Adaptive Packet Classification Algorithm Based on IXP2800 Network Processor  

作　　者：CHEN Zheng ZHANG Zhenhua YU Nenghai TANG Xinan 

机构地区：[1]Department of Electronic Engineering and Information Science, University of Science and Technology of China, Hefei 230027, China [2]Intel Corporationt

出　　处：《Chinese Journal of Electronics》2008年第3期529-533,共5页电子学报（英文版）

摘　　要：Packet classification is a critical data-plane task for modern routers to support value-added services, especially for those requiring QoS and flow based processing. However, classification at 10Gbps or higher using an algorithmic approach is still challenging. New generation of Network processor unit （NPU） provides unprecedented processing power for network applications, and it opens a new venture to explore thread-level parallelism for attacking networking performance bottlenecks. This paper studies the implementation issues of how an adaptive clas- sification algorithm can be efflciently implemented on a multi-core and multithreaded NPU architecture. Our algorithm combines best traits of Recursive flow classification （RFC） algorithm and bitmap compression technique to achieve deterministic classification performance while keeping the memory growth checked. When mapping such an algorithm onto the Intel IXP network processor, we consider the characteristic of IXP architecture early in the algorithm-design phase to eliminate the potential perfor- mance bottlenecks. The implemented algorithm is highly efflcient and it can run at 10Gbps speed or higher on a real IXP2800 chip.

关 键 词：Packet classification Network processor Recursive flow classification （RFC） 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]