Core Minutes 6/12/2007
Present: Joanne Bogart, Toby Burnett,
Eric Charles, Jim Chiang,
Warren Focke,
Tom Glanzman,
Navid Golpayegani, Heather Kelly,
Michael Kuss,
Julie McEnery, Igor Pavlin,
Leon Rochester, Tracy Usher
- Service Challenge issues:
- Too-high energy events (Heather) Tracy tracked this one
down. (Tracy) The problem was due to a change made in the way we build
G4. As of late March G4 was rebuilt with VERBOSE mode off (to
improve performance) but there is a bug
in the version we're using so that this resulted in uninitialized
memory for cosmic ray background sources. The very-high-energy
symptom only manifested itself when we used full background sources.
Note we're several versions back in G4. Newer versions don't have
this bug. (Heather) We now have patched source.
Navid will build new G4 binaries for additional platforms as needed
and make a new tag of the G4 interface package. (Tom) Any runs done with this
version of G4 with hadron or heavy ion sources
are potentially suspect, even if they don't have
the very-high-energy problem. (Heather) Would be
good to somehow annotate those runs made with the "bad" Geant. There seems
to be no new tag; apparently the library was rebuilt in place. Hence the
affected runs cannot be identified just by checking the version of
GlastRelease used, but rather by when and where the job was run.
[Let this be a lesson to us!
Changes in build method or build options — in this case the value of the
VERBOSE macro — for external libraries need to be tagged
just as much as code changes. ed.]
(Leon) believes Beamtest processing was complete before
the period when the suspect Geant was in use at SLAC, but he will inform
other Beamtest folk just in case.
- Pulsar initialization bug (Heather) Max has a fix for this. He will
probably make a new tag later today.
- Interleave problem (Toby) Tom's Interleave test failed;
signs point to Interleave code.
(Tom) The failure is apparently
connected to new merit variables. (Toby) In particular, to one
called McRa. Its value must be calculated after the interleave
step; maybe it isn't being done right. Unfortunately the
Interleave test program, which might have
been a useful tool to debug this problem, is currently broken.
[post-meeting update from Tracy:
-
the test program was broken because it was
trying to use a stale test merit tuple which did not have the right ntuple
variables (e.g. obfCnoStatus is
now ofbHFCStatus)
- Tom's interleave job failed because the interleave branch was missing
the two algorithms (McCoordsAlg and ObfCoordsAlg) for calculating
various pointing-related items, in particular the McRa variable that caused the
initialization to fail.
It would appear we are not completely out of the woods yet: after running
a test Tom found a new issue which will need to be looked at (but at least
not a crash; jobs run to completion now).]
- Albedo background (Tom) Our sample may be too sparse;
need more.
- Random number limitations (Tom) With the current scheme
we are limited to 20,000 runs per task and 50,000 events per run;
otherwise we run the risk of duplicating events, at least as far as the
MC particles. Digitization and orbit orientation will tend to
introduce differences. Michael has some suggestions; may wait until 1-year
run before attempting to implement and incorporate them.
- Other status (Tom) The large background run is complete and
has been binned.
- Geometry (Heather) The new ACD geometry
involves changes to ribbons as well as to side tiles. Side ribbons
used to be modeled with a single segment per ribbon and face, but in
order to follow the contour of the shingled sides they had to be
broken into several segments according to one pattern for
x-measuring ribbons and a slightly different one for
y-measuring ribbons. Minor updates have already
been made to AcdDigi to handle the new
segmentation. Additional changes will be needed in
AcdUtil, most likely
depending on a new (as yet unwritten) utility in
detModel. We won't be able to make use of the
new xmlGeoDbs tag until this is done.
- MRStudio update: (Heather) Riccardo,
apologetic as always, expects to be working hard on MRStudio in a couple
days, as soon as the school term is over. He'll start with the Linux gui
problem (he now has a Linux box but up till now hasn't had a chance
to install everything needed on it) and go on from there.
- RM, Installer: (Heather) The sudden
doubling of size of RM-built releases is due to re-instituting -g
(request for debug symbols) in optimized builds. Binaries could be stripped
for the Installer and for Pipeline use.
[post-meeting Joanne argued in favor of keeping the full unstripped libraries
for Pipeline. There are other ways to reduce the size of the Pipeline
installation.]
- CMT Action Committee report (Heather)
Not much has happened since last week. A decision will be made on
which way to go (SCons+ or NPMS) no later than the end of this
month, very likely sooner. We need some
checklist
entries for the proposed SCons approach.
- Whither Doxygen (Leon) The RM links to
Doxygen output for more recent builds are broken. (Navid) It appears to
have stopped working on April 5th. He will investigate.
- Service Challenge Steering Committee
(Julie) ..will meet tomorrow. One outcome will be a status update.
J. Bogart, Last Modified:
01-Jun-2010 15:47:40 -0700