This website is no longer maintained. Its content may be obsolete. Please visit http://home.cern/ for current CERN information.
Summary of meeting with Atlas on 10/12/02
Present:
David Adams (phone), Dario Barberis, Daniel Froidevaux, Fabiola Gianotti,
Norman McCubbin (phone), Gilbert Poulard, David Quarry, Marjorie Shapiro (phone), Craig Tull (phone)
Interactive Analysis Environment
Current approach
Physicists produce Paw N-tuple
running simulation/reconstruction programs such as ATRECON and ATLFAST
and analysis them using Paw or Root.
Physicists find Paw and Root satisfactory and have good
experience with both of them.
Root is seen as a natural evolution of Paw, a good
tool to make the transition to the new software environment
and easer for the newcomer.
There is a plan to deploy an interactive version of
Athena/Gaudi which will embed a command-line allowing
both python and root.
A prototype of an interactive application embedding a
python command-line was build in the past.
It allowed access to sub-event objects from the python command-line.
This prototype has never been deployed to the physicists.
Analysis Tool
As said, Atlas physicists seem satisfied by Paw and Root
Besides the old Explorer evaluation,
no evaluation of other analysis tool (from HEP or elsewhere)
has been performed (or at least widely presented) in ATLAS.
ATLAS consider that LCG (and PI in particular) should NOT:
- Develop a new generic analysis tool (paw, root, jas like)
- Develop common reconstruction or particle-level analysis tools
(vertex, 4C fits)
- Develop statistical analysis tool (binned and unbinned
statistics analysis)
- Work on the visualisation and interaction with such tools, the
data and the result associated to them
In general the feeling was expressed that the choice of the actual analysis tool
to be used should be left to the physicist him/herself
similar to how today s(he) is free to choose among web-browsers and
their plug-ins.
The same seems to apply to the desktop environment and other
desktop tools.
Python
ATLAS does not see any problem in using Python as high level scripting
language in the area traditionally covered by shell and perl
scripts.
Although no experience at physicist level exists, using it
to manipulate objects not below the event level seems to be
accepted.
Scepticism still exist on the validity of using python at sub-event level.
AIDA
Athena/Gaudi provides AIDA (version 2) interfaces to histograms and n-tuple
through a Gaudi service.
Such a service uses AIDA-HTL transient representations and is able to
convert to/from native root representations.
It seems that this mechanism is not widely used by the physicists.
No evaluation has been performed of AIDA interfaces: neither for batch
nor for interactive use.
ATLAS has no experience of any sort with AIDA-3.
Interactive Analysis
It was strongly felt that the ability to run the same identical
algorithm (the very same code) in "batch" and
during interactive analysis is of the utmost importance.
The following FORTRAN model was evoked:
Event-data were converted from Zebra banks to a common block.
This common block was used in analysis algorithms.
The very same common block was made persistent as paw n-tuple.
The very same analysis algorithm was run in paw as a "UWFUNC".
Mixed feeling seems to exist about the ability to do the "same"
with the current frameworks:
the intimacy between algorithms and data-structures
that exists in both GAUDI and Root seems to prevent the exchange of
algorithms between the two environments.
It was asked if LCG (and PI in particular) can help
in this respect.
The possibility of exchanging object between GAUDI and Root using
the LCG-dictionary was suggested.
Besides the technical and performance problems, the role of PI
is unclear in this approach: the dictionary is part of SEAL,
the event-object model is proprietary of the experiment.
Event Display
ATLAS does not seem to have, at the moment,
an event display program integrated with their current reconstruction framework.
Geant4 simulation makes use of Geant4 provided event-viewer.
Many prototyped existed in the past: the Atlantis project was
considered particularly important as addressing key issues in the
graphics representation of event in a "dense" environment.
One more prototype used in the past was
Persint: a 3D event display implemented in F90.
None of these prototypes were deployed to the physicists
in a broad and massive way.
Interest was raised to a common project addressing interactive event
display.
Required functionalities include:
- Powerful interactivity with both model (event) and view (graphics
representation)
- dumb viewer of "scene-graph" generated in "batch"
- full "real-space" 3D view
- specialised 3D and 2D views (lego plot, detector-cuts, histograms etc)
- Data Oriented Projections (Hans Drevermann and collaborators work)
No special hardware constrains were raised w.r.t. event display
although the laptop-in-airport-lodge was evoked.
More in general ATLAS is still working on its computing model
and therefore the role and the requirements for the desktop
(tier5) have not been yet fully specified.
In particular no clear position exists on
the role of interactive facilities such as LXPLUS with "remote X-connections"
w.r.t. real desktop computers.
Grid Physicist Interface
ATLAS is involved in several grid projects addressing
the physicist's interface to the grid. In particular
GANGA and Grappa were presented.
The GANGA project (EDG founded, in collaboration with LHCb),
intends to provide an interface for running Gaudi/Athena on the grid.
In this context an evaluation of technologies relevant to a user-grid
interface has been performed (and a document published)
and a first design of Ganga made.
Prototyping work is going on and a first deployment is foreseen for March 2003.
Although currently targeted to Gaudi/Athena it is felt that Ganga
can be used to run any application.
The Grappa project (PPDG/GryPhyN founded):
is aimed to provide a web-based interface to the grid using the VDT paradigm.
A first prototype was demonstrated at SC2002.
Also Grappa is not Gaudi/Athena specific and it is supposed to work
with any application.
ATLAS suggests that Ganga and Grappa be the base
for a common project on a Grid Physicist Interface (essentially subdued by LCG/PI).
Final Comments
It seems that ATLAS does not consider the areas potentially covered
by PI neither as high risk nor as source of competitive advantages.
The item that raised more concerns was the (in)ability to share
identical algorithm-code between batch and interactive
environments.
Possible common projects:
- Event visualisation framework and toolkit
Atlas contribution: Data Oriented Projections
- Grid Physicist Interface
Atlas contribution: Ganga and Grappa
References
OO
Software
Applications including links to Athena, Atlfast, ATRECON
Software for
Physics and Detector simulation including links to Analysis tool
evaluations, Graphics and Event display
Grid
Activities including links to Ganga and Grappa
Vincenzo Innocente
Most recently modified on Mon Jan 13 09:33:37 MET 2003 by Vincenzo Innocente