On Fri, Oct 06, 2006 at 01:20:50AM +0100, toad wrote:
> Currently, the opennet proposal includes an announcement mechanism where
> the node to be announced chooses its location, sends off an announcement
> request, and is provided with a bunch of connections near that location
> (and leading up to it).
> 
> This is arguably insecure; freenet 0.5 had an announcement mechanism
> where the network would choose a random location for the node, and give
> it some keys and some connections somewhere near it.
> 
> However, it should be possible to target a specific location simply by
> (inserting and) requesting lots of keys near to it... With a large
> network this gets expensive as you have to either know lots of keys, or
> create keys which are close to the target.
> 
> Are routed announcements a problem? It is not possible to generate a
> location randomly then route to it, because an attacker would just use
> the latter stage. Routed announcements do have the advantage of getting
> the right connections right at the beginning; nodes can start opennet
> very quickly.

Ian suggested:
<sanity> toad_: how about the connect request is initially routed
randomly for a few hops, and then it tries to find its way back to the
source node - establishing connections along that path

We can't do this exactly as-is, because the node isn't integrated yet -
it won't be possible to find it.

However, perhaps the node could drop an "anchor" - a random-routed chain
of nodes, which then return the location of the last one to the
announcee, and remember for a while that they are involved (and who was
before them). The node then routes to that node, and the node references
return along the anchor chain. (*Not* down the routing path).

That should work, but would be unreliable and complex... (More reliable
at the expense of more complexity...).
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 189 bytes
Desc: Digital signature
URL: 
<https://emu.freenetproject.org/pipermail/tech/attachments/20061006/043df3b0/attachment.pgp>

Reply via email to