AF meeting 08 February 2007
Andreas Pfeiffer, Rene
Brun, David Quarrie, Pere
Mato, Oliver Keeble, Gabriele
Cosmo, Philippe Charpentier, Fons Rademakers,
- Line management issues
- SPI project reorganization
- Introducing Oliver Keeble as LCG GD contact
- Action items
- Request builds of GFAL, LFC client libraries
for SLC4 (Pere)
- Status of nightly builds
- Inclusion of the tests of POOL (Dirk)
- Windows builds (Andreas)
- Releases status
- ROOT support for DPM
- CERNLIB and 64 bit support
- Experiments feedback
- Platform issues
- Request for Qt4 for windows
- cx_oracle for x86_64
- Request for the Installation of fftw
- Topics for future AA meetings
version 1 (14/02/07 @ 10:40)
Line management issues
- Pere Mato has replaced John Harvey as PH/SFT group leader. Pere has started from the first of February.
- The junior fellow that the SFT group has been requested has been
granted. The candidate will be starting the first of April to work in
the SPI project.
- Andreas Pfeiffer will be leaving the SFT group and joining the
CMS experiment as was announced last week by Michael Doser (PH deputy
SPI project reorganization
- Due to the departure of Andreas, Pere presented the plan for the
SPI project. The main proposal is that Stefan Roiser takes over from
Andreas as SPI project leadership. The manpower available for the SPI
project remains very low. The new fellow starting in April will help a
bit but the situation remains in the limit. The proposal was agreed by
the AF. Stefan will replace also Andreas in the AF meetings
starting from next one.
Introducing Oliver Keeble as LCG GD contact
- Oliver Keeble from the LCG Grid Deployment (GD) area will be our
link-person between the AA and the GD with the main goal to improve the
communication between AA, GD and experiments. He introduced himself to
the members of the AF. He will be added to the AF mailing list.
- Philippe asked if he could rescue and continue the
discussions started by Flavia Donno and Andrea Sciaba concerning the
installation of the client libraries and their dependencies and the
control we have on the libraries installed in the worker nodes.An
update of the initial spreadsheet needs to be done. Olivier will be contacting Stefan to update the table.
- Request builds of GFAL, LFC client libraries for SLC4.
- LCG/GD are working hard to get the UI client done for SLC4.
Oliver said that yesterday a new tag was produced that could contain
the necessary libraries .... Need to be verified.
- SLC4 version of the UI is built against VDT 1.6. All the builds
should be there on this version. This is receiving constant attention.
- Whenever the versions are ready we will be asking the GFAL and
LFC teams to proceed with the installation in the AFS external area.
Nightly build system status update
- CORAL/POOL tests. All the POOL/CORAL tests (about 50 test)
are using OVAL package directly. These tests need to be adapted
to use QMtest instead. Equivalent work has been done and partially
completed for COOL. For the data regression
tests is not completely clear how to do a fine grain integration with
QMtest and the nightlies. There are issues to do with account settings,
the AFS account of the job running the tests, database node
security issues, etc. The time estimate for CORAL is to be ready by the
end of the week. Both Ioannis and
Giacomo are working on this. POOL will be done later.
Dirk reported that several bugs found in CORAL
in the later stage when releasing configuration LCG_50. This will be fixed and a new release 1.7.X will be prepared.
Dirk also mentioned problems with Mac OSX support for POOL/CORAL to do with the link/plugin libraries.
- Windows builds. Andreas reported that have been not much progress since last meeting (no much time during the
last 2 weeks). Andreas has already talked with Stefan on how to do it with the help of Bertrand.
- Andreas mentioned that some more fixes are being developed in order to support non-debug platforms. This will imply
some extra time for building in the platform for which the two flavors (debug and non-debug) are
- Andreas intents to build the final releases with the same
scripts that are used by the nightlies. Some instability may be
introduced when the new scripts will be put in place. Since this may
affect the current nightlies, which are being used by ATLAS, it was
decided to delay their introduction for a few days.
- ROOT support for DPM
- David understood that sites that are running DPM could run
latest version of ROOT. It seems that this should be possible if the
right suffix is added (rfio: ?). Concerning the common RFIO
library between Castor and DPM nobody knew the current status. In
principle, it was foreseen by the middle of October. Ask
again what is the status of this common library. Action ==> Oliver.
- CERNLIB and 64 bit support
- Rene reported that many people have problems with CERNLIB on
SLC4 64-bit (by using the h2root utility).
Rene has debugged the problem and it seems to be related to mixing 64
bit and 32 bit code. Ian McLaren has made available
a distribution with PAW built in 32 bit mode that works fine on 64 bit
nodes. But, the packlib library that was compiled
with 64 bit and is not really working. Ian has to look into it and
evaluate the work that is needed. Rene will see with Federico Carminati
in case a solution already exists. [added after the meeting: Rene
claims that they have solved most problems with h2root when calling
Fortran routines from CERNLIB. The main point is the following: in
fortran a real function declared like
real myfunc(...) must be declared double myfunc(...) in 64 bits mode.]
- The issue remains the official position vis-a-vis the
maintenance of PAW. This is not so much a problem for the LHC
experiments, which are migrated or have definitive plans to migrate
away from PAW. We could
envisage to remove the support for 64 bit if this implies a lot of
- LCG_50 status
- Done last week. ATLAS has used it already in the last two nightlies.
- CMS (notes sent by Vincenzo)
- Production is continuing with 2006 "end-of-the-year release"
(cmssw 1.2.0) pre-releases of the new series (1.3.0) still build
with LCG_49: LCG_50 went into the last nightly. The next 1.3.0
pre-release will be build with LCG_50.
- 1.3.0 will be released with G4 7.1 and G4 8.1 (old
MultipleScattering) on both slc3 and slc4 (ia32) for a full regression
and validation tests. CMS is waiting for a patch for G4.8.2, to be used
to build a 1_2_0_g4_82 release (but it is known to be buggy in the QGSP
physics list). CMS will keep official 1_3_0 (due end of February) with
G4.7.1p02, and have a parallel 1_3_0_g4_81 with G4.8.1p02 for
reconstruction studies. For 1_4_0 (due in march) CMS aims at G4.8.2.
- Using the glite UI released for SLC4, CMS is able to
submit SLC4_ia32 jobs to the grid from SLC4_ia32 machine
with CMS grid tools (CRAB, Boss) as they do not use glite API (only
- In the last nightly a problem was found in root (5.14.00b)
where the limit of 12000 typedefs in cint was hit. We will see if we
can reduce the number of typedefs internally, if not, we will ask the
ROOT team to provide a patch for this.
- They are still fighting the IO/Reflex problem in the Muon event
model. The problem could be isolated in a ROOT standalone program that
was reproducing the error thanks to Stefan and Axel. Philippe Canal is
into it. Rene will have a
meeting this afternoon with him. David re-iterated that this problem is
critical for ATLAS. [added after the meeting: the problem has been solved and a new patch version is going to be released]
- ATLAS is hiring from BNL a new collaborator to work on I/O issues together with
the ROOT team.
- LCG_50 is being used in the nightlies. David transmitted a question from
Richard Hawkings about the time scale for COOL
2.1.0. Dirk said that no time scale is known. Targeting it for
mid-March would be OK. Some more discussion is need for the final list
of items to be included in the release.
- LHCb. Production still ongoing. A bug was found in the way the
generator was configured and as a consequence some part of the
production needed to redone. In parallel, there are the developments of
the applications on top of Gaudi 19. Not using yet LCG_50 since it
implies some changes to accommodate the new API of COOL. No plan of a
production with this
Philippe asked about the status of Castor for windows. Pere reported
that Bertrand have done the the work but is not aware that the changes
have been given back to the Castor team.
- Alice. Nothing to report.
- Request for Qt4 for windows.
The installation into the LCG external area was agreed. The built is
already done by Bertrand. It takes about 1.5 GB (300 MB the
Rene would like to know what will be the usage of Qt3 and Qt4 by the
experiments. Maintaining both Qt versions for both available interfaces
(GSI and BNL) is double work.
- cx_oracle for x86_64 (python binding to Oracle). This was
needed to access Oracle from Python, but experiments are encouraged to
use the CORAL python binding that will allow similar functionality.
Dirk said that the OCI oracle client (C interface) should be there for
all the platforms. Dirk recommendation is
to stay away from OCCI. He will check
that OCCI (C++ interface) is not being used since this interface has
always been a problem with compatibility platform/compiler. Action
ROOT team will also check what the interface used is OCI and not OCCI
- Request for the installation of fftw into the LCG externals to be able to build the corresponding ROOT plugins.
The request some from the ATLAS workshop on analysis. The AF see no
objections to install it and to build ROOT with the new plugin.
- Two more python modules have need asked by CMS iPython (already
there), NumPy and MathPlotLib. Andreas said that it souldn't be a
problem to add them.
Topics for future AA meetings 
- Feb 14 - (no meeting)
- Feb 21 - COOL status by Andrea Valassi and GD presentation by Oliver Keeble
- Next AF will be February 22.