Core Minutes 4/19/2005
Present: Joanne Bogart, Anders Borgland, Toby Burnett,
James Chiang, Johann Cohen-Tanugi,
Seth Digel, Richard Dubois, Zach Fewtrell, Dan Flath,
Riccardo Giannitrapani, Berrie Giebels, Tom Glanzman,
Navid Golpayegani, Tony Johnson, Heather Kelly, Michael Kuss, Matt Langston,
Chuck Patterson, Igor Pavlin, James Peachey,
Leon Rochester
- Data handling: (Tony)
- System Tests Last week's release had some bugs. Expect another
version soon.
- Data server Thinking about integrating the various pieces.
- Data catalog Matt will be talking about using XML/XPath
with Oracle this Friday at the
Data Handling meeting, particularly as applied to implementing a general
purpose data catalog.
- Parallel processing: (Richard)
We can expect a 500 hertz trigger rate for 16-tower data-taking. Unless
the runs are very short, the resulting data files will be rather long.
We just about have all the technology necessary to split
the processing of a single file among multiple cpus: Navid has a means of
submitting multiple jobs and resyncing when they're done; event in a file
can be accessed randomly, so reading different parts of the file can be
assigned to different jobs, and so forth.
- EM releases: (Heather) EngineeringModel tag
v4r060302p13 was released on Friday, promptly causing System Tests to
crash. The problem, having to do with the fake clock option for CalibSvc,
was found and fixed over the weekend. However the new tag,
v4r060302p14, has its share of problems as well:
- There is some problem, most likely in tracker ToT service. So far
Leon has no idea what this might be.
- There are complaints
about invalid idents. [Since the meeting, Anders realized there may be
a mismatch of xml geometry specified and actual. Fixing the xml
geometry has eliminated the ToT-related crash, but the idents warnings
still come out.]
- There are warnings about negative dacs. Zach explained that
sometimes this is a truly benign warning [but he later discovered in
this case it's probably due to the
Calibration That Won't Die].
- Data-taking: (Richard)
It has been deferred. Test runs will start next week, to be followed by
calibrations and end-to-end runs. Is there anything else that needs
to go in the next EM release (other than bug fixed) before this cycle
of data-taking starts?
- Return of the [T]Clones:
Hiro
noticed that, the further a job gets into a file, the slower the processing
appears to be. (Heather) This was tracked down once before by Ursula:
continual allocation and deallocation of small pieces of memory causes
fragmentation. The cure is to use a TClones array or, if that isn't possible,
some other class which emulates its memory-managing behavior. If we go
ahead and convert existing classes, we need to weigh the extra work involved
in maintaining backwards-compatability,
so old files can still be read with new software, against the hassle of
having to retrieve an old software version in order to read an old file.
(Richard) We need input from I & T before making a decision.
- GlastRelease: (Toby) Last week, as was agreed
at last Tuesday's meeting, Toby made a GR tag. It took 3 tries to get
one that built correctly, but the last one appears to be ok.
(Richard) This
release did not automatically trigger System Tests. Although RM now
prepares the necessary configuration files for System Tests, a human still
has to flip the switch. Config files for HEAD builds are not generated
by RM, but Julie has documented it so that others could do it.
- RM, Installer news: (Navid) He is nearly
ready with a feature allowing someone to mark builds as good or bad
and optionally hide the bad ones in displays. The front end is done;
he needs to write a little script to put the information into the
database. Should be available some time this week. (Richard) Ability
to mark a release should be restricted. Should this be mediated by the
Tony/Matt roles-managing system? (Matt) It isn't quite ready yet, but
will be comprehensive in various respects. (Navid) He already has a
component for this functionality which is integrated into the application.
It wouldn't be easy to replace it with something else, but he agreed it
is worth taking a look.
Navid has a
list of desired features
for the next version of the
Installer. Tony has some ideas about how to make a gui which would hook onto
the Installer. Richard remarked that the different form of downloads Navid
proposes to provide (binaries only, source only, both) give the user the
choice of what to optimize: download size and speed, build time, or
flexibility.
- Towards developer documentation: (Richard)
So far Chuck's work has been primarily aimed at end users. To help accumulate
developer information, Richard suggested we
- Start a Confluence page for developer tips
- Make a JIRA wish-list issue.
- Science Tools Checkout 2: (Seth) See the
summary online.
J. Bogart, Last Modified:
01-Jun-2010 15:48:32 -0700
previous
