It used to be interesting to see the news about new ipv6 deployments, adoption ratings etc. Now I basically don't care.
I also don't think IPv6 is a "failed technology" at all. If you've used almost any zeroconf service on an Apple product recently (AirPlay, AirDrop, AirPrint, HomeKit, so on) then you've been almost certainly using link-local IPv6. If you've used Spotify Connect or Sonos, there's a very good chance that's been over link-local IPv6. If you have any Thread or Matter smart home appliances, they generally communicate using IPv6 too. Most devices connect to Chromecasts over link-local IPv6. You've quite likely been connected to IPv6-capable Wi-Fi networks before and you probably never noticed. IPv6 is actually pervasive in all sorts of places.
1. It works 2. The internet today would not function without it 3. There are several IPv6 only deployments (primarily mobile carriers)
It being "interesting" has nothing to do with it being successful. I'm not interested in any new 120v electrical deployments or any new 4g LTE deployments but both of those technologies are very successful.
IPv6 has been mature for over 15 years and is slowly making its way. Most major LTE carriers have deployed it, most home broadband providers have the infra in place, they're just waiting to flip the switch. We're in the boring, uneventful, slow deployment phase.
Meanwhile 45% of Google's worldwide traffic is IPv6, 53% in the US:
All that said I do sort of wonder when or if we'll ever move off dual stack. I think there's a strong argument that running two protocols at once is riskier then the combined risks of the two.
ISPs have deployed IPv6 quite successfully now. Barring some legacy connections (DOCSYS 2 or something mostly) that still get no IPv6, basically all new connections have IPv6 as a default, and old ones have either already been migrated (including DSL) or are in the (sometimes long) rollout process. Some tied the IPv6 migration with an IPv4 CGNAT move, which makes IPv6 very attractive: hobbyists have to ride on IPv6 for external reachability, while casual uses like gaming, voip, conferencing, and whatnot benefit from relay-less p2p connections by preferring IPv6.
For the four major ISPs, this report from 2021 shows 89, 71 and 49% active IPv6 on mobile (not just 5G), with a 1% outlier that should have gotten its act together by now, and 99, 89, 53 and 22% for landlines, and projects 99, 93, 80 and 40% deployment in 2023 (which sounds about right for today in my experience), and 99, 97, 95 and 85% deployment for landlines in 2025, at which point CGNAT won't be an issue for most and IPv4 basically becomes legacy technology provided things also move forward server side.
https://www.arcep.fr/cartes-et-donnees/nos-publications-chif...
I honestly can’t figure out why they don’t, as it has to be cheaper to just do it than make customers call and speak to a representative, given that the technology is clearly in place already. Honestly, though, it’s been over three months and I still haven’t called, as there has been absolutely no reason for me to bother doing so. Like you, I also used to be interested in the rollout of IPv6, but now I just don’t care.
The number of people asking for an ipv6 address on the average ISP can be approximated to zero.
Never is long time. I bet you that they will be providing ipv6 as default in 20 years from now if they are still in business.
I don't care if I have to write angry letters to senators to pass laws forbidding the sunsetting of ipv4, this is 100% an issue I will go down fighting tooth and nail until I die or ipv6 dies, whichever first.
IPv6 is simpler than IPv4 in that you don't need to setup infrastructure (e.g., DHCP) to get going.
First, the IPv4 address space is tiny. Say, a PING packet is 84 bytes, so pinging the entire IPv4 address space will take you all of 336GB of traffic each way (not even accounting for the reserved parts). That can be done in the space of an hour in a decent connection. So things can and are being trivially found by brute force. If a vulnerability is found in something interesting you can bet that by tomorrow lots of nefarious people figured out who can be broken into.
Second, IPv6 has the privacy extensions, which means that even if somebody knows your address today, this has a very limited lifetime to it. Privacy extensions regularly randomize your IP address.
Under IPv6 even normal people can get more address space than can be reasonably scanned. This makes it harder to identify machines permanently, makes it harder to figure out how many machines there are and makes brute force scanning extremely impractical.
With ipv6 privacy extensions, your IP address changes every few minutes (withoin the same /64 network) so it's not a lot different (privacy wise) than today, wher that whole network would be hidden behind a single IPv4 address.
You say this like it's a bad thing.
Why should it be necessary to have to deal with all the extraneous infrastructure of STUN/TURN/ICE just for basic connectivity?
* https://en.wikipedia.org/wiki/STUN
* https://en.wikipedia.org/wiki/Interactive_Connectivity_Estab...
You have your stateful firewall that does default-deny, only allowing in packets that are replies to internally-initiated sessions. If you need to do hole punching you have UPnP or PCP (which also allows for authentication of requests):
* https://en.wikipedia.org/wiki/Internet_Gateway_Device_Protoc...
Alas, publicly traded companies....
I’d expect GitHub to have moved to Azure since the buyout.
The IPv6 adoption seems more tied to Azure/AWS/GCP support than anything else.
It's handy in many situation where you want to assign specific devices specific addresses, or at least make a log of dynamically chosen addresses (e.g., networks with auditing requirements).
Also, does Android even use NTP servers that are advertised over IPv4 DHCP? I don't think so, and if not, then is this really a reason?
In 99.99% of the cases you have to intercept or reconfigure.
Jan 2018 - 513.07
Jan 2019 - 574.02
Jan 2020 - 989.25
Jan 2021 - 1,136
Jan 2022 - 1,207
Looking at Google's IPv6 stats:
July 2019 - 29%
July 2020 - 33%
July 2021 - 36%
July 2022 - 42%
July 2023 - 44%
What's the worldwide average. For the US specifically it's the majority of traffic at 53%:
How are these situations "throwing off" the numbers?
There seems to an underlying assumption in this statement that if you (e.g.) stream a Youtube video from your cell phone (over IPv6) it is "invalid" (less valid) but if you stream it on your desk/laptop from your (IPv4-only) home or work's ISP it is (more) "valid".
> Cellular had no choice but to go IPv6 due to sheer numbers of users.
Of course they have a choice: they could spent millions of dollars getting IPv4 address for their customers, or they give IPv4 shared space [1] to all the devices and spend millions on CG-NAT equipment. No IPv6 needed.
Every IPv6 connection is just as valid as every IPv4 connection.
My alma mater (FAU Erlangen, https://fau.de ) did have some internal and external IPv6 already back when addresses were still 6bone 3ffe::, 20 years ago. Don't know since when the main website has been IPv6-reachable, but it has definitely been over 10 years.
When I asked the IT staff at my university years ago, they said that there are still some very old routers without IPv6 support in use. And since everything seems to work with IPv4 for the university administration, there is no money for new ones.
Will there be an OS layer that can run an app in IPv4 compatibility mode of some sort?
[0]: https://datatracker.ietf.org/doc/html/rfc6877
[1]: https://en.wikipedia.org/wiki/IPv6_transition_mechanism#464X...
Oh boy.
I've been working in IT for 30 years, managed plenty of servers, host services at home, develop FOSS and whatnot but never got interested in IPv6.
And I must say that this is a scary world. Some of my devices suddenly were getting their DNS settings from something else than the DHCP, I had to learn quickly about RA and other anagrams. I was super worried about the exposition of my services - something I completely controlled in IPv4.
With this in mind, I think that IPv6 is too complicated. It does not have that sweet spot between "plug the green cable to the socket called ETH" and "I am going to try to squeeze some extra bits into the datagram".