8 ms·

Ask HN: Why is HN not reachable through IPv6?

34 points by codethief 3y ago

It's 2023 and it looks like news.ycombinator.com lacks an AAAA record. Why?

8chanAnon 3y ago
The biggest problem with ipv6 right now is that the DNS record system needs to be updated to support both simultaneously. Right now, it is necessary to perform the AAAA lookup alongside the A lookup which doubles the workload on the DNS. Also, ipv6 coverage is still spotty enough that it is sometimes more efficient to open both ipv4 and ipv6 in parallel in order to pick the one which performs better. The problem is that websites will, for the foreseeable future, have to be accessible via ipv4 else suffer a significant reduction in traffic. The issues that I've mentioned above are the reasons why nobody is currently rushing to implement ipv6. Depending on what websites you use, as little as 20% might be accessible via ipv6. This will hold true, I think, for 5 to 10 years. I don't believe that ipv4 will go away. One reason is because ipv6 is unwieldly in comparison. For one thing, the format is imprecise because it does not enforce consistency. There are dozens of different ways to write an ipv6 address because leading zeroes are allowed while leading zeroes are illegal under ipv4. Personally, I have a real beef with the use of the colon as a separator. My prediction is that, as ipv6 becomes more endemic on the Internet back-end and the pressure on the ipv4 address space eases, then ipv4 addressing will become MORE attractive on the Internet front-end (meaning the consumer).
- Dagger2 3y ago
  That's really not the biggest issue. DNS servers can handle this level of load just fine. People throwing out excuse after excuse (frequently wrong or ill-informed) for not doing v6 is a bigger issue. > For one thing, the format is imprecise because it does not enforce consistency. There are dozens of different ways to write an ipv6 address because leading zeroes are allowed while leading zeroes are illegal under ipv4 You'd be amazed how much worse v4 is on this front. Leading zeros are actually legal, but they turn the field into octal. Hex is an option, and so is combining fields. You can write HN's IP as 0321.216.230.240, or 0321.216.0xe6.240. Or 0xd1.216.0163360. There's about 40 different combinations like this. v6 clearly defines a canonical form, and doesn't have as much possible variation outside of it. > Personally, I have a real beef with the use of the colon as a separator. It had to be changed from . to avoid ambiguity with DNS. It also allows writing the right-hand 32 bits as a v4 literal (e.g. 64:ff9b::209.216.230.240 -- fortunately only in this form and not all of the other forms v4 has) which is convenient sometimes. Given their existing use in other networking protocols and things like MAC addresses, colons seem like the obvious alternative.
  - 8chanAnon 3y ago
    >You'd be amazed how much worse v4 is on this front. Leading zeros are actually legal, but they turn the field into octal. That depends on the platform (and, no, leading zeroes are not legal). I work with Node.js and leading zeroes in an ipv4 address will be correctly rejected. There is no chance of mistaking 127.00.0.01 for 127.0.0.1 (the former will fail the net.isIP test). With ipv6, Node is quite forgiving and this annoys me. Without deconstructing and reconstructing the address in proper canonical form, it is difficult to tell the difference between identical addresses due to the presence of extra zeroes (they will pass the net.isIP test). Unfortunately, Node does not include a method for normalizing an address.
    - Dagger2 3y ago
      $ ping 0000000000321.216.230.240 PING 0000000000321.216.230.240 (209.216.230.240) 56(84) bytes of data. 64 bytes from 209.216.230.240: icmp_seq=1 ttl=46 time=142 ms Platform-specific syntax, excellent. That sure makes it sound more consistent. You're right that you need to canonicalize addresses as part of comparing their textual forms. It's well-documented -- at least in v6 -- that you need to do that. > Unfortunately, Node does not include a method for normalizing an address. That sounds like a Node problem. In C you can do it via getnameinfo(getaddrinfo()) or inet_ntop(inet_pton()).
      - zimpenfish 3y ago
        $ ping 0010.000010.0000000010.000000000000000010 PING 0010.000010.0000000010.000000000000000010 (8.8.8.8) 56(84) bytes of data. 64 bytes from 8.8.8.8: icmp_seq=1 ttl=117 time=2.48 ms Oh the humanity.
aaron695 3y ago
[dead]
stop50 3y ago
Its not only hn. Many other sites are too lazy for v6. Even at my work its an full v4 network. Some processes are even technically unable for v6.
- xnyan 3y ago
  >…too lazy… I thought this too, until I tried to add IPV6 support for a fairly simple web application. The single biggest problem is the quality and consistency of IPV6 implementations is frequently very poor. I ran into many bugs, strange inconsistencies and poor handling of edge cases by middleware, clients, routers and firewalls that any advantage from IPV6 was simply not worth it for our needs. Mobile telcos and cloud providers are currently the only parties where the benefits of V6 outweighs the pain, and as such their software and services tend to support it quite well. While the cost of V4 is going up, for almost everyone it’s still far less than dealing with V6, and there’s nothing relevant (unless you have telco/cloud provider needs) V6 can provide you that V4 can’t.
pestatije 3y ago
i guess cause "if it aint broke..."
- gargalatas 3y ago
  But unfortunately nodays with the prices of IPv4s it's like it's broken. The best price you can lease is probably in the ballpark of around 0.5 USD / IP.
  - nik736 3y ago
    IP pricing declined constantly throughout the past few months. I sold IPs in December '21 for $55/IP and now they are at around $36-$40/IP.
    - simfree 3y ago
      What is the best place for someone looking to buy a /24 to go? Should people be looking at legacy resources?
      - nik736 3y ago
        There are certain auction platforms, I sold my IPs through https://auctions.ipv4.global/ https://auctions.ipv4.global/ which went smoothly. The registries have waiting lists for resources that get handed back to them (very rare). You need to be a member to buy IP space though.
  - xnyan 3y ago
    I love the idea of v6, but the reality is still not worth it for most situations even at that price. IP v6 implementations are still all too commonly buggy and inconstant, and that’s on top of all the work required to bring v6 online.
  - mikeodds 3y ago
    where’s a good place to lease the IPs directly, going through vps providers I’m paying ballpark 2-3 usd per month per ipv4
api 3y ago
GitHub is the worst offender. It’s a major reason why V6 only VMs and cloud is a pain even though it’s cheaper in some cases.
- justinclift 3y ago
  GitHub Pages started supporting IPv6 towards the end of 2021 (finally). Does GitHub not support IPv6 for repo activity though? eg clone, pull, push.
- KomoD 3y ago
  IIRC there were even issues where github actions couldn't reach github because no IPv6
Someone1234 3y ago
There's a lot more work for IPv6 support than just throwing a AAAA record on a domain. A few of them: - IPv6 Peerage. - An IPv4 or IPv6 broker. - First party software configuration. - Support from upstream and downstream from infrastructure/partners and configuration. It is a significant project. People think you just assign an IPv6 IP, throw in a AAAA record, and then it magically just works. More than likely you'd just create a block-hole where IPv6 traffic goes to die.
- bauruine 3y ago
  hn has two servers at m5hosting which according to their FAQ [0] "IPv6 support is currently available for Dedicated Servers or Colocation." so you ask support, they assign you a prefix and you add a few IPs from that prefix. done. My guess is their anti spam or something else in the software expects an IPv4. [0] https://www.m5hosting.com/support/cloud-faq/ https://www.m5hosting.com/support/cloud-faq/
- toast0 3y ago
  IPv6 is work, but it's not that much work if your networking or hosting provider already supports it (as is claimed in this case). Address classification changes are real work. I haven't done heavy benchmarks, but bigger addresses and everything else probably does add up to a measurable server performance drop, but that's balanced by real world networking performance improvements for users with nasty network stuff on IPv4 and less for IPv6. Otoh, I'd guess HN isn't bottlenecked by networking anyway.
- bombcar 3y ago
  For many sites all you need is the AAAA record. But something like HN probably has various moderation/spam preventing tools that assume IPV4. I believe people have made IPv6 gateways to the ipv4 content.
- housemusicfan 3y ago
  > More than likely you'd just create a block-hole It's still called a black hole. https://en.wikipedia.org/wiki/Black_hole_(networking) https://en.wikipedia.org/wiki/Black_hole_(networking)
zigzag312 3y ago
Why does it feel like IPv6 has failed? There's a real need for bigger address space, but it looks like IPv6 is too much of a hassle.
- toast0 3y ago
  It hasn't exactly failed. Depending on where you look, it's in the neighborhood of 40% of traffic, although variability is very high. 40% market share isn't really failing, although it sure does feel like it could have been better. IPv4 is still a must, and IPv6 is optional in most circumstances.
  - zigzag312 3y ago
    Failed in a sense that it was intended to replace IPv4. At current rate it's going to take additional 10+ years to replace IPv4. Also, many don't seem eager to adopt it. If that is due to the protocol or just the lack of tooling, I don't know.
    - copperx 3y ago
      IPv6 failing is a self-fulfulling prophecy. It isn't that hard to implement, but it has become a meme.
- ZeroSolstice 3y ago
  Yeah I wouldn't say it failed its just that there wasn't a financial motive to move towards IPv6. As more networked devices and the device data itself becomes important moving to IPv6 will follow. Residential ISP's and some cloud providers are the biggest hold-up as most mobile carriers already dual stack and Tier 1 transit providers have had IPv6 for at least a decade or more.
  - saghm 3y ago
    I wonder if part of the incentive problem was that the companies with the most traffic would have already invested heavily in securing sufficient IPv4 address space for their usage. If so, the ones with the most ability to influence everyone to swap over would be the ones with the least need to do so, and then the smaller players who might need IPv6 most for future growth wouldn't have enough users yet to incentivize the infrastructure changes needed.

34 points by codethief 3y ago

It's 2023 and it looks like news.ycombinator.com lacks an AAAA record. Why?

The biggest problem with ipv6 right now is that the DNS record system needs to be updated to support both simultaneously. Right now, it is necessary to perform the AAAA lookup alongside the A lookup which doubles the workload on the DNS. Also, ipv6 coverage is still spotty enough that it is sometimes more efficient to open both ipv4 and ipv6 in parallel in order to pick the one which performs better. The problem is that websites will, for the foreseeable future, have to be accessible via ipv4 else suffer a significant reduction in traffic. The issues that I've mentioned above are the reasons why nobody is currently rushing to implement ipv6. Depending on what websites you use, as little as 20% might be accessible via ipv6. This will hold true, I think, for 5 to 10 years. I don't believe that ipv4 will go away. One reason is because ipv6 is unwieldly in comparison. For one thing, the format is imprecise because it does not enforce consistency. There are dozens of different ways to write an ipv6 address because leading zeroes are allowed while leading zeroes are illegal under ipv4. Personally, I have a real beef with the use of the colon as a separator. My prediction is that, as ipv6 becomes more endemic on the Internet back-end and the pressure on the ipv4 address space eases, then ipv4 addressing will become MORE attractive on the Internet front-end (meaning the consumer).

Dagger2 3y ago
That's really not the biggest issue. DNS servers can handle this level of load just fine. People throwing out excuse after excuse (frequently wrong or ill-informed) for not doing v6 is a bigger issue. > For one thing, the format is imprecise because it does not enforce consistency. There are dozens of different ways to write an ipv6 address because leading zeroes are allowed while leading zeroes are illegal under ipv4 You'd be amazed how much worse v4 is on this front. Leading zeros are actually legal, but they turn the field into octal. Hex is an option, and so is combining fields. You can write HN's IP as 0321.216.230.240, or 0321.216.0xe6.240. Or 0xd1.216.0163360. There's about 40 different combinations like this. v6 clearly defines a canonical form, and doesn't have as much possible variation outside of it. > Personally, I have a real beef with the use of the colon as a separator. It had to be changed from . to avoid ambiguity with DNS. It also allows writing the right-hand 32 bits as a v4 literal (e.g. 64:ff9b::209.216.230.240 -- fortunately only in this form and not all of the other forms v4 has) which is convenient sometimes. Given their existing use in other networking protocols and things like MAC addresses, colons seem like the obvious alternative.
- 8chanAnon 3y ago
  >You'd be amazed how much worse v4 is on this front. Leading zeros are actually legal, but they turn the field into octal. That depends on the platform (and, no, leading zeroes are not legal). I work with Node.js and leading zeroes in an ipv4 address will be correctly rejected. There is no chance of mistaking 127.00.0.01 for 127.0.0.1 (the former will fail the net.isIP test). With ipv6, Node is quite forgiving and this annoys me. Without deconstructing and reconstructing the address in proper canonical form, it is difficult to tell the difference between identical addresses due to the presence of extra zeroes (they will pass the net.isIP test). Unfortunately, Node does not include a method for normalizing an address.
  - Dagger2 3y ago
    $ ping 0000000000321.216.230.240 PING 0000000000321.216.230.240 (209.216.230.240) 56(84) bytes of data. 64 bytes from 209.216.230.240: icmp_seq=1 ttl=46 time=142 ms Platform-specific syntax, excellent. That sure makes it sound more consistent. You're right that you need to canonicalize addresses as part of comparing their textual forms. It's well-documented -- at least in v6 -- that you need to do that. > Unfortunately, Node does not include a method for normalizing an address. That sounds like a Node problem. In C you can do it via getnameinfo(getaddrinfo()) or inet_ntop(inet_pton()).
    - zimpenfish 3y ago
      $ ping 0010.000010.0000000010.000000000000000010 PING 0010.000010.0000000010.000000000000000010 (8.8.8.8) 56(84) bytes of data. 64 bytes from 8.8.8.8: icmp_seq=1 ttl=117 time=2.48 ms Oh the humanity.

aaron695 3y ago

[dead]

stop50 3y ago

Its not only hn. Many other sites are too lazy for v6. Even at my work its an full v4 network. Some processes are even technically unable for v6.

xnyan 3y ago
>…too lazy… I thought this too, until I tried to add IPV6 support for a fairly simple web application. The single biggest problem is the quality and consistency of IPV6 implementations is frequently very poor. I ran into many bugs, strange inconsistencies and poor handling of edge cases by middleware, clients, routers and firewalls that any advantage from IPV6 was simply not worth it for our needs. Mobile telcos and cloud providers are currently the only parties where the benefits of V6 outweighs the pain, and as such their software and services tend to support it quite well. While the cost of V4 is going up, for almost everyone it’s still far less than dealing with V6, and there’s nothing relevant (unless you have telco/cloud provider needs) V6 can provide you that V4 can’t.

pestatije 3y ago

i guess cause "if it aint broke..."

gargalatas 3y ago
But unfortunately nodays with the prices of IPv4s it's like it's broken. The best price you can lease is probably in the ballpark of around 0.5 USD / IP.
- nik736 3y ago
  IP pricing declined constantly throughout the past few months. I sold IPs in December '21 for $55/IP and now they are at around $36-$40/IP.
  - simfree 3y ago
    What is the best place for someone looking to buy a /24 to go? Should people be looking at legacy resources?
    - nik736 3y ago
      There are certain auction platforms, I sold my IPs through https://auctions.ipv4.global/ https://auctions.ipv4.global/ which went smoothly. The registries have waiting lists for resources that get handed back to them (very rare). You need to be a member to buy IP space though.
- xnyan 3y ago
  I love the idea of v6, but the reality is still not worth it for most situations even at that price. IP v6 implementations are still all too commonly buggy and inconstant, and that’s on top of all the work required to bring v6 online.
- mikeodds 3y ago
  where’s a good place to lease the IPs directly, going through vps providers I’m paying ballpark 2-3 usd per month per ipv4

api 3y ago

GitHub is the worst offender. It’s a major reason why V6 only VMs and cloud is a pain even though it’s cheaper in some cases.

justinclift 3y ago
GitHub Pages started supporting IPv6 towards the end of 2021 (finally). Does GitHub not support IPv6 for repo activity though? eg clone, pull, push.
KomoD 3y ago
IIRC there were even issues where github actions couldn't reach github because no IPv6

Someone1234 3y ago

There's a lot more work for IPv6 support than just throwing a AAAA record on a domain. A few of them: - IPv6 Peerage. - An IPv4 or IPv6 broker. - First party software configuration. - Support from upstream and downstream from infrastructure/partners and configuration. It is a significant project. People think you just assign an IPv6 IP, throw in a AAAA record, and then it magically just works. More than likely you'd just create a block-hole where IPv6 traffic goes to die.

bauruine 3y ago
hn has two servers at m5hosting which according to their FAQ [0] "IPv6 support is currently available for Dedicated Servers or Colocation." so you ask support, they assign you a prefix and you add a few IPs from that prefix. done. My guess is their anti spam or something else in the software expects an IPv4. [0] https://www.m5hosting.com/support/cloud-faq/ https://www.m5hosting.com/support/cloud-faq/
toast0 3y ago
IPv6 is work, but it's not that much work if your networking or hosting provider already supports it (as is claimed in this case). Address classification changes are real work. I haven't done heavy benchmarks, but bigger addresses and everything else probably does add up to a measurable server performance drop, but that's balanced by real world networking performance improvements for users with nasty network stuff on IPv4 and less for IPv6. Otoh, I'd guess HN isn't bottlenecked by networking anyway.
bombcar 3y ago
For many sites all you need is the AAAA record. But something like HN probably has various moderation/spam preventing tools that assume IPV4. I believe people have made IPv6 gateways to the ipv4 content.
housemusicfan 3y ago
> More than likely you'd just create a block-hole It's still called a black hole. https://en.wikipedia.org/wiki/Black_hole_(networking) https://en.wikipedia.org/wiki/Black_hole_(networking)

zigzag312 3y ago

Why does it feel like IPv6 has failed? There's a real need for bigger address space, but it looks like IPv6 is too much of a hassle.

toast0 3y ago
It hasn't exactly failed. Depending on where you look, it's in the neighborhood of 40% of traffic, although variability is very high. 40% market share isn't really failing, although it sure does feel like it could have been better. IPv4 is still a must, and IPv6 is optional in most circumstances.
- zigzag312 3y ago
  Failed in a sense that it was intended to replace IPv4. At current rate it's going to take additional 10+ years to replace IPv4. Also, many don't seem eager to adopt it. If that is due to the protocol or just the lack of tooling, I don't know.
  - copperx 3y ago
    IPv6 failing is a self-fulfulling prophecy. It isn't that hard to implement, but it has become a meme.
ZeroSolstice 3y ago
Yeah I wouldn't say it failed its just that there wasn't a financial motive to move towards IPv6. As more networked devices and the device data itself becomes important moving to IPv6 will follow. Residential ISP's and some cloud providers are the biggest hold-up as most mobile carriers already dual stack and Tier 1 transit providers have had IPv6 for at least a decade or more.
- saghm 3y ago
  I wonder if part of the incentive problem was that the companies with the most traffic would have already invested heavily in securing sufficient IPv4 address space for their usage. If so, the ones with the most ability to influence everyone to swap over would be the ones with the least need to do so, and then the smaller players who might need IPv6 most for future growth wouldn't have enough users yet to incentivize the infrastructure changes needed.