Core Minutes 12/7/2004
Present: Joanne Bogart, Toby Burnett,
David Chamont, James Chiang, Johann Cohen-Tanugi, Seth Digel, Richard Dubois,
Warren Focke, Berrie Giebels, Navid Golpayegani, Tony Johnson
, Heather Kelly, Michael Kuss,
Matt Langston,
Francesco Longo, Chuck Patterson, James Peachey, Dirk Petry, Leon Rochester,
Tom Stephens, Tracy Usher
-
G4 news: (Francesco, Tracy) [Many
thanks to Tracy for filling me in on what I mostly didn't hear. Any
remaining inaccuracies are all mine. ed.] The new GR release (v5r0p1)
with new G4 appears to run much faster than GR v4r7. See the last
slide of the
presentation given at the Analysis Group meeting. See also
some statistics that Toby gathered. [however it now seems
likely that the build of old G4 used by v4r7 is debug/non-optimized
and the new build for v5r0p1 is optimized, which could account
for much of the huge difference].
The new G4 calculates effects of delta ray production and ionization
energy loss differently, which probably accounts for the differences we
McPositionHits in tracker and ACD: fewer with low deposited energy; more
with higher deposited energy. Francesco was able to reproduce the GR v4r7
System Test distributions by using old G4 versions of relative routines.
They plan to implement a switch in G4Generator so we can use either version,
similar to what was done for multiple scattering.
-
New RM: (Navid)
It's well on its way! See his
status page
for a terse description of essential elements.
Output of operations
goes into a database, to serve as input for the front
end. The critical pieces
-
processor to handle scheduling
- batch submission
- application operations like compile
have all been tested individually and, as of yesterday, together.
What's left to do: some of the more peripheral functionality,
such as creating Doxygen, and the full-featured front end [first draft just
seeing the light this afternoon.
Log in
using SLAC unix count and click on Release Manager].
New RM should have accumulated old RM functionality, more or less,
in about 3 weeks.
Then would be the time to start serious work on RM for Windows. How this
is done is partly dependent on status of LSF for Windows.
Navid would prefer to use it if possible, rather than writing what
amounts to our own batch system for Windows.
(Joanne) Marco and Riccardo have
agreed to improvements to
MRcmt (callable build functionality on Windows..MRvmct without
the "v") as needed.
-
Ban on moving tags: (Richard, Navid)
Moving or deleting tags confuses CVS history for human browsers, but it
also can cause problems for the Release Manager and hence for the Installer,
since it depends on Release Manager. Seems best to disallow these operations
altogether for normal users.
Navid wrote a little script which will prevent anyone except user
glast from moving and deleting tags. Although it is a bit unfortunate that
certain clearly harmless changes, such as updates to documentation, will
always result in a new tag with this script in place, consensus was that
it's worth it to avoid the very occasional but also very unpleasant problems
we've had when tags are moved. Navid will talk to Pat
about incorporating the script, probably today, and will send an
announcement when it is in place.
-
No more IExternal open season: (Richard)
Some time ago Navid took responsibility for IExternal and its subpackages.
Proper maintenance of it is crucial for correct operation of the Installer.
Please remember to consult with him before making any changes
- Doubles or floats
(Leon) Discussion was to be continued this week [..but wasn't. Try
again next week?]
∫ Flight(w, ..) dw
- EM tag imminent:
(Heather) Head of EM passes all unit tests (Johann, Joanne tracked
down CRflux problem). It's based on GR v4r7 so includes the new
Xerces. Would be nice to pick of the latest LDF. May also want
Riccardo's latest HepRep packages, if OKed by Anders.
When the tag is made Navid will build optimized binaries and
Julie will start up System Tests.
- About timestamps: (Heather)
There are two in LDF data. The original one (seconds + nanoseconds
since base time of Jan. 1 1970) is not necessarily unique, so a
second set of words were added which provide more precision. They're known as
upper and lower PPC words. They can be used in conjunction with the
absolute time to construct a unique timestamp for each event. However,
when doing so, the result need not increase
monotonically with event number for events whose original timestamps
are identical. This can cause problems for TriggerAlg when it attempts
to calculate deadtime. One solution would be to put the deadtime
calculation somewhere else, where it could more easily be turned
off in case the input is real instrument data.
- Release controls on the horizon:
(Richard) Soon, perhaps as early as immediately after imminent EM tag,
changes to production EM will have to go through the newly-instituted
Change Control Board. We have to be able to move quickly
as we discover new "features" in hardware or software, but we also need
independent oversight. For now the release control applies only to the
EM container, not to GlastRelease.
J. Bogart Last Modified:
01-Jun-2010 15:45:10 -0700