Doesn’t remove the need for nat - my wired IsP might be able to bgp with me, but my backup 5g won’t, and when I want to choose which to send my traffic through with PBR that means natting.
My router doesn’t support 64, so I have to use my isp’s which is speed constrained compared with native 4. Ok that’s on my setup. Haven’t tested my 5g provider and where 64 occurs, I’d hope in their network, but how do I configure my dns64.
Still need to provide v4 at the edge and thus 46 nat so I can reach internal v6 only servers from v4 only locations
Perhaps lost of that is because my router doesn’t do 64, but again that just shows that v4 is still essential. I haven’t found a single service that’s v6 only, so if I have to run a v4 network (even if only as far as a 64 natting device) why bother running two networks, double the opportunity for misconfiguration and thus security holes. Enabling dual v6 on my IoShit network would allow more escape routes for bad traffic, meaning another set of firewall rules to manage. Things like SLACC make it harder to work out what devices are on the network, many end user devices are user hostile now and keeping control of them on v4 alone is less work than in v4 and v6.
Yes, it does. You just have each of your routers (wired and 5G) advertise the /64 prefix delegated by each of your ISPs. Your hosts will self-assign a v6 address from each prefix.
To control which link the traffic uses, you just assign router priority in the router advertisement (these are all standard settings in radvd.conf).
> Things like SLACC make it harder to work out what devices are on the network
Again, not true. If you really don’t trust your devices, then DHCP isn’t going to save you. Malicious hosts absolutely can self assign an unused v4 address, and you’ll be none the wiser if you just look at your DHCP leases.
> To control which link the traffic uses, you just assign router priority in the router advertisement (these are all standard settings in radvd.conf).
Have you done this? Did it actually work for you?
When I tried it, clients would regularly send to router B with an address from router A, and often ignore the priorities. As I understand the RFCs/client behavior, the router priority field is only relevant if multiple prefixes are in a single advertisement, otherwise most recent advertisement wins.
Once you need to aggregate the advertisements, you may as well NAT66, cause it will be easier.
Could be DirectAccess. Microsoft's earlier built-in VPN solution before Always On VPN. DirectAccess works only with IPv4 inbound so you can't use IPv6 only stack. Under the hood it uses a combination of v4-v6 transition and translation protocols, but it still requires the Windows client machines to have IPv4 addresses.
If you can run PowerShell commands on the laptop and if "Get-DnsClientNrptPolicy" returns some DirectAccessDnsServers then it's DA laptop.
For example, I recently attended the IETF meeting in Montreal--practically the epicenter of v6 thinking--which offers a by default v6-only network. My Mac worked fine, but my son's school-issued Chromebook had glitchy behavior until I switched to the network that provided v4.
For example, I recently attended the IETF meeting in Montreal, which offers a by default v6-only network. My Mac worked fine, but my son's school-issued Chromebook had glitchy behavior until I switched to the network that provided v4.
I went to IETF a few years ago and ran into issues on their IPv6 only network because I host some stuff from home, and my residential ISP doesn't support IPv6 at all. It made me really want to get all that fixed.
Nope. Large scale DCs are IPv6 only underneath, exascalers like Google and Meta have stated that multiple times. I.e. https://www.youtube.com/watch?v=Q3ird3UDnOA also see various NANOG talks https://www.youtube.com/@TeamNANOG/videos
Hyper scalers != cloud computing.
Unlike IPv4, my LAN addresses include the prefix, so every time they change it, all my LAN addresses change.
Combined with the lack of DHCP6 support in many devices, this means reverse DNS lookups from IP to hostname can't be done, making identifying devices by their IP essentially impossible.
So I don’t have the changing ip issue. I do however have an issue if I want to change ISP as it’s a whole mess of rules to update rather than a couple of dns entries and two dst nat rule (one per public IP)
I believe the idea in v6 if you have multiple prefixes on the same network - including a local fc00::/7 one for local services. Layers and layers of things to break.
Using Openwrt which pretty much all home routers are built on, all I have to do is tell my router which offset to give my subnets from the prefix and it does the rest.
Both for subdividing up the prefix from the ISP and my ULA prefix I use for internal devices.
I have changed ISPs I think 3 times with no ill effects. Plus it works when my ISP occasionally gives me a new prefix.
The only tweaking I had to do was when I went from an ISP that game me a /48 to one that only gave me a /56. I had been greedy and was handing a /56 to my internal router. I changed that to a /60 and updates it's expectations about which subnets it could hand out and all was good.
But I expect two layers of home routers without NAT is a bit of an exception.
IPv6 also has local addresses, but a lot more of them. Anything starting with fd00::/8 is a local address with 40 bits available as the network number. So you can set up your local network with the prefix fdXX:XXXX:XXXX::/48 (where the Xs are chosen randomly) as the prefix and still have 16 bits left over for different subnets if you want. These addresses do not change when your ISP changes your public prefix.
And if you want to add reverse dns for SLAAC addresses then just have your router listen for ICMPv6 Neighbor Announcement addresses and use them to update your DNS server as appropriate. Or configure your servers to use stable addresses based on their MAC address rather than random addresses (which are better for privacy), and then just configure the DNS as you add and remove servers.
This conversation is going in circles.
(same for ping)
Yes, a topic of active discussion at the IETF. See perhaps BCP RFC 9096, "Improving the Reaction of Customer Edge Routers to IPv6 Renumbering Events":
* https://datatracker.ietf.org/doc/html/rfc9096
And informational RFC 8978, "Reaction of IPv6 Stateless Address Autoconfiguration (SLAAC) to Flash-Renumbering Events":
* https://datatracker.ietf.org/doc/html/rfc8978
A few drafts, like "Improving the Robustness of Stateless Address Autoconfiguration (SLAAC) to Flash Renumbering Events":
* https://datatracker.ietf.org/doc/html/draft-ietf-6man-slaac-...
Using ULA seems to be what a lot of folks recommend:
If setting your client machine MTU to 1280 (`ip link set mtu 1280 dev eth0` or equivalent) magically fixes it, that's your problem.
Until you want to like, use GitHub.
In all seriousness, I have IPv6 enabled and GitHub works just fine for me. Though at a slower speed sometimes because the IPv4 CGNAT is heavily congested in my area.
https://en.wikipedia.org/wiki/Embrace,_extend,_and_extinguis...
for example:
Microsoft Word DOC. Due to the market dominance of Word, it is supported by all office applications that intend to compete with it, typically by reverse engineering the undocumented file format. Microsoft has repeatedly internally changed the file specification between versions of Word to suit their own needs, while continuing to reuse the same file extension identifier for different versions.
It is also impossible to report IPv6-specific outages. CenturyLink technical support is the worst of the worst, with agents utterly incapable of doing more than pushing a "check ONT" button on their end and scheduling a technician visit with a multiday window. If you ask them for the 6rd configuration information, they act like you're speaking an alien language.
Even among their technicians, IPv6 knowledge is rare. Imagine the guy installing hundreds of dollars of gigabit fibre equipment at your demarc staring you like an idiot because you spoke two extra syllables between "IP" and "address". I'd think the term "IPv6" is chatbot poison if it weren't for the fact it's a human physically in front of me.
The result is their service is effectively IPv4-only.
Interestingly, if I pay for their IPTV service the internet side becomes a bare ethernet port over which I can do DHCP for the upstream interface and number the downstream subnet out of my /28.
I have debated paying for TV service as a sanity fee.
Most do not.
There are far more single person, small, and mid sized companies that do not.
This includes b2b, regional ISPs, etc.
It works plenty well. I access everything accessible via IPv6, and the rest through their 464XLAT, transparently.
My LAN still has IPv4, because some ancient network printers don't know IPv6. OpenWRT on my router supports IPv6 just fine. Of course I do not expose any of my home devices to the public internet, except via Wireguard.
From my perspective:
• CGNAT is a feature, not a bug. I'm already deliberately behind a commercial VPN exit node shared with thousands of others. Anonymity-by-crowd is the point. IPv6 giving me a globally unique, stable-ish address is a regression.
• NAT + default-deny inbound is simple, effective security. Yes, "NAT isn't a firewall", but a NAT gateway with no port forwards means unsolicited inbound packets don't reach my devices. That's a concrete property I get for free.
• IPv6 adds configuration surface I don't want. Privacy extensions, temporary addresses, RA flags, NDP, DHCPv6 vs SLAAC — these are problems I don't have with IPv4. More features means more things to audit, understand, and misconfigure.
• I already solved "reaching my own stuff" without global addressing. Tailscale/Headscale gives me authenticated, encrypted, NAT-traversing connectivity. It's better than being globally routable.
So yes, my parents are using IPv6 to watch Netflix. They're also not thinking about their threat model. I am, and IPv4-only behind CGNAT + overlay networking serves it well.
"It just works" isn't the bar for me to adopt IPv6. "It serves my goals better than IPv4" is the bar, and IPv6 doesn't meet it. Never has, never will.
IPv6 wasn't designed as "IPv4 with more bits." It was designed as a reimagining of how networks should work: global addressability as a first-class property, stateless autoconfiguration, the assumption that endpoints should be reachable. That philosophy is baked in. For someone like me, whose threat model treats obscurity, indirection, and minimal feature surface as assets, IPv6 isn't just unnecessary, it's ideologically opposed to what I want.
Want me to adopt a new addressing scheme? Give me a new addressing scheme, don't impose an opinionated routing philosophy on me.
Only for IP based trackers. Any webpages embedding facebook/twitter/microsoft/google trackers have already deanonymised you through a variety of fingerprinting techniques. This includes if you use private browsing sessions, and even qubesOS. You get a fuzzy feeling doing the things you do (and I do these things too), but that battle is lost.
> NAT + default-deny inbound is simple, effective security … That's a concrete property I get for free
Depends on your definition of “free”. Is it cheaper to lookup just a connection state table, or is it cheaper to look up both a connection state table and a NAT table?
> IPv6 adds configuration surface I don't want … More features means more things to audit, understand, and misconfigure.
100% agreed. More complexity, more attack surface, more things to go wrong.
> I already solved "reaching my own stuff" without global addressing … It's better than being globally routable.
I do something like this too. It’s more private and more secure. It adds more complexity, and it restricts my ability to access things from terminals I don’t personally own & control unless I create another exposed vector though. “Better” is subjective based on metrics being optimised for.
> IPv6 wasn't designed as "IPv4 with more bits." It was designed as a reimagining of how networks should work: global addressability as a first-class property
Apologies, but global addressability as a first-class property is exactly how the internet was designed. NAT was originally deployed as a hacky add-on to temporarily alleviate the lack of addressing space in IPv4 until a successor could resolve that.
That said, the internet of the 90s was a very different beast to the internet of today. A lot of your concerns and perspective is absolutely valid and extremely reasonable given the internet of today.
> "It serves my goals better than IPv4" is the bar, and IPv6 doesn't meet it. Never has, never will … Want me to adopt a new addressing scheme? Give me a new addressing scheme, don't impose an opinionated routing philosophy on me.
IPv6 can absolutely be configured in ways that just gives you a new addressing scheme and does away with a lot of the other complexity. You’re just very much straying off the happy path, removing complexity by introducing … other complexity.
FWIW, I’m operating my home networks much the same way you do. I’ve also been dual stacking networks since the 2000s. Things have come a long way since the original pure-dogma introduction of ipv6.
I bet OP has already blocked at least 3 of them. Private browsing is only a partial solution, blocking/unblocking domains, scripts, etc. on a case-by-case basis is a more reliable way to defend your right to privacy against abusive practices (I'm talking about fine grained adblockers such as uMatrix/uBlockOrigin) daily.
I admit it can be a hassle sometimes, in particular if one explores the net every day, but staying away from bad actors (such as some of those 4) is one way to maybe eventually stop them - even if "vote with your clicks" feels as pointless as "vote with your feet" when you're just one in many millions.
To be fair about fingerprinting, there's no such thing as "bulletproof", but I do have a pretty robust setup. DNS level ad and tracker blocking, browser extension level ad and tracker blocking, LibreWolf's extensive anti-fingerprinting measures, kernel-level measures like kloak, I block all third party JS by default, etc. My goal isn't to become invisible and untraceable to nation states (which is essentially impossible when 90%+ of all global ISPs can and do sell netflow metadata, enabling timing and packet size correlation even across multiple hops, even with background traffic forgery / traffic pattern obfuscation), but rather to frustrate lower-level tracking efforts, and mostly to reduce attack surface for security reasons, and to reduce the total amount of information I'm sending to adversaries, even if it technically increases uniqueness. For instance, WebGL, JS JIT, WASM, WebRTC, and even SVG rendering are similarly disabled by default on my browsers, and I may very selectively enable them on a case-by-case basis depending on how important I feel the web property I'm trying to access actually is. I'll spoof my UA, my screen dimensions, and use residential SOCKS5 proxies, one by one, to identify which fingerprinting measures are being used to block me with YouTube, for instance, without enabling JIT compilation or SVG rendering. This approach absolutely does make me more distinctly identifiable (less anonymous), but doesn't necessarily make me less private, nor less secure, if e.g. ad network JS never even runs on my box in the first place. Security is the base of the pyramid, it is the prerequisite for privacy, but doesn't guarantee it. Privacy is the middle layer, it is the prerequisite for anonymity, but doesn't guarantee it. I'm aggressively climbing that pyramid where I can while accepting some tradeoffs where the net benefit is positive to me. I don't think of any of these - security, privacy, or anonymity - as binary properties, but rather a unified journey I am on to enhance gradually and iteratively over time. Switching to IPv6 would greatly complicate and regress my path through much of the journey I've already completed.
If I could leave you with a couple questions: What tangible benefits have you reaped from IPv6 that simply weren't possible on IPv4? Has the ROI for you on going dual stack outweighed the costs on your time, attention, and configuration work required for securely handling edge cases, dealing with weird or unexpected routing issues, for straying from the happy path?
Personal networks: Globally unique addressing. That then lends itself to not needing any kind of split DNS for services, or worrying about addressing clashes with whatever LAN I happen to be on with my own network.
Work networks: Increased revenues.
> Has the ROI for you on going dual stack outweighed the costs on your time, attention, and configuration work required for securely handling edge cases, dealing with weird or unexpected routing issues, for straying from the happy path?
Personal networks: Absolutely not. I removed the dual stacks and went back to IPv4 only everywhere.
Work networks: That's a question for the bean counters.
It is well supported, easy to configure, private, secure.
...and I don't have to configure and secure ipv6 in parallel
IPv4 in the home is dead easy. You only need to remember the last digit (unless you've got multiple networks, but most won't). You can ssh to any device by remembering that ".1" is router, ".2" is NAS etc. Firewalls are simple.
You can buy a cheap domain and use it as your home DNS (eg "router.myhome.net" -> "192.168.0.1") so it works anywhere! In the home or roaming (over VPN). I don't really need to run DNS at home. My domain runs on Cloudflare DNS, my devices use NextDNS (with rebind protection disabled for my home domain).
I run OpenWRT and preallocate DHCP addresses for all known devices. Then I shrink the DHCP pool to a blacklisted range. A script automatically creates DNS records for all preallocated devices. If a new device appears in the blacklisted DHCP pool, I can manually allocate its MAC address a proper IP.
It's easy to get TLS certs for any service in the house using the ACME DNS01 challenge.
Tailscale is sexy and it worked fine until one day while roaming it wouldn't connect without "admin work", so I instantly dropkicked it. I'm now using the very unsexy OpenVPN Cloud (free for limited use) and in over two years it has never failed me. Plus it doesn't fuck with the IP addresses with fancypants tailnet addresses - I access devices directly using their DNS names which resolve to private addresses.
So, from inside or outside the home I can access the NAS to watch a movie, sync photos to Immich, print a document, check my IP cameras or ask my wife to put a document on the ancient scanner and access it via the raspberry pi phpscan website (which is on https://scanner.myhome.net)
I'm sure there's a very good reason not to do this and someone will now point it out.
Exactly. I randomly try to "upgrade" to ipv6 in my home once in a while and i always give up because I'd have to do the whole enterprisey setup for no good reason.
Edit:
Basically ipv6 is too complex and automated to hold your home network's whole configuration in your head without effort.
So the techies don't set it up at home unless they have a fetish for overcomplicated setups. They're not familiar with it so they don't push for it at work either.
Adoption is solely driven by ipv4 address space exhaustion. There is no "new toy!" feeling involved.
You could try fd00::1, fd00::2, ... for short internal static addresses. You don't have to use a random prefix in that range - it's just policy (for good reasons that might not matter for a small network).
Yeah, and my Windows box is again accessible from the outside with whatever services MS deems to run by default...
Yes, there are firewalls, but isn't it better if a potential attacker doesn't even know what's behind my router?
P.S.: Since webrtc showed up to do whatever it wants with my network, peer to peer has started to mean "donating resources to some company" to me.
On v6, you don't use NAT and networks are /64. Finding every server requires scanning 65536 ports on all 2^64 IPs, which is about 72 billion petabytes of traffic. There are ways to prune this down somewhat, but however you do it the search space is still far larger.
If you want attackers to not know what's behind your router, you want v6.
Enterprise thinking. It's not the publicly accessible servers i worry about, it's the other boxes that shouldn't be publicly accessible...
# You can buy a cheap domain and use it as your home DNS (eg "router.myhome.net" -> "2003:123:4:5::1") so it works anywhere! In the home or roaming (over VPN). I don't really need to run DNS at home. My domain runs on Cloudflare DNS, my devices use NextDNS (with rebind protection disabled for my home domain).
# I run OpenWRT and preallocate DHCP addresses for all known devices. Then I shrink the DHCP pool to a blacklisted range. A script automatically creates DNS records for all preallocated devices. If a new device appears in the blacklisted DHCP pool, I can manually allocate its MAC address a proper IP.
# It's easy to get TLS certs for any service in the house using the ACME DNS01 challenge.
There is literally no difference between v4 and v6 here.
I'd love to test all the internet services I host to make sure everything works over IPv6, but I can't. At least, not without using a 4to6 relay of some sort - but that adds latency to everything I do.
I just checked - apparently my ISP is "evaluating IPv6" because they're running out of IPv4 addresses and want to use CGNAT for everyone. I suppose its not the worst reason to switch to ipv6. But they've been making excuses for years. I really wish they'd get on with it.
2. Actual implementation MUST be ubiquitous (it never will be) some examples - glo fiber in Virginia, and while I can get pfsense assigned a ipv6 address, there is usually no upstream gateway (meaning that if I disable IPv4, I will not have internet). I say usually because of four times I've checked, once I did get assigned a gateway which was unresponsive even to icmp.
Starlink roam - assigns ipv6 but no bridge so if you disable v4 you lose access to most internet.
Frontier FiOS in Florida - does not support ipv6 at all on my node. I have seen business nodes in Orlando/Tampa assign addresses with bridging but again, without browser or dns translation it's not a practical solution.
3. 'Everyone' is not using ipv6, everyone plugs in or logs into a device that has whatever network stack it has. Those users are not suddenly going to jump through hoops simply to avoid CGNAT and get a unique network address
4. Infrastructure; I have two modest half racks on the east coast at decent sized datacenters (esolutions and peak10), neither of those hosts offer ipv6 routing blocks by default. No provider I have gotten quotes for offers ipv6 by default
The more technically knowledgable you happen to be on the subject, the more you realize IPv6 is some unreliable thing when compared to IPv4. Perhaps no longer niche though.
It's unfortunately still an afterthought for many backbones - and not just US-centric ones. There is a noticeable difference in performance metrics from clients served via IPv4 endpoints vs. IPv6 for web assets in the same locations from the same transit providers.
It is pretty much the opposite of "just works" depending on your definition of "just works". It results in more Traffic Engineering per bit served by a large factor compared to IPv4.
Where can we read some examples of this
I've read commentary about pros and cons of IPv6 over the years but never anything that suggested IPv6 was "niche" or "unreliable"
In order to have inspired the quoted statement these examples would have to be found in forum comments published before the quoted statement was made
Comments made in response to, i.e., after, the quoted statement would not qualify
My own home is Verizon, and they simply do not offer IPv6 in my area (nearby Washington DC).
In my direct experience, in the USA, at least Spectrum, AT&T, and Xfinity (Comcast) still run IPv4, of course, but they also have IPv6 working and on by default on their home internet offerings.
All mainstream computer and mobile OSes support it by default and will prefer to connect with it over IPv4.
‘Everyone’ in many areas is using it. For many of us, our parents are using Facebook and watching Netflix over it. Over 50% of Google’s American traffic is over it. It just works.