Core Minutes 10/27/2009

• ScienceTools: (Jim) gave the report; see it in Confluence.

• FSSC: (Eric) is working on v9r15p5 (latest ST tag); just one file to go. (John) Nothing to report on data access (i.e., it's working).

• Documentation: No Chuck, hence no report this week.

• OBF support evolution (Richard [heavily paraphrased since I didn't take good enough notes for anything else. ed]) As we move to newer compilers and 64-bit architecture, there is no obvious path forward for OBF. In theory there are several possible options, none of them clean or easy and some of them probably impossible. (Heather) has described them, along with some comments from concerned parties, in a Confluence page. (Richard) What might timescale be for the last option [stop using echt fsw OBF; make our own which would not require a pile of supporting non-filter fsw code]? (Tracy) This should be discussed next week along with effort to be put into Pass8.

• GR Pass7.2 reprocessing: (Tom) Six weeks of data were reprocessed in one night. On Thursday embarked on the next step: reprocessing the full year. An unexpected performance problem manifested itself. The data for the first batch of jobs were all being served by 3 xrootd servers and they became smothered in disk i/o, tending to squeeze out net i/o. He had to throttle incoming jobs. The reprocessing continues, but at the current rate, it will take a couple months. Actions being considered which might help include

  Change the code doing ROOT i/o to only read in what's needed

  He has approached Leon and Heather about this. Only a small part of digi data is used. (Heather) could look into just reading some branches You have to know in advance what they are. (Leon) is concerned that there is no standard response in our code when sought-after data are not available. In some cases the job might crash; in other cases it will quietly do without. Any scheme cutting down on the input will require careful verification.

  Upgrade OS on the xrootd servers

  Computer Center staff believes this is likely to help. However, since we have no spares, it will be time-consuming and difficult to keep all the data accessible while machines are offline.

  (Joanne) Would it help if data were more spread out among servers? (Tom) Yes, but normally it is more or less chronological. Data written back to HPSS, then recovered, tend to be randomized, but we're not in this mode.
  (Joanne) Could jobs be submitted to not be so chronological? (Tom) Yes, but that wouldn't be easy, either, and there are drawbacks. With the current scheme one can readily enlarge the time interval.

• Core talk:

  • GR (Heather) For the record latest tag for Pass7.2 is v17r35p1. Latest tag for use in L1 processing is v15r47p12gr08

  • rhel4 OBF (Heather) The native 32-bit rhel4 build looks ok so far. It's been run through sys tests. Expect to incorporate it into a new GR tag soon.

  • RM (Navid) got email from Eric concerning test program failures on rhel5. His suggestions should take care of most of the failures. Typically they involve changes only to the test, not to the rest of the code belonging to the package.

    He has also managed to get most Windows units tests to work, but some are complaining about not finding python25.dll. That's odd since there is no such thing in the Windows distribution of python; it contains only python25.lib.

  • SCons and Windows (Joanne) found a work-around for the problem she was having (which no one else had) with checkpoint release 1.2.0.d20090919. The problem goes away if Windows-only tools like mslink.py and mslib.py are initiallized explicitly, even though SCons documentation claims it's normally not necessary. Could be the confusion comes from mingw being in her path — to be confirmed.

    Now another problem has cropped up, or, rather, has been noticed. It probably was a "feature" of vc90 builds all along. Under certain circumstances, SCons loses track of certain dependencies, such as the dependence of xmlBase on facilities.lib. This only happens when the code generating project and solution files is executed.

  • ASP (Joanne) Once the method for writing ASP wrappers was in place it took very little time to convert the remaining ASP packages to build with SCons. There are a couple remaining minor issues to be worked out with Jim; then it will be ready for testing on SLAC unix.

  • GoGui on Windows (Joanne) A new build is available in ftp space which saves GoGui configuration to a file. Previously the unmodified default Qt scheme was used: configuration was saved to the registry on Windows, and to a file on other OSes. But this default doesn't work on the UW terminal server because it requires permissions users mostly don't have.

  • Python 2.6 (Toby) proposes we consider upgrading. 2.5 is pretty old. (Joanne) is not necessarily opposed but is worried that we're already spread rather thin. (Jim) is in favor of moving forward. If it is in extlibs he can start building against it. (Navid) 2.6.2 is installed in GLAST_EXT for redhat4 32bit.

previous

minutes index

next

J. Bogart, Last Modified: 01-Jun-2010 15:47:04 -0700