一种高效率的大流提取方法

发布时间：2018-03-25 19:54

本文选题：网络测量　切入点：大流　出处：《计算机研究与发展》2013年04期

【摘要】：随着网络带宽的不断提高,在线识别大流对于拥塞控制、异常检测等网络应用具有重要意义.提出了一种提取大流的算法FEFS(flow extracting with frequency size),能够通过在线识别和淘汰小流,把大流信息保存在有限的高速存储空间中,从而快速提取大流.该算法利用LRU(least recently used)定位更新频率低的流,并进一步用流尺寸因子s和自适应调节因子M标记其中相对较小的流,最后用新到达的流将其替换.FEFS把LRU策略和尺寸因子s相结合,同时考虑了流的近期更新频率和累积报文数量,因此能够准确在线识别大流.LRU策略和尺寸因子都利用了流大小的重尾分布特征,因此FEFS能以很低的存储代价保存和更新大流信息.模拟实验表明,在限定存储条件下,FEFS的平均相对误差率明显低于经典的multi-stage filter算法,而平均报文处理时间也短于multi-stage filter算法.
[Abstract]:With the increasing of network bandwidth, it is very important to identify large flows online for network applications such as congestion control and anomaly detection. An algorithm for extracting large flows, FEFS(flow extracting with frequency size, is proposed, which can identify and eliminate small streams online. Large stream information is stored in limited high speed storage space to extract large stream quickly. The algorithm uses LRU(least recently usedto locate the stream with low update frequency, and further uses flow size factor s and adaptive adjustment factor M to mark the relatively small stream. Finally, the newly arrived stream is replaced by .FEFS to combine the LRU strategy with the size factor s, taking into account the recent update frequency of the stream and the number of cumulative packets. Therefore, the large stream. LRU strategy and size factor can be accurately identified online using the heavy tail distribution feature of the flow size. Therefore, FEFS can save and update the large stream information at a very low storage cost. The simulation results show that, Under limited storage conditions, the average relative error rate of FEFS is obviously lower than that of the classical multi-stage filter algorithm, and the average packet processing time is also shorter than that of multi-stage filter algorithm.
【作者单位】：山东大学计算机科学与技术学院;中国科学院信息工程研究所;
【基金】：国家自然科学基金项目(60803142) 山东省优秀中青年科学家科研奖励基金项目(BS2009DX018) 山东省自然科学基金项目(2009ZRB019ON)
【分类号】：TP393.06

【参考文献】