From: Kevin Wampler (wampler@cs.washington.edu)
Date: Tue Nov 09 2004 - 23:05:01 PST
Although there are many situations in which a multicast protocol seems as
if it would be a useful technique, there are details in such a protocol
that should be resolved. One such critical detail is the regulation of
the bandwidth of a multicast transmission on a per-receiver basis. A
solution to this problem is presented in "Receiver-driven Layered
Multicast".
This paper describes a system in which receivers of a multicast
transmission achieve varying rates of transmission by subscribing to more
or fewer layers. As in TCP (though the actual mechanisms differ) an
optimal rate is achieved by detecting congestion through dropped packets.
As has been noted, such a system must cause congestion to find an optimal
bandwidth. Although this is unavoidable with such a system, the authors
describe an explicit mechanism of testing for congestion and for
coordinating these tests amongst groups of receivers. This allows users
to subscribe to the appropriate level of the transmission in a manner
which is efficient and scalable.
The system described seems to function well on a typical network. Even
given this, there is still the criticism to be made that it relies too
heavily on the structure current network. For example, the method depends
on a random packet drop policy instead of a priority drop policy. It is
interesting to see an approach which actually takes advantage of naive
network policies, but if such a multicast protocol became widespread it
could hamper developments to improve the functionality of a network by its
reliance on the more naive functionality.
This archive was generated by hypermail 2.1.6 : Tue Nov 09 2004 - 23:05:02 PST