I have observed a strange problem where the RNF percentage rises as the troughput of my nodes increases. This seems to relate somehow to overloading the routing table - i.e. at about 50 000 requests/hour and 400 000 KB/s of outgoing traffic all local requests start to RNF. Also the success percentage of the routed requests declines. I have tried to increase rtmaxnodes, but it doesn't seem to help.
I have not found any good description of the routing table and how the variables rtmaxnodes rtmaxrefs maxnodeconnections affect the routing table or if there are any other relevant configuration options that should be tuned according the the traffic a node is processing. Increasing both maxnodeconnections and rtmaxnodes leads to freenet running out of memory relatively quickly, despite having 416m of java heap to go with. I'm not certain, but seems strange that moving from default 100 rtmaxnodes and 200 connections to 220 and 110 results in so much increased memory usage - with the defaults the nodes can run well with at least 384m of RAM. A bug/a different default? Using freenet stable 5106, Sun java 1.5_b05. Compiled freenet using ant on the same java JDK.