PROOF on the Cloud for ALICE using PoD and OpenNebula

In order to optimize the use and management of computing centres, their conversion to cloud facilities is becoming increasingly popular. In a medium to large cloud facility, many different virtual clusters may concur for the same resources: unused resources can be freed either by turning off idle virtual machines, or by lowering resources assigned to a virtual machine at runtime. PROOF, a ROOT-based parallel and interactive analysis framework, is officially endorsed in the computing model of the ALICE experiment as complementary to the Grid, and it has become very popular over the last three years. The locality of PROOF-based analysis facilities forces system administrators to scavenge resources, yet the chaotic nature of user analysis tasks deems them unstable and inconstantly used, making PROOF a typical use-case for HPC cloud computing. Currently, PoD dynamically and easily provides a PROOF-enabled cluster by submitting agents to a job scheduler. Unfortunately, a Tier-2 does not comfortably share the same queue between interactive and batch jobs, due to the very large average time to completion of the latter: an elastic cloud approach would enable interactive virtual machines to temporarily subtract resources to the batch ones, without a noticeable impact on them. In this work we describe our setup of a dynamic PROOF-based cloud analysis facility based on PoD and OpenNebula, orchestrated by a simple and lightweight control daemon that makes virtualization transparent for the user.