Research On Elephant Flow Identification In High-speed Network

Network traffic measurement is the basis of network security, bandwidth controlling and charging. But in high-speed network, it is very difficult to do some accurate measurement to network traffic. The traditional method that facing packet-level can’t be used any more. Now the main measurement method for network traffic is facing flow-level. Research shows that, network flow generally follows a heavy tailed distribution, that is to say, a few flows occupy most of the traffic. So by focusing on the information of elephant flow, we can learn enough about the situation of network.But in high-speed network, packets arrival very fast. A large number of mice flow suddenly arrival will impact the accuracy of elephant flow identification. In this paper, we learn about hash technology and LRU technology, combining with their own advantages, a new algorithm is proposed. It uses Counter Bloom Filter to filter mice flows,then uses FEFS replacement scheme to select which flow should be dropped. And it can protect elephant flow as soon as possible.In order to reduce the false negative rate to improve the accuracy of elephant flow identification, we learn about sampling method. In this paper, we add recycle identification to dropped flows. We sample on the dropped flows based on their potential to be elephant flow. When certain condition, they can go back to be a candidate of elephant flows. The emulational experiment results show that our algorithm has low false positive rate and false negative rate in crurrent environment.At last, we summarize the research job. Combined with the problems we met, we point out the next step for future research.