Grid Infrastructure and Large Scale Production for ATLAS

In preparation for the startup of the LHC collider at CERN, the ATLAS collaboration is performing computing challenges of increasing complexity and scale. Monte Carlo production is ongoing with some 10,000 concurrent jobs worldwide, including at Canadian collaborating institutes. TRIUMF hosts one of the 10 ATLAS 'Tier-1' data and computing centres, where 270TB of data per year will be transferred, by high performance dedicated networks, processed and redistributed for analysis. We discuss recent large scale tests of all aspects of middleware, hardware infrastructure and procedures to support the 24x7 operation of the Tier-1 centre, required for LHC running.