ATLAS Distributed Computing on the way to the automatic site exclusion

This paper details the different aspects of ATLAS Distributed Computing experience after the first 1.5 years of LHC data taking. We describe the performance of the ATLAS Distributed Computing system and the lessons learned during the 2010 - 2011 runs, pointing out parts of the system which were in good shape, and also spotting areas which required improvements. We discuss improvements of the ATLAS Tier-0 computing resources, and study data access patterns for Grid analysis to improve the global processing rate. We present recent updates in the ATLAS data distribution model.