Data handling and processing for the ATLAS experiment

The ATLAS experiment is taking data steadily since Autumn 2009, collecting so far over 2.5 fm-1 of data (several petabytes of raw and reconstructed data per year of data-taking). Data are calibrated, reconstructed, distributed and analysed at over 100 different sites using the World-wide LHC Computing Grid and the tools produced by the ATLAS Distributed Computing project. This paper reports on the experience of setting up and operating this distributed computing infrastructure with real data and in real time, on the evolution of the computing model driven by this experience, and on the system performance during the first two years of operation.