摘要
基于深包检测技术的流量分类方法可以达到95%以上的识别率和准确率.然而,由于计算复杂性高、存储消耗大等原因,主流软件方法只能提供百兆(线速率)处理能力,而且不能处理大量流并发的情况.文中提出一个基于深包检测技术的芯片级流量分类架构RocketTC,通过对应用层协议特征、匹配引擎和流管理策略进行优化,使其支持万兆级数据吞吐率.RocketTC具有两个核心模块:基于FPGA的流管理器和动态可重构的分类引擎阵列,前者实现万兆吞吐率下的流表管理,后者快速检测流量特征并支持动态协议特征更新特性.文中提出的分类方法使用轻量级DPI方法,通过缩小检测范围和特征长度进一步减少计算复杂度和存储消耗.我们使用Xilinx Virtex-5FPGA对上述设计进行实现与在线流量测试,结果表明RocketTC可以对92种网络协议进行识别,准确率达到97%,而且稳定提供20Gbps线速处理能力.
Deep packet inspection （DPI） based traffic classification methods could achieve more than 95% accuracy and recognition rate. However, due to considerable computation and storage expenditures, existing software - based solutions could not offer sufficient processing capability for widely deployed high speed networks and massive concurrent flows. This paper proposes Rock- etTC, a high performance FPGA-based architecture for traffic classification with optimized DPI method, flexible and scalable classification engines and flow management scheme. Specifically, RocketTC architecture introduces two key elements to achieve high performance： an efficient flow management scheme using only on-chip BRAMs for storing the flow table, and a parallel and pipelined classification engine array supporting partial dynamic reconfiguration （PDR）. We implemented RocketTC on a Virtex-5 FPGA based platform to evaluate its actual performance. Experimental results show that the prototype could offer a sustained throughput of over 20Gbps and achieve high accuracy above 97% for classifying 92 popular applications while regarding L7-filter as the ground truth. Additionally, it is easy for RocketTC to update for the purpose of classifying more applications.
出处
《计算机学报》
EI
CSCD
北大核心
2014年第2期414-422,共9页
Chinese Journal of Computers
基金
国家自然科学基金(61272510
60803002
61070198
61379145)资助~~
