[p2p-hackers] paper "rarest first and choke algorithms are enough"

Daniel Stutzbach agthorr at cs.uoregon.edu
Thu May 4 13:21:00 UTC 2006


On Thu, May 04, 2006 at 11:04:40AM +0200, Arnaud Legout wrote:
> We identified that when the performance of a torrent is not optimal, 
> then it is in a transient phase. That means that the seed has not yet
> sent a copy of each piece. If you use network coding of whatever coding 
> you can imagine, we will not improve that fact that information is missing
> in the torrent and that is this missing information that causes the 
> decrease in performance.

Coding may be able to recover the missing information more quickly
when the system is in a transient phase.  I can think of two scenarios
where this may occur:

1) Early in the life of the torrent when the seed has uploaded each
   piece only once or twice.  However, due to the fact that many users
   abort their downloads, some of these uploaded pieces will be
   missing and the seed does not know which pieces were reproduced
   into the rest of the system and which were not.  Coding
   may allow the system to get out of the transient phase more rapidly
   since the seed does not risk uploading redundant information.

2) Late in the life of the torrent when all the seeds have departed.
   Coding improves the probability that the set of remaining
   peers have enough information among themselves to reproduce the
   whole file (and become new seeds), thus extending the life of the
   torrent.

-- 
Daniel Stutzbach                           Computer Science Ph.D Student
http://www.barsoom.org/~agthorr                     University of Oregon



More information about the P2p-hackers mailing list