由于在网络测量中存在不可避免的数据损失,网络监测数据通常是不完备的甚至是稀疏的,这使得大象流的精确检测成为一个具有挑战性的问题.本文提出了一种基于数据补全的离线大象流检测方法.为实现对于大象流的精准检测,首先实现了 一个基于矩阵分解的数据补全算法,将流量数据补全问题转化为一个低秩矩阵奇异值分解问题.其次,在此基础上进行高阶扩展,引申出张量补全模型,利用张量CP分解实现数据补全,将原问题转化为通过最小化张量秩来恢复缺失条目的张量补全问题.最后对上面使用的矩阵补全算法和张量补全算法进行了仿真实验,对比了各算法精准度,评估了超参数,并展示了张量补全算法的时间开销.实验结果证明该方法取得了较好的效果.
Due to the inevitable data loss in network measurement,network monitoring data is usually incomplete or even sparse,which makes the accurate detection of elephant flows a challenging problem.This paper proposes an offline elephant flow detection method based on data completion.To detecte elephant flow accurately,a data completion algorithm based on matrix decomposition is imple-mented firstly,which transforms the traffic data completion problem into a low-rank matrix singular value decomposition problem.Sec-ondly,a high order expansion is carried out on this basis.The tensor completion model is extended,the tensor CP decomposition is used to realize data completion,and the original problem is transformed into a tensor completion problem that restores missing entries by minimizing the tensor rank.Finally,the matrix completion algorithm used above is used.Simulation experiments were carried out with the tensor completion algorithm,the accuracy of each algorithm was compared,the hyperparameters were evaluated,and the time overhead of the tensor completion algorithm was shown.The experimental results show that the method has achieved good results.