A Survey of Congestion Control in the Internet
Luo Wan · Chinese Journal of Computers · 2001
This is a review paper on recent works about TCP/IP congestion control of Internet. A summary of the techniques considered in the literatures is presented. Several reasons cause Internet congestion control. The typical techniques of congestion control include TCP and IP congestion control. The traditional end to end congestion control mechanisms of TCP which employs an Additive Increase Multiplicative Decrease (AIMD) algorithm have been a critical factor in the robustness of the Internet.Modern implementations of TCP contain four intertwined algorithms: slow start, congestion avoidance, fast retransmit and fast recovery. In this paper, the details of the above four algorithms and the improvements to TCP, such as Reno, Tahoe, SACK, Vegas, are investigated. Self similarity, efficiency and fairness of TCP congestion control are also discussed. However, the Internet is no long a small community, and it is no longer practical to rely on all end nodes to use end to end congestion control for best effort traffic. Similarly, it is no longer possible to rely on all developers to incorporate end to end congestion control in their Internet applications. The network itself must now participates in controlling its own resource utilization. Assuming the Internet will continue to become congested due to a scarcity of resources such as bandwidth and buffer, this proposition leads to several possible approaches for congestion control. One approach is to improve the continued use of end to end congestion control, as the primary mechanism to share scarce bandwidth and buffer. A second approach is to rely on the financial incentives or pricing mechanisms. A third approach involves the deployment of packet scheduling disciplines in routers that deal with packets, as much as possible, according to different priorities of packets. There are several existing mechanisms including Random Early Detection (RED), Explicit Congestion Notification (ECN), Fair Queuing (FQ), Weighted Fair Queuing (WFQ), etc. The above algorithms are valuable for congestion control but have deficiencies in the aspects of robustness, efficiencies, and fairness respectively. Improvements are needed.Finally, research topics such as congestion control of IP multicast, Differentiated services (Diffserv), and congestion control in the special fields, e.g., ATM and satellite networks, are investigated. Research directions and open problems in this area are also discussed.