From: Tom Christiansen (tomchr@ee.washington.edu)
Date: Sun Oct 10 2004 - 21:43:33 PDT
The Byers paper presents a method for broadcasting data to multiple network
nodes at once. Examples for uses of such a protocol would include: service
pack releases, product updates, movie distribution, etc. The main challenge
of this is to ensure reliability of the distribution system with very
little feedback/acknowledgement from the recipients (basically avoid
flooding the server with re-transmission requests on packet drops). The
suggested solution to this problem is to encode redundant data into the
data packets to allow the recipients to reconstruct damaged or missing data
packets. While this method might not completely eliminate re-transmissions,
it will decrease their frequency dramatically.
The strong point in this paper is that the Tornado Z (invented/implemented
by the authors) is a vast improvement compared to other contestants in the
field (Reed-Solomon codes mentioned). Both encoding and decoding is several
orders of magnitude faster with the proposed method than with Reed-Solomon.
The paper further suggests a layered subscription approach to allow several
nodes of different speeds to subscribe to the same communications service
("digital fountain"). This particular part of the protocol uses a clever
but simple scheme for congestion control.
The weak spot of the paper seems to be the fact that if this is implemented
on larger networks (such as the Internet), the network will be flooded with
broadcasts which may or may not be relevant to all users. Hence, bandwidth
is wasted. However, for use on internal networks (such as LAN's) this
protocol would likely excel.
Tom Christiansen
This archive was generated by hypermail 2.1.6 : Sun Oct 10 2004 - 21:44:01 PDT