Russell Cagle wrote
> In Freenet, nodes can be 'near' or 'far', depending on how many hops you
> need to get there. Near and far also change based on popularity and
> key-affinity. However, getting data from a Freenet-near node
> might use more
> network resources than a Freenet-far node, because of the way the physical
> network is set up. Since the configuration w/r/t reality is random, and
> there are many more bad routing configurations than good, I'd imagine that
> bad routing overwhelms the network, eg even though you find a
> document in a
> handful of hops, the data is in taiwan while you're in ohio.
>
> This can't be a new idea.
It isn't a new idea. Its just in experimental small networks it is not such
a big issue, but I think it will become one.
To tackle this I sugested a while back to have several parrallel requests
instead of just one, and then use the quickest response for the refference,
this would introduce a bias towards nodes that are best suited. All factors
would then count (key closeness, datastore size but also physical network
topology, bandwith, usage, etc.).
Offcourse this would (dramaticly) increase the network traffic. Although not
exponentially becouse freenet routing
still focuses the request paths.
As a compromise Ian sugested to introduce random splitting of requests. You
still have the network bias towards 'fitter' nodes, but it is possible to
more or less contain the network traffic.
I am not sure if/when this will be implemented (0.4?)
Neil
_______________________________________________
Devl mailing list
[EMAIL PROTECTED]
http://lists.freenetproject.org/mailman/listinfo/devl
