High bandwidth concurrent processing on commodity platforms

The I/O bandwidth and real-time processing power required for high-energy physics experiments is increasing rapidly over time. The current requirements can only be met by using large-scale concurrent processing. We are investigating the use of a large PC cluster interconnected by Fast and Gigabit Ethernet to meet the performance requirements of the ATLAS second level trigger. This architecture is attractive because of its performance and competitive pricing. A major problem is obtaining frequent high-bandwidth I/O without sacrificing the CPU´s processing power. We present a tight integration of a user-level scheduler and a zero-copy communication layer. This system closely approaches the performance of the underlying hardware in terms of both CPU power and I/O capacity