On Wed, Oct 24, 2012 at 10:30:21PM +0200, Claudio Jeker wrote: > On Wed, Oct 24, 2012 at 10:12:33PM +0300, Jussi Peltola wrote: > > On Wed, Oct 24, 2012 at 02:43:14PM -0400, Simon Perreault wrote: > > > What you need to multihome is either BGP or NAT. Exactly as in IPv4. > > > Nothing has changed. The only new thing with IPv6 is that there's > > > more bits. > > > > Oh? I have two internet connections plugged directly into my desktop box > > at home, it is multihomed and there is no BGP or NAT. This does need > > some policy routing to work with uRPF filtered access lines. > > This is just the tip of the iceberg.
This is a very common setup for bastion hosts with multiple dsl lines for redundancy; it is extremely robust against all kinds of failures, unlike other forms of multihoming. > > With IPv6 multihoming should work trivially: plug two access lines into > > a switch, get RAs from both, get addresses from both on your end-host, > > and your end-host needs to select the proper route for each source > > address. Again, no NAT or BGP. Applications will need to support hosts > > having multiple addresses in the future, and happy eyeballs seems to > > have made browsers do that. > > Ha ha ha ha, this will work for a single host but how will you manage > multiple ones. Bonus question, how do you think the host router with no > knowledge of the underlying network topology will choose a route? > This setup is one of the biggest mistakes made in IPv6. Roaming single hosts are a very large subset of all hosts; server-type systems usually have static configuration anyway. I really don't see how multiple hosts wouldn't work if one does... > > There is also a considerable advantage against "multihoming" where hosts > > only have 1 address configured: if the application tries to use all > > source addresses available, you can get to google even if one of your > > access lines has no connectivity to them; with BGP multihoming you will > > not, with v4 NAT style multihoming you possibly can if it does > > round-robin and you try again. > > > > Add SCTP to this puzzle, and you should be able to roam seamlessly from > > WLAN to 3G to WLAN without your ssh sessions breaking. mosh already more > > or less does this. With multiple addresses and default routes per host, > > and SCTP or multipath TCP, you should also be able to load-share one > > connection among multiple internet connections. > > Hey, you forgot to mention shim6 and all the other crap ideas that already > died. SCTP is a monster and it is over engineered like IPv6. I wonder when > the first SCTP hacks will apear that take down host and maybe networks. > If I want persistent login sessions I use tmux. Yes, with a while loop trying to ssh and re-attach to screen or tmux forever you can get pretty close, as with web apps that do transient http requests. Again, this has absolutely nothing to do with ipv6; exactly the same problems and solutions exist on ipv4. > > End hosts need to get smarter, instead of the network adapting to their > > stupidity. But I'm not holding my breath. > > Nope. End hosts need to stay stupid. They can not handle the truth their > poor little mobile cores would just explode the moment they try to grasp > the real world. What exactly is your proposal? Infinite DFZ growth?
