Benchmarks and implementation of the ALICE High Level Trigger

The ALICE High Level Trigger combines and processes the full information from all major detectors in a large computer cluster. Data rate reduction is achieved by reducing the event rate by selecting interesting events (software trigger) and by reducing the event size by selecting sub-events and by advanced data compression. Reconstruction chains for the barrel detectors and the forward muon spectrometer have been benchmarked. The HLT receives a replica of the raw data via the standard ALICE DDL link into a custom PCI receiver card (HLT-RORC). These boards also provide a FPGA co-processor for data-intensive tasks of pattern recognition. Some of the pattern recognition algorithms (cluster finder, Hough transformation) have been re-designed in VHDL to be executed in the Virtex-4 FPGA on the HLT-RORC. HLT prototypes were operated during the beam tests of the TPC and TRD detectors. The input and output interfaces to DAQ and the data flow inside of HLT were successfully tested. A full-scale prototype of the dimuon-HLT achieved the expected data flow performance. This system was finally embedded in a GRID-like system of several distributed clusters demonstrating the scalability and fault-tolerance of the HLT. Index Terms--Data acquisition, data processing, distributed computing, real time systems, software fault tolerance, triggering.