基于可编程硬件平台的高速网络流量非线性采样测量系统

详细信息    查看官网全文

• 英文论文题名：A Programmable Hardware based Non-linear Sampling System for High Speed Traffic Measurement
• 论文作者：武昊 ; 黎阳 ; 张楚文 ; 刘斌
• 年：2016
• 作者机构：清华大学计算机科学与技术系;
• 论文关键词：网络测量 ; 自适应非线性采样 ; 硬件测量系统
• 英文论文关键词：network measurement ; adaptive non-linear sampling ; hardware-based measurement system
• 会议召开时间：2016-06-24
• 会议录名称：2016年全国通信软件学术会议程序册与交流文集
• 语种：中文
• 分类号：TP393.06
• 学会代码：ZGTH
• 会议名称：2016年全国通信软件学术会议
• 会议地点：中国陕西西安
• 主办单位：中国通信学会
• 学会名称：中国通信学会
• 页数：6
• 文件大小：1201k
• 原文格式：O
• 会议级别：全国

摘要

针对测量高速大容量网络链路面临的如何平衡测量速度、测量精度与存储器容量的问题,研究自适应非线性采样测量算法,根据实时流量,动态调整采样概率。在此基础上,利用高速FPGA硬件平台实现针对高速网络流量的测量系统FAST(Fpga-based Adaptive Sampling measuremenT)。通过控制状态机设计、存储器功能划分和并行回导设计,FAST原型系统可在较高的计数器压缩率下,保持99%以上的平均测量精度,分组测量吞吐量达27.4Mpps。
Aiming at the problem that how to trade off the access throughput,the accuracy and the capacity of counter memory in high-speed network measurement,the non-linear sampling measurement algorithms are studied,which adaptively adjust sampling rate based on the real-time flow volume.Based on the idea of non-linear sampling,the network measurement system FAST(Fpga-based Adaptive Sampling measuremenT) is implemented on FPGA platform,aiming at measuring the traffic on high speed links.The key designs of FAST include the centralized state machine,the separate function banks in counter memory and the parallel export operations.Through the theoretical analysis and numerous evaluations,FAST prototype can reach a high counter compression rate with more than 99%average accuracy rate.The throughput is 27.4Mpps.

引文

[1]Internet工程任务组(IETF).实时业务流测量规范.[EB/OL].https://datatracker.ietf.org/wg/rtfm/charter/,2000-10-10.
    [2]RFC-2720,IETF标准备忘录2720.
    [3]开放式网络基金会(ONS).OpenFlow交换机规范v 1.4.1[EB/OL].https://www.opennetworking.org/images/stories/downloads/sdn-resources/onf-specifications/open flow/openflow-switch-vl.4.1.pdf,2015-3-26.
    [4]sFlow.[EB/OL].http://www.sflow.org/packetSamplingBa sics/index.htm,2016-1-29.
    [5]IBM BNT RackSwitch G8264.[EB/OL],https://knovopress.com/tips08157cm_mc_uid=20849938537214557704009&cm_mc_sid_50200000=1458282294,2015-6-9.
    [6]KERR D,BRUINS B.Network flow switching and flow data export:USA,US 6243667[P].2001-06-05.
    [7]WANG H;ZHANG D;SHIN K.Detecting SYN flooding attacks[C]//Infocom 2002.New York:IEEE Press,2002:1530-1539.
    [8]张雪锋,范九伦.基于线性反馈移位寄存器和混沌系统的伪随机序列生成方法.物理学报,2010,59(4):2297.ZHANG Xue-Feng,FAN Jiu-Lun.Pseudo-random sequence generating method based on LFSR and chaotic system[J].Acta Physica Sinica.2010,59(4):2297.
    [9]Altera Corporation.Altera Stratix IV E FPGA DevelopmentKit.[EB/OL].http://www.altera.com/products/devkits/altera/kit-stratix-iv-e.html,2014-11-20.
    [10]HICK P.The CAIDA Anonymized 2013 Internet Traces:equinix-chicago 2013-12-19 14:00 UTC,Direction B.[EB/OL].http://www.caida.org/data/passive/passive 2013dataset.xml,2013-12-19.
    [11]CAIDA.网络链路平均分组长度的统计.[EB/OL].https://www.caida.org/da ta/passive/trace_stats/,2016-2-23.
    [12]QDR SRAM Co-Development Team.SRAM存储芯片规格.[EB/OL].http://www.qdrconsortium.org/qdr-product-fa mily.htm,2016-1-1.
    1根据CAIDA的统计~([11]),OC-192链路的IPv4平均分组长度远大于此。