Why IPv6 Hasn't Taken Off

I gave IPv6 a serious try. I disabled IPv4 altogether and used only IPv6 mainly to learn it. After running it for a while I discovered some of the shortcomings that are holding it back from widespread public adoption.

[petruspotgieter]

IPv6 is around 30 years old, but it still has growing up pains. It was designed before mobile broadband, before small businesses and home users started multihoming. Too many IPv6 cheerleaders saw NAT as a weakness of IPv4 instead of a flexible tool which goes well beyond the "temporary" fix for public addresses exhaustion. Then there are the dozen different ways of v4 to v6 migration and interworking. Also Apple, MSFT and google pushing different paths. Although you can get a PI (provider independent) /48 adress from your RIR, it is just not scalable for the hardware on Internet core routers to handle routing tables with a billion entries which wold result from everyone getting this.
IPv6 wihout translation only really works well for big institutions with a fixed PI allocation and BGP multihoming to ISPs or smartphones with temporary /64, which also allows for temporary hotspot.
For small business and home internet I think the solution is ULA (ptivate IPv6) for the LAN with stateless network prefix translation to the WAN prefixes from ever changing ISPs, which could even be multiple concurrently.

[jonathancrowder]

So basically local NAT went from "we have to do it" to "it's a feature I rely on"

[vanderhooftamvl]

No video cuts, no animation, no notes, yet so clearly explained!

[GregInHouston]

I worked with IP6 a little bit 20 years ago. Problem 1: equipment that can't handle it. Problem 2: those address are so damn long. Unless your address has lots of continuous zeros in it, it is very difficult to remember. Problem 3: they keep making changes. Once, they had a standard way to translate IP4 addresses to IP6. Problem 4: If NAT fails on IP4 then nothing gets in. If the firewall fails on IP6, all of your computers are now internet addressable.

[leandrotami]

IPv6 would allow people to easily self host services like photos in their homes instead of eternally paying a subscription or depending on external providers. I'd love to see it happenning but as time passes what I see is that providers like me don't use them, instead they have their own big private networks and do NAT to give you public internet access. They charge me extra for a public IP (I mean public, not static)

[fletchzz]

You need to control your DNS. The IP stack is important but jumping up a layer and controlling your DNS, that's the key.

[deathgripsonline]

Just discovered your channel through this video popping up on my feed. Just beginning my career, bout to finish uni. Interesting stuff, gives me an insight as to why IPV6 wasn't adopted earlier. They certainly push "we're running out of time to switch" message in school.

[swelarra]

I think the biggest problem with your idea is that if people started doing that, the bgp table would become way too big for any normal router to handle. Even today, not all ISPs keep a full bgp routing table since it consumes more ram than their equipment could handle. Imagine if we were to keep track on every customer subnets in one bgp routinf table, that would get impossible to manage. This is what I think, but I might be wrong, it happens a lot.

[bobthecannibal]

My question is, why are they delegating a whole 80 bits worth of /48 to an individual? Wouldn't it be smarter to delegate 24 bits at a time? (a whole /8 in ipv4.) I, as a home user, use 10/8 internally and could probably assign static addresses for every IP-capable thing I own, will own, or have *ever* owned and not run out in my lifetime. That's 16 million addresses for crying out loud.)

It reeks of the same thing that the initial classful IPv4 routing stank of "There's plenty of room. Here, Ford, have 18/8. Sure, it's 1995 and you only sold 317, 621 vehicles and you're making cheap and practicallly disposable disgusting sh*tboxes without any computers that few will want in 30 years time, but go ahead and have 1/255th of the address space. (Oh, and we're going to sit on 1/8th of the address space for "future use" that won't ever happen because of artificial scarcity.)

If you want IPv6 to take off, lobby to have IPv6 delegations *decreased* in size and deprecate IPv4, telling all of those hogs sitting on their /8s that they can keep their addresses, but IPv4 is going away and "your reserved netblocks are turning into a tiny delegation of half of a 64 bit prefix, GL;HF. Congrats, your puny 16 million address netblock is practically valueless now that there's over 340 undectilion addresses available."

[PingJerry]

While I love the idea of everyone gets their own /48, I think you're forgetting one of the biggest issues faced with ipv4, route table bloat. While an ipv4 entry only takes 12 bytes, an ipv6 entry takes 48 bytes, while that sounds small, this would be for every single user that wants their own subnet. Is that sustainable for even a moderate sized isp? pppingme

[MaidLucy]

You are doing IPv6 Prefix Delegation the wrong way. The idea is that your router manages an address pool that you can use to delegate smaller prefixes for your local networks from. I get my prefix via DHCPv6 from my ISP and my router will then announce different prefixes delegated from the pool to subnets on the LAN side. The important part is that your router will adapt a new prefix and distribute that when it changes.
Also, DynDNS has existed for years and I have been behind a dynamic IP address for years. The same applies to IPv6, just that instead of your router taking care of the DynDNS, now your individual clients have to take care of that.
On top of that, dnsmasq for example is a great solution to keep a dynamic DNS in your local network. It will detect changes in IP addresses and it will then distribute the new IPs with the DNS requests.
And if you set this up properly, you will never have to remember an IP address ever again and you will never have to set up static DNS for any machine.
Also, my internal IPv6 network also has a private prefix which makes really handy and short IPv6 addresses like fd97::1
That's even easier to remember than IPv4, isn't it?

[HenryLoenwind]

To be honest, one of the major "features" of IP4 that still makes me stay with it is that using a private address range and NAT seals my network. I don't have to trust a firewall to block things I don't want; that's the default. Unless I manually add a port forward, nothing gets in that's not a reply to an outgoing request. I don't have to worry about the watering timer on my garden faucet being hackable. It "physically" cannot be reached from the outside, no matter how good or bad the config of my firewall is.

And that's with a network I---someone who knows how to configure a firewall reasonably well---am managing. The same goes doubly with 99% of consumers. I still remember the times of Windows 98, when people were dialling in with their PCs (so no router or firewall involved), and you couldn't even set up a new PC and download all the updates without it being loaded to the brim with malware in the meantime.

Directly routing incoming internet traffic to a device that wasn't built specifically for it is folly.

[IslandHermit]

15 years ago I was excited about IPv6. Finally I would be able to get permanent, global addresses for all of my systems. Then I discovered what you talk about here, that the gate keeping on the IPv6 addresses is just as bad, if not worse, than IPv4. I haven't bothered with IPv6 since.

[JohnGotts]

Over the past decade, I've added IPv6, HTTP/2, and TLSv1.3 support to network software I've worked on. If customers ask for it, companies will sell it to you. If not, it will remain on developers' laptops and never be released publicly. Don't be afraid to ask for better IPv6 support. If enough tickets roll in your use case will become supported.

[fredflintstone]

I disabled IPv6 years ago on my network, in my router. I was noticing laggy behavior with my win 10 machines. I started monitoring a bit with Wireshark and noticed a whole lot of IPv6 traffic, and a lot of it going to Microsoft..
After turning off IPv6 on my machines and the router my networking performance increased quite a bit. To the best of my knowledge I haven't noticed any problems as a result.
I don't have an issue with NATing.
My ISP does traffic shaping so the upload speeds are pitiful by comparison. Of course this is going to work best for customers that do a lot of streaming. When calling my provider they suggested that I pay for a commercial account if I want faster upload speeds which costs 3 times as much per month than the standard home user account that I am using. I came up with some other solution which put my servers external to my network at a fraction of the cost.

[wojciechmikoajewicz]

You cannot have what you want. It's the same reason why Internet doesn't work on MAC addresses but on IP addresses. IP addresses are assigned to countries and then to ISPs. Having your own IP address means that addresses would have to be randomly distributed across the world. So Internet routers would have to remember each individual IP address - where to route traffic to that address. Switches do exactly that - they remember all MAC addresses assigned to a switch port. Customer switches can have memory for let say a thousand MAC addresses.
Internet routers work completely different. They work on addresses classes, not individual addresses. They do something like that - I have packet to 130.133.x.x - oh it's a Germany - I should send it through my D port. It doesn't have to remember each individual IP address, it remembers whole address classes. It's the only technically possible solution.

[stargazer]

Internet routing would be impossible if everyone had a permanent portable ipv6 prefix. Route tables would have to be ridiculously large.

[elslopez]

This is an EASY question to answer and it is the same reason the internet does not allow /32 IPv4 prefixes to be BGP routed across the internet. In short the routers that actually route the internet could not physically handle the amount or routes you would be talking about in this massivly deaggregated scenario. The current size of the IPv4 global table is just over 900K routes and the IPv6 global table is about 175K routes. You start allowing people to have their own /48 that they own and can take to what ever ISP they like, this will explode the size of the routing table many orders of magnitude. I do not know of any commercial grade routers that could currently hanld a fraction of this, and certainly 99% of the ISPs will not want to spend the money even if the hardware did exist. Maybe in another 20 years... we'll see!

[daffy]

i guess there is a reason BGP is not very open to public. I can only imagine how routing tables would blow up if people would be given private Ipv6 ranges.

[crabby]

I am no expert, but I did a couple of cisco networking courses over the years and it was noticeable how the attitude towards ipv6 had changed over those years. First time it was all enthusiasm and everyone would get personal ips because there were plenty to go round etc etc. Years later they were much more muted and basically came to the conclusion that there were just as many security issues under 6 as there was under 4, Nat was still a good idea for security, and people weren't going to get personal numbers except for the self configuration number which may possibly cause their own problems.
Then there is the issue of all the ipv6 traffic that seems not to be noticed by some security software, so you have unknown automatic activities occurring that you don't know are occurring if you normally concentrate on ipv4 still.
Throw in the long ipv6 numbers and the autoconfigure oddities, it is perhaps not surprising that anyone apart from the big boys find it easier to stick with ipv4. Nat means the address shortage isn't as acute as first stated.
It would perhaps have been more sensible to just double the bit length of the addresses rather that go mad.