Present: Joanne Bogart, Toby Burnett, Richard Dubois, Marco Frailis, Riccardo Giannitrapani, Navid Golpayegani, Heather Kelly , Michael Kuss/Johann Cohen-Tanugi, Sean Robinson, Leon Rochester, Alex Schlessinger, Tracy Usher
RootIO and ROOT: (Richard, Heather) Reprocessing events for Bill Atwood has uncovered some problems. Still have a memory leak (memory usage heads towards 1 Gbyte with a couple hundred thousand events) which causes UW Terminal Server to crash (however SLAC Linux can handle it), hence Bill can't do his own reprocessing. Richard has been rebuilding on Linux and doing it for him, awkward for all concerned.
Heather has added support for chains of input files. She has also committed a rough and ready version of the pruner ROOT macro into CVS. One project which has languished for lack of priority is to allow random access to events from the gui. Should not be that big a deal to implement, just need some data structure relating (run number, event id) to sequential location within file. [At the time of the meeting, there was a suspicion that run numbers were not getting written out, since shown to be pilot error. ed.]
DetDisplay resurrected: (Toby, Tracy, Leon) The display of read-in events was unpleasantly primitive, so Toby modified and incorporated code from Recon and G4Generator into the DetDisplay package, which had been moribund for a couple years. Mods were required to get data from TDS rather than from local classes. Recon has now been removed from GlastRelease and display code has been removed from G4Generator. It all looks OK. Leon noted there are some hardwired numbers which should be obtained in some other way, but there don't appear to be any significant inaccuracies at the moment.
Some of displayed events look weird, most likely because of trimming the MC tree. Tracy has been tracking this down, and meanwhile discovered that he cannot reproduce events by specifying random seeds, as used to be possible some months back when Karl first installed and tested the GlastRandomSvc. Random seeds are being set correctly, but the events generated look nothing like the originals (true for both Windows and Linux). Changes made since to GlastRandomSvc should be examined to see if they could be responsible.
FRED, HepRep and friends : (Riccardo) After some refactoring to provide insulation from the concrete implementation, everything is working well on Windows. See the how-to to try it out. Toby would like to get it going on the Terminal Server. We need builds of certain libraries, notably ACETAO, before we can use all parts of the system on Linux. For now Riccardo will add those packages which can be used on both Windows and Linux to GlastRelease. The rest will be added when there is support for both systems, not expected to be a big project. He is also working on adding more information (from Reconstruction) to the HepRep representation.
Move 'em out: (Toby) There are 16 packages currently in the Latest build but not yet in GlastRelease head. Hearing no objections, he proposes they all be promoted.
Release Manager update: (Alex) The enhanced system, allowing builds according to CMTCONFIG value, is working; he'll be committing to CVS shortly. In particular this will allow parallel Release Manager builds on Windows, e.g. to the shared GLAST space on the V disk. This could supplement the UW Terminal Server as a platform for Windows developers. Currently the VCMT-generated project files are used to do the Windows build. Toby suggested it would be advantageous to move to nmake.
Alex has a set of external libraries from Johann, compiled with gcc 3 but on a Red Hat 9 system. Since Red Hat 9 has an incompatible glibc, these can't be used on SLAC's Red Hat 7.2 and 7.3 systems. For now might go ahead and build gcc 3 binaries for these systems, but in the not-so-long run we need to move to Red Hat 9 since this is where SCS is headed. Red Hat itself will be ending support for 7.* by the end of the year.
Linux notes: (Alex) Although our standard Linux environment is that of the machines in the noric (not noric-new) pool, which run Red Hat 7.2, everything should also work on 7.3. There is a typo in our group.cshrc [make that "there was a typo". ed.] which could cause the wrong compiler to be used when gmake is invoked directly rather than via glastpack.
CMT and more: (Toby) We would benefit by moving to CMT v1r14. It supports the concept of a Release Area which can be used to simplify the distribution of our releases. Binaries and public includes are copied (Windows) or symlinked (Linux) to a common space.
He has been revising GlastPolicy to remove its dependence on GaudiPolicy, instead incorporating into GlastPolicy just those parts of GaudiPolicy that we need.
Onboard filter code: (Navid) See the new package OnboardFilter [newly renamed since the meeting, at which time it was OnlineFilter. ed.], including JJ's onboard c code and the file filter.cxx which acts as a bridge to our (Gaudi, c++) environment. This packages uses another new package, EbfWriter, which converts TDS events to an ebf-file-like format suitable for input to the filter.
David Wren is exercising the code and has found some bugs. He's talking to JJ about them. Also somewhere there is a substantial memory leak. Might be a good candidate for exploration with ValGrind.
Pipeline update: (Richard) We're waiting for Opus source code. We have officially and bureaucratically requested it through our designated bureaucrat, Jay Norris, and so far have gotten the standard bureacratic response: nil.
Workshop coming up: We will try to squeeze a discussion of Steve's request for hooks into the creation of ntuples into an already-full agenda. It's not yet clear what he wants and for what class of users.
Steve and Richard will be unavailable (drafted for subsystem manager meetings) on Thursday and perhaps Friday morning. These meetings are fall-out from LAT's recent request (which, it is rumored, had better be its last) for an additional $16 M and 3 months' delay of the schedule.
previous | minutes index | next |
J. Bogart Last Modified: 01-Jun-2010 15:46:19 -0700