Steady State Analysis of Random Early Detection Gateway with Controlled Traffic by TCP

In a packet-switched network, a feedback-based congestion control mechanism is essential to provide data transfer services efficiently. Its main objective is to prevent packet losses in the network, and to utilize network resources effectively. The current Internet uses a windowbased flow control mechanism in its TCP (Transmission Control Protocol), as the feedback-based congestion control mechanism. As an example, a version of TCP mechanism called TCP Reno uses packet losses in the network as feedback information from the network since packet loss implies congestion occurrence in the network [1, 2]. Until packet loss occurs in the network, TCP Reno gradually increases its window size. As the window size exceeds its available bandwidth, excess packets are absorbed by the buffer of an intermediate gateway for some period. If the window size increases further, the buffer of the gateway overflows, leading to packet losses. The source host detects an occurrence of packet losses in the network by, for example, receiving duplicate ACKs, and quickly reduces its window size. After reduction of the window size, congestion in the network will be relieved. Once congestion in the network is over, the source host increases its window size again. In short, the congestion control mechanism of TCP first increases its window size, and as soon as it detects packet losses in the network, it reduces its window size. The congestion control mechanism of TCP repeats this process indefinitely during the connection.