Re: Shortlived nodes: the route to position clustering?

On Monday 20 August 2007 22:40, vive wrote:
> On Mon, Aug 20, 2007 at 09:07:30PM +0100, Matthew Toseland wrote:
> > Not sure I understand your graphs. What is N1000_initial vs 
> > N1000_shuffled_initial? If it's an idealised kleinberg, it should be flat, 
> > no?
> 
> Both are initially the Kleinberg model. The _shuffled_ graphs shuffle
> the initially assigned positions instead of randomizing new ones (as the
> other simulations do, as Freenet nodes do). Shuffling positions means
> essentially randomizing the location on the circle, taking the position
> from the available and evenly spread ones. You can see this difference
> in the _initial_ plots.
> 
> The reason to compare these cases is to see whether randomization of
> position clearly affects the position clustering or not. I found that
> there often is a relation, but that it happens in other places too. And
> it happens even if the positions taken before inflow of shortlived nodes
> are evenly spread out. So there is no easy relation between the two.

Okay, so in *shuffled*, we keep the original sequential locations, but 
randomize them. Whereas in the others, we assign new random locations to each 
node. So the gap between shuffled and non-shuffled should narrow 
significantly as the network gets bigger, right? Also ongoing creation and 
destruction of locations through churn should act to heal the network and 
bring it closer to the ideal - if it wasn't for the clustering problem.
> 
> regards,
> Vilhelm
>

_______________________________________________
Tech mailing list
[email protected]
http://emu.freenetproject.org/cgi-bin/mailman/listinfo/tech

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.6 (GNU/Linux)

iD8DBQBGyh9gA9rUluQ9pFARAn3rAKClk5ymNtpqyQYEjZ09iOi61Z00MQCcCDrL
8AIKuGBuyh7dFzOlfiG+uq0=
=50cp
-----END PGP SIGNATURE-----