Congrats on the rollout of v6, guys! Can't wait for Dallas to be turned up soon.

One question, though. Your FAQ states:



Does that really mean you're assigning a single /128 for free? If so, that seems *really* silly and somewhat going against how IPv6 was designed to be implemented.

Yup. Each IPv6 enabled Linode gets one address assigned.

You'll be able to have multiple pools of IPv6 addresses assigned to your account that are bound to a specific datacenter. These are inherently 'shared' across all your IPv6 enabled Linodes in that datacenter.

We'll also be rolling out support to have an entire /64 routed to one of your IPv6 Linodes, which you can then route wherever you please.

-Chris

Okay, thanks for clarifying, Chris!

Are there plans to support reverse DNS for IPv6 addresses?

That's how it is supposed to be used.
Hope you don't plan to charge extra for this.

Yeah. I've been a strong proponent of "if you need IP6 now, use HE tunnels" on the understanding that linode would eventually produce an IP6 solution of their own.

But one address is... umm.

It's 99.9% likely that your customers will never need more than 1 IP6 address but given that IP6 is generally given out as a /64 (or Panix a /96 since they're oddly subnetting) this single IP6 address is... weird. Given the size of linode as a company you could easily get a /48 for each DC and that'd allow 65k /64 customers in each DC. Heck even a /64 would allow 2^32 /96 customers (which is what Panix has done).

I admit to being boggled.

_________________
Rgds
Stephen
(Linux user since kernel version 0.11)

I think the allocation plan is good, so long as the prices are kept reasonable. For instance, $1 for a block of 16 and $5 for an entire /64. I don't want to see IP addresses handed out like candy like they were in IPv4, with large blocks of allocated but unused space.

Why not?

IPv4 addresses were handed out in large chunks because people thought that they had plenty of IPv4 addresses to go around. Turns out they were wrong. Very very wrong.

There's no guarantee that the same won't happen with IPv6 addresses, if people start handing them out like candy. Seriously, what would you do with 2^64 addresses, other than waste them?

Relevant: http://xkcd.com/865/

What are you talking about? You do know the size of the IPv6 address space, right?

Let's get some big numbers going here.

Size of 2000::/3 IPv6 global unicast space:
42,535,295,865,117,307,932,921,825,928,971,026,432 unique addresses

As noted earlier, Linode has a registered /30, which means roughly speaking they could hand out 17,179,869,184 /64 blocks off of this space alone.

You might think "well, that might not be enough! What if they need *another* /30!?"

Off of the total IPv6 global unicast space mentioned before, IANA could give out 134,217,829 /30's.

Essentially, this means that off of Linode's *single* registered /30 space, they could hand out a single /64 to every person on the Earth, to the estimulated population of the earth around 2025, *twice over*, still have room, and could buy another /30 to do it all over again.

I'm not implying that Linode, or any existing hosting company for that matter, does not have enough IPv6 addresses to hand out for the foreseeable future.

I do, however, agree with @BarkerJr that it's a bad idea to hand out IPv6 addresses like candy. This is more of a theoretical concern than a practical one, though it might become a practical problem in a few decades. 2^128 might look like an awfully large number right now, but people thought the same way about 2^32 only 30 years ago, and they were wrong. History has a tendency to repeat itself, no matter how unlikely you think it might be. Just wait until people start walking around with billions of anti-cancer/anti-diabetes/whatever nanobots in their blood stream, each with a /64 or even a /96 carelessly allocated to them.

There is no reason for any device to have more than 1 IP address allocated to it, except in certain circumstances such as failover or multiple SSL web sites. In fact, I found it rather disappointing that IPv6 was designed in such a way that makes it convenient for hosting providers to allocate /64 to each and every server. So I'm perfectly fine with having 1 IPv6 address per Linode, as long as it is easy and not too costly to acquire more addresses whenever needed.

Multi-homing is quite acceptable in the v6 world and it's very possible you end up with 3 or 4 different IPv6 addresses, depending on your setup. (Temporary Global, Random Global, Manual Global, Private-NonRoutable)--per machine based on RA and OS settings.

You list two reasons why a device must have more than one IP address, but there are also a lot of good and even more subjective reasons why a server should have more than one address. Obviously most situations can be made to work with a single IP address, we do it today with IPv4, but then again we also have NAT in our IPv4 world.

I don't think IPv6 was designed such that a hosting provider would allocate a /64 to each and every server much less VPS, but with the way it is currently in use today globally, it will probably be more hassle than its worth to allocate less than a /64 per customer / per physical location. An example scenario might be an administrator dealing with abuse, first time around they'll block the individual full IPv6 address - subsequent abuses will probably result in the network being blocked which they'll have no choice but to assume its an entire /64 at a minimum.

Personally I'm glad Linode has decided to go forward with their native IPv6 implementation in all data centers where they can do so, but I must admit I too also hope a more streamlined approach can be found for all those involved. I know what my first question will be, how much for the additional addresses?

Even a /116 would give most customers more than enough addresses, and Linode could hand out 7.74*10^25 of them.

This isn't really about protecting the internet, it's about making money.

I think it's hard to appreciate just how mind-numbingly large 2^128 is.

I know for myself (having grown up with IPv4 and even participating in the working group that led to CIDR, so I've worried about address exhaustion for a while) I feel bad being assigned a /64 when I know I just need a few addresses. It just feels so - inefficient.

At the same time there's just so many of them it's hard to grasp. And there are efficiencies within the network by assigning larger blocks to be used locally (and even if sparsely), not to mention the freedom of easier management of local usage. In other words, handing larger blocks out "like candy" can actually have benefits.

You can do all sorts of calculations (and there are a bunch of web pages that show sample comparisons), but just to piggy-back on your example. Wikipedia says the median estimates for carrying capacity of the earth is about 10 billion people. Let's double that to 20 billion. Even at that point, you could still give out almost a billion /64s to each person on the planet, or almost 2.5x more addresses than the 7x10^27 estimate for atoms in a human body a quick google search turned up. (I haven't tried to fully account for the various reserved bits of the space, but the numbers are still astronomical, and besides - the nanobots would probably share a link-local space rather than actually needing global addresses)

So I think your nanobots are safe. Plus every possible electronic component designed for use by each human, plus any possible device they may need to interact with, and so on...

Maybe expanding to other planets might stretch things more, but I suspect we'll be able to have an "off world" prefix with its own space somehow - if IP still exists in any sort of incarnation at that point.

-- David