The RTT Distribution of TCP Flows in the Internet and its Impact on TCP-based Flow Control

It is well known that the performance of a TCP flow is affected by its round-trip time (RTT), i.e., the elapsed time between the instant a packet is released by the source to the instant the corresponding ack is received by the source. The distribution of RTTs can dramatically affect not only the data rates realized by individual flows sharing a link but also the utilization of Internet links. In this paper we present a comprehensive study of the RTT distribution for TCP traffic in the Internet. Because access to measurement points is so limited, one typically has access to a node that may be neither the source nor the destination of most of the TCP flows passing through it. Thus, our primary challenge is to infer the RTT distribution by looking at traffic in only one direction (either source-to-destination or destination-to-source, but not always both since the ack (reverse) path may differ from the data (forward) path). We present three different methods of estimating RTT and show that all three methods provide a consistent description of the RTT distribution. Further, we use this data to validate the use of fluid models that have grown pervasive in TCP analysis. Finally we make a study of burstiness of TCP flows.