The ATLAS TDAQ DataCollection Software

The Large Hadron Collider, which is currently under construction at CERN near Geneva, will collide protons with a center-of-mass energy of 14TeV. This high energy offers the possibility to discover particles with masses on the TeV scale. Bunches of 1.15 10^11 protons will cross at a rate of 40 MHz. 23 proton-proton collisions will happen at every bunch-crossing, which results in a total proton-proton interaction rate of almost one GHz. The biggest part of these interactions do not contain new physics but mostly QCD background. Therefore the detectors to discovery physics, such as ATLAS, need to select the ~100 bunch-crossings with the biggest discovery potential out of the 40 10^6 bunch-crossings per second. In case of the ATLAS experiment this reduction will be achieved on a three level trigger system. The first level trigger runs on custom hardware, the two higher trigger levels run as software algorithms on farms of hundreds of commodity PCs. The second level trigger will run at a rate of up to 100 kHz on a subset of the event data (16 kB in average) and the third level trigger will run at a rate of around 3 kHz on the full event data (1.2 MB in average). The DataCollection subsystem is a part of the data acquisition system and has to provide the event data for the second level trigger (1.6 GB/s), for the third level trigger (3.6 GB/s) and in case of selected events to mass storage ~120 MB/s. The event data is obtained from 144 or 1600 data sources (depending on an open choice on the architecture of the data acquisition system). This thesis explains the infrastructure which is needed to fulfill the demanding task of the DataCollection subsystem, focusing on the software aspects. In addition, performance studies on the DataCollection software and studies on the open choice in the architecture are presented.