Router Mechanisms to Support End-to-End Congestion Control

This paper considers the potential negative impacts from an increasing deployment of non-congestion-controlled besteffort traffic on the Internet. These negative impacts range from extreme unfairness against competing TCP traffic to the potential for congestion collapse. To promote the inclusion of end-to-end congestion control for best-effort traffic, we propose lightweight router mechanisms for identifying and restricting the bandwidth of high-bandwidth best-effort flows that are using a disproportionate share of the bandwidth in times of congestion. Our method does not require per-flow state based on packet arrivals, but instead relies on the history of packet drops from a queue with RED (Random Early Detection) queue management. Starting with high-bandwidth flows identified from the RED drop history, we describe a sequence of tests capable of suggesting flows for bandwidth regulation. These tests additionally identify a high-bandwidth flow in times of congestion as unresponsive, “not TCP-friendly”, or simply veryhigh-bandwidth. An unresponsive flow is one failing to reduce its offered load at a router in response to an increased packet drop rate. A flow that is not TCP-friendly is one whose long-term arrival rate exceeds that of any conformant TCP in the same circumstances. A very-high-bandwidth flow uses a disproportionate share of the bandwidth relative to other flows. Simulations show the results of regulating the bandwidth of these unresponsive, TCP-unfriendly, or veryhigh-bandwidth flows in times of congestion. We end with a comparison between this approach and others using per-flow scheduling for all best-effort traffic.