5 ms·

Global IPv6 Deployment Progress Report

38 points by fcambus 12y ago

axaxs 12y ago
Interesting. The TLD status isn't that telling as full IPv6 compliance is required for all new tlds. A good thing, but doesn't show 'willful' adoption nor use in the grand scheme. The AAAA glue to A glue ratios are still extremely low, and is probably a good indication of adoption and everyday use.
calinet6 12y ago
All I know is that I have native IPv6 at home now thanks—and I can't believe I'm saying this—to Comcast. Progress perhaps? Also, cute: Facebook's IPv6 address is 2a03:2880:f00d:401:face:b00c:0:1
- 2bluesc 12y ago
  I've had Comcast native for a while too. Impressed that it just works. Even more excited that I can request a prefix delegation.
  - simoncion 12y ago
    What does it mean to "request" a prefix delegation? I know that they use DHCPv6-PD to allocate v6 blocks to end-user networks, and that you can (on residential service) request blocks as large as /60. Is there something more going on?
_jomo 12y ago
My ISP is having issues with IPv4 via DS-Lite, so I often cannot connect to IPv4-only servers and it's frustrating how few servers are using IPv6 yet. twitter, heroku, amazon, github, HN, they all don't have IPv6. Is there a reason for the lack of IPv6 support? It shouldn't be that hard to get a v6 address and add an AAAA record.
- taf2 12y ago
  i've found it much harder to block unwanted IPv6 traffic compared to IPv4... but part of that is just the addresses are so much longer... and I need to get better at iptables
  - simoncion 12y ago
    It might help if you built your IPv6 firewall rules to deal with networks, rather than individual IPs.[0] [0] Hell, with the horrorshow that is IPv6 "Privacy" addresses, it's all but impossible to recognise a particular host.
- rectang 12y ago
  > Is there a reason for the lack of IPv6 support? Yeah -- the failure to design IPv6 as a compatible extension of IPv4 in order to ease migration. See http://cr.yp.to/djbdns/ipv6mess.html http://cr.yp.to/djbdns/ipv6mess.html
- unsignedint 12y ago
  My ISP (Comcast) support IPv6, but I had to turn it off because the router doesn't work with it.[1] [1] http://forums.comcast.com/t5/Home-Networking-Router-WiFi/Netgear-R6300-TG862-Bridge-IPv6-is-unstable/m-p/1966373/highlight/true#M62643 http://forums.comcast.com/t5/Home-Networking-Router-WiFi/Net...
tedunangst 12y ago
This report is really hard to interpret. Is there no way to know how many of the top 100 sites support IPv6 except by counting them up by hand? I get that google.XX supports IPv6; not sure I needed to actually see all their repeated addresses though. Similarly, if I wanted to know that 2.35% of .com domains have AAAA records, why do I need to divide the numbers myself?
stalinone 12y ago
I personally think that Path MTU Discovery is one of the major obstacle of IPv6. It is quite unstable to me. My home WiFi router has v6-ready certification but cannot properly tell my Windows PC which MTU size should be chosen. I have to manually lower down MTU size but it keep reverting. Directly PPPoE from my PC is better but there's still some website I have to set my MTU size to 1280 to reach.
- zAy0LfpBZLC8mAC 12y ago
  Are you sure the problem is the uplink MTU of your DSL(?)? That's the only thing where your router would be responsible for telling your PC about the bottleneck (if it receives a packet from your PC that doesn't fit through the link, it should send an "ICMP6 packet too big" to the PC), and if you get any working TCP6 connections at all, that most likely is working just fine. If you get stuck connections with some websites, the problem most likely is with that website: Their server sends you a big-ish packet, the router at the provider's end of your DSL drops the packet and sends an "ICMP6 packet too big" back to the server, but some incompetent admin configured their firewall to drop ICMP6, so their server never learns that its packets are too big, and just keeps resending the big packet until it times out.
  - simoncion 12y ago
    It really bothers me that some network admins choose to drop all ICMP traffic. I wish that the Shields UP! crowd had sent the message that ICMP is a very important part of IP signalling; rather than sending the message that eeeevil hackers might discover your network and Hax your Gibsons, so you'd better drop everything you can on the floor to throw them off of your trail.
    - ghshephard 12y ago
      With IPv6, admins learn within a week or two that they have to allow ICMPv6, or their networks don't work. With IPv4 you could (sort of) get by without ICMP, with IPv6 it's mandatory.
zAy0LfpBZLC8mAC 12y ago
There is one big problem I see with IPv6 deployment: Hosting providers don't get it. Yes, you can increasingly get IPv6 connectivity, but it's common that you get a single /64, or even just a single address. Now, I want to run a bunch of virtual machines on my server, with a virtual network or two between them ... and already, I have to use ugly hacks to get by instead of just assigning addresses using the standard way of doing things with IPv6. In the best case, I just have to ask for more address space, but even that is completely pointless overhead that only works as an incentive to use fragile network constructs instead of just assigning addresses where they are needed. People seem to have internalized that addresses are scarce and therefore precious and therefore have to be conserved at all cost - they don't even notice that that was the problem of IPv4 (or at least the most important problem of IPv4) that IPv6 was supposed to solve, and which it indeed does solve very well, and that their efforts at conserving IPv6 address space cause more harm than they help. Lots of hosters around here are RIPE members who have their standard /32 allocation (which they could extend to /29 without any further documentation required), so if they chose to assign /56 networks to customers, they would have space for about 16 million customers before they would need to request more address space from RIPE, and for 128 million before they need to justify any further need. But apparently they consider it better use of that address space to just leave it mostly unused and make it difficult for their customers to use IPv6, not even noticing that there is a reason why RIPE assigned them a /32 ...
- mappu 12y ago
  Correct me if i'm wrong, but doesn't an IPv6 /64 give you 2^64 globally routeable addresses? How is that not sufficient?
  - simoncion 12y ago
    You can't subnet a /64. This RFC [0][1] describes the IETF's thoughts on why allocating a block larger than /64 is to be strongly encouraged. [0] https://tools.ietf.org/html/rfc6177 https://tools.ietf.org/html/rfc6177 [1] Particularly, see page 3
  - matteotom 12y ago
pmontra 12y ago
A more interesting report would be which ISP by country let their customers use IPv6, landline and mobile. If servers are IPv6 but people can access them only over IPv4 then we're still at IPv4.
- p1mrx 12y ago
  This data shows the fraction of users with IPv6, per ISP: http://www.worldipv6launch.org/measurements/ http://www.worldipv6launch.org/measurements/ And here is some similar data, but per country: http://www.google.com/intl/en/ipv6/statistics.html#tab=per-country-ipv6-adoption http://www.google.com/intl/en/ipv6/statistics.html#tab=per-c... The US just recently crossed 10%.

38 points by fcambus 12y ago

axaxs 12y ago

Interesting. The TLD status isn't that telling as full IPv6 compliance is required for all new tlds. A good thing, but doesn't show 'willful' adoption nor use in the grand scheme. The AAAA glue to A glue ratios are still extremely low, and is probably a good indication of adoption and everyday use.

calinet6 12y ago

All I know is that I have native IPv6 at home now thanks—and I can't believe I'm saying this—to Comcast. Progress perhaps? Also, cute: Facebook's IPv6 address is 2a03:2880:f00d:401:face:b00c:0:1

2bluesc 12y ago
I've had Comcast native for a while too. Impressed that it just works. Even more excited that I can request a prefix delegation.
- simoncion 12y ago
  What does it mean to "request" a prefix delegation? I know that they use DHCPv6-PD to allocate v6 blocks to end-user networks, and that you can (on residential service) request blocks as large as /60. Is there something more going on?

_jomo 12y ago

My ISP is having issues with IPv4 via DS-Lite, so I often cannot connect to IPv4-only servers and it's frustrating how few servers are using IPv6 yet. twitter, heroku, amazon, github, HN, they all don't have IPv6. Is there a reason for the lack of IPv6 support? It shouldn't be that hard to get a v6 address and add an AAAA record.

taf2 12y ago
i've found it much harder to block unwanted IPv6 traffic compared to IPv4... but part of that is just the addresses are so much longer... and I need to get better at iptables
- simoncion 12y ago
  It might help if you built your IPv6 firewall rules to deal with networks, rather than individual IPs.[0] [0] Hell, with the horrorshow that is IPv6 "Privacy" addresses, it's all but impossible to recognise a particular host.
rectang 12y ago
> Is there a reason for the lack of IPv6 support? Yeah -- the failure to design IPv6 as a compatible extension of IPv4 in order to ease migration. See http://cr.yp.to/djbdns/ipv6mess.html http://cr.yp.to/djbdns/ipv6mess.html
unsignedint 12y ago
My ISP (Comcast) support IPv6, but I had to turn it off because the router doesn't work with it.[1] [1] http://forums.comcast.com/t5/Home-Networking-Router-WiFi/Netgear-R6300-TG862-Bridge-IPv6-is-unstable/m-p/1966373/highlight/true#M62643 http://forums.comcast.com/t5/Home-Networking-Router-WiFi/Net...

tedunangst 12y ago

This report is really hard to interpret. Is there no way to know how many of the top 100 sites support IPv6 except by counting them up by hand? I get that google.XX supports IPv6; not sure I needed to actually see all their repeated addresses though. Similarly, if I wanted to know that 2.35% of .com domains have AAAA records, why do I need to divide the numbers myself?

stalinone 12y ago

I personally think that Path MTU Discovery is one of the major obstacle of IPv6. It is quite unstable to me. My home WiFi router has v6-ready certification but cannot properly tell my Windows PC which MTU size should be chosen. I have to manually lower down MTU size but it keep reverting. Directly PPPoE from my PC is better but there's still some website I have to set my MTU size to 1280 to reach.

zAy0LfpBZLC8mAC 12y ago
Are you sure the problem is the uplink MTU of your DSL(?)? That's the only thing where your router would be responsible for telling your PC about the bottleneck (if it receives a packet from your PC that doesn't fit through the link, it should send an "ICMP6 packet too big" to the PC), and if you get any working TCP6 connections at all, that most likely is working just fine. If you get stuck connections with some websites, the problem most likely is with that website: Their server sends you a big-ish packet, the router at the provider's end of your DSL drops the packet and sends an "ICMP6 packet too big" back to the server, but some incompetent admin configured their firewall to drop ICMP6, so their server never learns that its packets are too big, and just keeps resending the big packet until it times out.
- simoncion 12y ago
  It really bothers me that some network admins choose to drop all ICMP traffic. I wish that the Shields UP! crowd had sent the message that ICMP is a very important part of IP signalling; rather than sending the message that eeeevil hackers might discover your network and Hax your Gibsons, so you'd better drop everything you can on the floor to throw them off of your trail.
  - ghshephard 12y ago
    With IPv6, admins learn within a week or two that they have to allow ICMPv6, or their networks don't work. With IPv4 you could (sort of) get by without ICMP, with IPv6 it's mandatory.

zAy0LfpBZLC8mAC 12y ago

There is one big problem I see with IPv6 deployment: Hosting providers don't get it. Yes, you can increasingly get IPv6 connectivity, but it's common that you get a single /64, or even just a single address. Now, I want to run a bunch of virtual machines on my server, with a virtual network or two between them ... and already, I have to use ugly hacks to get by instead of just assigning addresses using the standard way of doing things with IPv6. In the best case, I just have to ask for more address space, but even that is completely pointless overhead that only works as an incentive to use fragile network constructs instead of just assigning addresses where they are needed. People seem to have internalized that addresses are scarce and therefore precious and therefore have to be conserved at all cost - they don't even notice that that was the problem of IPv4 (or at least the most important problem of IPv4) that IPv6 was supposed to solve, and which it indeed does solve very well, and that their efforts at conserving IPv6 address space cause more harm than they help. Lots of hosters around here are RIPE members who have their standard /32 allocation (which they could extend to /29 without any further documentation required), so if they chose to assign /56 networks to customers, they would have space for about 16 million customers before they would need to request more address space from RIPE, and for 128 million before they need to justify any further need. But apparently they consider it better use of that address space to just leave it mostly unused and make it difficult for their customers to use IPv6, not even noticing that there is a reason why RIPE assigned them a /32 ...

mappu 12y ago
Correct me if i'm wrong, but doesn't an IPv6 /64 give you 2^64 globally routeable addresses? How is that not sufficient?
- simoncion 12y ago
  You can't subnet a /64. This RFC [0][1] describes the IETF's thoughts on why allocating a block larger than /64 is to be strongly encouraged. [0] https://tools.ietf.org/html/rfc6177 https://tools.ietf.org/html/rfc6177 [1] Particularly, see page 3
- matteotom 12y ago

pmontra 12y ago

A more interesting report would be which ISP by country let their customers use IPv6, landline and mobile. If servers are IPv6 but people can access them only over IPv4 then we're still at IPv4.

p1mrx 12y ago
This data shows the fraction of users with IPv6, per ISP: http://www.worldipv6launch.org/measurements/ http://www.worldipv6launch.org/measurements/ And here is some similar data, but per country: http://www.google.com/intl/en/ipv6/statistics.html#tab=per-country-ipv6-adoption http://www.google.com/intl/en/ipv6/statistics.html#tab=per-c... The US just recently crossed 10%.