Resource utilization in the ATLAS Data Acquisition System

Data taking with the ATLAS experiment at the Large Hadron Collider at CERN has started. The three-level trigger and data-acquisition system of the experiment is fully functional. In 2009 and 2010 large samples of cosmic ray and collisions data have been and are expected to be collected with it. The smooth operation of the system relies on a tuning made on the basis of test-system measurements and modelling performed prior to installation. It is now possible to compare these predictions with measurements made with the system in active use during data-taking and to extrapolate to performance at higher luminosities. In the system events to be analyzed offline are selected by means of a hardware first-level trigger, receiving input data via dedicated paths, and of two levels of software trigger, implemented on commercially available server computers embedded in the data-acquisition system. Data of events accepted by the first-level trigger are received and buffered in other computers forming together the ReadOut Subsystem (ROS). Within the data-acquisition system event data are transferred via dedicated Gigabit Ethernet links and switches. A distinguishing feature of the second-level trigger is that it, via the data-acquisition network, requests from the ROS for the initial processing step or steps only event data produced in regions of the detector indicated by the first-level trigger. Understanding as well as prediction of the da ta traffic therefore requires detailed knowledge of the actions of the second-level trigger. Relevant information is obtained by gathering of details on trigger processing (e.g. trigger menu item, processing steps, processing time, event data request patterns) for each event during data taking. Similar information is extracted from trigger simulations making use of Monte-Carlo generated event data, allowing to extrapolate to future operation.