So, it's been about five years since we rolled out extensive use of IPv6 on our internal network, and, I can report, that it just works. OSPF with IPv6 is, well, like OSPF with IPv4. Address Assignment from routers to downstream devices works reliably. After a few years, the address "2001:1868:209:FFFD:0013:50FF:FE12:3456" is pretty much as straightforward to parse as 67.228.107.138 (and, in some ways, is actually much easier to understand and remember)
We now have about 10 million devices happily communicating via IPv6 (mostly using RFC2893s IPv6-in-IPv4 to hop through carrier IPv4 networks) over corporate networks to back office applications - and all the processes we developed for IPv4, such as anycast, and variable-cost route advertisements, carry over just fine to IPv6.
As with any technology - you need to learn some things the hard way (the one that bit us hard was "Do not hard code your IPv6 router the way you did in the IPv4 into your hosts if you use the EUI-64 auto-promotion of router advertisements against the VLAN on your router - either (A) statically assign a 64 bit address to that router interface, or (B) use auto-router detection on your hosts") - but that's probably true of any technology.
My feeling is IPv6 is a pretty straight forward and simple technology, that has been slow to ramp, and won't achieve 100% penetration for quite a while - but has a great, and guaranteed future ahead of it. Looking forward to 6/6!
First, my perspective is that IPv6, to some degree, abandons 15+ years of VLSM (Variable Length Subnet Masks) - so, at a glance, without referring to a subnet mask, you can identify what position of the address is a network, and which portion is the host element. There is a 64 bit network and 64 bit host field.
Second, The network portion for most enterprises is going to be a 48 bit super net (which network admins will memorize within a few days of working within a particularly company) followed by a 16 bit network portion. So, in the above example, the 48 bit super net is 2001:1868:209 and the network portion is FFFD.
So - now, when I see an IPv6 address, my eye automatically scans the network portion, much, much more quickly than I did in the IPv4 world, where we had 30 or 40 IPv4 networks, many of them with their own strange /30, /28/ /26 /25, etc.. net masks. I've been doing this for 16 years, and I still can't transparently do the math in my head for all variants of IPv4 - and even 2-3 seconds of translating /27 into 5 bits into 32 addresses into boundaries of 0,32,64,96,128,etc... can break your concentration. That problem no longer exists with IPv6. It's a 48 bit super net + 16 bit network for most enterprises.
Now, for those who aren't using the privacy elements, you also have advantage of embedding the MAC address in the address, so, for those of us who are OUI geeks, we can also identify the equipment in the address so (it's the first 24 bits of the host portion, with a bit of bit twiddling the in 7bit Local/Global spot) - and, we can also see whether proper EUI-64 expansion has taken place (in the above example, it hasn't - the proper EUI-64 expansion would be 0213:50FF:FE12:3456)
So - at a glance, I see the network, sub network, and hardware, skim over the FF:FE filler, and, the actual address - 12:3456 is simple to remember.
Finally - if you have a lot of hosts you are trying to put on a single network, you don't have to play games. We routinely put 25,000 devices on a single routed link, and we aren't concerned about rolling out 2,000 of those 25K device routed links for a particular instance. And we can (and have) rolled out a hundred instances of such configurations, with no overlap in addressing between any of them. Try that without gymnastics in the IPv4 world. In IPv6 it just follows the simple rules of 48 bit supernet + 16 bit network + 64 bit host field.
It takes a couple years (or at least it took me a couple years) - but once the brain's rewired - the additional structure is in _some way_, easier to understand and remember -but obviously this is all within a particular hardware and enterprise context. Clearly a Random 32 Bit IPv4 address is easier to memorize than a random 128 Bit IPv6 address.
One can chose addresses that maximize compressible sequences of zeros. If I were allocated a subnet of 2002:/, I could make my computer's address 2002::2 and my router 2002::1.
Realistically the first 64 bits are going to be very ugly, so I disagree that a v6 will ever be much easier to remember than a v4 address. Especially with the 0 compression being back-ported to v4 stacks, running on the 10. network - You can ping 10.1 as 10.0.0.1 on many stacks now.
My guess: Since any given organization's IPv6 allocation is so large, there's no need to densely pack the addresses that you're using, so it's much easier to stuff human-readable information into the address. For instance, one of the segments of $work's IPv6 addresses is an ID indicating what facility the machine is located at.
Awesome info! Would you be interested in being part of a case study about IPv6 migration? (or if not you directly, perhaps someone at your company?) I'm working on IPv6 case studies for our Deploy360 site at the Internet Society. Please contact me at [email protected] if interested.
I have IPv6 at my San Francisco office provided by Cogent. Unfortunately, Cogent has quite a few peering disputes which means I can access say, Facebook over IPv6, but not Google.
I can only hope that Google flipping on AAAA records permanently will change this, but I wonder how fractured the rest of the IPv6 world is.
Cogent is its own little fractured IPv6 world. If you're relying on them as your sole transit provider, you will likely never have a "complete" view of the IPv6 Internet. This is true for all single-homed networks, but it's even more true if you're single-homed to Cogent.
Also, Google adding AAAAs to their hostnames won't change Cogent wanting to turn peers into customers by shutting off peering connections. This (now removed) news post on ESnet's website when they were depeered by Cogent is pretty telling, especially the bit about Cogent lying to their own customers about the nature of ESnet's unreachability:
If you want to see this attitude change, vote with your wallet -- stop purchasing transit from Cogent, and then tell your former Cogent sales rep exactly why you're no longer purchasing transit from them. Until their bottom line is impacted, they have no incentive to change.
Web and email servers that are publicly accessible, and your desktop or laptop that you connect at home or at a Wifi point, will probably still be IPv4.
IPv6 will be used internally by large companies, by large DSL ISPs and Comcast, but the interface the end-user will see, will be IPv4.
I think 99% of the usage will occur in non-visible ways - Comcast will have all their home-subscriber modems put out IPv4 on the side you see, but v6 on the pieces you don't, such as the management interface that lets them remotely configure and manage your modem.
Combined with large 6to4 gateways they will run on their edge (where they connect with non-v6 enabled systems), most people will not really notice anything.
The tipping point will come, but not for another 5 years I think, that is, 2 more hardware upgrade generations.
Mobile of course, will go to v6 first, since no-one cares as to how their phone accesses the web.
If the customer sees an IPv4 WAN address in the medium distance future, it's likely to be a private address. Many ISPs are going to do "carrier grade nat", which just means that the NAT your current router is doing will be aggregated at the ISP. Your router will be more of a simple bridge device. (It's bad to perform double network address translation of packets, and the relevant DS-Lite rfcs discuss this).
I think this will be the tipping point where I start seeing a v6 address as the first class citizen. When I have to start bugging my ISP for a port mapping, I will prefer to just have the entire session occur natively on v6 because it will no longer be the toy, it will have graduated to the tool.
So? Large corporate networks are where a lack of IPv6 will hurt the least. They can do large-scale NAT with no trouble, because they have a centralized authority that knows which machines need port forwarding, and can guarantee that most clients don't need any port forwarding. They can also afford to buy more IPv4 addresses as needed.
Folks who use CloudFlare already have IPv6 as a free option just be flicking the IPv6 switch in the control panel. Even if the real web server is IPv4 only it's possible to go IPv6 immediately with that.
I could be really snarky here, but I won't. My earlier comment was a bit of a response to the article's claim that 1% of network traffic will somehow change the world forever. That there's an 'IPv6 Launch Day' for a protocol that's been around since the 90s and still failed to gain significant traction tells you that this won't really change the world at all.
on June 6, CDNs like Akamai will enable IPv6 too. As the exposure grows by enabling AAAA records and as people have IPv6 enabled locally, traffic will only grow from now on.
We're solving a chicken and egg problem by providing more and more chicken so eggs ought to follow.
I've tried and failed to set up IPv6 on my MBP a few times now. HE seems to provide free tunneling, but I can't figure out if it's compatible with my AT&T U-Verse RG.
Is there a way to enable IPv6 tunneling on OS X in a way that doesn't require firewall configuration?
HE provides a 6in4 tunnel, that is it uses protocol 41, hence your router/firewall has to be able to forward proto41.
While SixXS provides proto41 tunnels too, they also provide AYIYA tunnels, a (standardized) encapsulation protocol able to pass NATs and handle IPv4 local endpoint changes with ease. For this you need to use their companion (open-source) tool called aiccu. Another advantage of aiccu is that it uses a protocol for autoconfiguring the tunnel (TIC), so you don't need to do some ifconfig incantations. Just put your creds in the config file, do aiccu start or load the plist and you're IPv6 enabled. I've written a homebrew formula for aiccu [0] and made a pull request, so it should be available straight from brew search/install aiccu.
Set up a tunnel on a machine: Teredo for a single host, or 6to4/6in4/AYIYA via HE or SixXS, then use that machine as a router to distribute a /64.
For those with an Apple Airport Extreme or a Time Capsule, using the Airport Utility 5.x you can trivially set up a 6to4 tunnel if you have a public IPv4 address on its WAN side, or easily set up a HE 6in4 tunnel.
It takes me from 5min to 15min to enable a machine or a whole network with internet-enabled IPv6.
There's a test at http://test-ipv6.com that will give you a aggregated (or detailed if you wish) report of influence of IPv6 upon both IPv4 and IPv6 in your configuration. Its goal is to help you troubleshoot some problems even if you have only IPv4.
I'm very familiar with 6to4. It's basically a slower way to reach stuff that you could already reach over IPv4, so there's no point in using it. None of the practical IPv6 transition scenarios include 6to4 or Teredo or tunnel brokers, so if you're "testing" that stuff you're not really helping.
In the mist, dark figures move and twist Was all this for real or just some kind of Hell?
Not sure ipv6 qualifies as "some kind of hell," but I'll take any reason for an Iron Maiden reference.
That said, the "hell" part seems to be getting big ISP's to roll out ipv6 support. Last I checked, Time Warner Cable still didn't have ipv6 available to RoadRunner customers, at least not here in NC. :-(
June 6th is a gr8te day in IT this year. It's also the day version 2.0 of the grOOvy programming language is being released to the world, at their annual European conference (http://gr8conf.eu).
We now have about 10 million devices happily communicating via IPv6 (mostly using RFC2893s IPv6-in-IPv4 to hop through carrier IPv4 networks) over corporate networks to back office applications - and all the processes we developed for IPv4, such as anycast, and variable-cost route advertisements, carry over just fine to IPv6.
As with any technology - you need to learn some things the hard way (the one that bit us hard was "Do not hard code your IPv6 router the way you did in the IPv4 into your hosts if you use the EUI-64 auto-promotion of router advertisements against the VLAN on your router - either (A) statically assign a 64 bit address to that router interface, or (B) use auto-router detection on your hosts") - but that's probably true of any technology.
My feeling is IPv6 is a pretty straight forward and simple technology, that has been slow to ramp, and won't achieve 100% penetration for quite a while - but has a great, and guaranteed future ahead of it. Looking forward to 6/6!