Hi,
I’m writing for two reasons: - The new 10KiB/s minimum bandwidth is too high for some users. - With 1474 few users are able to keep more than 70 connections.¹ To fix these problems, we can reduce the scaling constant used for calculating the target number of peers (this might already have been the better action in 1472 but I did not dare to try it back then). I see two options: A risky one and a safer one. - safer: https://github.com/freenet/fred/pull/559 - risky: https://github.com/freenet/fred/pull/560 The safer one reduces the scaling constant from 12 to 7 and reduces the minimum bandwidth to 7 KiB/s. Nodes with 7 KiB/s will try to get 7 peers. Nodes with 10 KiB will try to get 8 peers. Nodes with 140K upload will go for 31 nodes and nodes with 2.5MiB/s will go for 132 nodes. The more risky one reduces the scaling constant from 12 to 4 and reduces the minimum bandwidth back to 5 KiB/s (where it was before 1472). With this change nodes with 5 KiB/s will go for only 4 peers — one long connection and 3 short ones. This is risky because 5 KiB/s nodes will contribute much less to routing (roughly as much as a 10 KiB/s node did before the link length fix). However it ensures that they will actually be able to keep their connections. In this more risky change, nodes with 10K will go for only 6 connections (2 long, 4 medium or short), nodes with 140K upload will go for 24 nodes and nodes with 2.5MiB/s upload will go for 100 connections. Even the high speed connections which were common in Japan in 2014 will only go for 122 nodes — so few nodes will actually hit the upper limit. With the strong reduction of the scaling constant, requests might also need more hops to reach the target, though I hope that this will be prevented due to more longlived connections. Also, for very fast nodes, FOAF cutoff at 100 (IIRC) already stopped routing from taking into account all peers. I’d like to ask your opinion on these two options. Both options reduce the required peers per bandwidth, so they can serve as a short term fix for the reduced peer count¹. Also they reduce the required minimum bandwidth for taking part in Freenet without heavy connection flapping. But they might be risky since they reduce the outdegree of nodes. I’m not sure whether they actually reduce the effective outdegree, since more nodes will be able to keep their connections stable. On the other hand, the optimization for small world routing should work better when less requests go through huge centralized nodes. Best wishes, Arne ¹: See https://asksteved.com/stats/plot_peer_count.png — this used to have a peak around 100 before 1474, and people report that their nodes cannot keep their connections anymore. _______________________________________________ Devl mailing list [email protected] https://emu.freenetproject.org/cgi-bin/mailman/listinfo/devl
