From: Masaharu Kobashi (mkbsh@cs.washington.edu)
Date: Sun Oct 10 2004 - 16:47:01 PDT
1. Main result of the paper
The paper formalizes the notion of an ideal digital distribution system,
"digital fountain", which is a fully scalable protocol for applications
which distribute bulk data to a large number of heterogeneous autonomous
clients. It also demonstrates that Tornado code, a fast erasure code, can
better approximate to the ideal digital fountain than previous systems
by experimenting with a prototype system based on the code.
2. Strengths in this paper
One of the valuable contribution by the paper is they formalized the
notion of digital fountain. It is a great help for the research community
to have a formalized notion, since it can facilitate precise
discussion and
development work in the future.
They also propose a multicast approach, as opposed to the conventional
unicast, based on the idea that the use of an erasure code, Tornado code,
can efficiently eliminate the cost of retransmission.
Their multicast is also special because it is layered.
It is a good argument style that their claim of the performance is based
on both simulation and empirical tests.
3. Limitations and suggested improvements
The most prominent weakness of this paper is it lacks the perspective
of "the total convenience measure" to the participants especially
clients.
It determines the value of the digital fountain.
The authors talk a lot about encoding/decoding efficiency. They also
claim the advantages of multicast. But they do so in separate
discussions.
They do not talk about the whole of the digital fountain in terms of user
convenience such as total download time in comparison to the other
alternatives like on-going unicast distribution.
It can be possible that the single source distribution by the digital
fountain can take more time for clients to download than well designed
multiple mirrored conventional unicast distribution. It is the key point.
In terms of network congestion, there is possibility that the digital
fountain can fall behind a well designed multiple mirrored distribution.
Also multicast can have adverse effect at gateways depending on hardware.
Overall they stress too much on the superiority of Tornado code, which
is in an extreme way of saying "off the topic of networking", and miss
the key point.
4. Relevance today and future
I wonder if it is so promising as the paper sounds, by now there may
be a lot of implementers of the digital fountain with Tornado code.
But I, not being an expert in networking, do not know any such move.
I wonder there may be some hidden shortcomings in the proposed method.
I want to know the reality at the class.
This archive was generated by hypermail 2.1.6 : Sun Oct 10 2004 - 16:47:02 PDT