On Fri, Aug 17, 2007 at 05:55:17PM +0100, Michael Rogers wrote:
> vive wrote:
> > Should be correct directly after generated from Kleinbergs model. But what 
> > do
> > you mean with current locations? Referring here to the current locations on 
> > the
> > circle, where closeness does not have to mean being close in the network 
> > after
> > having swapped and done destination sampling (where positions affect 
> > swapping).
> > Is your suggestion to introduce nodes in roughly the same part of the 
> > network
> > where others just disappeared?
> 
> Not necessarily - wherever the new node appears on the circle, isn't the
> important thing the relationship between its distance (on the circle)
> from other nodes and its probability of being connected to them?

In the ideal Kleinberg model it is. But when you have randomized positions
and swap for positions as in Freenet; then no. Circular distance can no longer
be used in the same way by peering with nodes just based on which positions
they have taken later on. When you randomize positions you lose the relation
between circular distance and distance to "remote parts" of the network. A key
point: two nodes separated many steps in the possible shortest route between
them may still have ended up with close circular positions. So they are far
in the network, but close on the circle. Swapping hopefully gives assigned
positions back to a usable level for routing, but not perfectly.

Try to see the network from the Kleinberg model as a platonian ideal form.
We can try to replicate it by running the swapping algorithm, where nodes will
prefer positions based on how they clustered in the original network, but the
generated positioning will never (for all practical timescales) be more than a
shadow of the perfect network. Hopefully it works rather well (surpringly well
in simulation! :)) but still with its small faults and glitches in the
assignments which affects the routing.

This is also why #freenet-refs between strangers and the darknet model (without
opennet) does not give a good result for routing. The ideal topology is not
there from the first case. People in the real world seem to form connections
so that networks are navigable; #freenet-refs just adds people in a long
sequence. The "ideal" topology will not be there from the beginning, not even
something that can be used for short routes. Opennet gives something resembling
the Kleinberg model, since it adds/rewires connections to other parts of the
network, and this is the strong argument for running opennet as long as the
network does not grow well enough to resebmle real social clustering.

> Why would the connections between the other nodes make a difference?

Sorry, I didn't understand that.

Kind regards,
Vilhelm
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 187 bytes
Desc: not available
URL: 
<https://emu.freenetproject.org/pipermail/tech/attachments/20070818/3ee6841a/attachment.pgp>

Reply via email to