On Aug 10, 2017, at 5:48 PM, Markus Stenberg <markus.stenb...@iki.fi> wrote:
> - 3.3
> - it implies that homenet exposes DNS outside home (by default?) and uses
> instead custom dns server logic to handle .home.arpa from ‘outside’; why not
> just firewall it and be done with it (or listen only on e.g. ULA prefix)
No, it doesn't say that.
> - why filter out global IPs?
Because if you don't, then when you lose your global prefix, you lose access to
> - 3.5 (PVD madness)
> - WHY? can’t we get just rid of split horizon DNS madness and use _a_ DNS
> instead of N DNS servers?
If you tell me how to implement that, I will be excited. Otherwise, not
solving this problem will produce brokenness. The simplest way to solve it is
to have the network advertise only one external prefix on the homenet if the
homenet is multiply-homed. Is that your preferred solution?
Also, this isn't necessarily split-horizon madness. It's valid for the same
query to yield different answers; this is operationally normal, and trying to
do something to stop it in homenet isn't going to work. Like you, I would
prefer to do CDNs in a way that doesn't involve all of this brokenness, but if
you want to watch Netflix on your homenet, we need to make this work.
> - round-robin = bad (think why happy eyeballs came up for example of why)
DNS resolvers use round-robining. That's how the protocol works. I can
think of ways to improve on that, but they all involve changing the DNS packet
format. So I don't think that's in scope. The draft just specifies how DNS
round robining should work in the context of mpvd on a homenet—it didn't invent
homenet mailing list