I. Introduction
CONGESTION is an inherent property of the currently best-effort Internet. Consequently, transport protocols commonly implement congestion control, which refers to end-to-end algorithms executed by a protocol in order to properly adapt the sending rate of a network flow to the available bandwidth in the network. Protocols with window-based end-to-end flow control utilize one or another version of TCP-friendly congestion control, which includes Jacobson's modifications to TCP [1], [15], TCP-like congestion control (e.g., [43]), binary-feedback algorithms (e.g., [2], [3], [6], [12], [21], [28], [38], [49]), and equation-based methods (e.g., [11], [40]). These algorithms are shown to work well in the environment where the sender relies on “self-clocking,” which refers to the use of positive acknowledgment not only to recover lost packets, but also to slow down the sender during congestion [15].