On Mon, 22 Jul 2002 20:37:29 -0700 "Scott G. Miller" <[EMAIL PROTECTED]> wrote:
> On Mon, Jul 22, 2002 at 09:05:38PM -0400, Michael Wiktowy wrote: > > Greetings, > > > > I was just wondering if Ian and others feel that multi-area > > specialization is a requirement for Freenet to work (i.e. stems > > from a conscious design decision) or just something that is a > > consequence of the current Freenet implementation. > > No, I don't think its necessary except in very small networks (which > of course might characterize the current network). > > > > > It seems to me that Freenet could be designed to work either > > way and having one dynamic keyspace specialization might > > simplify routing quite a bit; maybe even enough to make it > > more reliable while retaining its anonymity. > Theres nothing that should prevent this currently, in fact, my node > has two distinct peaks. > > > There is also a line to be drawn between first party > > specialization (what the node itself tries to specialize in via > > selective caching) and third party specialization (what other > > nodes think another node's specialty is and attempting to > > reinforce that via more requests and inserts to that node). > > Separating these is very very bad. If the network thinks your good > for something and you think its something else, the node will > constantly be fighting the network. > Not to mention the potential for (intentionally) screwing up the network. The idea is that you don't know/care what _your_ specialization is, only those of your neighbors (the whole point of refs). Then routing turns your neighbors' ideas of your specialization into a self-fulfilling prophecy, because you end up caching more keys in those areas. When you state it this way, it's fairly easy to see that there really isn't any difference between one and multiple peaks on the histogram. I'm thinking that on a big enough network, most nodes would eventually flatten out more to one peak, but that doesn't account for the fact that inserts happen all the time. Anyway, it's all good. Or would be if routing and contacting would actually work. :) --hobbs
msg03473/pgp00000.pgp
Description: PGP signature
