From: Jenny Liu (jen@cs.washington.edu)
Date: Sat Oct 09 2004 - 02:25:28 PDT
In "A Digital Fountain Approach to Reliable Distribution of Bulk Data", the authors present the notion of a digital fountain, and then show how a near-ideal digital fountain can be achieved using tornado codes. They also present experimental data from an implementation of their ideas.
The idea presented in this paper is awesome. A digital fountain created using Tornado codes outperforms a digital fountain created using Reed-Solomon codes in time required for encoding and for decoding, and requires only slightly more information to recreate the original message. It also approaches the "ideal" multicast distribution protocol outlined in the paper, particularly when implemented with the proposed layering across different multicast groups: it is reliable, efficient, on demand, and tolerant.
Reed-Solomon codes can be used as error-correcting codes, so a digital fountain created using Reed-Solomon codes can tolerate some errors getting through to the receiver. In contrast, it seems if packets encoded using Tornado codes are transmitted over a network and a single error gets through to the receiver, that one error could be compounded into many errors as a chain of substitutions occurs. Further, because the epsilon factor in the run time of Tornado based digital fountain transactions depends on the random decisions made during the random graph construction during encoding, the runtime might be bad if an unlucky random graph is constructed.
The work could be improved by incorporating a different code that is less suspectible to error propogation, or somehow error-correcting the packets sent by the Tornado code.
This archive was generated by hypermail 2.1.6 : Sat Oct 09 2004 - 02:25:29 PDT