Abstract submitted to the Second UK e-Science All-Hands Meeting,
Nottingham, UK, 2-4 September 2003.

Ganga: a user-Grid interface for ATLAS and LHCb

N. Brook (Bristol), K. Harrison (Cambridge), R.W.L. Jones (Lancaster),
W. Lavrijsen (LBNL), P. Mato (CERN), A. Soroko (Oxford), C.L. Tan (Birmingham),
C.E. Tull (LBNL)

Keywords: Grid portal, Component architecture, Job management,
Distributed analysis, High-energy physics applications

Ganga is a user interface that gives access both to local resources and to the
Grid, and provides job-configuration and data-management tools matched to the
computing environments of the particle-physics experiments ATLAS and
LHCb. These experiments will study the products of high-energy proton-proton
interactions at the Large Hadron Collider currently under construction at CERN,
Geneva, with startup scheduled for 2007.  Both experiments involve many
hundreds of physicists, from tens of institutes, and will require analysis of
data volumes of the order of petabytes per year.  The experimental data,
recorded at CERN, will be shared between sites for processing.  Data for the
large samples of simulated interactions needed for a full understanding of
detector behaviour and of the physics effects of interest will also be
distributed between multiple locations. Ganga will exploit Grid services to
allow the physicist user transparent access to the globally distributed
datasets, and to the decentralised computing resources available to each
experiment.

Ganga is based on a component architecture, with components interacting via a
software bus.  The implementation is in python.  Components of general
applicability deal with tasks such as workflow definition, script generation,
job submission, file transfer to and from remote sites, and monitoring.  This
core functionality can be supplemented as needed by components tailored to meet
the requirements of specific groups of users, so that Ganga is readily
extensible.  Specialised components for ATLAS and LHCb incorporate knowledge of
the experiments' common software framework and configuration-management system,
and take care of tasks such as application configuration, job splitting, the
setting-up of the run-time environment, and output collection.  The software
bus builds on the functionality of the python interpreter, adding to the
possibilities for connecting, removing and configuring components.

A first version of Ganga has been developed, and has been used successfully to
run and monitor jobs on the European DataGrid and on local batch systems.  All
functionality can be accessed either from the command line, for example in
scripts, or from a graphical interface that simplifies use of the available
tools.  To make it feel immediately familiar to users, the graphical interface
is modelled on the interface to a mail system, and many of the operations
performed using a mail client find analogues in Ganga.  From the user's point
of view, submitting a job to the Grid becomes as easy as sending a mail
message.

In this presentation, we give details of the Ganga design and implementation,
we report on the first use of Ganga in ATLAS and LHCb, and we outline plans
for future improvements.