David Rousseau, Dirk Duellmann, Fabrizio Furano,
Mato, Benedikt Hegner, Rene Brun, Andrea Valassi, Gabriele
Gowdy, Predrag Buncic, John
Harvey, Liz Sexton-Kennedy [remote], Fons Rademakers, Peter
Kelemen, Marco Cattaneo, Marco
Clemencic, Pete Elmer [remote]
Line Management Issues (John)
Xrootd planing (Dirk)
Proposal for releasing ROOT in parts (Fons)
Report from the LIM meeting (Benedikt)
News from projects
version 2 23/3/2011 @ 17:00)
Management Issues (John)
- Better document the actual contents of a
- Release notes pointers are now there. The database
contents is not correct yet. Keep action open.
- Prepare and present a plan for
releasing ROOT in parts
- Follow-up with OpenLab on the
status of the performance study of the Intel data compression
- Rene contacted Sverre Jarp. Rene will be sitting
with Marc d'Arcy in April to
continue the work and perform some more tests and measurements and to
verify that the library is not buggy. In his mind this still
makes a lot of sense. Created a story from this (ROOT-107). Action
- Install the gcc 4.3.x compiler and use it in
one of the nightly slots
- Not yet done. If the priority would be increased it
will be done earlier.
Provide an answer to the
Savannah feedback document produced by ATLAS
- Not yet done. It would be nice if ATLAS could also
comment about JIRA issue tracker.
Xrootd planning (Dirk)
- This report from Dirk has been triggered from a discussion with
Liz and the need to keep the AF informed.
- The consortium agreement between CERN and SLAC has been
finalized. Few people working
together, monthly meetings, infrastructure being setup (git repository,
schema for the new releases, build infrastructure based on
teamcity, etc.). Being discussed recently the many flavors of XROOT
that exist. Typically clients are picking up what they want, and
therefore there is the need to have a better packaging in place (to be
implemented in a few months).
- Next release of XROOT (3.0.3) will be added at the next ROOT
patch release. Dirk clarified that the client part is always going to
be released with ROOT.
- In the development plans are:
- Looking at the 'read ahead' and 'caching'. The caching
mechanisms are better
in the common client side (i.e. ROOT) and not directly associated to a
given protocol. Asynchronous I/O is implemented at the
level of ROOT. In a recent GDB meeting it was presented the advantages
of this and there was general agreement. There is a patch provided
above ROOT (already included in the development version of ROOT).
- Developing additional optimization in the ROOT side to
cache the I/O data blocks using local disk storage which then can be
re-used between sessions and users.
- Removing specialties of XROOT. The idea is to simplify the
xroot-client by removing some of the code that is currently in the
client with the goal to be
thread safe. Planning this for the end of the year.
- Having a look at the low level part of XROOT I/O with the goal
to make the code more understandable.
- Dirk also informed us about the EOS (xroot disk
- There has been 3 months of production period with ATLAS. The
current CASTOR disk-only pool will migrate to EOS. There are some
extensions and the most important thing is the in-memory namespace
good for writing) based in hash tables. Several copies of the data
kept at different disks in EOS. Automatic replication of files when a
- Currently ramping EOS as an IT 'service'.
- Pere asked if other experiments are involved in the definition
and testing of EOS. Yes, ATLAS and CMS have regular meetings for
the setup and testing of their requested infrastructure. LHCb and Alice
are welcome to join in case there is interest.
- Liz asked about the documentation of the cmsd protocol (xroot
cluster management) which was requested by the d-cache
developers to possibly re-implement that protocol in the d-cache
system. It seems that not much has been done, but Dirk will get
in contact with d-cache people (Patrick) to understand how concrete
this development plan is.
Proposal for releasing ROOT in parts (Fons)
- Fons clarified that the 'parts' that make sense to release
separately are TVMA, RooStat and RooFit. These are the packages that
end-user physicists has most interest to get latest versions while
keeping compatibility with the experiment framework that is lock on a
given version of ROOT.
- The proposal is to build these packages using its own Makefile
(separated from Module.mk) against a released version of ROOT. It
will be needed to make sure the compatibility between the various
versions and eventually back or forward port changes. A number of
combinations of versions of these packages and ROOT base will be
exercised regularly in the 'nightlies'.
- TMVA comes already with its own Makefile. So, what is needed is
to develop similar Makefiles for the other packages.
- The trunk version of these packages should be made compatible
releases of ROOT.
- Pere was concerned about the duplication of effort in oder to
support platforms not habitually used by developers (i.e.
- Fons will discuss with the authors and develop such Makefiles.
- Pete mentioned that at a given moment CMS was interested to
have a separated release of Eve. This is not the case at this
- The proposal was agreed. Fons will contact the developers to
develop the required makefiles and setup the necessary infrastructure.
A new story ROOT-108
has been created.
Planning Preparation (Pere)
- Pere explained that the initial plans for each of the AA projects
has been made available in form of a list
of stories. The idea is to ask the Experiments to have a look at
this list and provide us feedback by next AF (2 weeks).
- Details can be obtained by clicking at each issue.
- If the story is not clear, we can add more information to
explain it better.
- What we are interested to know is whether anything important is
it still missing or whether there is anything that the experiments are
concerned or do not agree.
- Are the priorities right (blocker, critical, high, medium, low)
from your viewpoint?
- The initial number of items to review this time is quite large
and we will not spend a lot of time on each one unless there are
objections. The experiments can either write comments to Pere or if
they which to get an account for SFT JIRA they could do it directly
into the issue.
- Once stories has been discussed (i.e. the experiments had the
opportunity to have their saying) the label 'AF-for-discussion' will be
removed and replaced by a label 'AF-agreed' or similar.
Report from the LIM meeting
- For details see the minutes of last
- Benedikt reported on
- Continuous problems with AFS
- A new configuration LCG_60b was released requested by ATLAS,
which includes a fix of a bug found in ROOT. In addition, there has
been also minor changes in POOL and CORAL.
- Benedikt raised the question of what is to be considered a patch
release. Often when building patch released additional features
and other fixes are included taking the opportunity of the the release.
This was felt not to be a good practice because of the risk of adding
additional delays in the process and potential side effects of the 'new
features'. David stated that ideally if a patch release is requested in
urgency for a given fix, the release should include this fix and
nothing else. In practice this is not very easy since the 'patch
branch' typically contains other fixes of problems detected by other
users. The conclusion was that the contents of the release should be
made very clear and any doubtful fix or additional feature should be
communicated to the requester.
News from Projects
- Released a version of CernVM-FS to Grid sites.
- Started the migration of the
CernVM-FS repositories with the new structure. CMS has been the
first one, and the rest of the experiments will migrate one by one.
Both repositories will be available in parallel for a while, so that
users having previous versions of CernVM-FS will still be able to
access the software.
- Progressing in the setup of CernVM-FS replicas at various
- RAL is already operational.
- BNL had some problems and are being addressed.
- CERN replica started working yesterday.
- Once the replicas are operational the users will be well
shielded from our infrastructure. Therefore, there is more time for
migrating our pilot infrastructure to IT. Predrag estimated that
another two months will be needed for this. Stephen argued that we
cannot call 'production' level with only the replicas being setup since
the pilot infrastructure will still be required for making updates to
the experiments software repositories.
- New version of CernVM is in preparation.
Peter reminded us that we should be looking for the next IT
representative since he is going to step down in a couple of AF