Extending ATLAS Computing to Commercial Clouds and Supercomputers

The Large Hadron Collider will resume data collection in 2015 with substantially increased computing requirements relative to its first 2009-2013 run. A near doubling of the energy and the data rate, high level of event pile-up, and detector upgrades will mean the number and complexity of events to be analyzed will increase dramatically. A naive extrapolation of the Run 1 experience would suggest that a 5-6 fold increase in computing resources are needed - impossible within the anticipated flat computing budgets in the near future. Consequently ATLAS is engaged in an ambitious program to expand its computing to all available resources, notably including opportunistic use of commercial clouds and supercomputers. Such resources present new challenges in managing heterogeneity, supporting data flows, parallelizing workflows, provisioning software, and other aspects of distributed computing, all while minimizing operational load. We will present the ATLAS experience to date with clouds and supercomputers, and describe efforts underway to automate and scale the utilization of such resources. In particular, we will describe the successful use of these resources through the ATLAS workload management system, PanDA, and future development plans.