[BitTorrent] Implementing Vivaldi?

jamuraa at base0.net jamuraa at base0.net
Thu Jul 15 01:59:31 EDT 2004


On Thu, Jul 15, 2004 at 01:17:10AM -0000, shad0wmatt3r wrote:
> I was wondering if anyone here had attempted to incorporate Vivaldi
> into BitTorrent, or whether Vivaldi has been considered for inclusion
> in the 2.0 spec?
> 
> Vivaldi is a de-centralized algorithm/manner in which peers on a P2P
> network can assign themselves synthetic k-dimensional coordinates
> (typically, k = 5). The Euclidian distance between the coordinates of
> two peers is a reasonable estimate of the latency between them. You
> can go to http://www.pdos.lcs.mit.edu/papers/hotnets:vivaldi/paper.pdf
> for the full paper.
> 
> Because peers in BitTorrent already exchange messages with one another
> constantly, and because Vivaldi data is meant to be piggybacked on
> "normal" messages in a P2P system, it seems like BitTorrent is an
> ideal application for Vivaldi. I'm no BitTorrent expert, but I was
> thinking it could be like this:
> 
> * When a peer contacts the tracker for the first time, to get an
> initial list of peers, the returned list is random.
> * When the peer contacts the other peers in this list, it gets their
> Vivaldi coordinates and, combined with pinging, can establish it's own
> approximate Vivaldi coordinates.
> * Each successive time the peer contacts the tracker for peers, it
> includes its own Vivaldi coordinates. The tracker responds with n
> peers that are close to its coordinates (but not necessarily _the_ n
> closest peers as determined by the Euclidian distance -- that could be
> computationally expensive). These peers are expected to have, on the
> average, lower ping times than the peers in the initial random list,
> and so we would expect, on the average, faster downloads.
> 
> Any thoughts?
> 
> Michael Parker

Peer latency discovery systems have been deemed not useful for the
BitTorrent application, because we are not concerned with latency as
much as we are pure bandwidth.  If a peer has high latency but can
push 1MB/s down our throats we'd prefer them over one with low latency
that can only give us 1KB/s.  The mainline client already does
bandwidth-based choosing based on the speeds it has received from
peers in the past, so this should generate the kind of high-bandwidth
network that advantageous to moving data around.  This also means it 
finds peers on local networks and uses them instead, which are
inherently more low-latency, but again, it finds them based on
bandwidth. 

-- 
Michael Janssen --- Jamuraa --- jamuraa at base0.net --- janssen at debian.org


------------------------ Yahoo! Groups Sponsor --------------------~--> 
Make a clean sweep of pop-up ads. Yahoo! Companion Toolbar.
Now with Pop-Up Blocker. Get it for free!
http://us.click.yahoo.com/L5YrjA/eSIIAA/yQLSAA/dkFolB/TM
--------------------------------------------------------------------~-> 

 
Yahoo! Groups Links

<*> To visit your group on the web, go to:
    http://groups.yahoo.com/group/BitTorrent/

<*> To unsubscribe from this group, send an email to:
    BitTorrent-unsubscribe at yahoogroups.com

<*> Your use of Yahoo! Groups is subject to:
    http://docs.yahoo.com/info/terms/
 



More information about the BitTorrent mailing list