Present: Joanne Bogart, Toby Burnett, Xin Chen, Johann Cohen-Tanugi, Eduardo do Couta e Silva, Richard Dubois, Sean Robinson, Leon Rochester, Alex Schlessinger, Tracy Usher, Karl Young
New packages: (Toby) The new package astro encapsulates some fundamental astronomical concepts such as Julian date, celestial/galactic positions, and Earth geodetic coordinates. Also included are classes to report on positions of the Sun and Moon, and a detailed Earth orbit model, code due to Gino Tosti and collaborators. A search turned up no other such set of tools in C++.
The new package RootPolicy will centralize proper handling of rootcint for those packages which need it. It looks fine in test, but Toby will await Heather's return and seal of approval before unleashing it for standard use.
Random report: (Karl) Tests have been completed. He has investigated three scenarios:
As per proposal, have separate streams (one per application using randoms), each starting with a new seed every event.
Generate a single long stream.
Separate streams, but starting with new seeds only at start of run.
He used the Diehard test suite (written by George Marsaglia at Florida State) to evaluate the quality of the randoms so generated. The Gaudi default, RanLux, did miserably. HepRandomJames, which is what we've been using (since it's used by CLHEP) is better but still not great. However there are other alternatives which did well.
Do-it-yourself - writing a Gaudi tool: (Richard) Faced with the problem of designing and testing multiple models of light taper, Richard opted for a Gaudi tool. The Gaudi Developer Guide (chapter 12) does a reasonably good job of describing tools and how to implement them, but the description is still a bit sketchy for many of us. Richard wrote up his experiences which should smooth the way for future tool writers. The bottom line is that a Gaudi tool is appropriate for this particular task and Richard was able to implement it without undue difficulty. [For a good discussion of the fuzzy line between algorithms and tools, touched upon during the meeting, see the opening sections of chapter 12 of GDG, especially 12.2. ed.]
Nightly builds/Release manager: (Alex, Richard) Builds of Gleam are ongoing. Several enhancements/additions are envisioned:
web page history of builds of each constituent package
Hall of Shame
when package fails to build or test properly, email not only owner but also all recent committers.
run system tests at release time; save and display results
keep track of resources used per package ((Toby) speaking of which, it appears we have a moderately serious memory leak, probably associated with ROOT i/o, which will become more critical as longer runs are routinely needed)
Thanks to the improved ability to compare results from different builds provided by the facility, Toby has noticed a degradation in PSF which should be attended to.
PDS/TDS discussion: (Richard) There are two important items to be addressed:
Determine the contents of the PDS, given that it has to serve two purposes: the traditional one, as mirror of the TDS, and also it should be a self-contained data store appropriate for use in a possibly non-Gaudi analysis environment like command-line ROOT.
Revive some form of summary n-tuple.
For thoughts to date on these issues, see these notes.
Incomplete identifiers: Riccardo has made changes to G4Generator to use the new GlastDetSvc services. All relevant changes (detModel, GlastSvc, G4Generator) are still untagged, pending completion of testing. Joanne embarked on such testing, but got held up before even trying a non-LAT top volume by a problem when tracker recon code encountered some weird digis. This was correctly diagnosed by Leon as being due to tracker digi code padding identifiers which it assumed to be incomplete when they were actually from the ACD. In the new world we can do a better job here because all identifiers will be complete [Leon has already put in a fix and made a new tag for TkrDigi since the meeting ended. ed.]
AOB: (Leon)
Perceived problems with AIDA appear to have been due to pilot error. It's working now.
Many thanks to Toby for the new version of VCMT, which allows streamlined builds as advertised.
previous | minutes index | next |
J. Bogart Last Modified: 01-Jun-2010 15:47:53 -0700