On Sun, Nov 17, 2002 at 04:10:12PM +0000, Matthew Toseland wrote: > On Sun, Nov 17, 2002 at 02:03:07PM +0100, Oskar Sandberg wrote: < > > > ...that are not running a NAT or who know how to configure it (which > > rules out the popular NAT boxes, windows connection sharing etc etc). > > And just having broadband rules out the overwhelming majority of the > > Internet's population already - the countries with the highest broadband > > penetration are still in the twenties. > > Yeah, blergh. Aren't there supposed to be some protocols for talking to > NATs?
There are some RFCs, but it isn't implemented anywhere AFAIK. Anyways, I wouldn't hold my breath - SOCKS for example already contains a bind command that allows the proxy to accept connection for hosts inside it, yet almost all SOCKS proxies I have run into don't allow that command. The client/server Internet is very entrenched at this point. <> > > I have experimented quite a bit with this. After the first > > announcements, it takes about 24-48 hours for the node to establish > > itself and start getting a constant stream of traffic. If the node goes > > down for an extended time during this period, it is pretty much back to > > square one. Once a node is established, if it goes down for about 10-12 > > hours or so, it seems to take about 16 hours or so for it to return to > > it's previous traffic. (I have a plan to use nukes on the moon to slow > > the earths rotation, but that is not planned until 1.0.) > > Hmmm. What's wrong with moving non-working references to an "inactive" > secondary routing table and polling for the next ARK insert? And then do what when the node comes back online? Restore all it's references removing any that were put in while the node was gone? There are a maximum of 2500 references in the routing table, which means they can be expected to be flushed completely for an active node during a single night. How do we figure out which to restore and which remove when we have nodes blinking in and out? Freenet doesn't just work locally, it is based on global effects. We really don't know how to try compensate for a node being missing from the network for a long time (note that the observation that transient nodes function worse then permanent ones indicates that the network shifts quickly and that routing tables need to be constantly updated.) > > > > (d) Not starved for bandwidth or other resources. > > I disagree. Nodes limited to 5K/second are still useful as long as they > can actually serve 5K/second (these are probably the majority of > permanent nodes). Especially w.r.t. splitfiles, and small files in > bunches (i.e. most freesites). The only thing that loses is single files > of several hundred K. Latency, on the other hand, matters a lot, so > again we're talking unmetered broadband, which is not that rare. The global mean according to my node is about 6000 qph. That is about 1.6 queries per second. On my node, a query averages a little over six messages (sent and received) which should add up to 3 kB or so. And then add to that that my node establishes and receives ~13000 connections per hour at 10kqph, so at least half, or about 2 per second, should be expected for the mean node. That should be more than enough to fill up your 5K/second without ever sending any data. > > 30 MB??? My freenet node alone eats several gigabytes a day - and I > > never use it myself. It also uses between 35 and 50 megs of RAM, and > > about half the CPU of a PIII 600 that is dedicated to it. (And all that > > P3/600 is about 1/4 the speed of modern chips, by megahertz alone. But it probably represents about the median system. And most people are not willing to dedicate the machine to freenet like I am. <> -- Oskar Sandberg oskar at freenetproject.org _______________________________________________ devl mailing list devl at freenetproject.org http://hawk.freenetproject.org/cgi-bin/mailman/listinfo/devl
