Towards optimized packet processing for multithreaded network processor

Abstract

With the evolution of the Internet, current routers need to support a variety of emerging network applications while the high packet processing rate is still guaranteed. As a result, the network processor has become a promising solution for network devices due to its computation capability and programming flexibility. However, developing the network applications on network processors is not easy. How to efficiently program multiple processing elements and utilize various memory modules as well as the hardware resources on network processors are always challenges. In this paper, we investigate several optimization issues and programming techniques that should be considered by the developers to achieve higher packet processing rate on network processors. We use an existing packet classification scheme called hierarchical binary prefix search (HBPS) [1] as the benchmark to test and evaluate these optimization techniques. The experiments conducted on Intel IXP2400 network processor show that the overall performance of HBPS can be improved about 42% while these techniques are adopted.