The ATLAS Read-Out System Performance with first data and perspective for the future

The Readout System (ROS) is the ATLAS DAQ element that receives the data fragments from the ~1600 detector readout links, buffers them and provides them on demand to the second level trigger processor or to the event building system. The ROS system is implemented with ~150 PCs each one housing in average 4 custom-built PCI mezzanine boards (ROBIN) and a 4-port PCIe NIC. Each PC runs a multithreaded OO-software framework managing the requests for data coming through the NIC and collecting the corresponding fragments from the physical buffers. At LHC luminosity of 10^33 cm-2s-1, corresponding to an average Level1 trigger rate of 75 kHz, the ROS has to concurrently service up to approximately 20 kHz of data requests from the Level2 trigger and up to 3.5 kHz of requests from event building nodes. The system has been commissioned in 2007 and since then has been working smoothly. For the most of 2008 the main activity has been data taking with cosmics in which the Level1 trigger rate is much lower with respect to LHC operation. However, the commissioning of the detector requires a much larger data size per event and a very limited rate reduction between Level1 and Level2, pushing the ROS to operate in unforeseen conditions. We will describe here the ROS performances under these conditions, analyze them in view of the 2009 data taking conditions and discuss evolutionary scenarios of the ROS system for improving its flexibility and enh ance its performances, especially in view of the super LHC upgrade.