一种用于小流估计的数据包公平抽样算法

现有数据包公平抽样算法通常根据到达数据包所属流大小的估计值设置包抽样率,令大流所含数据包抽样率低,小流所含数据包抽样率高,缺点是算法的优劣依赖于不同方法对流大小估计的准确性;小流估计误差较大。针对此问题,利用大流持续时间长且到达速率高的特点,提出一种基于时间分片的用于小流估计的数据包公平抽样算法(MFEPS)。该算法将测量时间分割成片,抽取每个流在每个时间片内的第一个数据包,而不需要估计数据包所属流的大小。理论分析和实验结果均表明,与已有算法相比,对于小流估计,MFEPS算法在相同的CPU资源消耗条件下,具有更高的准确性和良好的扩展性。

A Fair Packet Sampling Algorithm for Mice Flow Estimation

In most existing fair packet sampling algorithms, the sampling probability is usually set according to the estimation of the size of flow which the arriving packet belongs to, so the accuracy of the algorithm depends on the accuracy of the method to estimate the size of the flow and existing algorithms have a high estimation error for mice flow. To solve this problem, a new fair packet sampling algorithm which is based on time sectioning and used to estimate mice flow is proposed according to the characteristic that elephant flow has a high arrival rate and long alive time. The algorithm samples the first packet of every flow in a fixed time section while do not need to estimate the size of the flow. Theoretical analysis and experiments results show that packet sampling for mice flow estimation (MFEPS) method has a higher accuracy and a better scalability at the same CPU resource consumption in estimating the size of mice flow compared with existing sampling algorithms.