On Wed, 2004-05-05 at 09:59, Pekka Savola wrote: > On Wed, 5 May 2004, JORDI PALET MARTINEZ wrote: > > That's true if both hosts try to use the same Tunnel server. > > Obviously, as we want to implement a discovery protocol, these would > almost always be the same. Unless the protocol does load balancing, but we must assume that there could be multiple hosts behind the same NAT and talking to the same server, may that be a TB, teredo, dstm.. whatever. > > But I also believe that when there are several host behind a NAT, or > > in the same LAN, is more interesting to use proto-41 or any other > > mechanism to provide a single prefix to all the LAN. It provides a > > lot of advantages. > > Obviously, but may be non-trivial to set up, and might not be > supported at all by the non-registered mode (if it only provided a > /128, which would remain to be seen). Next to that there are ISPs who NAT their complete userbase, say 100k hosts behind a single IPv4 address. It is simply not possible to put up an RA behind it, well it would work but for that single part of the LAN. > > It should not be very difficult, as part of the implementation of > > the tunneling protocol (TSP or whatever), detect this situation > > automatically and even configure one of the host as "router" for the > > rest of the network. > > Actually, I would argue that this _would_ be very challenging. Note > that the NAT may be done by the operator as well, not just the user. > It would be simpler to try to figure out whether there are other v6 > nodes behind the same NAT, but that's deep magic and failure-prone as > well. Also that would make one host suddenly be a 'transit' for the other hosts, automatically. I don't think many users will like this setup. Paying for others, especially with bandwidth/traffic limitations is not something most people like to do. > Proto-41 forwarding seems work within certain specific scenarios > (i.e., the only NAT is done by the customer's CPE, a host would be > willing to serve as an IPv6 router, and that a prefix delegation is > available), but it doesn't seem to be possible to generalize its > detection and use for the protocol. I even would stretch it as far as saying: directly connected: proto-41 NAT: udp Just to keep it simple. Anyone written a IPv6-over-UDP draft yet btw? Greets, Jeroen
Attachment:
signature.asc
Description: This is a digitally signed message part