From: Michelle Liu (liujing@u.washington.edu)
Date: Sun Oct 10 2004 - 16:25:28 PDT
Review of "A Digital Fountain Approach to Reliable Distribution of Bulk Data"
Jing Liu
This paper presents an approach to distribute multicast and broadcast data with reliability, efficiency and scalability.
The main contribution of this paper is the proposal of Tornado codes for the receivers to reconstruct source data with a subset of packets received. The author compares Tornado codes with the existing Reed-Solomon codes and interleaved codes in encoding and decoding efficiency, encoding and decoding times and scalability. Although the Tornado codes require more redundant packets to decode than Reed-Solomon codes, the encoding and decoding times are much lower than those of Reed-Solomon codes by using exclusive-or operations instead of the field operations. With regard to interleaved codes, the decoding inefficiency is a random variable that depends on the loss rate, loss pattern, and the block size, while the decoding inefficiency of Tornado is relatively fixed by simulation results.
The author provides quite a lot of performance figures and tables to show the properties of Tornado codes and also the results of implementing Tornado codes on the layered multicast protocol. All the experiment results show clearly the good performance of Tornado codes and Tornado codes could be practically used in different network environments.
One of the weaknesses of the paper is the calculation of Tornado codes' decoding inefficiency 1 + e. The author only deducts the number close to 1.054 by simulation and experiment. Since this number depends on the packet loss pattern and the random choices used to construct the code, the certain simulation setup can not prove that the decoding inefficiency number is low at all network setups and all the other networks except the Internet, such as wireless networks and satellite networks. It would be nice if there were some theoretical deduction for 1 + e and a range of e can be given.
The other things that could be improved are it would be nice if simulations of Reed-Solomon codes and interleaved codes on layered multicast protocol could be provided.
This paper shows a very good example of simple ideas, but function very well. The choice of exclusive-or operations in the encoding is very nice. The other point is that in the encoding process, random choices are used to construct the code. We find that randomization sometimes could give amazingly good performance while at the same time require only simple implementation. The future work of the paper could be the theoretical deduction of the decoding inefficiency of Tornado codes and implementing Tornado codes on other networks besides the Internet, such as wireless networks and satellite networks.
This archive was generated by hypermail 2.1.6 : Sun Oct 10 2004 - 16:25:42 PDT