[ih] History question -- multi-homing

[John Day]

For years in my courses and when relevant to a talk, I always tell the story of the cold day in February 1972. One morning Grossman came in late from a ARPANET PIs teleconference and told me that Tinker AFB wanted to join the Net and wanted redundant links. I started to say, “O great!” But stopped myself mid-syllable and instead said, “O, crap! that won’t work!” But immediately knew that the solution was to route to the host, not just to the interface.

We left it there. We had work to do, code to write. It was so obvious others would see it as well. We didn’t write up an RFC on it or anything. But it was clear that multihoming committed a network architecture to routing to the node, not the interface.

I guess we should have written something up.  

> On Sep 12, 2026, at 02:04, Tony Li <tony.li at tony.li> wrote:
> 
>>>> Destinations are not the issue, sources are.
>>> 
>>> Not really.  They are all addressed to the same address, the destination.
>> 
>> They are because when I change interfaces on current hosts, the host’s source address changes. That’s breaks all sorts of things (or requires the gymnastics of MP-TCP, e.g.).
>> 
>> If you recognize that having two paths to the rest of the world means a router, all that mess goes away.
>> 
>> We implemented that when we double-homed hosts throughout DIv7 on both Ethernet and Myrinet - that’s what the 1997 paper describes.
> 
> 
> If you recognize that hosts need addresses, then the problem also goes away.
> 
> Tony
>