Re: [homenet] Configuration must not be carried by the routing protocol

[Michael Thomas <[e-mail]>]

On 06/26/2013 03:56 PM, Juliusz Chroboczek wrote:
>
> do we need to invent a separate distributed database for each class
> of information?
> Yes, because different kinds of information have different
> requirements.  There is information that needs to be reliably
> synchronised in a timely manner (the link-state DB), there is
> information that must be distributed in a timely manner but need not
> be synchronised throughout the network (configuration information),
> there is global information that is too large to be distributed except
> on demand (DNS), there is local information that is not distributed
> except on demand (mDNS), and probably many other kinds that I'm missing.
>

I'm not sure that I have an opinion on this debate, but I'd like to
point out that there are probably other classes of information that
have little to do with routing, but a lot of desire to be replicated
across a homenet's routers. For example if I have a DNS repository,
I'd sure like to be able to take a dead router, throw it in the trash,
plug in a new one, all the while where my network still was serving
up my names.

DNS in particular has ways to transfer the zones, but I'm less certain
about whether there's anything that would orchestrate that hot standby
kind of desire, and I'm even more dubious that it would satisfy zero/littleconf
requirements even if there were.

Picking ULA's seems to be another kind of problem that needs some
kind of config replication/coordination too at the router level if it's
to be zeroconf.

Whether routing protocols would be a good fit, I don't know but it seems
like an odd way to think about the problem to me.

Mike