Theo writes, regarding random announcement: > A priori it seems like it might be a problem, but in practice when I did > simulations it seemed to work well enough. Certainly it's a huge > improvement over inform.php -- the inform-style announcement really broke > down over 10,000 nodes, and no tweaks I tried could get it any higher, > whereas the random announcement sailed up to 250,000 nodes (when I ran out > of memory).
Do you recall how many nodes received the random announcement when you simulated this large network? > There could be two reasons for this: first, the nodes aren't entirely > random, they are still linked to one another, although perhaps not by the > keys you would ideally like; second, the insert mechanism is still > operating. Also, for Alice's own purposes, it seems she would rather get > references to a random set of nodes than a closely related group, so that > she has more choice in routing her own requests. I thought the idea was to reverse the direction of the insert pointers once random announcement was in? The rationale being two fold: first, that insert pointers are the way they are now because it is the only way to announce new nodes, so with a real announcement protocol they can change; and second, that it is more logical for the insert pointers to go the other way (towards the center rather than the edges of the network) so that all pointers point towards nodes that specialize in the keyspace, consistent with DataRequest caching. Did you try any sims with insert pointer reversal plus a random announcement protocol? > However, if you can think of a way to announce to a related group of Bobs, > we can certainly test to see which one works better. That sounds tough. Hal _______________________________________________ Devl mailing list Devl at freenetproject.org http://www.uprizer.com/mailman/listinfo/devl
