On Sat, May 26, 2001 at 11:48:07AM -0700, hal at finney.org wrote: > Isn't it still just a matter of choosing the node which holds the > key closest to the target one?
yes > I thought the tree based search was just a speed optimization. The > basic heuristic is still to define relative closeness, right? We can > tell if A is closer to B or to C. This should be irrespective of what > other keys D, E, F, ... are present on the node. Yes. Actually we can now tell if A < B (which is needed to build a tree). > > 6. Potential problems > > > > The model mentioned in section 2 suggests a potential problem. In order to > > be a good self-organising network that routes requests to nodes containing > > their associated data, it is necessary for nodes in the vicinity of the > > insertion point to have many requests routed through them. > > I don't understand this, and it may be based on the false assumption that > most requests "should" be satisfied by a node near the original insertion > point of the data. But Freenet is not designed to follow this principle. > It is entirely possible that most requests will be satisfied by a node > far from the original insertion point. In that case there is no reason > for nodes near the insertion point to see many requests for that > >data. I think by 'nodes containing their associated data' he meant the epicenter, not the insertion point. > > This has the profound effect of moving the data away from the point of > > insertion. ie. It starves the routing mechanism of data. > > Yes, to the first, but no to the second. The routing mechanism is not > starved, it is still informed about where the data is. Yes, and requests *should not* route to the point of insertion for very good reasons. > Furthermore this HTL escalation makes the problem worse because it > makes the search space flatter. Nodes can't intelligently choose their > neighbors when every neighbor has the same data. Routing paths become > random and data which is not widely spread through the network is not > found. Hopefully, under 0.4/0.5 with the announcment system, nodes will try to position themselves at their conter-of-keyspace (as found by the annoument protocol). The network should route requests to them because the initial links to them will be for the same center. Then, maybe, things start working nicly because nodes can enter the closeness of a document's key to their own center of keyspace, in the decision about dropping the document. AGL -- The herd instinct among economists makes sheep look like independent thinkers -------------- next part -------------- A non-text attachment was scrubbed... Name: not available Type: application/pgp-signature Size: 240 bytes Desc: not available URL: <https://emu.freenetproject.org/pipermail/devl/attachments/20010526/4582f961/attachment.pgp>
