Event streaming in the online system

The Large Hadron Collider (LHC), currently in operation at CERN in Geneva, is a circular 27-kilometer-circumference machine, accelerating bunches of protons in opposite directions. The bunches will cross at four different interaction points with a bunch-crossing frequency of 40MHz. ATLAS, the largest LHC experiment, registers the signals induced by particles traversing the detector components on each bunch crossing. When this happens a total of around 1.5MB of data are collected. This results in a data rate of around 60 TB/s flowing out of the detector. Note that the available event storage space is limited to about 6 PB per year. With an operational period of about 20 million seconds per year, this requires a data reduction factor of 200:000 in the trigger and data acquisition (TDAQ) system. Events included in the recording rate budget are already subdivided and organized by ATLAS during data acquisition. So, the TDAQ system does not only take care of data reduction, but also organizes the collected events. This procedure is known as "event streaming", which is defined as 'the splitting of raw data into logical entities by the online system based on decisions taken by the trigger'. We describe the event streaming functionality as implemented in the ATLAS online system. Use cases, requirements and implementation of the main features are discussed. Furthermore, we present the commissioning and monitoring activities related to ev ent streaming. Finally, we report on streaming strategies and first experiences with single beam and first collisions.