Trigger monitoring and rate predictions using Enhanced Bias data from the ATLAS Detector at the LHC

A data-driven method for performing offline rate and CPU usage predictions for any algorithmic selection in the ATLAS High Level Trigger at the Large Hadron Collider is described. To assure statistical sensitivity in the most relevant kinematic regions, a mix of events is selected by the Level 1 trigger system that emphasises higher energies and object multiplicities. This sample, referred to as `enhanced bias', is constructed in such a way that the selection bias is removable with event weights. The use of enhanced bias data to calculate the rates of HLT trigger chains along with complex combinations such as group rates, the total rate and unique rates is described, along with methods for performing extrapolations of rates to different instantaneous luminosities and for performing predictions of trigger CPU usage. The process by which ATLAS collects and processes monitoring data in the High Level Trigger is outlined, this allows for CPU and readout system resource utilisation within the trigger to be studied at differing levels of granularity.