一种面向不均衡网络流的综合抽样方法  

An Integrated Sampling Method Over Imbalanced Network Flows

作　　者：李景富[1] 杨志强[1] 

机构地区：[1]黄淮学院,河南驻马店463000

出　　处：《火力与指挥控制》2015年第12期74-79,共6页Fire Control & Command Control

基　　金：河南省科技攻关计划基金(122102210510);河南省教育厅科学技术研究重点基金资助项目(13A520786)

摘　　要：针对互联网流量中短流数量多但承载信息少、长流数量少但承载报文数多的特点,提出了一种面向不均衡网络流的综合抽样IS(Integrated Sampling)方法。IS方法首先采用容量固定的高速缓存实现有限时间窗口内报文的实时归并,在此基础上,IS方法采用可部分重构的哈希函数实现单报文流聚类,采用流长和时间组合赋权的权值更新模块和频繁项模块共同实现频繁项流的抽取,对于网络中的其他流量,IS方法通过多个蓄水池模块实现蓄水池分段抽样。实验证明,相对于单一的抽样方法,IS方法在相同缓存下能够抽取出更为丰富地网络流信息,算法能够实时应用于高速网络中。Aiming at the property of huge quantity and little information of small flows and small quantity and high payload of big flows,an Integrated Sampling（IS） method over imbalanced network flows is proposed. The fixed size of high-speed memory is used to aggregate packets into flows within limited time windows on line in IS algorithm first. Then,the partially reconstructed hash functions are used to cluster single packet flows,and the weight-updating module in which the flow weight is assigned by length and time together and the frequent-item module are cooperated to mine frequent items. Lastly,the multi-reservoir modules are introduced to sample flows in a stratified way for the rest flows. The experiment on real traffic shows that IS is capable of sampling more abundant information of flows comparing with other single sampling method at the same memory cost and it can be applied in the high backbone network on the wire.

关 键 词：网络流 流抽样 哈希聚类 频繁项提取 蓄水池抽样 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]