The ATLAS trigger: high-level trigger commissioning and operation during early data taking

The ATLAS experiment is one of the two general-purpose experiments due to start operation soon at the Large Hadron Collider (LHC). The LHC will collide protons at a centre of mass energy of 14~TeV, with a bunch-crossing rate of 40~MHz. The ATLAS three-level trigger will reduce this input rate to match the foreseen offline storage capability of 100-200~Hz. After the Level 1 trigger, which is implemented in custom hardware, the High-Level Trigger (HLT) further reduces the rate from up to 100~kHz to the offline storage rate while retaining the most interesting physics. The HLT is implemented in software running in commercially available computer farms and consists of Level 2 and Event Filter. To reduce the network data traffic and the processing time to manageable levels, the HLT uses seeded, step-wise reconstruction, aiming at the earliest possible rejection. Data produced during LHC commissioning will be vital for calibrating and aligning sub-detectors, as well as for testing the ATLAS trigger and setting up the online event selection (or trigger menu). This will be done initially using zero-bias and minimum-bias collisions for system calibration and checks. As the LHC luminosity increases, the trigger menu will need to adapt so that the available output storage rate is used optimally to maximize the ATLAS physics potential. In preparation for LHC collision data, the functionality of the data-acquisition system and the HLT software was demonstrated using part of the readout chain to supply simulated events in raw-data format to a subset of the HLT farm. These tests included a complete set of physics triggers as well as configuration and monitoring. We give an overview of the ATLAS High Level Trigger focusing on the system design and its innovative features. We then present the ATLAS trigger strategy for the initial phase of LHC exploitation, up to a luminosity of $10^{31}$ s$^{-1}$cm$^{-2}$. Emphasis will be given to the full trigger menus, including physics and calibration triggers. Finally, we report on the valuable experience acquired through in-situ commissioning of the system where simulated events were used to exercise the trigger chain. In particular we show critical quantities such as event processing times, measured in a large-scale HLT farm using a complex trigger menu.