A more complete discussion can be found at (surprise!) Wikipedia: http://en.wikipedia.org/wiki/Multihoming

I'm leaning more toward John's position, but it's still not entirely clear to me.

I think that article is pretty crummy, even by Wikipedia standards.

A multihomed host sits in two different subnets. By that definition, Trevor's box is multihomed. So is my Linux box at home with one Shaw connection and one private LAN connection -- it has a routing decision to make, albeit very simple. Two connections to the same L2 network like described in the wikipedia article isn't multihoming it's layer 2 redundancy. Don't even get me started on their "two addresses, one interface"

But I haven't heard that definition used in years. Nowadays, multihoming implies two different carrier connections and BGP, giving both outbound **and inbound** redundancy (DNS-fu doesn't count).

Put another way, at my job we've got two big connections to the same carrier (active/passive), using BGP to handle the failover and route advertisement, and we still don't describe ourselves as multihomed.

Sean