I get the feeling something like this has probably been suggested before, but I can't find it. If it has, a simple explanation of why I'm out of line would suffice.
What this is, is an idea for decreasing the number of nodes that data passes through on freenet, and so also freenet's bandwidth usage. It would not be a compatible change, but it's just an idea. :) Why not: * On requests, pass a "data target", which is the identity of the node to send the data back to. Nodes initiating requests would set this field to themselves. * When passing a request along, there's a chance (tweakable, possibly influenced by load or bandwidth usage) that we reset the DataTarget to ourselves, and remember the old one. If we do, then things are as normal: we remember the request, and when the data (or the DNF) for it comes back, we pass it back to the original DataTarget. * If we did _not_ reset the target to be ourselves, then once we've sent on the request, we forget about it entirely. It will jump backwards "beyond" us in the chain. Implications: * The reverse paths that data follow will be shorter than the paths of requests, by a factor of (1 / reset probability). * I suppose this would replace pcaching -- as only some of the nodes on the chain will ever see the data. If pcaching were to remain, the probabilities would need to go way up to make sense. * What's the bandwidth penalty for sending an identity on every request? * Anonymity/deniability shouldn't be affected, really, as any node still might or might not be forwarding the request. * It would be possible to "spam" nodes with StoreDatas, in a way similar to the "smurf" attack -- is there any way around this? Hrm. Identities involve a pubkey, right? I think that should do it. * This might make it necessary to open connections to new nodes more often. On the other hand, it provides another opportunity for learning about new nodes. * If data passes through fewer nodes, then bandwidth usage should go down also, leaving more for messages -- which decreases latency and rejected queries. * Possible implications for NGRouting data gathering? Too tired to think about that one at the moment. * Anyway -- I think that's all the points I can come up with at the moment. Discuss among yourselves. ;) --Hobbs
pgp00000.pgp
Description: signature
_______________________________________________ Devl mailing list [EMAIL PROTECTED] http://dodo.freenetproject.org/cgi-bin/mailman/listinfo/devl
