Performance of the final data-logging system of the trigger and data acquisition for the ATLAS experiment at CERN

The ATLAS experiment at the Large Hadron Collider at CERN is getting ready to observe collisions between protons at a centre of mass energy of 14 TeV. These will be the highest energy collisions in a controlled environment to-date. The first single beam circulated on the 10th of September 2008, while the first collisions are expected for Spring 2009. The ATLAS Trigger and Data Acquisition (TDAQ) system selects events online in a three level trigger system in order to keep those events promising to unveil new physics at a budgeted rate of ∼200 Hz for an event size of ∼1.5 MB. This paper focuses on the data-logging system on the TDAQ side, the so-called “Sub-Farm Output” (SFO) system. It takes data from the third level trigger, and it streams and indexes the events into different files, according to each event’s trigger path. The data files are moved to CASTOR, the central mass storage facility at CERN. We report first on the functionality and performance requirements on the system and we describe the solution we opted for; to use a small farm of commercial PCs with 3 RAID controllers on each. The data-acquisition software application running on these PCs has been designed to take maximum advantage of the independent RAID controllers, allowing fast I/O. Together with a high speed network, this minimizes the size of the needed farm. Finally we focus on the behaviour of the full data-logging system during the “continuous ATLAS data-taking mode” with cosmics since August 2008 and the first beam events, since performance and reliability will be of equal importance during the physics data-taking period.