The analysis of the complex LHC data usually follows a standard path that aims at minimizing not only the amount of data but also the number of observables used. After a number of steps of slimming and skimming the data, the remaining few terabytes of ROOT files hold a selection of the events and a flat structure for the variables needed that can be more easily inspected and traversed in the final stages of the analysis. PROOF arises at this point as an efficient mechanism to distribute the analysis load by taking advantage of all the cores in modern CPUs through PROOF Lite, or by using PROOF Cluster or PROOF on Demand tools to build dynamic PROOF cluster on computing facilities with spare CPUs. However using PROOF at the level required for a serious analysis introduces some difficulties that may scare new adopters. We have developed the PROOF Analysis Framework (PAF) to facilitate the development of new analysis by uniformly exposing the PROOF related configurations across technologies and by taking care of the routine tasks as much as possible. We describe the details of the PAF implementation as well as how we succeeded in engaging a group of CMS physicists to use PAF as their daily analysis framework.

No abstract