Contents I. Introduction
A good end-to-end congestion control (CC) algorithm for the Internet should achieve consistent high performance, including high throughput, low queuing delay, fairness, and TCP friendly. Typically, CC algorithms, from Reno [1] to and Cubic [3], rely on packet loss as the fundamental congestion signal and fill the buffer for high throughput. However, with the rapid increase of the link capacity, the impact of delay on user-experience becomes more and more significant [5], [6]. The large queuing delay incurred by the loss-based CC algorithms become intolerable to many delay-sensitive applications, such as online meeting and interactive web applications, and the buffer-bloat problem [8] becomes a hot research topic. The delay-based CC is the elixir for low queuing delay, as the delay signal is easy to obtained and fine-gained enough to reflect the degree of congestion.
