The ATLAS Trigger System

Bunches of protons in the LHC cross at the centre of the ATLAS detector every 25ns or at a rate of 40 MHz. At LHC design luminosity, an average of 22 p-p collisions will occur per bunch crossing. Each interaction produces numerous secondary particles which will be "photographed"/detected in the ATLAS detector. Each "photograph" or event has a size of about 100 Megabytes this giving some 4 - 6 PetaBytes per second (or approximately 1 million DVDs per second). Clearly such a vast amount of data per second can be neither stored in a permanent manner nor analysed. In addition, most of the data is not interesting or pertinent to the physics goals of ATLAS.
A multi-level online computing system (trigger system) has been developed and installed to reject unwanted events and increase as much as possible the quality of selected events appropriate for the ATLAS physics programme. The first level (LVL1) of the system is built in custom electronics based on FPGAs and ASICs, and analyses data from the Calorimeter and Muon detectors in a time latency of less than 2.5 microseconds. It analyses the event topology and identifies regions of interest (RoIs) in the detector where activity has been seen. Analysing these regions, it takes a decision to accept or not the event. The event accept rate from LVL1 is 75 kHz. The second trigger level (LVL2) comprises standard PCs containing Intel based dual-quad core processors. The LVL2 receives the RoI addresses for accepted events from LVL1 and uses them to access event data in the RoI regions which is stored in several of a total of 1600 holding buffers. It then analyses these data in a typical time of a few milliseconds and decides to accept or not the event. The event accept rate from LVL2 is ~3kHz. Detector data from events accepted by LVL2 are assembled together in a complete data record having a typical size of 1.5 MB and then passed to the final trigger level, the Event Filter (EF), also comprising standard PCs containing Intel based dual-quad core processors. The EF analyses the complete events using algorithms almost identical to those used in the final offline data analysis. Events accepted by the EF are then sent to permanent storage in the CERN Computer Centre. Detailed analysis of these daya are then performed both at CERN and at major computing centres around the world connected to the LHC Computing Grid. The LVL2 and EF systems together constitute a farm of ~2300 PCs, installed at ground level near to the ATLAS control room.
CPPM has been instrumental in the design of the software infrastructure and monitoring system for the EF, coordinating the overall design, implementation and testing and authored the majority of the software.
Introductions to the Data Acquisition, Trigger and Data Analysis can be found on the detector web pages and more detailed information can be found on the ATLAS TDAQ twiki pages.