一种大象流两级识别方法

A two-level method for elephant flow identification

基于大象流的识别准确度高且开销低,对于解决SDN流量管理过程中控制器单点故障问题具有重要意义。针对现有大象流识别方法识别开销大的问题,提出一种大象流两级识别方法。该方法在第一阶段提出基于TCP发送队列的可疑大象流识别算法,在第二阶段提出基于流持续时间的真实大象流识别算法;第一阶段是在端系统中识别可疑大象流,用于降低第二阶段真实大象流识别过程中SDN控制器所需监测的网络流数量。实验分析表明,在保证大象流识别的高准确度前提下,大象流两级识别方法较基于采样的大象流识别方法可以降低约85%的控制器识别开销。

The high accuracy and low overhead of elephant flows identification have a great meaning on solving the controller＇s single point of failure problem in SDN traffic management. Aiming at the problem of high overhead of the existing elephant flows identification method, a two-level method for elephant flows identification was proposed which included a suspicious elephant detection algorithm based on TCP write-queue in the first stage and areal elephant detection algorithm based on flow duration in the second stage. During the first stage, the suspicious elephant flows were identified in the end systems to reduce the amount of flows monitored by the SDN controller at the second stage. Analysis and simulation prove that,under the premise of ensuring the accuracy of elephant flow identification, the two-level method for elephant flows identification reduces about 85% overhead of identification compared with sampling identification method.