AF meeting 1st April 2010

Attendees

Andrea Valassi, Pere Mato, Predrag Buncic, Sebastien Binet [remote], Rene Brun, Pete Elmer [remote],  Oliver Keeble, Peter Kelemen, Gabriele Cosmo, Vincenzo Innocente, 

Agenda

• Announcements
• Line management issues
• Action items
• Status of releases
• Platform and external libraries issues
• Experiments feedback
• R&D project status
• AoB

version 3 (9/04/2010 @ 9:30)

Line management issues (Pere)

• The IC post for the SFT group has been published. Boards will be scheduled most probably in May.

Action Items

• Provide test suites for dcache, xrootd, etc. in roottest (Rene)
• Rene reported that a standalone test for xrootd exists in rootest. 
• Stefan reported [by mail] that order to run the test suites for dcache we need to be part of a VO. This is currently being discussed with the IT/ES team who is owning some "test VOs" which we could probably join.  A meeting for how to proceed on this is scheduled for mid April.
• Include the standalone test suites into the nightly builds (Stefan)
• This is pending since it depends from the previous one.

Status of Releases

• Configuration LCG_56f was requested by ATLAS a couple of days ago. It was build yesterday by Stefan and is ready to be used. It contains among other changes:   
• A couple of fixes in POOL and one fix in ROOT.
• Additional fixes in COOL and CORAL (see the list of version numbers from Stefan's mail to the AF).
• Andrea asked about the new version of Boost, which is kept unchanged. Sebastien said that there is a problem with ATLAS to keep it backward compatibility with TDAQ.
• Software installations for Theory (Stefan)
• We are currently testing a new way of deploying the AA software stack in order to make it usable without configuration management tools - i.e. "flat deployment".
• There was a meeting with TH who showed interest in this and a first prototype has been made available to them. The reactions on this were quite positive and SPI will continue to pursue this.
• This generated some skepticism if this is the proper way to distribute large projects such as ROOT. It was agreed that Stefan explains the main goals and implementation details at the next AF meeting. --> Action Stefan
• Geant4 is preparing a patch for version 9.3 (9.3.p01). Geant4 is preparing a patch for version 9.3 (9.3.p01). This includes fixes in various parts of the code, addressing also some optimization issues reported by the CMS performance team.

Platform and external libraries issues

• Python modules: the requests for python modules are steadily increasing, we have decided to "group" python modules into the packages "pytools", "pyanalysis", "pygraphics". These module groups are now available for all platforms and have been also released in the last "58 series" LCG 58b. Modules consist of
• pytools:         processing, ipython, pytest, mock, scons, 4suite, pyxml, cx_oracle, mysql_python
• pyanalysis:     numpy, scipy, matplotlib
• pygraphics:     pyqt, sip, pydot, pyparsing
• Mysql: in order to fix one problem on a windows test Andrea proposed to upgrade the version of the mysql installation from 5.0.18 -> 5.1.45. This is currently being done.
  Andrea asked if we need to continue to maintain the mysql back-end for CORAL. CMS has dropped it some time ago. He will send a mail to the experiment contacts to get an answer. -->Action Andrea.
• Oracle. Finally the SElinux bug on OCI library 64 bit has been fixed with only 12 months delay! :-) The 32 bit OCI was already fixed in default version 11.2.0.1.0. Actually the difference is simply an additional -fPIC option for a single C file in the makefile. Still two issues remain:
• The OCCI is still not fixed for 64 bit. ---> another 12 months? :-)
• The fix for the ppteron/quadcore still missing [added after the meeting: a fix to solve this issue for ATLAS has been received]
• Python: We are currently testing Python 2.6 on a separate nightlies slot, i.e. the python 2.6 package with the rest of the dependencies on 2.5.4. In order to run the full validation e.g. by ATLAS this would need a porting of all dependent packages. QUESTION: If successful, can this be seen as a mid/longterm perspective of moving the whole AA software stack in production to Python 2.6? No answer yet. --> Action all

Experiments Feedback

• ATLAS
• Happy with the POOL fix. Nothing else to report.
• LHCb
• The LHCb experiment has an interest in the Neurobayes product. See AoB.
• CMS
• Concerning TMVA they are going through a number of issues (change in the interface, message verbosity, memory and optimization). The authors are going to provide a number of changes to the package.
• Nothing additional particularly new.
• ALICE
• Nobody

• Workshop preparation (dates, goals, etc.)
• The dates has been converged to June 20 and 21. It will be a one and half day event starting Monday afternoon.
• A still empty agenda has been prepared.
• First announcement will be done to the experiments. --> Action Pere to send the pre-announcement to the experiments.
• Virtualization (Predrag)
• He promised the version 2.0 beta end of March and didn't happen. Next date will be in 2 weeks. It almost works but need some polishing and it takes time to iterate.
• Multi-core (Vincenzo)
• There was a meeting this morning with several reports.
• Fork and COW is well established.
• PSS as a measure of memory usage. Next step is to involve some friendly sites to test it and produce a solid technical proposal. Peter volunteer to run on some CMS site and get some numbers and the feedback. --> Action Pete and Vincenzo to write a half-page proposal.
• Performance monitoring.
• Using intrinsics to make use SSE operations. It is a trade-of between clarity and using intrinsics.  

AoB

• There is an interest from CMS and LHCb physicists to use the Neurobayes software for their reconstruction and analysis code. ATLAS is already using it for some evaluation. This is a commercial software package (code delivered in binary form) that in principle CERN and the CERN's experiments can get almost for free. It was agreed that the negotiations of the conditions with the company is done in common for all experiments under the auspices of the PH-SFT group.
• Next meeting in two weeks. Pere will be absent and will ask Gabriele to chair the meeting.