Identification of TCP congestion control algorithms with convolution neural networks
Takahiro Nogiwa, Kouji Hirata · 2018
Currently, various high-speed congestion control algorithms such as CUBIC and Compound TCP are used in order to accommodate sufficient bandwidths over high delay-bandwidth networks. CUBIC and Compound TCP have been implemented as the default congestion control algorithms in Linux and Microsoft Windows OS, respectively. However, Compound TCP drastically decreases its throughput when traffic flows of CUBIC and Compound TCP share a bottleneck link. In order to resolve this problem, it is necessary to identify them at the bottleneck link and suppress the throughput of CUBIC flows. In this paper, we propose an identification method of congestion control algorithms with image recognition using convolution neural networks. Through simulation experiments, we show the effectiveness of the proposed identification method.