CERN: J.Apostolakis, G.Cosmo, V.Ivantchenko, G.Folger, I.Mclaren
SLAC/US: M.Asai, D.Wright, J.Perl;
KEK/Japan: K. Amako, T.Sasaki (partial), H.Kurashige, H. Yoshida,
INFN Genova: M.G.Pia (partial)
Editor(s): J. Apostolakis
Draft version: 0.4, 6th March, 2008, 15:45 CET
The minutes of the SB meeting of Dec 11, 2007 were approved
John reported briefly from the OB meeting of January 17th. The new Fermilab representative, Amber Boehnlein, was introduced. Mauro Morandin (INFN) reported on ongoing steps to define the form of future Geant4 contributions. Richard Mount (SLAC) reported on the foreseen impact of DoE budget situation. Planning for Delta Review, potentially in November 2007, was started.
The next TF meeting is scheduled for Tuesday, February 26th, 2008.
Gunter and Ian reported on the expected reduction of effort in system testing, after Ian's retirement in April 2008. The effort is expected to be reduced over the medium and long term, from the level of about 0.75 FTE average in 2007. An effort is planned to streamline the current testing scheme, by migrating the scripting for running the build and test programs from custom scripts utilising Tinderbox to the scripting system used for nightly test builds for several other projects by the SPI project of LCG at CERN's PH/SFT group.
Ian reported that the largest part fraction of the effort in system testing is in reviewing the results of tests undertaken. Additional challenges included the need to undertake Windows (and currently Mac) testing manually, the very old and weak Solaris machines. The number of tests which continue to give warnings for a long time is a significant concern. The current algorithm for determining whether a test worked (in case the execution did not crash) considers only the size of the error log. This is compared with the expected size, which is different on each platform. There is a strong need to improve the criterion in each test regarding for its sucess or failure.
Action: Agreed to review all the system tests, to address reported failures, reduce potential redundancy, cover fuctions or models not tested and create clear conditions for success.
John reported on the requests from LHC experiments for backporting fixes to the version(s) utilised in large scale production during 2008 and beyond. The current process for backporting fixes to older releases relies exclusively on the release and testing teams; this results in significant difficulties in particular when several revisions have been undertaken within the relevant file(s) or module, and greatly increased testing load. (In particular four versions were under testing in late January: candidate patches for release 9.1, 9.0 and 8.3 and the reference tag for 9.1.) STT noted inability of the current system to group tags proposed for testing contributed to this too.
The challenges of creating patches for production releases and testing them in significant depth,
Action: Gabriele to propose clear process for gathering tags with fixes for previous releases. (The candidate process was to use a branch tag for each release for which fixes are provided.)
The effects of revisions attempting to improve physics performance, and in particular their past track record and the potential for unexpected side effects were discussed. There is a need to improve the stability of established results for physics list used in production (where changes are neither desired or intended).
Note: This does not contradict the need for further improvements in physics lists, to address open issues identified in test beams and other comparisons, which Albert de Roeck noted. The physics list task force clarified that new or revised models have been regularly introduced in physics lists. Typically they are first added in new physics models, and later retrofitted into existing ones - examples include Chips stopping, elastic (in release 8.2). When warranted, faster introductions are made: for example the introduction of quasi-elastic in 8.3 in most production physics lists, and backup lists, which did not include the revised models, are provided for a transitional period.
The work items were proposed as candidate items or work plans for 2008 for each WG:
To address users' needs for stability, it was agreed that fixes to Geant4 9.1 would be provided for all of 2008.
A scheduled release in December 2008 was agreed.
A decision regarding a potential minor release in June 2008 will be taken in April, taking into account the status of developments and the available capabilities and capacity for integration testing.
Agreed to round FTE numbers to the nearest integer, in order to determine the number of representatives for the Steering Board.
Welcomed Vladimir Uzhinskiy (JIRN, hadronics) and Anton Ivantchenko (CERN, STT) to the collaboration.
The date of the review is not confirmed. Tentatively it was scheduled for November 2007.
The meeting adjourned at 12:45.