Core Minutes 9/25/2012
Core Minutes 9/25/2012SeeVogh: Another attempt today with mixed results.
ScienceTools: (Jim) Is working on implementation of a scheme to set correct IRFs automatically when running tools downstream of gtselect.
FSSC: (Tom S.) Alex has managed to build the new ST tag on rhel5 and rhel6. Tom has run into problems on Mountain Lion because of ROOT. Our current production of ROOT won't build there. When he tried building ST without ROOT (which is supposed to be supported) there were problems with EarthPhenom, which must have ROOT available. (Jim)There is code in celestialSources which should exclude EarthPhenom for ROOTless builds. (Tom) will see if he needs a new version of something. Once Heather finishes the migration to the new ROOT this will become a non-issue, since that version of ROOT is ok on Mountain Lion.
(Liz) Solar system tools are included in this tag. Is there a thread in the workbook yet? (Richard) Doubtful. (Jim) will coordinate with solar system guys.
Hardware (Tom) Five new Dell servers for xroot have been installed and are in use.
nfs partitions (Tom G.) Wilko has been draining wain32. wain31 is being rsynced. We'll wait for M.E.'s return (next week) before cutting over.
Reprocessing: (Tom G.) We're waiting for IRFs and new calibrations for the next interval (post July). (Leon) We get new CAL calibrations every 2-3 months and new TKR dead strips every 2 months. The current suite of calibrations is good through September.
pass 7, Pass 8 (Leon) Tracy found and fixed the problem with CAL calibrations not being updated automatically. Leon's suspicion that ACD might suffer from the same disease was borne out. Tracy has fixed that as well.
(Leon) has been working on making a pipeline task for the generation of overlay events. The collection generated will be more inclusive than in the past. (Tracy) Discussions are ongoing concerning exactly which events should be used. He is working on code to defer selection to the job that does the overlaying, based on conditions placed on trigger bits.
(Tracy) All of the above will go in a GR release expected for about a week from now, following the imminent tag for the new ROOT version. Expect lots of CVS activity and new package tags between now and then.
(Tracy) Another item to be included is some tuning of the cosmic ray track finder. Bill had noticed that two-track events have better resolution and better handling of backgrounds, so requirements for two-track events were loosened. However, it seems to have back-fired; we'll retreat to something similar to the original.
Core talk:
ROOT upgrade (Heather) It's ready to go for rhel5. Path for rhel6 has been more complex. She first had to create shareable versions of python libraries for ROOT to link to, meanwhile keeping the statics around since the python executable needs to link to them. If these or other problems continue to hold up the rhel6 build she may skip it for now and only upgrade ROOT on other platforms. While she was at it (making a new python build) she included PyYAML, something which had been requested a while back.
Skimmer problems (Heather) This problem, concerning skims with large numbers of files, is still with us (see JIRA FDH-31). David Chamont will try to find time to take a look; Heather might have another go at running in the debugger.
(Tom G.) There is a small issue with skimmer finding shareables. The situation changed with the change from CMT builds to SCons builds. It maybe can be resolved by updating documentation.
lsf replacement (Tom G.) The motivation for replacing lsf is the cost. The task force has now narrowed down possible candidates to just two: an offshoot of the Sun (now Oracle) Grid Engine (it's open source so another entity, Univa, has picked up maintenance and development), and something rejoicing in the name SLURM, used at Livermore. SLURM has no Windows support and is not likely to ever have it. Cost for these would be 1/3 to 1/2 of what we're now paying for lsf. Next step: pilot projects with these candidates. Original goal was to have this all in place within about 6 months (lsf licenses need to be renewed in the spring), but the Task Force thinks a year is more realistic.
(Heather) Other possibilities are: SLAC keeps a few lsf licenses for our use (who manages those machines? TBD), we use Hudson/Jenkins to do batch allocation, or we use Hudson/Jenkins even more extensively. If SLAC does ultimately replace lsf we could ask for manpower from the Computer Center to help us adapt. (Tom G.) Basic command structure (i.e., kinds of things you can do) for the two candidates is similar to lsf though the details are entirely different. (Tom S.) It might not be too much work to convert RM to use a different batch system [assuming it handles all OSes of interest] if it just means changing batch submission commands; needs to be looked at. Or perhaps we could use Hudson/Jenkins to launch processes which would run the existing 6 or so RM programs that do the work.
To keep up with this issue you might want to set a watch on the LSF Replacement Confluence page Heather has started.
SCons RM (Heather) There is still a problem with LATCalibRoot on Windows. (Tom S.) Yes, the fix he put in apparently needs to be fixed; no test programs are running. [He fixed the fix later this afternoon.] (Joanne) There are two other issues on other platforms. vc71 GR builds are failing because they're looking for the wrong version of geant. rhel6 GR builds are mostly failing in the check-out step, and not always in the same place. Messages sent to the relman list indicate the RM gave up after 30 minutes. (Tom S.) The check-out normally only takes a couple minutes, so it should not be timing out. (Joanne, guessing) Could this be due to contention among all the checkouts for different platforms occurring more or less simultaneously? If so, would it be possible to delay start of checkout step by varying amounts, depending on platform? (Tom S.) would have to think about it. still with us:
Windows developer environment (Joanne) The supersede support described last week has all been committed and tagged, but apparently has no users yet, so I have no news. (Heather) will incorporate the new stuff in a tag.
AOB
previous
|
minutes index
|
next
 |
J. Bogart, Last Modified: