Abstract:

Traditional approaches to receiver-driven layered multicast have advocated 
the benefits of cumulative layering, which enables coarse-grained congestion 
control and complies with TCP-friendliness equations over large time scales.
In this talk, we qualify and quantify the costs and benefits of using
non-cumulative layering and present a scalable multicast congestion 
control scheme which provides a fine-grained approximation to the behavior of
TCP additive increase / multiplicative decrease (AIMD).  

Our results show that careful design of non-cumulative layering sequences 
can provide an effective, fine-grained, TCP-friendly congestion control
solution.  
In contrast to the conventional wisdom, we demonstrate that fine-grained rate 
control can be achieved with only modest increases in the number of layers and 
aggregate bandwidth consumption while using a small constant number of join and leave operations per rate adjustment.

We will demonstrate how our technique integrates naturally with existing
methods for content delivery using reliable multicast implemented with fast
forward error correcting (FEC) codes.

Joint work with Michael Luby (Digital Fountain) and Michael Mitzenmacher
(Harvard).