From: Ioannis Giotis (giotis@cs.washington.edu)
Date: Tue Nov 09 2004 - 17:13:44 PST
Multicast has drawn a serious attention in the past few years and
various techniques have been proposed mainly to optimize the large
bandwidth requirements needed for a “naďve” multicast. In this paper,
the authors focus on multimedia multicasting. Multimedia content allows
the use of layered encoding protocols which encode the stream into
various layers, a subset of which is decoded at the receiver to obtain
variable quality. The goal is to match the receiver’s bandwidth capacity
while minimizing the load on the server.
The main idea is that the server sends out all the stream layers but
does not initiate a direct connection to all the receivers which could
lead to scaling issues. The receivers are responsible for choosing their
appropriate level of quality/bitrate they can obtain without congestion
occurring. The mechanism behind the receivers’ choices is an
experimentation driven technique where each receiver determines the
maximum possible subset of layers it can obtain by periodically sampling
the network’s condition.
The authors propose a scheme that has a lot of benefits, and devote a
significant part of the paper to deal with all the small details that
would otherwise render their scheme inefficient such as the interference
between multiple experiments by different nodes. Their explanations and
justifications for their decisions are intuitive and simple. They also
address issues that could arise in the overall multicasting mechanism
and demonstrate their good understanding of the problem as a whole.
However, the simulated data to support their claims is lacking and
probably not enough to convince someone that it would be worth
implementing. And despites their explicitly stated efforts to make their
design easily deployable it still requires a significant effort. That
was probably the reason why a similar scheme is not used in practice
today, although multi-bitrate multicasting is used somewhat.
Still, their results remain relevant today and the need to address the
issue will be perhaps more critical in the near future as the demand for
multimedia multicasting increases.
This archive was generated by hypermail 2.1.6 : Tue Nov 09 2004 - 17:13:44 PST