摘要： 在网络大流识别中,突发的大量小流会影响大流识别准确度。为此,结合基于频率和大小的流提取(FEFS)算法与基于计数型布鲁姆过滤器(CBF)算法,提出一种新的大流识别算法,即FEFSCBF算法。该算法采用三级存储结构,运用CBF结构存储小流,将达到过滤阈值的流移至筛选区(LRU)中,当LRU满载时,使用FEFS机制选择一个符合条件的流淘汰,并及时隔离大流。仿真结果表明,该算法的误报率和漏报率均较低,存储开销较小,可以运用于高速网络链路的大流识别中。

关键词: 大流, 识别算法, 频率和大小的流提取, 计数型布鲁姆过滤器, 高速网络

Abstract: Aiming at the impact of elephant flows identification accuracy caused by a sudden arrival of mice flows,based on Flow Extracting with Frequence & Size(FEFS) and Count Bloom Filter(CBF),and a new algorithm FEFSCBF is proposed.It uses threelevel structure which is EFS list,Least Recently Used(LRU) list and CBF array.FEFSCBF uses CBF to store mice flows,and a flow is moved to LRU list if its count is greater than filterthreshold.In LRU list,it uses FEFS replacement scheme to select an eligible flow to drop when LRU is full.And it can separate elephant flows to EFS list to avoid false positive as far as possible.Simulation experimental results show that FEFSCBF has lower false positive ratio and false negative ratio,and it costs lower storage.It can be used to identify elephant flows in highspeed network.

Key words: elephant flows, identification algorithm, Flow Extracting with Frequence & Size(FEFS), Count Bloom Filter(CBF), highspeed network

中图分类号: 

• TP393