FairCast: Fair multi-media streaming in ad Hoc networks through local congestion control

Multicast streaming is gaining increasing importance in wireless ad hoc networks, in part because ad hoc scenarios often include team activities and the requirement for distribution of audio, video and situation awareness to the members. At the network level, techniques for routing the multimedia streams are quite mature. Much more challenging is the allocation of resources, the fair sharing among streams and the control of congestion. While in rate adaptive UNICAST streams congestion control and fair sharing are accomplished with end-to-end feedback techniques inspired to TCP, the feedback does not scale well in MULTICAST. In fact, it leads to the well knows ACK/NAK "implosion" problem and unfair penalties for heterogeneous receivers. These limitations can be overcome using backpressure from congestion points to the sources - but this approach suffers of latency and cannot rapidly adjust to changes in traffic. Another solution is multilayer adaptive coding. Namely, the encoding adaptation is done locally by dropping layers. It does not require end-to-end feedback nor changes in input rates. Multi-resolution codes are now becoming attractive due to the progress in technology; we expect these to become the prevalent techniques in large scale media distribution. One issue, however, that still remains to be resolved is the fair sharing among competing multicast streams. In this paper we address the congestion control AND fair sharing in a multilayer multicast scenario. We show that lack of proper fairness provisions in the "local adjustments" can lead to serious capture situations, especially in heterogeneous traffic mixes (e.g. voice and video). We then propose a FAIR local adjustment that targets a fair dropping of packets in each interference domain. We show that the scheme can be interpreted as a distributed implementation of a utility function minimization, where the utility is the packet loss subject to fairness bounds across flows. This formulation guarantees stability and convergence of the distributed algorithm. The main contributions of this paper are the low overhead design of the local fairness enforcement algorithm, the utility function framework and the demonstration of convergence via simulation in representative scenarios. Copyright 2008 ACM.