I decided to commit to using IPv6 a couple of years ago and got my personal servers serving my sites in a dual-stack. I also got my home router set up for it, getting my subnet from my ISP and making sure that all of my home devices were getting addresses.
It was only after doing this that I realized the problem with IPv6 adoption in the US: the network sucks! Most ISPs (including mine) are using 6rd gateways to get their customers onto the IPv6 backbone. The gateway servers are poorly placed (rarely local) and totally overloaded. The performance was so bad that it was making my everyday web browsing experience feel broken. I killed v6 at home and haven't looked back.
Maybe some day ISPs will do a proper deployment with real routing and I will try it again.
That is generally the opposite of reality. IPv6 is generally faster than IPv4.
I don't know what "a couple of years ago" is, but all tier-1 ISPs have IPv6 backbones. There should be no reason for your access ISP to have to tunnel anything to get to a tier-1.
Your statement "Most ISPs (including mine) are using 6rd gateways to get their customers onto the IPv6 backbone" is kind of a red flag. If ISPs aren't connected to a backbone, can you even call them an ISP?
Some ISPs (I don't know if it's accurate to say most) have last-mile or middle-mile equipment that does not support IPv6 and they can't afford to upgrade it, hence kludges like 6rd. Of course good ISPs don't use kludges, but then you're getting into no true Scotsman territory.
This is the experience I had with CenturyLink, they support IPv6 via 6rd and it sucks. It's annoying I don't have native IPv6 support with CableOne either (and they're too cheap to bother upgrading anything beyond the DOCSIS head-ends to charge you more money), but I'd rather use a Hurricane Electric tunnel if my ISP can't give me native v6 in the first place.
I think you're conflating "backbone" with metro distribution.
I'm pretty sure that all of the big US ISPs have native IPv6 on the backbone, but very few can bring native IPv6 to your home.
Most of the big US ISPs are using 6rd: AT&T, CenturyLink, and Cox. As far as I know, Comcast is the only one with true IPv6 to the curb but I might be wrong about that.
Comcast is a dual stack network, and their IPv6 network is faster and better than their IPv4 network. Faster RTT between two endpoints, and overall I am happy to have IPv6 turned on.
On the last graph, it identify the current period as the begining of the early majority starting to switch over to ipv6. ISP are not the fastest institutions out there, and it will likely exist a large variance where v6 is either much greater or much worse than v4 traffic.
There exist this project that track performance for v6 vs v4. In some countries v6 has above performance to v4, in others places it is worse. You seems to had the bad luck of getting the worse end of the variance.
That's the same problem: your IPv6 is worse than your IPv4.
The lesson learned from 6bone, HE, 6to4, 6rd, and Teredo is that we need native IPv6 or nothing. Tunneled IPv6 is worse than no IPv6 due to its unreliability.
Tunneled IPv6 through HE is more reliable than the average big ISP 6rd gateway but it doesn't solve the problem of latency. More than likely, you're still crossing a significant amount of network distance just to get to their gateway and then however much farther to your destination. The latency just isn't worth it.
For home users, Netflix blocks Hurricane Electric (sometimes?) because it can serve as a proxy for non-US users. I haven't had it happen in every case, but it blocked users at work from connecting and at home I got US Netflix instead of Canadian.
Seems like one of the largest sources of IPv6 adoption is mobile phone networks. I'm on AT&T in the US and if I check "what is my ip" when on LTE it shows me an IPv6 address. Mobile now accounts for well over half of all web traffic, so if/when mobile networks are universally IPv6 then we'll basically be over the hump.
I like to change to IPv6 and my current ISP can make it happen (Ziggo, The Netherlands). However, I have currently my own externally accessible IPv4 address and I will lose that when I opt-in to IPv6 because Ziggo only offers IPv6 with DS-lite. DS-lite means I cannot access my network anymore over IPv4, so I will not change to IPv6.
(Hopefully obsolete) anecdote: in France, the ISP 'Free' offered IPv6 ahead of the competition (2009? I forget). I was glad to hop onto the bandwagon and enable that.
A while later (2014?), I was annoyed by my poor bandwidth. A bit of sleuthing revealed that IPv6 was the culprit: turning that off in the ISP's control panel doubled my bandwidth!
I know that a lot of internet routers didn't have IPv6 acceleration. IPv6 packets jumped out of the (fast) data plane to be processed by the (slower) control plane. At the time this made sense considering the rarity of IPv6. I suppose Free, being a cheap ISP, were still using these older routers.
In 2013, the telephony-focused router company I worked for still hadn't implemented data-plane processing of IPv6, because there was no requirement from it.
Hopefully in 2017, driven by Comcast or other IPv6-dependant phone operators, most routers do IPv6 acceleration.
Even worse, there are often huge bugs in routers or switches with regards to IPv6 that aren't often discovered because IPv6 is so rarely used.
I remember reading about Facebook's migration as a great case study. One example problem they had was that their switches couldn't handle getting a BGP feed with IPv6 addresses when they had not been configured for IPv6.
In other words, when they added IPv6 to their BGP feeds they crashed an entire network segment worth of switches.
This is another massive headache w.r.t IPv6 switchover: all the bugs and edge cases we've found with IPv4 thirty years ago are all brand new wild country problem with IPv6.
I'm not aware of any routers that can forward IPv4 in hardware, but pass IPv6 to the control plane.
Even the discontinued Catalyst 6500, which was used (and still is) by a lot of ISPs with the SUP720 supervisor, can do native IPv6.
I think you are completely wrong about this. Routers with native IPv6 were available 15 years ago, and are already getting pulled out to be replaced by the new stuff.
The reality is that, that ISP tier 1s use either Juniper MX or Cisco ASR9000 routers. These routers have good IPv6 performance.
A few years ago I remember hearing all about the IPv4 crisis and how we were going to run out of addresses very soon. How did we misjudge this so badly? It seems like it should be some pretty simple mathematics to figure out when exhaustion would occur?
There was no misjudgement. IPv4 addresses are still running out. Adoption of IPv6 has helped the issue. Some parts of the world are aggressively going IPv6 only. IPv6 adoption has seen a massive uptick in the last year. Check the google graphs.
Also some under-utilized IPv4 address blocks have been recently re-claimed. MIT recently sold off half of its class A to Amazon:
> MIT recently sold off half of its class A to Amazon:
And garbage like this is why I cannot wait for the IPv6-only future, CIDR let us stall IPv4 address exhaustion but it has caused severe bloating in routing tables. Considering an IPv6 route at the internet level is capped at 64-bits (the max length of a proper network prefix, you can technically do smaller ones but WHY, and nobody will take your BGP advertisement for it either) I expect even with a 2X increase in size occupied by each individual route the size of a full BGP table is going to be dramatically smaller.
We have already run out of IPv4 addresses. IANA allocated all blocks in 2011, Asia ran out in 2011, Europe 2012, Latin America 2014, and North America in 2015. The address exhaustion predictions were pretty accurate.
Everybody used the band-aid known as NAT and broke everything in the universe in order to not deploy IPv6 for a couple of years.
Now, we're adopting IPv6 in earnest, but everybody is whining that they want to hang onto NAT.
So, before IPv6 got deployed, everybody shot themselves in the left foot by deploying NAT. And, now that IPv6 is deployed, they continue shooting themselves in the right foot by hanging onto NAT.
There's a fixed pool of IP Addresses that can be allocated. When that allocation pool runs out, no more can be allocated. Allocation occurs in blocks at a very high level and then at subsequent levels more sub block allocation happens.
The INAN ran out of IPs to hand out. That doesn't mean that all of the IPs have been assigned to physical devices attached to the internet. For example the US Government Agencies have tons of IPs they aren't using but that had been previously allocated to them decades ago and can't be taken back.
It was only after doing this that I realized the problem with IPv6 adoption in the US: the network sucks! Most ISPs (including mine) are using 6rd gateways to get their customers onto the IPv6 backbone. The gateway servers are poorly placed (rarely local) and totally overloaded. The performance was so bad that it was making my everyday web browsing experience feel broken. I killed v6 at home and haven't looked back.
Maybe some day ISPs will do a proper deployment with real routing and I will try it again.