Core Minutes 8/16/2005
Present: Pol d'Avezac, Joanne Bogart, Toby Burnett,
Eric Charles, Sasha Chekhtman, Jim Chiang,
Seth Digel, Richard Dubois, Warren Focke, Riccardo Giannitrapani,
Tom Glanzman, Navid Golpayegani, Heather Kelly, Matt Langston,
Julie McEnery, Chuck Patterson, Igor Pavlin,
Leon Rochester, Tom Stephens, Tracy Usher
- DC2 code readiness:
(Tracy) Problem with tracking at high energies is fixed.
Remaining issues: about 10-12% jobs are failing, for a couple different
reasons:
- Something most likely in or near Onboard Filter. Heather will
be looking into this.
- Somewhere in Pattern Recognition. Tracy is close to full understanding
of this one (affects events where the CAL axis is close to horizontal
so predicted position is very far away). [late breaking news: Tracy thinks he
has solved this one.]
(Richard) Still more issues:
- There is a suspicion that Onboard Filter is
not working correctly. This needs to be investigated since we depend on
it to reject backgrounds. [also now understood: appears to be due to hard-wired
pedestals and gains]
- Francesco has found that crashes in G4 in background generation
have to do with particle energies > 10 TeV. One possible work-around
is to cap these energies.
(Sasha) He reported
on a comparison of calorimeter code performance in GlastRelease HEAD builds
1615 and 1616. (Note:
1616
contains older versions of
CalXtalResponse and
CalDigi than 1615.) Using a restricted source
(muon pencil beam into center of a crystal), he saw no differences in
reconstructed energy or position. There was a difference
in CAL lo and hi thresholds between the versions (attributable to
calibration parameters, not executable code) which could cause a
bias in event selection. To be investigated.
(Toby) would like to review tagged packages not now in GlastRelease HEAD
to determine which if any should be promoted for the DC2 runs. Leon made
a case for including recent changes to
TkrDigi and GlastSvc
addressing JIRA GEO-6.
We probably don't need the latest tags of HepRepSvc,
IExternal/OmniOrb,
CalibData or calibUtil.
Probably will not pick up newer tags of RootIo and
reconRootData, or at least not without thorough
verification first. Toby will pursue the decision-making process via email.
When do we freeze for DC2?
- EM: (Eric) He and Heather found and fixed
last week's problem: the unbounded memory usage by ROOT. A reset was being
applied in the wrong place (mcRootWriterAlg) so that for jobs which didn't
write
out mc, some per-event memory never got released. They moved the reset
to a spot where it will always get executed, for any kind of job.
- I & T: (Richard)
- They'll be taking calibration data todayish.
- ACD arrived as planned on Saturday.
- An alternate vendor may be able to make the flex cables that were
going to hold up tower delivery (current vendor has about 10% throughput).
If so, will be able to install towers a couple at a time rather than
having to do all the remainder in one go, at the last minute.
- RM, Installer: (Navid) The Windows LSF queue is
still down, so Windows RM builds are on hold. Everything needed for the
Windows bin directory—both creation during
build and handling by
the Windows Installer—is done, but can't be fully tested until
Windows RM builds are back. Next will be a common
lib directory and
a pfiles directory for Science Tools.
Navid may turn a lemon into lemonade by
putting the new changes for lib
into production now, so
that he can deal with problems one OS at a time. There were some related
questions and comments:
- (Jim) Where do Python scripts go? (Navid) is thinking about setting
up a CMT pattern for them and so forth, but hasn't had a chance to
work out all the details yet.
- bin directory just has scripts in it which
invoke the actual binaries,
still stored in their usual place. Libraries really will be moved to
a different location once the new stuff is in. (Riccardo) doesn't
believe MRvcmt will be affected.
- (Toby) Do the new directories allow for a distinction by tag so that,
for example, it's possible to keep VC8 and VC8Debug builds in the same
development area? (Navid) Doesn't do this now, but could be made to.
- (Leon) Do changes need to be made to a package's .cvsignore to avoid
committing these binary directories? (Navid) No, because the
bin and
lib directories are outside any package.
- (Toby) Should this go in before big DC2 runs? (Richard) It needs testing,
but isn't that big a risk in the sense that, if it fails, it won't be
in some subtle way which only affects the science.
J. Bogart, Last Modified:
01-Jun-2010 15:45:10 -0700
previous
