On 2009-03-23 08:26, Iljitsch van Beijnum wrote: > On 20 mrt 2009, at 14:40, Brian E Carpenter wrote: > >>> NAT does not offer ANY multihoming benefits whatsoever, in fact, NAT >>> breaks multihoming because after a rehoming event, the addresses are >>> translated differently. > >> It's correct that NAT changeovers break existing sessions. But your >> blanket >> statement isn't true. NAT-based multihoming has the major benefit that >> the number of extra BGP4 routes caused by a multihomed site is exactly >> zero. > > No. What you're talking about is multiaddress multihoming.
That's true too, but it isn't the same scenario. If it's NAT-based, the site can use a nice home-made ULA prefix and never has to think about it again. Multi-prefix based multihoming doesn't have that convenience factor for the site's IT manager. See draft-carpenter-renum-needs-work for some of the consequences. > > Then you add NAT to hide the changes to addresses from the hosts. But > IPv6 hosts can work with multiple addresses anyway (well, there's the > ingress filtering issue) so NAT is largely orthogonal to the multihoming. In fact, there's the exit router selection issue as a result of the ingress filtering issue. Certainly a site with many exits gets that problem in any case, but I suggest that it's less acute in the NAT model because in the end, any exit point will do. > > Also, shim6 gives you actual multihoming where sessions survive rather > than the watered down thing where you only get to reestablish new sessions. Correct. That's why we're standardising shim6. The question isn't there; it's about what gets deployed. > >> Also, NAT-based multihoming has value for large international corporate >> networks with dozens or hundreds of interconnection points to >> the public network. It basically solves their address management >> problem when dealing with multiple ISPs in multiple locations. That's >> running code today. > > People run whatever they can get away with. Doesn't mean it's a good idea. > > However, I do agree that it's useful to have stable internal addressing > when external connectivity is subject to change. That is a legitimate > advantage of NAT (66) which we haven't managed to make work without NAT. > We could though, by making sure that ULAs are used for local > connectivity regardless of the external connectivity. Yes. So how can we persuade IT managers to adopt that as standard practice? > > On 21 mrt 2009, at 16:07, Brian E Carpenter wrote: > >> Suppose you're operating a large international network with (to take >> a random example) IPv4 1/8 as its PI prefix. > >> You can't just advertise 1/8 in BGP4, because in fact it is split >> up into many longer prefixes for various kinds of use and various >> geographies. > > Then what is the point of having a single prefix? Mainly historical, or to say it another way, a large corporate network acquires its own routing swamp over many years. Suppose you sell a department of the company off to another company, for example, but the cost of renumbering is considered too high? (I am not making any of this up, although 1/8 is an example.) > >> So how do you connect your internal users to the Internet? > > Same way as everyone else, return the /8. Not if you want to do traffic engineering, so that traffic for the Hong Kong office doesn't enter the Internet in New York. > >> You have (I'm making this up) 100 different interconnects to the >> public Internet around the world, across a variety of ISPs. If you >> advertise longer prefixes out of 1/8 through those ISPs, life gets >> highly complex if you want multihoming. Certainly you won't be able >> to advertise *all* those prefixes through *all* those ISPs, so you'll >> need >> a complex worldwide management system just for your BGP4 advertisements, >> to decide which prefixes are advertised where, and what the desired >> backup >> paths are. It can be done, but the OPEX is high. > > Cost for the community is also high because a single organization puts a > whole bunch of prefixes in the routing table. Yes > >> So instead, you run NAT at every ISP connection. > > Ok, I said they didn't need the /8 before, but now you've completely > lost me. What is the point of having that prefix now?? None, by now; it's become a private swamp. Brian _______________________________________________ Ietf mailing list Ietf@ietf.org https://www.ietf.org/mailman/listinfo/ietf