Globally Distributed User Analysis Computing at CDF

Abstract

To maximize the physics potential of the data currently being taken, the CDF collaboration at Fermi National Accelerator Laboratory has started to deploy user analysis computing facilities at several locations throughout the world. At the time of this writing, over 700 individual users are signed up and able to submit physics analysis and simulation applications directly from their desktop or laptop computers to these facilities. These resources consist of a mix of customized computing centers and a decentralized version of our Central Analysis Facility (CAF) initially used at Fermilab, which we have designated Decentralized CDF Analysis Facilities (DCAFs). The goals of this project are to reach a total of 25% of the experiment’s overall computing through off-site resources by the end of 2004, and to expand this to 50% by the end of 2005 through grid-enabling these resources. We report on experience gained during the initial deployment and use of these resources for the summer conference season 2004 that have allowed us to meet the 2004 goal. During this period, we allowed MC generation as well as data analysis of selected data samples at several globally distributed centers. In addition, we discuss our plans for developing a migration path from this first generation distributed computing infrastructure towards a more open implementation that will be interoperable with LCG, OSG and other general-purpose grid installations at the participating sites. INTRODUCTION AND BACKGROUND The CDF collaboration at Fermi National Accelerator Laboratory is currently engaged in analysis of physics data resulting from operation of the Tevatron collider. Like all operating experiments, the experiment faces a computing ∗Alan.Sill@ttu.edu load that includes elements of initial processing of raw data, application of calibration and good run filtering information, splitting into physics data sets by candidate process selection, organizing and cataloging of the results, and further processing to produce ntuples used for detailed analysis by the physicists. In addition, Monte Carlo generation and production must take place in large scale for simulated data sets for each of the physics analyses to develop and refine analysis techniques, check backgrounds and acceptance, model detector response, etc. For CDF, the resulting computing load divides into the following steps in terms of analysis workflow: • Initial production processing, calibration and physics dataset splitting is handled by a step called “production” on a dedicated processing farm at Fermilab. • Monte Carlo generation and detector simulation allow simulated data to be prepared, which are then taken through a similar production processing path. • The resulting data sets are catalogued and made available for processing into physics analysis ntuples by the physics groups. During the past year, CDF has taken steps to move a large portion of the latter two categories of work above, which constitute the majority of our computational needs, into a state such that they can be conducted off site, both for convenience of the physicists and to lessen the load on CDF central computing resources. In addition, the development and widespread current adoption of grid computing methods throughout the world has created opportunities for CDF to position its workload and resources in a way that will enable us to take advantage of such resources in the future. In this document, we discuss the current state of deployment of CDF global computing and the steps that have been required to reach this state. The total computational power of the resources deployed so far represents 35% of the experiment’s capacity in terms of compute cycles, placing us well in the target range of the desired goals. We also map out a short plan for grid-enabling these resources and merging this system with the developing worldwide high energy physics grid in the future.