An adaptive queue management method for congestion avoidance in TCP/IP networks

In mid 1980s, congestion control algorithms were implemented in TCP, realizing that the original flow control mechanisms were not capable of protecting IP networks from severe congestion. In recent years, the limitations of endpoint congestion control have been realized. The active queue management method RED and the feedback scheme explicit congestion notification (ECN) have been proposed to complement the TCP congestion control with the goal of preventing congestion instead of curing the network after congestion occurred. We evaluate the performance of the RED/ECN congestion avoidance system, and show that it is not dynamic enough to provide successful congestion avoidance in different traffic conditions. As an alternative to random early setection (RED), we propose an adaptive queue management scheme that can maintain high throughput and low roundtrip delays under dynamic traffic loads.