An ensemble method for estimating TCP throughput on application layer
Natsuki Kai, Hiroshi Yoshida, Koichi Nihei · 2018
Transmission Control Protocol (TCP) throughput is estimated by using the data size and the transferring time when large data is transferred from a sender to a receiver. However, if divided data is intermittently transferred such as in live video streaming, the estimated TCP throughput has an error because the actual transferring time is unobservable on an application layer. Other methods have been proposed for estimating TCP throughput, such as equation-based estimation that uses network information such as round trip time (RTT) and packet loss rate. Nevertheless, the accuracy of estimation lowers when RTT becomes unexpectedly much longer or when the burst loss occurs. We propose using an ensemble method for estimating TCP throughput on the basis of information that is observable on the application layer. Our proposed method can reduce error of the estimation. An evaluation experiment using actual data obtained on a Wi-Fi network showed that our proposed method estimates TCP throughput more accurately than conventional methods.