Re: [v6ops] draft-ietf-v6ops-multihoming-without-nat66 WGLC

[Fred Baker <[e-mail]>]

On Feb 28, 2011, at 12:19 PM, <[e-mail]> wrote:

> For what I know multihoming is very commonly used to describe a scenario where a host has simultaneously multiple uplink ISP connections active. 

There are 185 RFCs and I-don't-know-how-many Internet Drafts (that may or may not ever become RFCs) that discuss multihoming. And five IENs.

The oldest use of the term I could find is in IEN 110, dated 1979. It uses the term without definition (it must have been in the air at the time), but states that

   3.  Hosts which are deliberately multihomed on distinct networks
   should be able to recover from interface failures, but by mechanisms
   above the TCP/internet layer, not within them.

So, clearly the expectation there was that "multihomed" implied "[multiple] distinct networks" and the ability for a host to recover from failures.

IEN 178 has a section with that title. It's introductory comment is that

             A subscriber  may want to have multiple  connections to a
        communication  system  for reliability or performance reasons.

It goes on to describe having multiple addresses, the ability to use any of them, and the ability to optimize its use of them, among other things.

The first definition I find that can be described as a "definition" is in RFC 1122; RFC 1123 continues the discussion.

         A host is generally said to be multihomed if it has more than
         one interface to the same or to different networks.  See
         Section 1.1.3 on "Terminology".

And by the way has a second 3.3.4 on "local multihoming". Since RFC 1122 is "host" requirements, it doesn't discuss the multihoming of a network. What it addresses (no pun intended) is a host that is part of two or more subnets, whether that means two or more interfaces or one interface with multiple subnets.

RFC 1127, which is Bob Braden's "Perspective on the Host Requirements RFCs", goes on to discuss "AS Multihoming", which he defines as 

   multihomed AS: an AS that has more than one connection to other ASs,
      but refuses to carry transit traffic.




Bringing this back to the draft in question, my understanding is that the authors are working from the model in which:

a) a residential/SOHO/other-edge-network has multiple upstreams
b) said network has at least one CPE
c) the multiple upstreams provide a prefix each
d) as a result, the residential/SOHO/other-edge-network has multiple prefixes overlaid in it. 

Every host, by RFC 1122's definition, is multihomed (it has more than one subnet, regardless of physical connectivity), and the network is multihomed (it has more than one upstream.


Did I miss something? Randy's description of multihomed networks he runes is perfectly accurate. But BGP is not a requirement for multihoming; what is a requirement is "multiple addresses" or "multiple upstreams".