Re: Automatically connecting stub networks...

[Ted Lemon <[e-mail]>]

On Jul 12, 2020, at 5:25 AM, Simon Hobson <[e-mail]> wrote:
> So take a large(ish) house, and because ti was convenient for the telco or developer, the phone/internet line comes into one end of the house. User gets some (say) home automation stuff, and the hub doesn't connect reliably to the main WiFi. User happens to have a repeater (which is actually a stub router, but the user doesn't know that) which is able to connect to the main WiFi and provide a signal that the HA hub can connect to. So HA hub is a stub router, with another stub router as it's upstream connection. The user, who knows nothing about routers and "technical stuff", expects it to work …


This is not a use case for stub networks. If the user does that, then the stub network doesn’t work. This is actually a better outcome than what we do today, which is to double-NAT and then have weird and unexpected behavior that doesn’t match the user’s mental model. Sure, the user might be disappointed that when they plugged these two things together, they didn’t work, but at least they didn’t sort of work.

This whole topic has been discussed at some length in the Homenet working group. Homenet came up with a solution to the problem you’re talking about. Nobody implemented it—it was too complicated, and not baked enough. Meanwhile, the market moved on: nowadays, if you want to address the use case you are talking about, you buy one of a variety of Wi-Fi mesh routers. You plug them in, and they take care of this problem. If you want to extend your network, you buy another mesh hub. Your Home Automation network plugs into whichever hub is nearest—you don’t even know or care which hub it’s talking to.

I’m not in love with the home Wi-Fi mesh solution, but the one thing I can say for it is that it got the job done. Nobody is wondering how to set up home networks in large homes anymore.  But we definitely do have a problem of what to do when we want to plug a home automation network into a Wi-Fi network, for example.

It’s quite possible that, if we decide to standardize a way to do stub networks, that later on we may want to explore building hierarchical networks. I’m on the fence about that—the reason we went with mesh in homenet rather than hierarchy was that people thought it wasn’t sufficiently general, and would fail the “mental model” test. So that’s a question the IETF has already considered and decided, and the decision did not go in the direction you’re suggesting. If we want to re-open that issue at some point, I would not object, but right now I’m trying to address just this very specific problem, which I think is a problem that we need to address in the short term, for which the market doesn’t already have a solution, and, importantly that we _can_ address in the short term.