So we need to extend the UPNP protocol so that when the local NAT box receives a request to open up an external port, it relays the request to the carrier NAT. Or we could do what we did last time and pretend that nobody will deploy carrier grade NAT if we don't specify a way that it can work without pain. On Sun, May 30, 2010 at 11:02 AM, Arnt Gulbrandsen <arnt@xxxxxxxxxxxxxxxxxxx> wrote: > On 05/30/2010 04:44 PM, Sabahattin Gucukoglu wrote: >> >> BitTorrent is popular, yes. People at home *are* behind NAT boxes, with >> all the usual pain that implies *. It's just that BitTorrent, being a >> straightforward TCP protocol with no embedded payload addresses **, can >> operate behind NATs, and those NATs can be configured either manually or >> automatically by users or their client software ***. If the NAT should move >> to the ISP, it seems possible that this is no longer true. > > Not quite. > > 1. Bittorrent clients connect to each other via TCP. Each connection is > incoming at one end. Torrent clients mostly use UPNP to accept incoming > connections. > > 2. UPNP is an ethernet-level protocol (it uses UDP/IP broadcasts), so it > works only if the USER is on the public internet. Hence, NAT within the > user's network is now very different from NAT within the ISP's network. > > That's why I said the wide popularity of bittorrent shows that USERS are on > the public internet. > > Arnt > _______________________________________________ > Ietf mailing list > Ietf@xxxxxxxx > https://www.ietf.org/mailman/listinfo/ietf > -- Website: http://hallambaker.com/ _______________________________________________ Ietf mailing list Ietf@xxxxxxxx https://www.ietf.org/mailman/listinfo/ietf