From: Jonas Lindberg (jonaslin@kth.se)
Date: Sun Oct 10 2004 - 18:22:34 PDT
A Digital Fountain Approach to Reliable Distribution of Bulk Data
John W. Byers, Michael Luby, Michael Mitzenmacher, Ashutosh Rege
Reviewed by Jonas Lindberg:
In this paper the authors present an approximation to a "digital fountain",
which is an ideal protocol for distributing bulk data (e.g. software
updates) over a network. The characteristics for such an ideal protocol are
that it is: scaleable; reliable; efficient; on demand; tolerant and not
relying on feedback from the clients.
The suggested approximation uses a new kind of erasure codes, called
"Tornado codes", which require less encoding and decoding time than the
older "Reed Solomon codes" but still is almost as efficient. For this paper,
the authors have actually implemented their suggested protocol and presents
results from various tests, which of coarse is a good thing. For example
they present results from benchmarking the encoding and decoding times for
Tornado and Reed Solomon codes and proves a substantial time difference.
The design of the protocol seems to be motivated by looking at how others
have done before and by using the Tornado code instead of Reed Solomon code.
In this paper, the design choices are explained fairly well, and it is
assumed that the basic ideas (like using multicast) are correct.
The authors write that their suggested protocol is not optimal for streaming
media and do not discuss how it could be used for this purpose. However,
streaming media being such an important application, I think that the paper
could be improved by such a discussion. Maybe it is possible to use the
protocol if a reasonable buffer is used for the playback?
I find this article relevant since the need for efficient distribution of
data over networks is needed every day. As the authors write, their
suggested solution is not final and needs a little more research before it
can be used in practice. The article would be even more interesting and
relevant if it was updated with some new research results.
This archive was generated by hypermail 2.1.6 : Mon Oct 11 2004 - 02:22:51 PDT