From: Rosalia Tungaraza (rltungar@u.washington.edu)
Date: Mon Oct 11 2004 - 02:31:37 PDT
This paper is about a data distribution protocol termed "the digital
fountain". This protocol is based on the idea that a server wishing to
broadcast or multicast its data to many clients each accessing the data at
the same time, would do so by injecting a stream of that data into the
network. The clients on attaining just parts of that data can reconstruct
the full package without having to wait for the missing pieces (i.e.
provided the packets that they have already received satisfy the
protocol's requirements for encoding the overall message).
The authors describe an implementation of the digital fountain protocol
using the Tornado code. They justify the use of that algorithm using
experimental results that depict Tornado codes having slightly higher
decoding inefficiency compared to standard codes (Reed-Solomon codes), but
smaller encoding time and decoding time. They not only provide numbers to
depict these differences, but also attempt to explain how the experiments
were carried out. In other words, what machines were used and which
variables were either held constant or varied in order to assess a
specific criteria. All these add to the strength of the paper.
Nevertheless, there are some ideas that I think weren't treated well. For
instance, according to the authors, the Reed-Solomon codes are among the
frequently used standard codes. Hence, they used them for the comparative
experiments. I do wonder though, how different the other standard codes
are from the Reed-Solomon or the Tornado codes? I think a small table
outlining their similarities or differences to the above codes would help
answer that question.
Apart from that, I think the paper would be more readable if they had
provided more diagrams. For instance, when they were explaining the
Tornado code implementation in terms of graphs G1 and G2. Moreover, they
could place some of the equations and mathematical terms in more open
sections of the paper as they did with the tables and graphs.
Finally, the paper does propose a very useful protocol for contemporary
distribution of mass data via different types of networks. I especially
think that this protocol should be explored further in relation to
wireless networks. In terms of the tornado code, as the authors suggested,
more work needs to be done to maximize its efficiency.
This archive was generated by hypermail 2.1.6 : Mon Oct 11 2004 - 02:31:47 PDT