A Stable and Flexible TCP-friendly congestion control protocol for layered multicast transmission

We propose an improvement of our RLS (Receiver-driven Layered multicast with Synchronization points) protocol, called CIFL for “Coding-Independent Fair Layered mulaticast”, along two axes. In CIFL, each receiver of a layered multicast transmission will try and find the adequate number of layers to subscribe to, so that the associated throughput is fair towards TCP and stable in steady-state. The first improvement is that CIFL is not specific to any coding scheme. It can work as well with an exponentially distributed set of layers (where the throughput of each layer i equals the sum of the throughputs of all layers below i), or with layers of equal throughputs, or any other scheme. The second improvement is the excellent stability of the protocol which avoids useless join attempts by learning from its unsuccessful previous attempts in the same (or better) network conditions. Moreover, the protocol tries and reaches its ideal TCP-friendly as soon as possible by computing its target throughput in a clever way when an incipient congestion is confirmed.