Michael J Freedman
mfreed at cs.nyu.edu
Sat Mar 11 15:49:26 UTC 2006
On Fri, 10 Mar 2006, Jeff Rose wrote:
> Date: Fri, 10 Mar 2006 11:08:57 +0100
> From: Jeff Rose <rosejn at gmail.com>
> Reply-To: Peer-to-peer development. <p2p-hackers at zgp.org>
> To: Peer-to-peer development. <p2p-hackers at zgp.org>
> Subject: Re: [p2p-hackers] clustering
> It seems like people are always putting arbitrary restrictions on p2p systems
> and simulations in terms of connectivity, but is this really necessary?
> Unless you are trying to use NATed nodes (assume we can punch or route
> through a neighbor),just about any pair of computers on the internet can be
> neighbors. In essence the internet is a fully connected overlay graph.
The problem is that "just about every" and "every" node being able to
communicate are not quite the same thing. Indeed, it's precisely the
difference in these two assumptions which actually raises a lot of
problems when actually deploying DHTs in the wide-area.
We recently presented a short paper at WORLDS '05 which discusses the
real-world problems that arises from non-transitivity in Internet routing:
A can speak to B, B can speak to C, but A can't speak to C
as we all independently discovered from running CoralCDN, OpenDHT, and i3.
(Firewalls and NATs are actually an easier problem that this, as they
express routing constraints much more symmetrically.)
I sent an email about this paper to this mailing list a few months ago,
and my apologies for the repeat. However, as our main audience for this
paper was actually meant to be the hacker community, as opposed to the
academic one, I thought it bears re-mention.
More information about the P2p-hackers