Fons, Andreas, Dirk, Gabriele, Vincenzo, Rene, Philippe,
Pere, David
Agenda
Line management issues (John)
Action items
Feasibility of the migration to SCRAM v1 (Andreas)
Status of the writing of SMatrix objects using Double32_t (Rene)
Releases status
New ROOT development release 5.11.04
Scheduling a new configuration (LCG_45?)
Experiments feedback
Topics for future AA meetings
AoB
version 2 (7/06/06 @ 16:00)
Line management issues (John)
The two CERN Fellows assigned to the SFT group (ROOT and Geant4 projects) have accepted.
The EuroPython conference is going to be held at CERN this year
from 3rd to 5th July (http://www.europython.org/) The SFT group is
acting as the local organizer. Experiments and projects with high level
of contents in Python are encouraged to propose presentations of their
work, in particular for the Science track. LHC experiments has been
contacted. ATLAS and LHCb are preparing a proposal each. Also a Geant4
presentation could be a possibility.
Action items
Feasibility of the migration to SCRAM v1 (Andreas)
Andreas was only aware of this action yesterday. The bulk of the work
is in the adaptation of the ToolBox and BuildFiles. He thinks that with
the help of Shaun and Stefano should be feasible, as CMS has
already promised to help to do the migration of the projects once.
Andreas will organize a meeting with the relevant people soon. The
migration can be done independently for each project up to certain
degree, but needs to end with a complete release of all LCG AA
software. The timescale for doing the migration discussed last
meeting (basically during July) seems to be adequate.
Status of the writing of SMatrix objects using Double32_t
(Rene)
Rene distributed a note before the meeting with the current
understanding of the problem and some proposed solutions. He started by
saying that the problem with using the typedef Double32_t to
indicate
to the I/O subsystem to store doubles as 32 bit numbers only affects
templated classes that have Double32_t as one of the parameters. For
the "normal" usages of this trick there is no problem. He mentioned
that there is a solution if the dictionary for these classes and
classes using them are generated by rootcint within different
compilation units. This would allow to have both instantiations, one
with double and one with Double32_t at the same time in a running
program.
Rene is going to organize a (phone) meeting with the relevant people to
see if the proposed solutions are realistic and can be applied for the
concrete use case of CMS and discuss the technical implications. The
deadline from CMS is June 15th for having a workable solution in
oder to be ready for the production scheduled July 1st. Pere reminded
that the motivation for CMS is to have a gain (almost a factor 2) in
the size of their event data files.
Releases status
New ROOT development release
5.11.06
It was supposed to be released yesterday but there was redone with a
new version number due to last minutes problems. This version
includes massive code changes needed in order to be complied
with the flags that will generate warnings about violations of the
style guidelines from Scott Meyers' More Effective C++ book. It also
includes a new package for multivariate analysis. Detailed release
notes available at http://root.cern.ch/root/Version511.news.html.
This version also fixes some of the CINT warnings observed by CMS and LHCb.
package old version new version
root 5.11.02 5.11.06
sqlite 3.2.8 3.3.5
oracle 10.2.0.1 10.2.0.2
castor n/a 2.1.0-0
CMS is requesting that the classes for which the dictionary is
available in RELAX be listed somewhere in order to avoid to add these
classes in their dictionaries.
ATLAS would like to see in LCGCMT for the
configuration LCG_45 the changes to cope with installation area
for external libraries. David Quarrie and Florence Ranjard will
get together to discuss the details that would make the changes work
for ATLAS and LHCb.
Castor Problems
There was an incident during the deployment of the latest release
of Castor libraries. The old versions of the libraries were removed
during the version upgrade causing old built experiment
executables/libraries not be loadable in nodes where the new version
was installed. This was first detected by ATLAS and caused some
moments of panic. The communication channel to
inform/announce these kind of upgrades was not clear to the
experiment's architects.
The installation of the Castor libraries in the external area,
which will be activated for the first time for configuration LCG_45,
should help in these cases. The upgrade of version in the configuration
(used for newly built software) can only be done when the version
has been fully deployed in all nodes/farms. There was a concern about
the fact that some sites (IN2P3?) are patching the libshift.so library
to make it compatible with their mass storage system.
The -lshift should be removed (with the corresponding changes in
the libraries to be used) for the CERNLIB interface package for CMT to
avoid explicit link of any library using CERNLIB with the Castor
libraries.
Experiments feedback
CMS: They are waiting for the fix/solution for SMatrix with Double32_t
They are also concern about some of the C++ macros in SEAL that do not
accept a semicolon when they are used, producing in some cases a double
semicolon, which gcc 3.5 does not like or an invalid C++ statement
(e.g. semicolon after namespace). In general, macros should not contain
the last ';' and needs to be added when the macros are used. --> Action Lorenzo. He should look into it.
ATLAS: They thanked for the rapid response from the ROOT team on
fixing the dcache problem and producing a new patched release for
version 5/10.00 this week. They have started to deploy their software
based on this new production version. The full validation has been
completed.
They are waiting for LCG_45 to move their development release based on it.
They would like to see the level of verbosity in the POOL catalog reduced. Details to be communicated to the POOL team.
LHCb: There is interest coming from the on line system to
try the Intel compiler.
They claim that a level 30-40% improvement could be achieved with
icc 9.1, thus saving a lot of real money for the online farm if this is
confirmed. This new version of the icc compiler could be installed
in AFS. -->Action Andreas. The compiler should be installed.
There was also mentioned that some of the hardware architecture flags
available in gcc could be tried to see if there is an effect in the
performance. In particular, this could be tried for the packages that
count most in terms of CPU time in their trigger applications.
Platform issues
Andreas announced that CERNLIB version 2005 is now available for all platforms. He reported that in same cases it is needed
to add the compilation flag -fPIC (as it was required for the GENSER libraries).
Topics for future AA meetings [1]
June 14: Reliable database applications and application fail over with POOL/ CORAL.
June 28: ??
July 12: EuroPython highlights and summary
Precompiled headers?
AoB
There was a request from Paolo Calafiura to comment on the new
proposed package for Boost called PQS (physical quantities system). It
seems that this package covers an area similar to the SIUnits package
(developed by Walter Brown/FNAL). The later package was evaluated some
years ago by the experiments and arrived to the conclusion that was
impractical for imposing it consistently across all experimental
software. The general opinion is therefore not very positive on these
initiatives. One example that was given was the treatment of covariance
matrices, for which each element in the matrix could have
different physical dimensions.