Core Minutes 1/29/2013

• ScienceTools: (Jim) Steve (Fegan) looked into memory issue in Likelihood and found a place where memory could be returned earlier; this has been fixed.

  (Jim) has been talking with Steve and Emilia about investigating alternate algorithms for diffuse response calculations. They also are looking into a massive refactoring in ST which could take a couple months.

  See Science Tools Development Notes for latest ST news.

• Reprocessing (Tom G.) There is no reprocessing news at the moment.

• Hardware and migration report (Tom G.)

  • We expect to start a new round of writing to tape in March and could run out of tapes by mid-May (or even earlier since this calculation does not take into account increased P8 recon data volume). We need to order more.
  • The first lnx box has been configured with two VMs. Wilko is poised to try it out. He is concerned there could be extra latency due to the use of VMs.
  • The first part of the PPA batch cluster (256 cores) has been opened for normal batch use. We're still waiting for word on the shared scratch space.

• Pass 8 Face-to-face (Tracy) Several people were late in arriving because of weather so the agenda was rearranged somewhat. However the goals of the meeting — to see where we stand, what we must do, what we would like to do,.. — were achieved. See a prioritized list of remaining issues in JIRA, project LPATE. High-priority ones include:

  • Unguarded sort causing segmentation error (LPATE-73) [as of Jan. 30 Tracy says "..which I guardedly think is fixed now"]
  • Adding the Tree-related TDS content to the PDS (LPATE-103)
  • GR Timing benchmark (LPATE-102) Track finding is not the major contributor to average event processing time; that honor goes to energy reconstruction. However, Philippe has made substantial progress. He believes he can save about 30% of the time spent. Tracking is the chief culprit where tails are concerned; individual events can take several minutes to reconstruct.

  Bill has been working on getting point resolution errors on an event by event basis from the track/vertex covariance matrix. This has been an issue for some time but with Pass 8 Bill has crossed the final hurdles and believes we are now achieving a better representation of the PSF with the covariance matrix than what is returned from the IRF's. There are a few pieces that still need to be put in place but its almost all there (so not a show stopper).

  The goal is to start reprocessing no later than a month after the Collaboration Meeting. This time around we're making a serious effort to get science validation started early.

• Core talk:

  • GR and still more CVS problems (Heather) had much difficulty getting out GR 20-08-05 due to CVS lock problems. Lessons learned:

    • do not attempt to make a new HEAD or release tag while LATEST checkouts are still in progress
    • remote users with flaky connections to SLAC should not attempt to tag; ask someone local to do it instead. Connections from Paris often were dropped within 10 minutes. CVS operations in progress would crash, leaving stale locks behind.

    (Joanne) In some situations it should be possible to clean up after a failure immediately after it occurs — if the process issuing the failed CVS command or a parent process is still around. Lock names contain process id and node so, when a process fails, any locks matching the correct pattern can be deleted immediately. Time permitting (i.e., not right away), she will see about implementing such a strategy in stag and tagCollector.

  • Data Catalog and Pass8 (Heather) Fixes have been applied so that Data Catalog can now handle Pass8 data.

  • New flight release (Heather, Joanne) There is a new flight sofware build, B3-1-3. We need to make a corresponding obf external and then a GR using it. Joanne expects to get to the first part either late this week or next week. Heather then needs about a week to generate and test a new GR on the L1 branch.

  • RM and Mac (Heather) Tony J. was, after some travail (which shed light on connection problems we've had in the past), able to get Jenkins to make a build on bldmac01. (Tom S.) Jenkins invoked RM programs to check out, build and test. What's missing is the triggering to do this automatically. He will see about creating a new Workflow in RM (the code is there but untested except for the one pattern used by RM currently). (Richard) Could this be used to do Windows builds? (Tom S.) Yes, if Jenkins runs there, which it most likely does. In fact, if the Mac experiment is successful, he would like to do this everywhere, dispensing entirely with lsf.

  • Windows and SCons (Heather) What is the status of support for the Bill scenario: supersede built against a public, read-only release? (Joanne) Has not looked into this for a while. Most recent Windows effort has been directed to problems Leon has uncovered which have frequently been peculiar to his situation. He is now able to work productively, albeit with a less-than-ideal development environment requiring some work-arounds. The Bill scenario is probably the highest priority Windows support item now; she hopes to get to it in a couple weeks. (Heather) volunteers to try it out on the terminal server.

previous

minutes index

next

J. Bogart, Last Modified: