End-to-end multicast congestion control and avoidance

IP multicast was first proposed by Steve Deering in 1989 in RFC 1112 [26]. It is an efficient mechanism to disseminate data from one source to multiple destinations simultaneously, and therefore is more suitable for various applications such as video conference, multimedia broadcasting, and so on. However, over a decade has passed, IP multicast still has not been deployed in the Internet. Congestion is one of the most important problems impeding the extensive application of multicast.; There are several major challenges in multicast congestion management. These challenges that are not present or are different in unicast context, and thus require different considerations for multicast. The challenges are listed below. (1) Scalability---A multicast group usually has multiple receivers. If the receiver population is large, whether the congestion management protocol can still work smoothly is a big concern. (2) Inter-session Fairness---Different multicast sessions and unicast sessions should be fair in terms of the average throughput they get. A multicast flow in a best effort network should not use too high or too low bandwidth compared with other flows. (3) Inter-receiver Fairness---Because the network is often heterogeneous, different receivers may have different desirable throughput. It is the inter-receiver fairness requirement that the sending rate of a multicast session satisfies faster receivers without overwhelming slower ones at the same time.; In this dissertation, we study the multicast congestion management problem subject to different function placements as follows. Since end-to-end congestion management protocols are preferred in the Internet, all the function placements that we have considered are of the end-to-end category. (1) No support particular for congestion management is provided by receivers. (2) There is support from receiver side, but only one multicast group is allowed for a multicast session. (3) There is support from receiver side, and unlimited multicast groups are allowed for a multicast session. (4) Congestion avoidance for the second situation above.; We have proposed novel designs that simplify the multicast congestion management mechanisms while providing good performance. For the first situation, we have proposed a purely source-based scheme called LE-SBCC. For the second situation, we have proposed a scheme called ORMCC that does not need to measure RTTs between all receivers and the sender, and therefore provides more flexibility and improves the scalability. For the third situation, we have proposed a fully adaptive multi-rate scheme where different receivers can achieve different throughput. For the last situation, we have developed an end-to-end mechanism to detect and react to incipient congestion. Moreover, all our designs are modularized, so that different rate adaptation policies can be used for the protocols we have developed.; Due to the extreme complexity of the real traffic on the Internet, the analytical study of the performance of multicast congestion management schemes is intractable. Consequently, we use simulations and experiments as the major methodology for performance evaluation while having theoretical analysis for some simplified situations. Large scale discrete event simulations are first used by us to verify the performance of multicast congestion management protocols.