The data is collected from users that visit the site and take part in the test. Keep in mind that the data isn't conclusive as the end-users location, and his hardware, effects the providers offerings and the statistics.
Yes migration to IPv6 is essential, but virtual hosting for websites and NAT for ISP subscribers provides a way to indefinitely postpone IPmageddon. This never seems to be mentioned. Also once a block is assigned the actual usage is unpredictable and not reported on that I'm aware of - I know several folks who have blocks assigned by their DC and they're only using a fraction.
So far as I can tell NAT for ISPs is brought up pretty much every time that IPv6 is mentioned on HN.
And every time people point out that it breaks the fundamentally decentralised nature of the internet and makes it harder to run P2P services.
And also, pretty much every time someone mentions unused blocks, and someone else points out that even if we reclaimed the largest known blocks that aren't publicly routed this would only expand the timeline for a few months.
Comcast has laudably been on the forefront of deploying IPv6. And several other major consumer ISPs in the U.S. are working on deploying it as well (and in the interim, AT&T is offering tunneling with 6RD).
Amazon, on the other hand, still does not support IPv6. I browse the Web from Comcast with a Firefox extension that tells me if a site is using IPv6 or not, and the lack of websites supporting IPv6 concerns me way more than the pace of IPv6 rollout by consumer ISPs. There's simply no conspiracy here.
There isn't much reason for servers to support IPv6 yet. Clients can use 6to4, and will use it as soon as they become IPv6 only, because... well, there isn't much reason for servers to support IPv6 yet. (Yep, that's circular.)
Servers will add support for IPv6 in due time, when admins get used to having it on their machines, and all network equipment supports it, and it stops breaking the sites for people on IPv4 only and people behind slower IPv6 tunnels.
And with that, the transition of the internet from a peer to peer network to a client/server model is finally completed.
/not directed at you specifically; I'm just depressed from how efficiently the empower, democratic, anybody-can-publish qualities of the internet are being undone
We've already completed that transition when ISPs started blocking ports from their customers.
Putting everybody in IPv6 will make things better, and odds are that it will create an entire new crop of descentralized services. But current http and smtp servers gain very little from supporting IPv6 just now.
Actually there is a nice reason for any reasonable sized stack to support v6. it makes life easier.
you run globally addressable ranges that just make life easier as you get to remove all those wonderful dnats and snats in your networks. as your internal ranges become easier to manage and more plentiful so you can layout your networks in ways that make sense. group services in IPs that make sense without ever really worrying about exhausting those ranges.
comcast for example moved to v6 because they ran out of rfc1918 to run their administrative networks in.
I think you mean NAT64/DS-Lite/MAP (which are all types of carrier-grade NAT), since 6to4 is just a means of connecting two IPv6 endpoints across an IPv4 tunnel.
When an ISP funnels multiple users through the same IPv4 address, you can no longer use IP-based bans without inflicting collateral damage. One way around this problem is for both the ISP and the website to support bigger addresses.
> Clients can use 6to4, and will use it as soon as they become IPv6 only
I think you mean NAT64. 6to4 is an IPv6-over-IPv4 tunneling mechanism, and thus would not be used by an IPv6-only host.
> Servers will add support for IPv6 in due time, when admins get used to having it on their machines, and all network equipment supports it, and it stops breaking the sites for people on IPv4 only and people behind slower IPv6 tunnels.
This is largely FUD. Network equipment has supported IPv6 for a while, IPv6 does not break sites for IPv4-only users, and there's no reason to believe that slow IPv6 tunnels are a significant problem. The performance and reliability myths were entirely debunked by the highly successful World IPv6 Day and World IPv6 Launch Day in 2011 and 2012, which is why forward-thinking companies like Google and Facebook have permanently enabled IPv6.
Running a box that smoothly maintains hundreds of millions of sessions (a household might would peak at about 1000 concurrent sessions) is a different expense than running boxes that just statelessly route the packets for those hundreds of millions of sessions.
You actually don't need state to share an IPv4 address, if you statically allocate port ranges to customers, and route packets based on the address+port.
But rather than build special routers that understand port ranges, it makes more sense to map each address+port into an IPv6 prefix, and use standard IPv6 routing. That's the basis of MAP:
Right now, ARIN IPv4 addresses are easier to get than ever before, as providers are handing out IPv4 addresses out like candy so that they have as much IPv4 space under their control as possible when ARIN runs out.
The next phase, as has already begun in the RIPE region, is where IPv4 addresses are bought and sold on the private market. Right now, they go for published rates of between $7-$15 per IP, depending on the size of the block they're sold as. This will probably go up dramatically once all the regional registries (and in particular ARIN) run out.
We'll probably then see the price of IPv4 addresses go up higher and higher, as utilization efficiency increases. It remains to be seen how high the costs will go, but it will probably get to a point where the costs of acquiring IPv4 will generally be higher than the costs of implementing IPv6. It will likely only be then that we see wholesale adoption of IPv6. Large portions of the Internet may begin using it before then, but the problem is that IPv4 will continue to be necessary until the whole of the Internet is on IPv6.
Generally, a /24 is the smallest block for which most networks will accept BGP advertisements. You'd generally purchase a /22-/20 or larger though. Note that in most cases, these are (part of) blocks that were assigned before the current regional registries existed, as blocks allocated through the registries aren't actually owned.
Funny you make that comparison and take away the completely wrong conclusion from it.
You do know IPv4 addresses are a mathematically limited set of numbers right? Once you're out of it, there's no "let's make up more numbers" or "Maybe we'll look in other dimensions" (funnily enough that's what NAT does). Not for oil, not for ipv4, and not for green parallel lines that form a red circle.
What really surprises me is that by this graph it really seems APNIC is better rationing its allocation the most of the other RIRs with it projecting still having small numbers of IPs available for crossover after 2020.
Part of this is that APNIC have limited all new applicants to a /22 when they hit a final /8 available. Existing registrants were offered an ADDITIONAL /22 as their final allocation, and due to successful hand-back of unused space, an additional /22 allocation is available for existing registrants.
It seems to ensure that small players (such as startups) can still enter the market and build a viable product without having to worry about it until they are using more the 1024 IPs.
Interestingly, the allocation rate has picked up again in the last month or so. I'm guessing that's related to the recent announcement about the "IANA IPv4 Recovered Address Space".
I find it very interesting how few /8s are actually advertised vs how many are handed out -- if those graphs are collect then the difference is a quarter of all the IPs???
27 comments
[ 2.0 ms ] story [ 81.3 ms ] threadThe data is collected from users that visit the site and take part in the test. Keep in mind that the data isn't conclusive as the end-users location, and his hardware, effects the providers offerings and the statistics.
And every time people point out that it breaks the fundamentally decentralised nature of the internet and makes it harder to run P2P services.
And also, pretty much every time someone mentions unused blocks, and someone else points out that even if we reclaimed the largest known blocks that aren't publicly routed this would only expand the timeline for a few months.
Well, those are two things ISPs don't care about...
Amazon, on the other hand, still does not support IPv6. I browse the Web from Comcast with a Firefox extension that tells me if a site is using IPv6 or not, and the lack of websites supporting IPv6 concerns me way more than the pace of IPv6 rollout by consumer ISPs. There's simply no conspiracy here.
Servers will add support for IPv6 in due time, when admins get used to having it on their machines, and all network equipment supports it, and it stops breaking the sites for people on IPv4 only and people behind slower IPv6 tunnels.
And with that, the transition of the internet from a peer to peer network to a client/server model is finally completed.
/not directed at you specifically; I'm just depressed from how efficiently the empower, democratic, anybody-can-publish qualities of the internet are being undone
Putting everybody in IPv6 will make things better, and odds are that it will create an entire new crop of descentralized services. But current http and smtp servers gain very little from supporting IPv6 just now.
you run globally addressable ranges that just make life easier as you get to remove all those wonderful dnats and snats in your networks. as your internal ranges become easier to manage and more plentiful so you can layout your networks in ways that make sense. group services in IPs that make sense without ever really worrying about exhausting those ranges.
comcast for example moved to v6 because they ran out of rfc1918 to run their administrative networks in.
This photo sums up the problem with relying on CGN: https://www.flickr.com/photos/jfesler/6751925977/
When an ISP funnels multiple users through the same IPv4 address, you can no longer use IP-based bans without inflicting collateral damage. One way around this problem is for both the ISP and the website to support bigger addresses.
I think you mean NAT64. 6to4 is an IPv6-over-IPv4 tunneling mechanism, and thus would not be used by an IPv6-only host.
> Servers will add support for IPv6 in due time, when admins get used to having it on their machines, and all network equipment supports it, and it stops breaking the sites for people on IPv4 only and people behind slower IPv6 tunnels.
This is largely FUD. Network equipment has supported IPv6 for a while, IPv6 does not break sites for IPv4-only users, and there's no reason to believe that slow IPv6 tunnels are a significant problem. The performance and reliability myths were entirely debunked by the highly successful World IPv6 Day and World IPv6 Launch Day in 2011 and 2012, which is why forward-thinking companies like Google and Facebook have permanently enabled IPv6.
But rather than build special routers that understand port ranges, it makes more sense to map each address+port into an IPv6 prefix, and use standard IPv6 routing. That's the basis of MAP:
http://en.wikipedia.org/wiki/Mapping_of_Address_and_Port
Of course, the scheme is useless if you don't have an IPv6 network in the first place.
BTW, speaking of MAP: the CPE-side code is now in OpenWRT trunk.
I have built a few images here, in case someone is interested to test: http://stdio.be/blog/2014-06-10-OpenWRT-MAP-images-from-pure...
(Needless to say, would love to hear feedback)
Right now, ARIN IPv4 addresses are easier to get than ever before, as providers are handing out IPv4 addresses out like candy so that they have as much IPv4 space under their control as possible when ARIN runs out.
The next phase, as has already begun in the RIPE region, is where IPv4 addresses are bought and sold on the private market. Right now, they go for published rates of between $7-$15 per IP, depending on the size of the block they're sold as. This will probably go up dramatically once all the regional registries (and in particular ARIN) run out.
We'll probably then see the price of IPv4 addresses go up higher and higher, as utilization efficiency increases. It remains to be seen how high the costs will go, but it will probably get to a point where the costs of acquiring IPv4 will generally be higher than the costs of implementing IPv6. It will likely only be then that we see wholesale adoption of IPv6. Large portions of the Internet may begin using it before then, but the problem is that IPv4 will continue to be necessary until the whole of the Internet is on IPv6.
You do know IPv4 addresses are a mathematically limited set of numbers right? Once you're out of it, there's no "let's make up more numbers" or "Maybe we'll look in other dimensions" (funnily enough that's what NAT does). Not for oil, not for ipv4, and not for green parallel lines that form a red circle.
Part of this is that APNIC have limited all new applicants to a /22 when they hit a final /8 available. Existing registrants were offered an ADDITIONAL /22 as their final allocation, and due to successful hand-back of unused space, an additional /22 allocation is available for existing registrants.
It seems to ensure that small players (such as startups) can still enter the market and build a viable product without having to worry about it until they are using more the 1024 IPs.
http://www.potaroo.net/tools/ipv4/fig27h.png
Interestingly, the allocation rate has picked up again in the last month or so. I'm guessing that's related to the recent announcement about the "IANA IPv4 Recovered Address Space".