该文分析了hardflow算法抽样率与大流抽样损失率的关系,提出两个改进算法以降低大流抽样损失率与抽样率的比值rd,并讨论了缓冲区大小、测量周期对相对误差的影响以及参数的选取。结果表明:与hardflow相比,在低抽样率(4.5%),两种改进算法的rd比hardflow分别降低1.4和6.6,当抽样率提高至一定水平时(〉11.3%),则3种算法的rd均趋于0。
This paper analyzes the relationship between the Sampling Rate (SR) and Sampling Loss Rate of Elephant Flows (SLREF) of hardflow. Then two improved algorithms are proposed to reduce the ratio of SLREF to SR( rd ). The relative error and the parameters setting are discussed with various sampling buffer sizes and various measurement periods. Compared to the hardflow, the results show that rd of the two improved algorithms decrease 1.4 and 6.6 respectively when SR is low (e.g. 4.5%). Furthermore, all rd incline to zero when SR is large (e.g. 〉11.3%).