Review - Receiver-driven Layered Multicast

Review of Receiver-driven Layered Multicast
by Ivan Hernández

The paper presents a rate-adaptation network protocol called
Receiver-driven Layered Multicast (RLM). This protocol use a
cumulative layering, where a signal (audio, image, video, ...) is sent
through different layers, each layer provides refinement information
to the previous layers. Each layer is sent in a different multicast
group. The sources takes no active role in the protocol; on the other
hand, the receivers implement all the logic of the protocol adapting
to the network conditions, i.e., each receiver adapts to congestion by
joining and leaving multicast groups. Finally, there is no explicit
signaling between the receivers and routers or between the receivers
and source.

In the first stage, a receiver search for the optimal level of
subscription by adding layers until congestion occurs and backs off to
an operating point below this bottleneck. Then, the receiver enters
into an adaptation stage where the receivers tries to determine if its
current level of subscription is too low, and therefore it will try to
add a new layer. In order to do this, the receiver performs an
experiment by adding the next layer (i.e., joins the multicast group),
if this causes congestion, then the receiver drops this layer (i.e.,
leaves the multicast group) and backs off exponentially before
experiment again; on the other hand, if the experiment does not causes
congestion during a certain time, then the receiver keeps the layer,
thus getting a more optimal operating point. The protocol allows
shared learning in a group, this is, before perform an experiment the
receiver notifies the entire group by multicasting a message with the
experimental layer; thus, the all receivers can learn from other
receivers when an experiment fails. Nevertheless, each receiver must
learn by itself when it may add another layer by performing an
experiment itself.

I liked this paper. The authors explain clearly all the concepts and
they provide examples of their ideas. One good thing of the solution
is that works with the current multicast technologies, i.e., the
solution works with out of the box routers with multicast
functionality. In the simulations, the results show a good performance
and are consistent with the predictions of the authors. On the other
hand, RLM does not provide mechanisms to achieve fairness allocation
of bandwidth. To provide this fairness (and congestion avoidance as a
plus), the authors propose use of RED instead of the drop-tail
mechanism that they used in the simulations.
Received on Mon Oct 30 2006 - 22:00:03 EST