From: Ioannis Giotis (giotis@cs.washington.edu)
Date: Sun Oct 10 2004 - 19:28:27 PDT
The problem at hand is to efficiently multicast data on various networks.
The ideal protocol would have to be efficient in processing time, redundant
data and reliable. The client should also be able to start the download at
any time and receive the data even with varying packet losses.
Error-correcting codes come into play as the basis of such a protocol.
However, the standard Reed-Solomon codes despite their theoretical
superiority do not perform well in practice with the processing time needed
their major disadvantage.
The authors propose and evaluate a much simpler scheme called Tornado codes
that requires less computations but also requires more redundant data to be
transmitted. Their tests indicate that their scheme performs well as
expected and they go further into examining some variations of their
protocol.
The main thing to notice from this paper is that there exists a tradeoff
between computational efficiency and transmitted data. It is clear that the
authors choose to sacrifice some redundancy in order to achieve good
performance. Overall, their tests seem complete and present a strong case.
On the other hand, since the paper tries to address the design of an "ideal"
protocol, I would expect more variables in the whole picture. For example,
at which point, with which networks and under what circumstances is the use
of Tornado optimal? I would imagine that when the network is very unreliable
even the small overhead of Tornado would become an issue. Although some
conclusions can be derived by their decoding diagrams, it is far from a
complete analysis.
Some other details are also missing; formal proofs of the decoding
inefficiency, interleaving and layering efficiency seem not too hard to
produce.
Despite the paper's weakness, the authors seem to address most of the
interesting cases and I was convinced that Tornado is highly useful in
practical applications.
This archive was generated by hypermail 2.1.6 : Sun Oct 10 2004 - 19:28:27 PDT