Todd Walton <todd at hardboot.org> writes: > Forgive me if I missed some point here, but I thought that the reason > specialization was good was simply because, realistically, a node *can't* > cache everything. If the node can cache everything, why does it need to > specialize? You are talking about specialization as in what types of keys > to hold onto, correct? > > -todd
Specialization to me is a clustering effect of three things together: a node is specialized in a portion of keyspace when, for that keyspace, it's got a higher than normal concentration of 1) data in its store, 2) references in its routing table, and 3) references in other people's routing tables pointing to it. Ian's plan is to have the node force (1) by decreasing the chance of caching data it doesn't think it specializes in. I see it being an interesting idea in terms of increasing the amount of different keys a network with all full stores. The only way I can figure that changing (1) would effect (2) and (3) is to slow down changes in (3). i.e. the network is constantly sending data to nodes, and as is, the nodes try their best to be experts at everything the network throws at them. With probabilistic caching, the nodes try to keep their current specialties, and are more immune to nodes "incorrectly" sending requests to them for content they don't specialize in. This _may_ be good if the dempening is slight, but too much probabilistic caching will result in routing becoming nearly static (definitely not a good thing for anonymity.) As it's definitely going to be implemented, all I can do is warn against the dangers of overusing this tool to decrease the effects of requests that are not in line with nodes' perception of their place in the system. (which is a major part of my disagreement with this ides) Thelema -- E-mail: thelema at swbell.net Raabu and Piisu GPG 1024D/36352AAB fpr:756D F615 B4F3 BFFC 02C7 84B7 D8D7 6ECE 3635 2AAB _______________________________________________ devl mailing list devl at freenetproject.org http://hawk.freenetproject.org:8080/cgi-bin/mailman/listinfo/devl
