Re: Anybody Using IPv6?

Nuno Silva <nunojsilva@invalid.invalid> wrote:

On 2025-05-21, Marc Haber wrote:

You don't have the most basic knowledge and still you feel yourself
qualified to judge about the protocol. That's Dunning-Kruger at work.

>
Yet you keep doing the same about judging people. That's hypocrisy at
work.

I just expect people to have a basic education about a topic that they
want do discuss in public. That is necessary to have a discussion on
eye-level¹.

There is zero evidence about IPv6 network speed being slower than IPv4
on a feature-par network.
>
Of course there are stupid ISPs who that send IPv6 on absurd detours.
Those need to be put out of business.

>
Can you provide evidence that such a delay will never happen without
broken routes? Or is it up to the implementation that asks for the
address?

I dont understand the question.

While I don't recall details, I think I've seen and read about this
behaviour too. Only for one case with my computers do I remember it
being a stale route or assignment. For those back in the past I don't
recall much.

We are talking about how a system with both IPv4 and IPv6 enabled
behaves on an IPv4-only network. Such a system will not have IPv6
routes going further than the automatically established link-local
networks and thus any attempts to use IPv6 will immediately result in
the network stack returning a "no route to host" error message.

A well behaved application is then expected to try the next IP address
it might know for the desired communications partner. This applies to
both IPv4 and IPv6. Sadly I don't know at the moment whether this
functionality is implemented in the network stack of whether the
application is expected to implement the necessary logic.

The suggested gai.conf change will, by the way, also only hit at this
stage, just pulling IPv4 in front of IPv6. Most IPv6 averse people who
are looking for reasons to disable it say that it slows down DNS, and
even IF they're right, the gai.conf change doesnt affect this part of
commnunication.

When I am talking about gai.conf here, I actually mean the in-kernel
address label table that is maintained by virtue of the ip addrlabel
command. Most modern Linux distributions only have gai.conf as kind of
a legacy interface that is not necessarily connected at all to the
in-kernel table that the kernel actually uses. I don't know if and
which distributions have code that reads gai.conf and uses the
contents to initialize the in-kernel table at startup time, since my
systems directly interface with ip addrlabel (often via
systemd-networkd).

I think that the local resolver should² also refrain from asking for
AAAA records if the local system doesn't have IPv6, but I don't know
whether this special-case handling is implemented at all. And I'm too
lazy to look that up.

But all this needs to be taken into account before someone can comment
about speed of one IP protocol compared to the other on a level that
is beyond passing myths.

Greetings
Marc

¹ please excuse me if that translation of the German "auf Augenhöhe"
was bad or invalid.
² not in the RFC2119 sense
--
----------------------------------------------------------------------------
Marc Haber         |   " Questions are the         | Mailadresse im Header
Rhein-Neckar, DE   |     Beginning of Wisdom "     |
Nordisch by Nature | Lt. Worf, TNG "Rightful Heir" | Fon: *49 6224 1600402