计算机与现代化

• 网络与通信 • 上一篇    下一篇

基于GPU的并行报文分类方法

  

  1. 南京航空航天大学计算机科学与技术学院,江苏南京210016
  • 收稿日期:2014-08-26 出版日期:2014-11-27 发布日期:2014-12-10
  • 作者简介:张唯唯(1988-),女,天津人,南京航空航天大学计算机科学与技术学院硕士研究生,研究方向:并行计算,计算机网络; 张玉洁(1991-),女,江苏盐城人,硕士研究生,研究方向:高性能计算,云计算。
  • 基金资助:
     国家863计划资助项目(2009AA044601); 国家自然科学基金重点项目(61139002); 江苏高校优势学科建设工程资助项目; 南京航空航天大学基本科研业务费专项科研项目(NP2013308)

Parallel Packet Classification Method Based on GPU

  1. College of Computer Science and Technology, Nanjing University of Aeronaoutics and Astronautics, Nanjing 210016, China
  • Received:2014-08-26 Online:2014-11-27 Published:2014-12-10

摘要:  报文分类是网络设备的基本处理模式,通常采用报文过滤系统对每个报文进行分类。传统报文分类难以适应当今越来越高的网络流量,分类处理速度低于报文到达网络接口的速度,无法实现实时分析。因此,本文提出使用GPU对大规模报文集进行并行分类的方法,利用GPU的线程级并行处理能力加速报文分类吞吐率,并对其性能及优化方法进行详细分析。实验结果表明,GPU加速的Linear Search和RFC报文分类算法与纯CPU系统执行相比可达到4.4~132.5倍的加速比。

关键词: GPU, CUDA, 报文分类, 并行计算, 优化

Abstract:  Packet classification, which is the basic processing model of network devices, commonly uses packet filtering system to classify each message. Traditional packet classification is difficult to adapt to todays increasingly high network traffic. Its classification processing speed is lower than the speed of packet to reach the network interface, so that it cannot achieve real-time analysis. Therefore, we proposes a method that uses GPU to classify the large scale packet set parallelly. The thread-level parallel processing capability of GPU is made use to accelerate packet classification throughput. Its performance and optimization methods are analyzed in detail. The experimental results show that compared with processing by pure CPU system, GPU-accelerated Linear Search and RFC algorithm achieve a 4.4x~132.5x speedup.

Key words: GPU, CUDA, packet classification, parallel computing, optimization