Estimating TCP Congestion Control Algorithms from Passively Collected Packet Traces using Recurrent Neural Network
Ohzeki Naoki, Ryo Yamamoto, Satoshi Ohzahata, Toshihiko Kato · 2019
Recently, as various types of networks are introduced, a number of TCP congestion control algorithms have been adopted. Since the TCP congestion control algorithms affect traffic characteristics in the Internet, it is important for network operators to analyse which algorithms are used widely in their backbone networks. In such an analysis, a lot of TCP flows need to be handled and so the automatically processing is indispensable. Thin paper proposes a machine learning based method for estimating TCP congestion control algorithms. The proposed method uses a passively collected packet traces including both data and ACK segments, and calculates a time sequence of congestion window size for individual TCP flows contained in the trances. We use s recurrent neural network based classifier in the congestion control algorithm estimation. As the results of applying the proposed classifier to ten congestion control algorithms, the major three algorithms were clearly classified from the packet traces, and ten algorithms could be categorized into several groups which have similar characteristics.