On May 16, 2003 02:46 pm, Toad wrote: > On Fri, May 16, 2003 at 12:10:57AM -0700, Ian Clarke wrote: > > -----BEGIN PGP SIGNED MESSAGE----- > > Hash: SHA1 > > > > On Fri, May 16, 2003 at 01:15:49AM +0100, Toad wrote: > > > Observe > > > http://127.0.0.1:7888/servlet/nodestatus/diagnostics/incomingHopsSinceR > > >eset/hour > > > > > > Mine shows nothing over 1.94, many under 1. > > > > > > Hence, pcaching is not working, presumably because there are too many > > > old nodes on the network. Hence, pcaching is not responsible for the > > > current problems! So upgrade already! > > > > Well, we don't need to infer how many old nodes are on the network, > > we can look at the node version histogram - which tells us that > > people are actually keeping pretty up-to-date. > > Hrmm. Well then another possible explanation is that very many nodes are > underloaded, and so the request won't get far before it reaches a node > that resets the datasource... load balancing interfering with > probabilistic caching... hrmm. I am reluctant to weaken the load > balancing, because it would seem a reasonable way to try to reduce the > network imbalance... the other possibility is to not (or not usually) > reset hopsSinceReset on a datasource reset... which means that a node > can distinguish a reply from the original datasource from a reply from a > node that just happened to reset the datasource... which probably isn't > a good idea.
Does load balancing take bandwidth limits into account? If not its going to think some nodes are under utilized when they are working as hard as their limits will allow. Ed Tomlinson _______________________________________________ devl mailing list devl at freenetproject.org http://hawk.freenetproject.org:8080/cgi-bin/mailman/listinfo/devl
