Core Minutes 9/11/2012

• ScienceTools: (Jim) ST tag 09-30-00 was created on Aug. 29th, largely to collect all the updates FSSC wanted. There will probably be a patch tag soonto incorporate a new version of SolarSystemTools. (Richard) Does 09-30-00 handl time-dependent IRFs? (Jim) No. That is expected no earlier than January of next year.

• nfs partitions (Tom) We're in the midst of migrating. See details in this spreadsheet. It's being done in 4 parts. The first one (group="User") is complete and operating without incident. The next one to be tackled ("Code") includes space used by both the RMs, the Workbook, and various other things, comprising 13 different partitions. This one is more problematic. rsync, used to copy files from the old to the new partition, will take several hours for some partitions because of the large number of files involved (call these slow partitions. Most of the rest will take under 30 minutes. Two strategies have been proposed for the slow partitions:

  1. Make old partition read-only during rsync. When it's done, switch over to the new partition for both reads and writes. No writes are allowed while rsync is going on, but all the old information is available for read at all times.
  2. Perform rsync in the background, while people continue to write to the old space. This rsync process will omit files created while it's in progress. Then turn on writes to the new partition and, finally, do another background rsync to pick up files written to the old space during the first rsync. This preserves the ability to keep writing (somewhere) at the cost of possible confusion. Applications such as the Installer will be broken during the second background rsync.

  Proposal is to use the second method for those slow partitions where it would be disruptive to disable writing for a long period (e.g., u35 and u52, containing RM builds) and the first method for other slow partitions. The standard method (shut off access to old partition, copy, turn on access to new) will be used for the remainder. Proposal is to do this on Thursday. (Heather) is ok with it. She is not expecting to make a GR tag in the immediate future. (Tom) will send email to the Core list [already sent] and other interested parties.

• Hardware (Tom) 450 Tbytes of new xroot space has arrived at SLAC and is being racked.

  (Richard) We have permission to spend fiscal 2012 money on batch nodes. (Tom) Stuart Marshall came up with 3 different configurations for batch nodes. Scratch space per core ranges from 6 to 50 Gbytes, which might not be enough for us. (Warren) A typical recon file is 35 GBytes, so we might need as much as twice that. But most jobs only use a couple GBytes.

• Pass 8 (Tracy) The goal coming out of the Pass 8 F2F meeting is to use Pass 8 for the 5-year catalog. See implications for schedule on slide 18 of Philippe's presentation. Short summary: a suitable Pass 8 release has to be ready in about 6 months.

• Core talk:

  • GR Pass 8 (Heather) The tag GlastRelease-20-05-00 includes all the no-headers-install code for Windows and little if anything else new. She ran into a snag with systest creation so is redoing them.

    Heather and Joanne each gave infrastructure update talks tailored to the F2F audience.

  • ROOT upgradeWe'll upgrade to version 5.34, the last or nearly so in the 5.. series. (Initial 6.. versions will not support Windows so we'll leave that alone for now.) She is building 5.34 for all interesting OSes> They will be installed in SLAC GLAST_EXT so they will be available for LATEST builds.

  • Pass 7 branches We can't support Pass 7 on vc90 because we can't upgrade G4 (it wouldn't really be Pass 7 any more) and can't get the old version of G4 to build on vc90. This leaves Leon (and anyone else wishing to work on Windows with Pass 7) somewhat in the lurch. For the time being CMT RM will continue to run for Windows — rhel4 CMT builds were shut off at the end of August — but the build machines it uses are getting quite decrepit. Heather had been thinking of turning them off around the end of the year if they don't fall apart first.

  • Windows SCons RM annoyances (Heather) The following are still with us:

    • LATCalibRoot is not properly defined so various test programs fail when RM runs them. The same programs run fine when LATCalibRoot has the right value.
    • Permissions are too restrictive on Installer zip files. The work-around is for someone — generally Heather — to set permissions to the correct value by hand every now and again.

  • Redhat 4 going, going... (Heather) glastlnx14 still lives, but JJ and Tony W may be the only ones with a legitimate need for it as JJ finishes porting the GRB tools.

  • Windows developer environment (Heather) Now that there is a GR tag with the no-install-headers code, she was thinking of installing it on Glast-ts but it's become clear that most Windows developers will not be satisfied with that tag. The problem is that "supersede" (area where a developer keeps packages s/he is actively working on, to be built against some other "base" installation) is broken on Windows in at least 2 ways:

    • Solution files generated in the supersede area should refer to projects in either base or supersede area as appropriate. Currently they don't know what's in the base area. This affects only those wishing to run and debug code from within VS. I made a merged solution file by hand — a messy business — to see what else might be wrong, and discovered
    • that the new no-install-headers code is supersede unaware. One of the things it does is to add include paths during compilation, so that if package A needs a header from package B it can find it. But the current version of this code assumes all packages belong to the same area. Hence packages in the supersede area will not compile if they need headers from packages in the base area. This affects both native SCons builds and builds from within VS.

    Neither of these things is going to be trivial to fix; at best I expect each will take about a week. At that point the fog will have cleared enough for me to see the next stumbling block.

  • Virtual machines (Joanne) A first version of GR Standard was released. In addition to the contents of bare-bones, it contains gdb (inadvertantly omitted from bare-bones; it will be added to the next version) a recent GR tag along with externals it requires and a somewhat pruned calibration archive. The lat user home directory has a set-up file glastSetup.sh which defines the GLAST_EXT environment variable, LATCalibRoot, etc. Suggestions on what else you might need may be found in comment lines. See also Installing and Using a GlastRelease Appliance.

    Initially I forgot to set protections correctly so the first person who tried to download the appliance was unable to do so. Protections have since been fixed but I don't think anyone has downloaded it yet.

  • SeeVogh (Heather) We're thinking about another trial soon.

previous

minutes index

next

J. Bogart, Last Modified: