Triggering on 7 TeV collisions with the ATLAS high level trigger

In 2010 ATLAS has seen the first proton-proton collisions at 7 TeV. This year a collision rate of nearly 10 MHz has been achieved. Events of potential interest for physics analysis are selected by a three-level trigger system, with a final recording rate of about 200 Hz. The first level (LI) is implemented in customized hardware, the two levels of the high level trigger (HLT) are software triggers. The selection is described by the Trigger Configuration in the form of menus, each of which contains about 400 signatures. Each signature corresponds to a chain of algorithms which reconstruct and refine specific event features. The HLT Steering receives information from the Configuration system, dynamically creates chains and controls the execution of algorithms and flow of information during event processing. The Steering tests each signature on LI-accepted events, and those satisfying one or more test are recorded for later analysis. To save execution time, the Steering has a facility to cache results, avoiding later recalculation. To control rate, prescale factors can be applied to LI or HLT signatures. Where needed for later analysis, the Steering has a test-after-accept functionality to provide the results of the tests for prescaled signatures. In order to maintain a high selection efficiency it is essential that the trigger can be dynamically re-configured in response to changes in the detector or machine conditions, such as the status of detector readout elements, instantaneous LHC luminosity and beam-spot position. This relies on techniques that allow configuration changes, such as LI and HLT prescale updates, to be made during a run without disrupting data taking, while ensuring a consistent and reproducible configuration across the entire HLT farm. We present the performance of the trigger system during collisions, showing the evolution with LHC luminosity and briefly describe the expectations for the next phase of LHC exploitation.