Delay-based congestion avoidance in computer networks

摘要

A packet data communication system employs a congestion avoidance method in which each node measures the round-trip delay occurring when it sends data to a destination and receives an acknowledgement. This delay is measured for different load levels, and a comparison of these delays is used to determine whether to increase or decrease the load level. The load level can be adjusted by adjusting the window size (number pf packets sent in to the network) or by adjusting the packet rate (packets per unit time). The objective is operation at the knee in the throughput vs. traffic curve, so that the data throughput is high and the round trip delay is low. Control is accomplished at each node individually, without intervention by the router or server, so system overhead is not increased.