From: Lillie Kittredge (kittredl@u.washington.edu)
Date: Mon Oct 11 2004 - 07:46:39 PDT
This paper presents a new efficient algorithm for distributing huge wads
of data to large numbers of clients which may join the stream at any time.
The major objective of this paper was to demonstrate a use for a previous
baby of two of the authors: "Tornado" erasure codes, which provide
redundant packets that allow the client to reconstruct the data even if
they miss some of the data packets. I was rather impressed with the
whole idea of erasure codes, that a stream of data and redundant data
could suffer packet loss but still allow the client to deal with it.
Tornado codes are an improvement in terms of efficiency over earlier
erasure codes, and they spend much of the paper discussing the advantages
of Tornado codes over the older ones.
I was unclear on the prevalence of Reed-Solomon erasure codes. If the
software for dealing with R-S erasure codes is widespread, they may
encounter resistance to Tornado codes when attempting to spread the use of
their fountain. However, the fact that Tornado codes are much more
time-efficient, and scale much better to large data objects, will probably
help them out.
I also would have liked to see more discussion of the fountain rather than
the Tornado codes - given that the fountain is the titular achievement,
the discussion of it feels rather tacked-on. More support for their claim
of "oh look how well we apply to heterogeneous clients" would have been
appreciated, too.
Relevance: I wonder if this is the sort of thing that will replace p2p.
This kind of technology probably appeals to record companies, so they can
take over the Internet and just start saying "here! consume this!" and
pouring new releases at us. But even without such orwellian scheming, yay
massive file distrobutions for code and media and so on.
This archive was generated by hypermail 2.1.6 : Mon Oct 11 2004 - 07:46:41 PDT