[Tech] Proposal: UP&P

[Florent Daignière (NextGen$)]

* Matthew Toseland <toad at amphibian.dyndns.org> [2007-03-06 16:56:37]:

> On Tue, Mar 06, 2007 at 12:09:10PM +0100, Florent Daignière (NextGen$) wrote:
> > * Matthew Toseland <toad at amphibian.dyndns.org> [2007-03-06 01:37:52]:
> > 
> > > However:
> > > - It would significantly improve connection reliability. If for example
> > >   all your peers are german and in germany all domestic IPs change every
> > >   24 hours, if you are down for 24 hours you are lost for good.
> > 
> > Do we have any stats regarding how many of our users are double-natted?
> > I know that in france most people had DSL connectivity before the
> > Wireless craze ; meaning that most users are likely to have two
> > different natting appliances (the routing modem and the wireless AP).
> 
> You are quite sure that it isn't possible to break out of a double-NAT?

UP&P is a mess involving multicast and http/xml streams over udp : no :
the multicast won't pass the nat.

> > 
> > > - Right now good connectivity relies on getting a few geek nodes - nodes
> > >   that are directly connected or port forwarded. UP&P would increase the
> > >   proportion of such nodes dramatically.
> > 
> > I am not sure it's an issue. I am idling on #freenet-refs on a regular
> > basis to see how the installer performs and to see where users get
> > stuck : most of them don't have connectivity problems.
> 
> Not initially. The problem here is that somebody gets a few references,
> they're all in Germany so their addresses are recycled every 24 hours.
> They take their node offline for a couple of days for whatever reason.
> They bring it back online and they have no connectible peers.

I don't think that's why people leave... but yes, it's an issue.

> > 
> > > - It would allow for all sorts of bootstrapping protocols. One-time
> > >   references are the tip of the iceberg: Anything that involves giving
> > >   your details well in advance of the actual connection attempt will be
> > >   greatly helped by UP&P support.
> > 
> > Will be greatly helped if UP&P works : indeed.
> 
> Right.
> > 
> > > - It would (mostly) eliminate the need to rely on centralised STUN servers.
> > 
> > it's assuming there is no double nat : again ;)