TLD <[EMAIL PROTECTED]> writes: > Is message send time a problem? I mean, AFAIK freenet is able to recognize > links with higher latency and use them as little as possible, thus reducing > the outbound traffic over those links in favour of "local" > (=not-so-limited) nodes.
I think message send time IS a problem, but only by observation. And my observation is that when messagesendtime goes up over 1000 ms my node starts rejecting most incoming connections and at 2000ms basically blocks everything until things settle down. I think this may be a separate issue from lower level latency. In any event, consider my node which has 10 connections which I'm happy to run at a shared 256 kbps and 40 connections which I (financially) need to cap at a shared 0.75 kbps. What my node seems to do now is look at my total message send time, and stop all incoming requests. Or to put another way, how does a node cope with connections with orders of magnitude differences in bandwidth? Not very well is my observation. > The other two possibilities, namely lower bandwidth for all and an add-on > to fred, look uninviting: the first because it's just sub-optimal, the > second because both it requires much work on fred (to implement the > different bandwidth levels and to test them -- how many nodes would benefit > from that?) and, for those who need the feature, does not significantly > reduce the amount of configuration work (compared to a QOS system). > > Please correct me if I'm wrong! :) Well I was thinking any management would have to be external to Fred but Toad at least seems to consider the possibility. I guess I'd like to see two channels. The other thing I've thought about is running two nodes - with a shared datastore. One only talks to high bandwidth domestic IPs and the other only talks to low bandwidth IPs. I'm nto sure if that is a good or bad idea. Oh and Toad if you are listening: rather than a humongous CIDR list of IP blocks, I'd also be quite happy with a DNS query like 1.0.0.127.zz.countries.nerd.dk which returns a number code for a country. Cheaper and nastier, but given the relatively small number of domestic IP addresses contacting me as a Freenet node (versus a list of all possible IP addresses) it may be easier to implement. It does make use of a third party service, but if the fallback position is to assume it's a high bandwidth cost IP, it's of no harm if it fails and may do some good. _______________________________________________ Support mailing list [EMAIL PROTECTED] http://news.gmane.org/gmane.network.freenet.support Unsubscribe at http://dodo.freenetproject.org/cgi-bin/mailman/listinfo/support Or mailto:[EMAIL PROTECTED]
