Speaker
D. Adams
(BNL)
Description
The ATLAS distributed analysis (ADA) system is described. The ATLAS
experiment has more that 2000 physicists from 150 insititutions in
34 countries. Users, data and processing are distributed over these
sites. ADA makes use of a collection of high-level web services
whose interfaces are expressed in terms of AJDL (abstract job
definition language) which includes descriptions of datasets,
transformations and jobs. The high-level services are implemented
using generic parts of these objects while clients and endpoint
applications additionally make use of experiment-specific
extensions. The key high-level service is the analysis service
which receives a generic job request and creates and runs a
corrresponding job, typically as a collection of sub-jobs each
handling a subset of the input dataset. The submitting client is
able to monitor the progress of the job including partial results.
The system is capable of running a wide range of applications but
the emphasis is on event processing, in particular simulation,
reconstruction and analysis of ATLAS data. Other high-level services
include catalogs and dataset splitters and mergers. The ATLAS
production system has been used to construct an analysis service
that makes production activities available to ATLAS users. An
analysis service with interactive response is provided by DIAL.
Another analysis service based on the EGEE middleware is being
constructed in the context of the ARDA project. All are accessible
from ROOT and python command lines and from the user-friendly
graphical interface provided by GANGA.
Primary authors
A. Soroko
(Oxford University)
C. Haeberli
(Bern)
C. Kannan
(Stonybrook)
C.L. Tan
(Birmingham)
D. Adams
(BNL)
D. Liko
(CERN)
F. Fassi
(IFIC Valencia)
F. Orellana
(CERN)
G. Rybkine
J. Fulachier
(LPSC, Grenoble)
J. Lozano
(IFIC Valencia)
K. Harrison
(Cambridge University)
M. Branco
(CERN)
N. Chetan
(Stonybrook)
S. Albrand
(LPSC, Grenoble)
V. Sambamurthy
(Stonybrook)
W. Deng
(BNL)
