Accurate Traffic Classification with Multi-threaded Processors

Abstract

Nowadays traffic classification is a fundamental process for Internet traffic management devices and Internet applications need accurate, high performance and scalable traffic classification. Traditional traffic classification is inaccurate and elementary because they are based on imprecise transport layer port method and have unacceptably memory access latency in packet processing. In this paper, we discuss an accurate multi-stage traffic classification in gigabits Internet traffic management systems using multi-threaded processor. Firstly, we address the problem of inaccurate packet classification and analyze payload of applications and standard protocols. Secondly, we present a multi-stage traffic classification using packet header fields and payload string. Finally, we present the software pipeline architecture and hardware design for our approach with network processor. We used our approach to monitor a carrier's backbone node for a month. Compared with traditional methods, the multi-stage traffic classification has 91% accuracy in a real network environment.