Ask HN: Why are we so reluctant to use IPv6?
15 comments
I have been working with IP based networks for close to two decades. I have yet to have a client who really wants and needs IPv6. The main reason being that the majority of nodes do not need to access any arbitrary node over the internet. Thus for the most part nodes happily exist behind RFC1918 private networks. Their edge NAT and the carrier's NAT provide accessibility in spite of the address space limitations of IPv4.
Of course, there are use cases for IPv6 and where it is the best solution it is being used. But where IPv4 gets the job done it is very hard to justify replacing equipment just for the sake of being IPv6 compliant.
The use of NAT64 is helping many organisations to delay migrating to IPv6. From a purists point of view it is probably not an ideal solution. But it seems to keep the pragmatists happy.
Of course, there are use cases for IPv6 and where it is the best solution it is being used. But where IPv4 gets the job done it is very hard to justify replacing equipment just for the sake of being IPv6 compliant.
The use of NAT64 is helping many organisations to delay migrating to IPv6. From a purists point of view it is probably not an ideal solution. But it seems to keep the pragmatists happy.
I have been working with IP based networks for over 20 years, and specifically IPv6 networks for over 17 of those. IPv6 is the main reason I got my current job, FWIW.
> The use of NAT64 is helping many organisations to delay migrating to IPv6.
That's...not how NAT64 works. NAT64 is fairly dependent upon deploying IPv6, like T-Mobile's 11+ million IPv6-only users who use their NAT64 platform.
I do agree that from a purist's point of view, it's not ideal, but it enables network operators to stay largely single-stack (on v6), while still facilitating access to the IPv4 internet.
> The use of NAT64 is helping many organisations to delay migrating to IPv6.
That's...not how NAT64 works. NAT64 is fairly dependent upon deploying IPv6, like T-Mobile's 11+ million IPv6-only users who use their NAT64 platform.
I do agree that from a purist's point of view, it's not ideal, but it enables network operators to stay largely single-stack (on v6), while still facilitating access to the IPv4 internet.
I tried IPv6 on my business modem years ago. Per the ISP's top engineer, many firmware bugs made a static setup impossible at that point in time.
Amazon only added it to EC2 instances a year or so ago.
Most corporate network and infosec guys I've talked to about it are dragging their feet as much as they can--network guys because if they don't have to have it, they don't want it, and infosec guys because it's one more attack surface, and is not as easy to enumerate as IPv4.
Personally, if it gets us back to old-school disintermediated communications, I'm excited about it. In the back of my mind, I fear the powers-that-be will come up with some bullshit to NAT the hell out of IPv6 as well. Disintermediation will de-rail quite a few gravy trains!
Amazon only added it to EC2 instances a year or so ago.
Most corporate network and infosec guys I've talked to about it are dragging their feet as much as they can--network guys because if they don't have to have it, they don't want it, and infosec guys because it's one more attack surface, and is not as easy to enumerate as IPv4.
Personally, if it gets us back to old-school disintermediated communications, I'm excited about it. In the back of my mind, I fear the powers-that-be will come up with some bullshit to NAT the hell out of IPv6 as well. Disintermediation will de-rail quite a few gravy trains!
Same for me: Had to use IPv6 via my home cable uplink. The IPv4 gateway was located at the ISP. This translation service was called "DSLite". The performance was horrible, connections frequently dropped. This was basically last year.
Now I pay more (like 35€/m instead of 20€/m) to get an IPv4 only connection.
Now I pay more (like 35€/m instead of 20€/m) to get an IPv4 only connection.
Funny. For me the carrier grade NAT via DSLite performs better. If i run full dual stack my router is getting warmer, and there are dropouts/pauses. IPv4 only is nothing special. Best performance of all is when only IPv6 is involved, like unbelievable pings and rtts over the atlantic. I tested that a long time ago because it could switch it at will, and after a reconnect it worked like ordered. Anyways, it also depends on the peering of your ISP and the involved devices.
Interesting, I never heard of practical advantages of IPv6 for the end user, such as dramatically better ping times. Given the simplified routing, I can imagine this to be true. Are there applications where this pays off, such as reduced delays in cross-atlantic video chats?
Because dual stack is a hassle. It doubles your operational workload - which is a big deal in a large network. IMHO, we'll only see large scale IPv6 adoption when it becomes possible to deploy an v6 only network and translate to/from v4 at the edge for legacy edge cases.
Facebook already does this - everything is IPv6 only inside their datacentres, and the only dual stack devices are their load balancers at the edge. They had to do this because they were running out of RFC1918 address space to use internally for IPv4, and the traffic was already going through the dual stack load balancers anyway.
On the access side, mobile networks are also going v6 only. EE here in the UK (sometimes) only gives you an IPv6 network and uses 464XLAT to NAT you when you want to reach an IPv4 only address. They can do that because they know what devices all their users are using, and can whitelist recent versions of both Android and iPhone which have CLATs in them which makes 464XLAT possible.
Fixed line residential and business connectivity is much harder - laptops and smartphones will work just fine with IPv6 only, but people expect their games console or old networked print server to work, and frequently they only support IPv4. What's needed there is islands of IPv4 on their LAN and a 464XLAT compatible CLAT inside the customer's router - then the entire ISP network can be v6 only apart from a few NAT64 devices at the edge. I'm sure that will come, but it hasn't happened yet.
It will happen - mass adoption is inevitable IMHO. There are new internet users and services being added every day, and there aren't enough IPv4 addresses for all of them. It's either IPv6 and 464XLAT or IPv4 only with CGNAT. And CGNAT is expensive - keeping state for all those connections makes for expensive boxes with lots of memory.
Facebook already does this - everything is IPv6 only inside their datacentres, and the only dual stack devices are their load balancers at the edge. They had to do this because they were running out of RFC1918 address space to use internally for IPv4, and the traffic was already going through the dual stack load balancers anyway.
On the access side, mobile networks are also going v6 only. EE here in the UK (sometimes) only gives you an IPv6 network and uses 464XLAT to NAT you when you want to reach an IPv4 only address. They can do that because they know what devices all their users are using, and can whitelist recent versions of both Android and iPhone which have CLATs in them which makes 464XLAT possible.
Fixed line residential and business connectivity is much harder - laptops and smartphones will work just fine with IPv6 only, but people expect their games console or old networked print server to work, and frequently they only support IPv4. What's needed there is islands of IPv4 on their LAN and a 464XLAT compatible CLAT inside the customer's router - then the entire ISP network can be v6 only apart from a few NAT64 devices at the edge. I'm sure that will come, but it hasn't happened yet.
It will happen - mass adoption is inevitable IMHO. There are new internet users and services being added every day, and there aren't enough IPv4 addresses for all of them. It's either IPv6 and 464XLAT or IPv4 only with CGNAT. And CGNAT is expensive - keeping state for all those connections makes for expensive boxes with lots of memory.
I disabled it a long time ago on Arch because of kernel/driver bugs and haven't had any reason to enable it since. There is no compelling need for ipv6, which is a shame, because the protocol is much simpler than ipv4. So much of the internet could be much better if we didn't have to keep compatibility with protocols developed in the 70's.
What bugs? I find Arch works nearly flawlessly with IPv6. The only "bug" I found was that wiki.archlinux.org did not respond via IPv6.
I don't remember. I think it was something with the iwlwifi driver. It could also have been the "broken dns servers breaks ipv6 in FF"-issue. Probably has been fixed by now but there is no reason for me to turn ipv6 on.
One reason it hasn’t been adopted very quickly is because the RFC basically rearchitected layer 3 in the OSI. The network topology for IPv6 changes significantly. It wasn’t just, “let’s make up addresses longer”. Adoption has been slow enough to call the RFC a failure at this point. I personally think it will slowly fizzle out and then be viewed as a waste of time and a less drastic step will then be introduced by a corporation that we’ll all end up adopting.
What? There's no alternative to IPv6. Nobody is making one. If someone did make one, it would be stuck on hundreds of vendors making thousands of new pieces of equipment in just the same way as IPv6. It would be stuck on the immense training costs of a new IP system, just the same as IPv6. It would get stuck on the inertia and change-aversion of network people and purse-holders, just the same as IPv6.
IPv6 was a clean break from IPv4 with a lot of changes because the authors knew that there was a very high 'minimum cost' to any IPv4 replacement. Even if they only added more octets to the addresses, there were still billions of networking devices which would need replacing, millions of people to retrain to some degree, mountains of software to modify or rewrite, and a similar conscious decision to switch. If we had to go through this pain anyway, then the obvious choice was to go all-out designing a greenfield new protocol which solved as many problems as possible and which would last for as long as possible. That's IPv6.
IPv6 is here to stay. It's already 30% of the internet, and has industry-wide buy-in and backing. All OS' support it, Apple even requires apps to work in IPv6 only environments to get on the app store. New IPv6 deployments go online every day.
IPv6 will win. One day. The costs of keeping IPv4 around are high and increasing. ISPs are spending millions on specialized CG-NAT infra to keep their dwindling pool of IPs usable, in return for tremendous capital and maintenance costs, reduced performance, increased support burden from angry customers who had their app broken by CG-NAT. And anyone who wants to buy an IP address has to deal with the price - which just keeps going up.
IPv6 was a clean break from IPv4 with a lot of changes because the authors knew that there was a very high 'minimum cost' to any IPv4 replacement. Even if they only added more octets to the addresses, there were still billions of networking devices which would need replacing, millions of people to retrain to some degree, mountains of software to modify or rewrite, and a similar conscious decision to switch. If we had to go through this pain anyway, then the obvious choice was to go all-out designing a greenfield new protocol which solved as many problems as possible and which would last for as long as possible. That's IPv6.
IPv6 is here to stay. It's already 30% of the internet, and has industry-wide buy-in and backing. All OS' support it, Apple even requires apps to work in IPv6 only environments to get on the app store. New IPv6 deployments go online every day.
IPv6 will win. One day. The costs of keeping IPv4 around are high and increasing. ISPs are spending millions on specialized CG-NAT infra to keep their dwindling pool of IPs usable, in return for tremendous capital and maintenance costs, reduced performance, increased support burden from angry customers who had their app broken by CG-NAT. And anyone who wants to buy an IP address has to deal with the price - which just keeps going up.
Or at least because people incorrectly think it did.
v6's L3 model is pretty much the same as v4's.
v6's L3 model is pretty much the same as v4's.
it just takes time. there are a _LOT_ of systems that need to be updated/ perhaps replaced. have not installed a ipv4 only system for a few years. but still have a huge amount of systems that is not migrated. and probably will not be until there is a revision or some reason to touch it.
I'd have to call my ISP to get it.
Why is it that were not using it more?