meeting 14th April
Nils Hoimyr (CVS Service), Pere
Cosmo, Antonino Sergi,
Andrea Valassi, Predrag Buncic, Stephen
Gowdy [REMOTE], Marco Clemencic, David Rousseau, Peter
Rademakers, Rene Brun, Helge Meinhard, Marco Cattaneo, Vincenzo
Innocente, Marco Cattaneo, Liz Sexton-Kennedy [REMOTE], Pete
Line Management Issues (John)
Migration of the LCGCVS service (Nils Hoimyr)
NeuroBayes license (Pere)
Report from the LIM meeting (Benedikt)
News from projects
version 1 21/4/2011 @ 11:00)
- Nils Hoimyr and Helge Meinhard are joining us for the
CVS migration agenda item.
- Helge Meinhard has been added to the AF mailing list as IT-PES
- 3 Technical Students requests for the SFT group have been put
- Install the gcc 4.3.x
compiler and use it in
one of the nightly slots
- The installation of gcc 4.3.5 is fine for LHCb and
ATLAS. Unfortunately the problem is that the distcc
cluster does not know about it and it cannot be used in real
production. There is now a open ticket in SNow after the initial
difficulty on how to submit a ticket when the service is not known to
- Helge clarified that the IT-PES group is committed
the distcc service.
- Action will be closed.
- SPI-33 Provide
an answer to the
Savannah feedback document produced by ATLAS
- A reply on a reply has been provided. Some of the
requested features could be
done in the current Savannah. Working on a general questionnaire.
Benedikt will report at next meeting.
- New IT contact person replacing Peter Kelemen (Pere)
- There is a proposal from IT on the IT-link person
and this will be finalized soon.
- Date for the next multi-core and virtualization
Migration of the LCGCVS service (Nils Hoimyr)
- The migration plans for the LCGCVS service to plain CVS (AFS
based) were presented by Nills. See
the slides. The bottom line is that the migration for the
repositories CMSSW, COOL, LCGTEST, PF is scheduled for Monday 18th of
April at 9:30.
- Vincenzo requested to make sure that the 'old' service is not
accessible anymore after the migration to avoid losses of code.
- Pere asked if performance measurements has been made recently
with the actual implementation of AFS as storage. The CVS service team
did checkout tests of the CMS repository (5 simultaneous checkouts) and
the checkout time gets about 50% longer for the complete CMSSW
repository (~90k files). This is acceptable for CMS. No timing in
tagging was performed (see
the details of the tests).
- The answer was no, but CMS did some and their conclusion was that
performance was adequate.
- Pere stated that the intention of a general license agreement
between CERN and Physics Information Technologies GmbH (<phy-t>)
concerning NeuroBayes software for scientific use is the simplification
of the paper work for the license agreements with individual
experiments/groups/universities. The initiative comes exclusively from
<phy-t>. The software is free for CERN but obviously is not 'open
source'. He asked the experiments to have a look at the agreement and
report any major concerns that they may have. The experiment
situation is as follows:
- ATLAS: There is indeed a license agreement between ATLAS and
<phy-t>, which is very similar to the CERN one. This was done on
request of one of the groups who has paid some licenses for
"NB-teacher". Both "NB-teacher" and "NB-Expert" are available on
itself does not pay anything. In practice, NeuroBayes is used in one
ATLAS analysis. So, for ATLAS this license is fine.
- CMS: [added after the meeting] Currently CMS don't have
analyses based on NeuroBayes. For the future, CMS physics coordination
will always advice to use "open source" code and in case a
NeuroBayes-based analysis will be presented they will ask
detailed cross-check studies (ie. using other multi-variate
techniques). In addition, NeuroBayes will not be used for
reconstruction code, based on CMS policy about code that can be used in
CMSSW, (i.e. source code should be available in order to be used in our
reconstruction software.) So, for CMS this license is fine but they can
live without it.
- ALICE: ALICE is not interested in this software and would
discourage any CERN involvement.
- LHCb: Assuming that the terms and conditions for the use of the
software are what was discussed one year ago then it is OK for LHCb.
happy that somebody signs such agreement in their behalf. LHCb
bought into this software since it is used in some of the the HLT lines.
- The conclusion was that we will proceed with the signature.
Planning Preparation (Pere)
- No feedback to the current 70 'stories' under discussion was not
provided before the meeting.
- CMS re-stated their interest to be added as 'customers' to some
of the 'stories' in the list. In particular the ones concerning the
ROOT I/O optimization, the work in the compression algorithms, LLVM
development, etc. Pere will make the necessary changes and add CMS on
- It was agreed that Pere will mark these initial stories as
'AF-agreed' and they will be monitored from now on. In addition, new
created ones will be added for discussion in subsequent meetings.
- There are no stories for the Persistency Framework project yet.
Pere asked for the time scale and it was agreed that they will be there
fore the next AF meeting.
Report from the LIM meeting
- Discussed the future handling of LCG releases in CVMFS and a
few technical implications/requirements. The outcome of the discussion
that SPI will 'mirror' the current structure of the AFS installation
every newly installed release. What is done for already installed
releases on CVMFS is currently under discussion on the LIM mailing
list. It was discussed also how to optimize the file
catalogues. After further
discussion with Predrag and Jakob, they will do per-version catalogue
splits for the biggest 10% of software packages and see how that
performs. In the meantime it has been agreed to leave what is in CVMFS
already as is.
- For what concerns the gcc435, they are waiting for feedback about
distcc service by IT. Once that is clear they do the switch.
News from Projects
- CernVM (Predrag)
- Version 2.3.0 was released last week. No major issues.
- Last week was full of meetings in which CernVM-FS was discussed
(GDB, ATLAS software week, etc.) Overall it was very positive outlook.
Likelihood that the CernVM-FS will be deployed to all Grid sites.
- Concerning the migration of the infrastructure to IT: The
so-called stratum 1 servers are in
place at RAL and CERN, BNL is under testing.
- Testing LXCLOUD infrastructure at CERN with OpenNebula (Amazon
API). The first impression is very encouraging. CernVM instances
can be contextualized to perform specific functions. Starting a test
- CernVM in Google Summer of Code. 48 candidate applications has
been received. 4 slots has been allocated Google. Very
good students. The goal is a final consolidation of both CernVM and FS.
- Multi-core (Vincenzo)
- Testing 24 core machines.
- May 6th Hepix meeting. Exploration of new ways of running
- Openlab/Intel Spring 2011 Multi-threading and Parallelism
Workshop in May.
- Seminar on gcc by end of May.
- Whole-node job submission Task Force (Pete)
- ATLAS is currently polishing the details and
validating their parallel applications
- LHCb can run in parallel some of the jobs. Validating the
output for physics has not really started.
- At least two of the experiments seems to be in well advance
state. Other discussions such as accounting could be started in
- There will be a status report at the next LCG Management Board
- Working on some changes in
Oracle access to protect against network glitches.
- Resuming some work on POOL needed by ATLAS.
- Fixing Coverity issues.
- Starting to attack the Coverity defect reports in an organized
manner (about 20k reports). Most of
them are real but not very likely.
- Starting a major production to reconstruct 10**9 events.
- Release based on LCG_60b is ready but not validated in a
- Testing gcc 4.6.
- Release based on ROOT 5.27.00b is there. The one based on ROOT
5.28.00 is not yet there but will come.