Abstract: This paper studies the problem of large-scale data flow statistics, and optimizes the problems in the replacement strategy of Elastic Sketch, which is a typical structure of large flow statistics. The optimization strategy solves the problem of cold flow being misjudged as hot flow inserted into the heavy part. In order to optimize the problem that the flow stored in the heavy part may not be the largest flow, a replacement strategy based on the maximum value and group connection is proposed to ensure that the largest flow stored in the heavy part is guaranteed to improve the accuracy of the large flow statistics. At the same time, the probability of thermal collisions is greatly reduced. Compared with the traditional measurement statistics method, the measurement accuracy is improved while the memory usage is reduced.

Key words: a large-scale data flow statistical, sketch, cold and hot flows separation