun-2010 15:48:23 -0700