Derived Physics Data Production in ATLAS: Experience with Run 1 and Looking Ahead (slides)

While a significant fraction of ATLAS physicists directly analyse the\nAOD (Analysis Object Data) produced at the CERN Tier 0, a much larger\nfraction have opted to analyse data in a flat ROOT format. The large\nscale production of this Derived Physics Data (DPD) format must cater\nfor both detailed performance studies of the ATLAS detector and object\nreconstruction, as well as higher level and generally lighter-content\nphysics analysis. The delay between data-taking and DPD production\nallows for software improvements, while the ease of arbitrarily\ndefined skimming/slimming of this format results in an optimally\nperformant format for end-user analysis.\n\nGiven the diversity of requirements, there are many flavours of DPDs,\nwhich can result in large peak computing resource demands. While the\ncurrent model has proven to be very flexible for the individual groups\nand has successfully met the needs of the collaboration, the resource\nrequirements at the end of Run 1 are much larger than planned. In the\nnear future, ATLAS plans to consolidate DPD production, optimising\nresource usage vs flexibility such that the final analysis format will\nbe more homogeneous across ATLAS while still keeping most of the\nadvantages enjoyed during Run 1.\n\nThe ATLAS Run 1 DPD Production Model is presented along with an\noverview of the resource usage at the end of Run 1, followed by an\noutlook for future plans.