(no subject)

This paper has proposed a framework for the transmission of layered
signals over heterogeneous networks using receiver-driven adaptation.
The source-based rate-adaptation performs poorly in a heterogeneous
multicast environment because there is no single target rate. This
approach is to combine a layered source coding algorithm with a
layered transmission system. By selectively forwarding subsets of
layers at constrained network links, each user receives the best
quality signal that the network can deliver.
RLM works within the existing IP model and requires no new machinery
in the network. RLM assumes that receivers can specify their group
membership on a per-source basis. A set of end-systems communicate
via a common set of layered multicast groups as a session. The
relationship among the information contained across the set of groups
in a session can either be cumulative or independent. This paper
focuses on the cumulative model because it makes more effective use
of bandwidth but RLM is also compatible with the simulcast model.
In the RLM protocol, the source takes no active role. The key
protocol machinery is run at each receiver, where adaptation is
carried out by joining and leaving groups. Conceptually, each
receiver will drop a layer when detect congestion, and will add a
layer on spare capacity. It's easy to determine if the receiver's
current level of subscription is too high or low, because congestion
is expressed explicitly in the data stream through lost packets and
degraded quality. The approach in RLM is to carry out active
experiments by spontaneously adding layers at "well chosen" times.
The spontaneous subscription to the next layer in the hierarchy a
join-experiment.
Join-experiments may cause transient congestion that can impact the
quality of the delivered signal. In order to properly correlate a
join-experiment with its outcome, how long it takes for a local layer
change to be fully established in the network and for the resulting
impact to the detected back at the receiver should be known. It's
called 'detection time'. Whether the experiment is successful depends
on whether the experiment last longer than detection time.
If each receiver carries out the adaptation algorithm independently,
the system scales poorly. Moreover, measurement noise increases
because experiments tend to interfere with each other. To avoid these
problems, RLM use "shared learning": Before a receiver conducts a
join-experiment, its notifies the entire group by multicasting a
message identifying the experimental layer. Thus all receivers can
learn from other receivers' failed join-experiments. Each receiver
need not run individual experiments to discover this on their own.
Authors evaluated the performance of RLM through simulation and
showed that it exhibits reasonable loss and convergence rates under
several scaling scenarios.

This paper is well written, and obviously it's the research work with
originality. The main contribution is that authors propose a new
scheme for multicast. It uses receiver-driven layered algorithm. RLM
is the first comprehensive instance of a receiver-driven multicast
adaptation algorithm. This paper is very solid with analysis and
simulation. One thing I remember after reading this paper is that
there is a new software developed by Microsoft Research Asia. When I
was in China, I got to know something about this software. It's a
video chatting software using in wireless networks. Users could chat
with each other after installing this software on their cellphones or
PDAs. It's like MSN video chat. Moreover, It also uses the layer
adaption algorithm. Because the wireless signal is not very stable
all the time, and the heterogeneity of wireless networks (WLAN,
Bluetooth, etc) is also a critical problem. Thus, this software could
adapt by sending different level layer of video according to users'
bandwidth. So I think this layer adaption algorithm could be used in
many applications, and solve heterogeneity problem very well.
Received on Sat Oct 28 2006 - 23:08:16 EDT