Use ATLAS@home to exploit extra CPU from busy grid sites

Grid computing provides the main resource for data processing of High Energy Physics experiments, and a study we conducted shows that typical grid sites used by ATLAS computing are not fully utilized with the regular workloads. In order to increase the CPU utilization of these grid sites, we use ATLAS@Home jobs as back ll. The results show an extra 15% to 42% CPU can be exploited by the backfilling jobs while the grid sites are filled with regular workloads and the overall CPU utilization can remain over 90% for the sites. We also measure the impact on the grid jobs and the quality of exploited cputime in the backfi lling model. The measurements prove that the backfi lling model has no impact on the grid jobs regarding the failure rate of the grid jobs, and the impact on the CPU efficiency of grid jobs varies from 1% to 11% depending on the con figuration of the site. This is mainly linked to the memory usage of the grid jobs and the resource allocation policy of the site, and this is tunable from both the ATLAS@home and site's resource allocation policy. In addition the throughput of back ll jobs in terms of cputime per event simulated is the same as for resources dedicated to ATLAS@Home. We believe this approach is generic enough to be extended to other clusters.