The real time event selection system for the ATLAS experiment

In the paper we present tasks of the ATLAS event filter (trigger) system and proposed implementation, which uses commodity computers and network components. The system should find events containing interesting interactions, and select not more than a hundred in every second out of a billion generated by an accelerator. To achieve this reduction rate, the system was split into 3 levels: the first one implemented in hardware, and the higher ones as a farm of commodity PC-based processors and a computer network in Ethernet technology. Using the information from the first level, which describes areas in the detector where an interesting interaction happened, allows for loosening the requirements for network throughput and computing power. It is estimated that required network throughput would reach 6 GB/s, whilst the second and third level of the trigger system will use approximately 500 processors with 4 GHz clock speed. Because of the size and costs of the project, building a full-scale prototype is not feasible. Estimations for the throughput and decision latency are based on results of computer modeling of the system. The system presented in the paper may serve as an exemplary implementation of a trigger system for high energy physics experiments on huge particle accelerators. (3 refs).