On Mon, Jul 29, 2002 at 05:13:59PM +0100, Matthew Toseland wrote: > maxRoutingSteps should not include steps rejected because they point > back to the requestor. > Pro: > Should prevent RouteNotFound, attempts were made to contact 0 nodes > Con: > Requests may get routed too far away from the ideal route. > > This is in the new build 490. > > Comments? Revocation notices? Flames (ian, that means you)? Death threats?
Looks ok, but I really think that we need to make it only do FSRR (First Step Random Routing) if a node receives a repeated request for the same data with a higher HTL. People have said that a hint of de-specialization is a good and useful thing, but nodes already have that in abundance (see http://hawk.freenetproject.org:8889/key_histogram_detail.txt). Oskar claims that he has observed what he calls a forked network (which I define to be a network which has split into two or more networks with no inter-connections, he has a different definition). It is possible that for a given key, multiple "attractors" for keys like it might exist, but since inserts are drawn towards one or the other, a request which follows the wrong attractor may not find its data. The solution is to merge these "divergent" attractors for the same key somehow. A brute-force solution would be for nodes to periodically send multiple requests for the same key to different nodes, then suggesting to the datasource nodes for the received data that they announce to each-other. Of course, we would need to take precautions to ensure that this mechanism could not be abused for malicious purposes. Ian. -- Ian Clarke [EMAIL PROTECTED] Founder & Coordinator, The Freenet Project http://freenetproject.org/ Chief Technology Officer, Uprizer Inc. http://www.uprizer.com/ Personal Homepage http://locut.us/
msg03553/pgp00000.pgp
Description: PGP signature
