On Friday 19 July 2002 19:56, you wrote: > > We seem to be getting a lot of RNFs when inserting, even at a HTL of 10. > This doesn't seem to make sense given that there are clearly at the > very least about 70 working nodes in the network (judging by my node's > routing table). > > Does anyone have any theories as to what might be causing this? > > I wonder can we use the Watchme network to test this... > > Ian.
Hi Ian: I think that there are at least 4 issues at work: 1) I added code to limit the maximum number of noderefs a request will try before giving up in order to cut traffic. That was done a while back (i.e. build 467). The default minimum number of steps is 10. Refs which are skipped because of CP or because of routing backoff (QR'd in the last minute) *are* counted. So the node can easily RNF if all the "right" routes are too busy or have poor transport level connectivity. The rationale was that if all the network is so busy that all the "close" noderefs are busy, there's no point in further increasing traffic by continuing to route. The default cutoff value 10 was a guess. Maybe it's too low, but I don't thing so. We really need stats on request success rates as a function of routing attempt..... 2) The overload prefiltering code discriminates against requests which are far from the node's estimated specialization when the node is overloaded, making them more likely to be QR rejected. 3) The routingTime based load estimate used to determine node load seems somewhat flaky. I'm not sure if the problem I'm seeing is specific to my node. Under load my node routinely has hourly mean routingTime values in excess of 2000ms. I often see it rejecting all incoming requests because of the routingTime load estimate even though there are many free threads. Is anyone else seeing this? I haven't had time to look into it. I'm not proposing that we get rid of this. I think it's a good idea. We just need to make it work better. 4) Transport level connectivity seems worse since 481/482. I come to this conclusion by looking at the "Successful Connections field" of the Routing Table Status. e.g. http://hawk.freenetproject.org:8889/nodestatus.html I don't know what this means. Maybe newer builds handle more requests, so they are more likely to be overloaded, or maybe something in the new threading code is causing the newer builds to time out more easily. I almost never see my node fail to answer incoming connections, which leads me to suspect that the problem is most likely with timeouts, not outright connection failures. --gj -- Freesite (0.4) freenet:SSK@npfV5XQijFkF6sXZvuO0o~kG4wEPAgM/homepage// _______________________________________________ devl mailing list [EMAIL PROTECTED] http://hawk.freenetproject.org/cgi-bin/mailman/listinfo/devl
