Timeout strategy of sampled flow in high-speed networks

Timeout is one of the most important features of flow recognition, which determines the maximum time interval between adjacent packets. Sampling technology has now been widely used in measurement systems to control the problem of processing overhead and memory consumption. When packet sampling is used, the timeout of original flow is not suitable for the sampled flow. The inappropriate timeout can lead to flow joining or flow splitting, and it may reduce the accuracy of measurement systems. In this paper, we propose a timeout strategy of sampled flow, which can produce an optimal timeout of sampled flow according to the original flow timeout and packet sampling rate. A reasonable timeout of sampled flow can reduce the probabilities of flow joining or flow splitting. The theoretical analysis shows that the number of flow joining and flow splitting can achieve a tradeoff when we adopt the timeout of sampled flow. And the experimental result also demonstrates that original flow length distribution can be obtained accurately by using the timeout of sampled flow.