Graeme Stewart
U Glasgow

The ATLAS PanDA Pilot Job Framework

I will describe and discuss the PanDA pilot job framework (Production and Distributed Analysis
system). PanDA is a pilot job system developed specifically to serve the needs of the ATLAS
experiment in grid computing. I shall describe the historical development of PanDA from its
origins in 2005 in US ATLAS to through its adoption in 2008 as the single unified ATLAS
production framework and its subsequent migration from BNL to CERN. I will discuss the details
of the PanDA architecture and how it benefits over a generic system from intimate knowledge of
the ATLAS data management system. I will also describe how PanDA supports user analysis,
brokering jobs and collating output data from different execution sites. I will spend some time
generally discussing the advantages and disadvantages of pilot (late binding) systems over other
early binding approaches previously tried in ATLAS. Finally, I will review the performance of
PanDA with the arrival of the first LHC data last year, discussing the first reprocessing
campaign, early data analyses on the grid and prospects for the system's future evolution.