[v6ops] Re: 464XLAT-only networks

[Lorenzo Colitti <[e-mail]>]

On Sat, 26 Oct 2024, 20:24 Soni "It/Its" L., <[e-mail]> wrote:

> I appreciate the "what can we do to drive adoption" sentiment, but I think
> deploying IPv6 without global connectivity is generally*worse* than staying
> IPv4-only. This is because it requires apps to code workarounds such as NAT
> traversal that are not needed on the vast majority of IPv6 networks. It's
> better to deploy IPv4 only and let those hacks die with IPv4.
>
>
> "NAT traversal" is usually used to describe firewall hole punching, which
> is very much still needed in IPv6.
>

It's not the same. NAT traversal requires that apps be able to deal with
the IPv6 assigned to the device not being the same as the address seen by
the server. With IPv6 end-to-end connectivity, this doesn't happen.

it's also nearly entirely irrelevant on a server. so aside from the
> hypothetical DNS lookup latency issue when suffering from packet loss, this
> is a very good way of running IPv6-only software on an IPv4-only network.
>

What IPv6-only software? Who write software that doesn't work on more than
half of the Internet?

Oh, and don't forget the latency penalty incurred on outgoing connections
as the stack tries IPv6 and falls back to IPv4.

any reason why we shouldn't make our customers use a SIIT-DC on their
> IPv4-only networks?
>

It's more complicated and less reliable than IPv4 NAT, more difficult to
operate, imposes latency penalties, and requires that apps deal with
partial reachability. And it provides no advantages over IPv4 NAT. Could
you make it work? Probably. But why would you want to?

>