Core Minutes 11/29/2011

• ScienceTools: (Jim) Jim addressed some plotting issues raised by Jean and David, and he has triggered opt builds for ST LATEST so they can be used to try the new diffuse response calculations in the catalog pipeline. See Science Tools Development Notes for more information on recent ST activity.

• FSSC: (Tom S.) FSSC has a new data server, now in the process of being configured. It's a 12-core, 24-thread machine with 64 Gbytes of memory and 9 Tbytes of disk. Write speed is at least 2.5 times faster than the old hardware; expectation is that improvement for read will be significantly greater.

• Systests: (Leon) The goal is to run Systests on an SCons build and compare to results from a CMT build of essentially the same code. There are two issues:

  1. Systests are hard-wired for CMT. There have been messages from Liz that she knows how to fix it but she hasn't had a chance to do it. (Richard) She's on maternity leave, due back soonish.
  2. Random number behavior is different. On Linux, the same seed is used by all clients. Exactly which seed is determined by which client starts up last. That happens to be one client for SCons builds and a different one for CMT builds (on Windows each client uses a distinct seed and the streams are kept separate). There is therefore no hope of event-by-event comparison between the two Linux builds. Leon is content with statistical comparisons.

  (Heather) We must get this verification (of SCons rhel4 builds) done forthwith. The plan is to submit by hand the jobs Systests would have run automatically and compare the outputs.

• Reprocessing: (Leon) A committee of luminaries has been assembled to decide about Pass7 reprocessing. The decision needs to be made in the context of plans for the next 12–18 months. The following are generally agreed upon:

  • If there were no cost to doing the reprocessing, we would certainly do it. There is no question that there are advantages.
  • If the cost were small, it would still definitely be worth it. Among other things, it would help down the road in making Pass7/Pass8 comparisons.
  • There is a significant cost

  If we decide to do the reprocessing, it would start no earlier than January — probably also no later.

  (Warren) We need to work an ft2 reprocessing in, too.

• Truncation (Leon) Tracy has been looking at events generated using new truncation schemes (e.g., read in entire layers from one end, alternating end as layers alternate. Then limit each layer to 28 rather than limiting each half-layer to 14). Meanwhile, Bill has come up with yet another alternative which looks very promising: keep split readout but do not impose artificial limits. Read in hits until the per-end hardware limit of 64 is reached. It seems counter-intuitive, but actually works very well.

  (Tracy) has been looking at 3 different truncation schemes. Bill's improves pointing resolution significantly. It also increases the number of large events which slows down tree-based tracking; recon for such events goes up by a factor of 10. Still to be investigated are effects on deadtime, onboard filter performance, etc.

  (Anders) Because of the limits now in place we never fill up the FIFO. If we implement Bill's suggestion we will get FIFO full errors. Fortunately there was a change (a year or two ago) to the compression algorithm so that compression should still occur even for an event with this error since such an error does not imply corrupt data.

  See this Confluence page on TKR truncations for a more comprehensive description of the problem and possible strategies (but not yet including anything on Bill's suggestions).

• Core talk:

  • GR, gaudi, and rhel5 (Heather) The new gaudi build has been created and installed at SLAC. The merge of new-gaudi-adapted code into GR HEAD (for Pass8) is in progress. The same will need to be done for the Pass7 branch. The merge is the last thing standing in the way of rhel5 GR builds. In January we will push to validate everything: new Gaudi, rhel5 and its new compiler, etc. The technical drop-dead date for rhel4 is end of February; we hope to be done well before that.

  • Upgrade to rhel6 is occurring TODAY for most glastlnx** machines. See Tom G's page for details. (Tom S.) This includes the machines on which the RM runs. All RM tools, including daemons, have been tested rhel6. He's not expecting any glitches if the upgrade completes before the next scheduled launch of daemons at 12:48 PM. (The daily launch is necessary so that our lsf jobs always have an unexpired afs token.)

    (Tom S.) RM is building ST on rhel6. We have LATEST and HEAD builds already and everything is set up to build release tags as well.

  • VMs (Joanne) has returned to exploring VirtualBox. It turns out that the API which looked so intriguing is not part of the standard rpm distribution. It comes in a separate zip file with almost no installation instructions, but it was reasonably clear what to do. I was able to build a sample C++ program. It fails partway through because it's expecting to find an ISO image in a certain hard-coded place. I'll edit, rebuild, and play some more.

• Announcements

  • SCons meeting Wednesday at 10:30

  • Richard is leaving for Belize and kayaking early Thursday morning, returning on the 12th of December. He expects to have essentially no internet access while he's gone.

previous

minutes index

next

J. Bogart, Last Modified: