[mnet-devel] Grid Of Trust -- pre-design

[Jim Dixon]

On 29 Nov 2003, Zooko O'Whielacronx wrote:

>  Jim Dixon <jdd at dixons.org> wrote:
> >
> > You imply that ring structures are a natural first choice. However,
> > this is an odd assumption.  In fact hypercubes have a long history.
> > People have been building hypercube-based computers for at least twenty
> > years and for a substantial part of that period hypercube machines have
> > been among the fastest in the world (Intel iPSC and Paragon XP, for
> > example).
> ...
> > Overall, performance seems likely to be better than that of Chord.
>
> So your answer is, for performance?  I had thought that this document was
> aimed at arguing for a certain kind of attack-resistance, but I see that it is
> also intended to have good performance.

I was simply trying to make it clear why Chord should not be an automatic
first choice.  For that matter I would not make DHTs an automatic first
choice.

> Are you part of the GOT project, jdd?

No.  The key Grid of Trust premise is that hash cash will provide a
reasonable defense against a variety of attacks.  However, any real
network will necessarily grow from a small number of nodes.  During that
phase virtually anyone will have sufficient computing power to scatter
nodes throughout the network.  If the network were ever to grow to a large
scale, say on the order of 2^20 nodes, it would still be possible for an
adversary with large resources to put a dummy node into each cell.  You
don't need to be a government to do this; many universities, for example,
have sufficient idle capacity.

On the other hand, it is difficult to believe that a large number of
people could be persuaded to invest a day of computer time just to get
into the network.  I think that most end users would not put more than
say tens of minutes into setting up a connection.  This means that the
primary effect of imposing this 1 CPU-day up front cost would be to keep
the network very small, perhaps to tens of users.

In other words, imposing a high hash cash cost of joining the network
would not keep out bad guys, but it would certainly keep out the casual
user.  Hypothesis: the higher the hash cash cost of joining, the higher
the percentage of bad guys in the network.

Hash cash might be useful as a defense against individuals with limited
resources.  It is not a practical defense against serious adversaries.
Of course, a large network might use hash cash to limit nuisance attacks.

> FWIW, I used to think that a hypercuboid (Kademlia) would be better than
> Chord, but this paper [1] has made me doubt that.

Will read it tomorrow.

> Regards,
>
> Zooko
>
>
http://zooko.com/reading.html#notes_The_Impact_of_DHT_Routing_Geometry_on_Resilience_and_Proximity

--
Jim Dixon  jdd at dixons.org   tel +44 117 982 0786  mobile +44 797 373 7881



-------------------------------------------------------
This SF.net email is sponsored by: SF.net Giveback Program.
Does SourceForge.net help you be more productive?  Does it
help you create better code?  SHARE THE LOVE, and help us help
YOU!  Click Here: http://sourceforge.net/donate/
_______________________________________________
mnet-devel mailing list
mnet-devel at lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/mnet-devel