Performance Evaluation of a Speculative Network Processor

摘要

Beside high bandwidth, real-time networking applications mainly require short end-to-end latencies. Up to now, network processors are focused in particular on application flexibility and throughput, but generally neglect fundamental new approaches concerning processing delay reduction. This paper describes the implementation and the evaluation of the parallel speculative packet processing architecture and of a reference architecture We show that a latency reduction of up to 26.3 percent can be obtained by the speculative approach in comparison to traditional implementations.