ATLAS foresees to produce 1 PB/year of raw data and ~200 TB of ESD
data; while for typical analysis a subset of raw data should be sufficient,
the full ESD sample is supposed to the needed, at least in the initial
phases of the experiment. The computing power required to analyse these
data is estimated in the range 500-1000 SpecInt95 (without taking into
account the Reconstruction and Simulation needs).
The Computing Model adopted by the collaboration is based on distributed
computing, implemented via a hierarchy of
Regional Centres: most basic ideas and building blocks of the model
are common to the LHC experiments and their description can be found in
the documents of the MONARC
project.
An efficient use of distributed computing resources requires however
tools for the management of the jobs, the data, the computing farms, the
mass storage in a integrated way, as well as tools for monitoring the status
of the network and computing facilities in the different sites. These tools
constitute the middleware layer that the GRID projects promise to develop.
Such developments are of crucial importance for enabling ATLAS to efficiently
implement the distributed computing model, which the Collaboration wants
to be functional from the very beginning of the experiment.
The ATLAS experiment is therefore participating in the DataGrid
project with the aim of providing the requirements and the feedback for
the development of the middleware. The model of interaction with the middleware
is based on the idea of iterative cycles and fast prototypes for the various
functionalities needed and requires phases of strong interaction between
the middleware developers, the physicists willing to get results from the
ATLAS application and the people who interface such application with the
GRID tools and infrastructure.
In the spirit of the iterative model mentioned above, the ATLAS applications
best suitable for providing requirements and feedback to the GRID middleware
development will be selected in each phase taking into account both the
status of the middleware and the needs of the ATLAS Physics and Detector
communities for getting useful results from the computing work going on
in the GRID framework. Only computing activities aimed at producing real
results and thus strongly driven by the interested ATLAS communities are
estimated to be useful HEP applications for GRID.
