After three more days of testing, I've started to come to the delusion that the number of references to dead nodes is really causing a problem.
Typically, data requests are successful if the data is within 5 to 10 hops, but after that it is almost impossible to retrieve anything. The number of QueryRestarted returns goes way up, incidentally meaning that the request takes forever to finish, and is very rarely successful. Even after waiting for data to propagate, the barriers put up by dead nodes don't seem to yield. No convergence. Considering the natural tendency of people to just check things out without making a serious commitment to running a node, I think that the strategy, at least for now, has to be very aggressive at eliminating unresponsive nodes from the references. Otherwise, we're all going to be stuck in little islands, only able to insert to and retrieve from a few, reliable, local nodes. The other problem that I have noticed is low performance nodes somewhere in the chain of a moderate to large hop retrieval, latency in these situations is a -big- problem, but not one for which I can think of a solution. Interestingly a data retrieval like that will hang, but if I interrupt it, and then try again in a few minutes, often the data is close enough to get it easily. The data send eventually gets there, but there is no point in letting the client sit and wait. Interesting, but time consuming, trying to figure these things out. I just hope the feedback is useful. David Schutt _______________________________________________ Freenet-dev mailing list Freenet-dev at lists.sourceforge.net http://lists.sourceforge.net/mailman/listinfo/freenet-dev
