Identification of TCP Congestion Control Algorithms from Unidirectional Packet Traces

Abstract

Recently, according to the diversification of network environments, various TCP congestion control mechanisms have been introduced. Since the TCP congestion control algorithms affect the traffic characteristic in the Internet, it is important to analyze which algorithms are used widely. This paper proposes a method to identify a congestion control algorithm from passively collected packet traces that include only unidirectional packet traces. In unidirectional traces, it is hard to map a data segment and the ACK segment corresponding to it and to estimate round-trip time. So, the proposed method uses the curve fitting for sequence number vs. time graphs by applying the least squares method with linear through quartic functions, and maps the first-order and second-order differentiations. This paper applies the proposed method to various TCP congestion control algorithms including TCP Reno, CUBIC TCP, Hamilton TCP, TCP Vegas, and TCP Veno.