Commissioning of real time algorithms in the ATLAS tau trigger system

The ATLAS tau trigger system runs very challenging real time algorithms on commodity computers (2 x Intel Harpertown quad-core 2.5 Ghz). Whilst in the second level trigger (L2) fast and specialized algorithms are used, in the third level trigger (Event Filter -EF-) sophisticated and detailed reconstruction algorithms run. For both types of algorithms, data from the whole detector can be used, and in fact there are dedicated separate algorithms processing the calorimeter data and the data from the tracking detectors.In this contribution we focus on the online performance of the L2 and EF algorithms during data taking period 2009 and 2010, showing our current understanding of the performance of such algorithms in real data. We present the overall performance and robustness of the calorimeter and tracking online reconstruction in the tau trigger during different data taking conditions, and first performance studies. In particular, a detailed comparison to Monte Carlo simulation is presented, and our first steps towards an optimization of the tau trigger in real data.