Core Minutes 23 October 2012

Core Minutes 10/23/2012

Present: Joanne Bogart, Jim Chiang, Richard Dubois, Elizabeth Ferrara, Warren Focke, Tom Glanzman, Heather Kelly, David Landriu, Leon Rochester, Alex Reustle

ScienceTools: (Jim) No development news for the post week. Jean reported an issue with excessive execution time for larger numbers of fixed sources, and he has been running a couple of tests that I suggested. Will work on that this week.
Hardware: (Richard) The IFC computing budget is currently under-running about $400k. With longer P8 processing times (Johan estimated 2x from recent AllGamma runs) and big MC runs expected, we figure we can get some time back by spending a good fraction of that money on batch cores. With current estimates that is 1024 cores, to add to the 400 we're already tabbed to get. I'm hoping the price will drop a bit and we can invest in some 5 TB tapes to start the migration of existing 1 TB tapes.
Scheduled outage: (Tom G.) This Thursday there will be a multi-hour outage, including L1, in order to move various servers to a high-availability rack. We will take the opportunity to simultaneously do the 3rd of 4 nfs server migrations.
Pass 7/8 (Leon) and Tracy were engaged in finding and fixing bugs, but now everyone is preoccupied with poster preparation.

There will probably be a Pass 8 meeting in Paris in late January. Because of new DOE rules this needs to get settled very soon or no one will be able to go.
obf B3-1-0 and GR (Heather) We were ready to go with a new L1 branch GR using obf B3-1-0, but the flight upload failed. (Richard) thinks CCB will be deciding today when to try again.
Core talk:
- New Pass8 tags (Heather) 20-06-04 has a CalRecon bug fix. Systests are done; see JIRA LPATE-61. There will be another GR tag soon to pick up a new TMine release with a fix affecting Windows only.
- Our ROOT packages and their dependencies (Heather) With CMT builds, stand-alone ROOT users could load the commonRootData library, then the primary library they were interested in (such as digiRootdata) and be on their way. This is not sufficient for SCons builds. Unlike CMT builds, they link explicitly against all dependencies. That could be changed, but there is general agreement that this is the proper way to link these libraries. The new ROOT, 5.34, provides a function AddDynamicPath which will add to LD_LIBRARY_PATH [as does current production ROOT, version 5.26, but in 5.26 it doesn't work!]. The proposal on the floor is to "fix" this problem by moving to 5.34, which we were going to do anyway. See JIRA ROOT-46 for more on this topic.
- Systests Team (Heather) sent email yesterday to likely candidates. No response yet; she doesn't expect there will be until after the collaboration meeting.
- RM issues (Heather) There still is a problem with file permissions on zip files of RM Windows builds. Joanne noticed they now have group read permission (or did they always?), but this isn't sufficient; we need world read.
  
  (Heather) In order to do systests on HEAD builds we need HEAD optimized enabled. This will require some changes to RM database.
- Windows developer environments (Joanne) Leon tried out supersede. There were no build problems, but Gleam immediately crashed without providing any diagnostic information. I followed Leon's recipe and got the identical result. Using the utility depends.exe I discovered libraries in the base release — in particular facilities.dll — could not be found. The PATH variable included the correct entry for these libraries and looked ok, but the form of that entry had double quotes around part of the path. One set of quotes somehow got lost and the loader choked. The quotes probably are not necessary (definitely not in the case I was testing, where the path had no embedded blanks) so I took them out. Then Gleam was able to load facilities.dll and the next problem revealed itself.
  
  Gleam could not translate environment variables like $GLEAMJOBOPTIONSPATH. This should be handled by the routine setupEnvironment in the facilities package, but it doesn't work properly for Windows. The code in there now can fake it, but only if there is no supersede directory. I have never been very happy with the implementation, even on Linux where it generally does the right thing. I plan to try a different strategy to make information known at build time (which packages are in the supersede area) available also at run time. I expect it will take a couple days to determine if the new approach is feasible and implement it.
- ROOT upgrade to v5.34.01 (Heather) The build for Linux is available in GLAST_EXT and the Windows build available at the ROOT site is more or less what we need, but she needs to rebuild Gaudi using this ROOT before we can make a GR build with it.

minutes index

J. Bogart, Last Modified: