It's so much cheaper and faster already today that it makes sense to do v6-only internally and deliver IPv6 to the large fraction of customers who can have IPv6 and deploy a transform layer for the (gradually shrinking) class of have-nots.
Your IPv4-only competitors are spending more money to deliver a worse service. Maybe they'll wake up and realise, but the story of the past couple of decades suggests most such outfits are not too bright and when offered IPv6 on new links they'll smugly decline. Meanwhile you get to keep the difference. Your competitor thinks this service costs 86¢ per dollar revenue and you're only spending 84¢ per dollar revenue that's a lot of extra profitability.
Fighting this is a losing battle, and I imagine that decades from now somebody is going to write a book about how it was cognitively possible for "leaders" to champion an idea that couldn't possibly work, that they intellectually knew couldn't work, and all the actual evidence said wasn't working, and yet they believed in it anyway. Or that book might be about climate change denial, either way.
Eventually by the way - and 90% global is close to and even perhaps at the threshold where that's plausible for some of them - many general ISPs won't bother providing actual IPv4 networking. If you want IPv4 you're a niche customer, go find a bespoke IPv4 Internet Service Provider. If you're IPv6 capable but you want to reach an IPv4 host you'll go via a translation box automatically. As the interest in global IPv4 shrinks, the transit companies will lose interest in that service too, and eventually - decades from now - the IPv4 Internet literally goes away, the RIRs stop "managing" the now largely unused space and the residual pockets of IPv4 cease to interoperate.
Absolutely not. Routers, OSes and applications still have much better support for IPv4 than IPv6.
Replacing large routers, fighting bugs and running pure IPv6 on internal networks is very expensive.
Also I'm not talking about routers worth 50 euro/$, but DC and carrier grade, worth 500K euro/$
ISPs and many enterprise sized orgs have been running v6 for a long time, so it's not like it's a checkbox feature that hasn't had a lot of testing.
Of course they do. They are hugely complex systems and troubleshooting and tuning is prioritized based on customers needs.
I’m also going to humor you on the IPv6-only thing. Can you actually point to a single ISP that is even considering IPv6-only?
Of course, IPv6 has problems in some parts of the world. IPv6-only usually means, there is IPv4 on some central gateway but your clients don't get any IPv4 assigned. There is a number of really big networks doing just that e.g. most of the mobile operators in the USA.
[0] https://developer.apple.com/videos/play/wwdc2020/10111/ [1] https://developer.apple.com/support/ipv6/
Without any hard data and like-to-like comparisons, the most that can be said is that IPv4 and IPv6 have different routing policies and this may attribute the differences in latency. As such what and where you test will affect your results.
That being said, Apple’s IPv6 policy is probably the best thing being done to encourage adoption.
However, they are not adopting the policy due to performance reasons, but due to necessity given who their ultimate customers are (cell phone companies).
IPv6-only as a term has been heavily abused. A truly IPv6-only network has no access to IPv4 resources. If it does, it isn’t really IPv6-only and relies upon IPv4 addresses and NAT to work.
But it doesn't stop there, you also no longer need subnet planning -- where under IPv4 it's very important whether this new subnet is 60 machines or 64 machines, in IPv6 you don't care and so nobody needs to manage that. Which translates to lower costs in your network team. Negligible when your "network team" is just the CTO doing a Google search, but significant at scale.
Your route costs fall too. Because of exhaustion (and because years ago nobody had any idea what they were doing, not having had an Internet before) IPv4 is now badly fragmented, which means a trivial route ("All the East Coast stuff") may become six or sixteen or sixty separate address blocks, but IPv6 blocks were deliberately assigned so as to reduce fragmentation, chances are "All the East Coast stuff" is one or two blocks.
As to speed, Facebook reports IPv6 is about 10% faster for them.
They did exactly what I described, they're v6-only internally and have translation at the edge for "legacy" IPv4 clients.
This is just silly. If you are an ISP then you already have an IPv4 allocation.
If you don’t then you join the waiting list and/or apply for a /24.
Once you have your allocation, RIR fees are the same regardless of IPv4 or IPv6 allocations.
> But it doesn't stop there, you also no longer need subnet planning -- where under IPv4 it's very important whether this new subnet is 60 machines or 64 machines, in IPv6 you don't care and so nobody needs to manage that. Which translates to lower costs in your network team. Negligible when your "network team" is just the CTO doing a Google search, but significant at scale.
If you believe this is in any way meaningful then I have a bridge to sell you.
> Your route costs fall too. Because of exhaustion (and because years ago nobody had any idea what they were doing, not having had an Internet before) IPv4 is now badly fragmented, which means a trivial route ("All the East Coast stuff") may become six or sixteen or sixty separate address blocks, but IPv6 blocks were deliberately assigned so as to reduce fragmentation, chances are "All the East Coast stuff" is one or two blocks.
You can’t buy smaller routers just because you deploy IPv6, so I have no idea what cost saving there are to be had here.
> As to speed, Facebook reports IPv6 is about 10% faster for them.
I fail to see how this is universally applicable to all ISPs.
> They did exactly what I described, they're v6-only internally and have translation at the edge for "legacy" IPv4 clients.
Having an IPv6 core is not the same thing as being IPv6-only.
Speaking of IPv6-only, you have yet to name even a single ISP that is even considering going IPv6-only. Cat got your tongue?
Still, while we're here:
> This is just silly. If you are an ISP then you already have an IPv4 allocation.
Due to exhaustion in fact you can't get new allocations in IPv4 in most of the developed world. If you want to expand, you're going to need to buy addresses on the open market for that sort of $40 price we talked about.
> ... Once you have your allocation, RIR fees are the same regardless of IPv4 or IPv6 allocations.
Nope. Let's take ARIN, suppose you're a large international company, hundreds of local corporate networks. With IPv4 this is quite a struggle, maybe you can get away with a /17 for which ARIN will charge you $4000 per year.
But with IPv6 this is easy, you'll qualify for their "3X-small" category with a /40 at only $250 per year, even if you choose a relatively wasteful address layout.
It gets worse! Unless you already had that /17 from years ago, chances are you had to cobble the addresses together from several distinct allocations. ARIN charges $150 per block aggregated in this way, if you need a dozen blocks that's $1800 extra because of the fragmentation. In contrast IPv6 isn't fragmented, it was deliberately allocated sparsely so it's easy to give you adjacent blocks to grow. Not that they'll need to when you're just an international company with a few hundred sub-networks, IPv6 is so wide.
As to the RIR fee thing, yeah, you are right about the ARIN thing. It differs by RIR and region, tho.
You are wrong about the allocation thing, tho. You can get new IPv4 allocations, even in ARINland, by joining the waitlist or to support a IPv6 deployment.
That is a goal, and perhaps the primary goal, but there are other goals as well. IPv6 makes the Internet itself more efficient by reducing routing overhead and it simplifies network management.
Of course, what will ultimately force ISPs and services to switch to IPv6 is IPv4 address space exhaustion. As the price of a public IPv4 address rises people will begin to more seriously consider an IPv6-only approach, even if it means losing a few users (especially since ISPs will have to make IPv6 available once such services become popular). It has taken longer than expected as various approaches have squeezed the last bits of utility out of IPv4, but it will eventually happen as several billion more people are connected to the Internet.
Similarly for hosting providers: as IPv4 addresses get more expensive, they may have to start charging their customers more and more for each one, at which point the customers may simply go only-/mostly-IPv6.
A single ISP can't do anything about the situation. They can provide IPv6 to their customers, but they still need IPv4 so their customers so they have a usable internet connection. It doesn't help them at all.
Which is part of the problem here: As long as IPv6 is not implemented by everybody the people who do implement it have no advantage.
This is how a lot of mobile/cell telcos already work. A T-Mobile presentation from 2018 (PDF):
* https://pc.nanog.org/static/published/meetings/NANOG73/1645/...
Here's a dude from T-Mobile explaining how they have (in 2017) over ten million end-customer devices without any IPv4 addresses assigned:
* https://www.youtube.com/watch?v=nNMNglk_CvE
For reaching services that are IPv6 only they use(d) DNS64 (with and without 464XLAT).
You cannot just deploy IPv6, take your IPv4 ball and go home.
ISPs will still need those IPv4 addresses, no matter how much they cost. ISPs may delay by deploying CGNAT, but once those IPv4 addresses run out there is no other option than to get out your checkbook or scale back your business.
Likewise with hosters. There is no choice but to use IPv4, damn the cost. It simply isn’t viable to go IPv6-only, as IPv4 is where the money/customers are.
This will not change until IPV6 somehow reaches critical mass.
Yes, that's what I'm saying. At some point going IPv6-only and paying for (CG)NAT64 may be cheaper than trying to buy more and more IPv4 addresses. ISPs would already have IPv4 addresses, but they'd simply use them for reaching the 'legacy' Internet, and they wouldn't be assigned to people's routers WAN interface.
> Likewise with hosters. There is no choice but to use IPv4, damn the cost.
It will probably not be the hosting company paying for the IPv4, but rather it will be the end customer, and they may not want to pay.
The first address may be free, but if you want multiple, then you could be charged for the 'extra' ones. Depending on the number of public services needed, some people may find it cheaper to use the IPv4 address(es) on the front-end of a load balancer and have their actual system be IPv6.
>It will probably not be the hosting company paying for the IPv4, but rather it will be the end customer, and they may not want to pay.
Of coarse it will be the hosting company paying for the IPv4 addresses as they will be the ones owning them.
Of coarse the end customers will complain and not want to pay, but pay they will as they have no other choice.
It’s neither here nor there whether the end customers run IPv4 or IPv6 internally, it’s the public IPv4 addresses that matter.
However, as I stated previously, they are only for niche uses such as hobby projects or internal services, not for use by the general public.
They will remain so until IPv6 reaches critical mass.
Ultimately the goal of IPv6 is to end IP scarcity. But it can only do this if close to 100% of users and services use IPv6. Seeing it like that it doesn't really matter much if IPv6 adoption is 30% or 60% or 90%. In none of these situations would an ISP connect its customers via IPv6 only or would anyone provide any service that requires IPv6.
I really wonder how IPv6 should ever become mainstream without any deployment strategy. "Let's tell everyone that IPv6 is nice and we need it due to lack of v4 addresses" obviously hasn't worked for more than 20 years.