I think the system we have been debating will adequately manage load, propagating load back to its original source, and preventing flooding.
On Mon, May 22, 2006 at 12:14:13AM -0700, Ian Clarke wrote: > We have a summer project that will hopefully be accepted which will > focus on load balancing, but that shouldn't prevent us from > discussing the issue. > > As I see it, we need some form of "tit-for-tat" system, as > popularized by BitTorrent, where nodes incur credit or debt with each > other every time they send requests to each-other. For example, if > node A sends a request via node B which is answered by node C, then A > incurs a debt to B, which incurs a debt to C - since C answers the > request, it incurs a debt to nobody and thus gets a net credit in the > network as a whole. Note that B's network debt remains unchanged as > it just forwarded a message. > > This approach means that nodes initiating requests incur debt to the > network, while those answering those requests incur credit. I think > we would probably handle inserts in the same way - a node initiating > an insert would incur a debt, but the node where the data gets stored > incurs a credit. > > So we keep track of how much each node is contributing, the question > then is how we bias in favor of nodes that we are in debt to - > considerations are: > > 1) A new node should be given the opportunity to build up credit with > the network, to do this it has to be able to make requests > > 2) We need to avoid a deadlock situation where all nodes are refusing > to talk to each-other > > 3) Ideally, we want to avoid any situation where nodes are just > sitting around waiting for each-other > > 4) We also want to avoid situations where nodes all end up being > forced to make poor routing decisions - as these simply increase the > load on the network by making requests go through more hops - > worsening the overload problem. This is the issue we ran into in > previous versions of Freenet. > > Ian. > _______________________________________________ > Tech mailing list > Tech at freenetproject.org > http://emu.freenetproject.org/cgi-bin/mailman/listinfo/tech > -- Matthew J Toseland - toad at amphibian.dyndns.org Freenet Project Official Codemonkey - http://freenetproject.org/ ICTHUS - Nothing is impossible. Our Boss says so. -------------- next part -------------- A non-text attachment was scrubbed... Name: signature.asc Type: application/pgp-signature Size: 189 bytes Desc: Digital signature URL: <https://emu.freenetproject.org/pipermail/tech/attachments/20060522/c4af904b/attachment.pgp>
