Re: [Rift] IPv4 vs IPv6

[Tony Przygienda <[e-mail]>]

Aha ok, what you say is reasonable. I was steered here originally by the
fact the even cheapest stuff today does v4 and if we have v6 support on the
fabric v4 forwarding over v6 nexthops is possible at basically zero config
cost and you have a great v4 to v6 migration story. but with the angle you
propose I see that writing "SHOULD support v4" would be better assuming
silicon will become v6 only over years and v4 will vanish.

We can't write of course "MUST support v6" since that makes the stuff
undeployable in foreseable future albeit we all wish the world would have
moved on for last 20 years :-}

Other opinions?

--- tony

"

On Thu, Jul 18, 2019 at 2:22 PM Mikael Abrahamsson <[e-mail]> wrote:

> On Thu, 18 Jul 2019, Tony Przygienda wrote:
>
> > Please dig out the archive where that discussion was brought up by Alvaro
> > and follow up there if you still have doubts.
>
> Since I cannot reply to that thread (I just subscribed to the list so I
> don't have an original email to reply to), that's hard.
>
> Anyhow, I think the IPv4 MUST should be removed. It's 2019 now, we should
> not mandate IPv4 support in anything.
>
> > not to the requirements _of devices runing RIFT_:  we cannot mandate
> > that everyone in reality will run IPv6 forwarding on the fabric so the
> > reality is that every silicon does IPv4 today and some does IPv6. And
>
> This fact doesn't warrant making IPv4 a MUST. Think 5-10 years down the
> line, there might be IPv6 only silicon and control plane. Why should
> future implementors be required to implement IPv4 if they do not want to?
>
> > then IPv4 over IPv6 is reality while IPv6 over IPv4 is not (since IPv6
> > does ND anyway so why would you).  So it is very good for v6 here to
> > basically make it an implicit transport for v4 (by saying running v6
> > does _automatically_ give you v4 forwarding) while making sure that v4
> > being-de-facto-what-e'one-can-do is the lowest common denominator of the
> > router's running RIFT. RIFT being only usable if a router/switch can do
> > IPv6 is simply not a practical proposition AFAIS.  We could arguably
> > split it into a "RIFT router/switch requirements document" but that
> > seems just paper bloat ...
>
> I am not proposing making IPv6 a MUST, but I am against making IPv4 a
> MUST. Better to be silent on the matter in that case so future
> implementors can decide for themselves.
>
> I just made me cringe when I read the IPv4 MUST in the document.
>
> --
> Mikael Abrahamsson    email: [e-mail]
>