The protocol that cried wolf

As we inch ever closer to IPv4 depletion and possible network chaos, the networking community as a whole seems to sitting back and watching with a simple smirk. Sure, there are pockets of activity; announcements from ARIN, presentations at policy conventions and the ramblings of blogs such as this. But one would be hard-pressed to say that any more than about one or two percent of the community is taking note…​which is probably about the current percentage of IPv6 deployment.

To confirm my belief I asked some random CLECs, ILECs and ISPs at a recent trade show where they were with IPv6 and how soon they thought general deployment would occur. Every one reacted basically the same way:

  1. Somebody within their organization is looking into it at a high level, but nobody’s expecting deployment anytime soon.

  2. They don’t believe demand will be coming anytime soon. Many of the people I talked to mentioned that everybody has been saying for the last decade that IPv6 was coming in 2-3 years. Since it still hasn’t happened they aren’t too concerned that they will get behind the eight ball.

  3. No customers are asking for it yet and, other than a few test sites, no real content is accessible via IPv6.

The vendors and distributors at this trade show echoed similar sentiments: fringe elements of the community have been pushing IPv6 and asking for us to support it, but our manpower is better spent producing products and features that customers need today.

While that is a valid business outlook, it doesn’t bode well for the Internet as a whole if the address squeeze really happens. While these vendors may be able to push out an IPv6 version of their products before addresses dry up in a couple years, I would feel a lot better if the software/hardware had a few years and release cycles under its belt before we become dependent on it in production.

I’m not sure I can really blame any of these vendors and carriers either. Industry leaders and speculators have been predicting IPv4 depletion for years and it hasn’t happened at the rates they predicted. This isn’t the fault of the predictors however, I believe the ebb and flow of the world economy has allowed the current pool to stretch much further most insiders thought possible.

Unfortunately, all these years of crying wolf may have really hurt our chances of a smooth, albeit delayed, transition to IPv6.

tags: IPv6
Cody Lerum - May 10, 2009

IPv6 Subnet Allocation Size - The Debate (Finally) Rages

It’s been over a decade now since IPv6 was approved as a standard, and people are still arguing about fundamental issues such as subnet allocation sizes and other customer-facing best practices. In my mind, the increased tempo of the argument is a very positive sign, albeit later in the game than most people would have preferred. The fact that so many people are arguing about such a basic operational issue as customer allocations, and that they are soliciting opinions and addressing plans from folks who’ve already rolled out IPv6 in their networks is a good indicator that more people are finally getting closer to actual IPv6 deployment. Even though with there is still not 100% consensus on some IPv4 "best practices", it will be interesting to see how things shake out — especially on some of the larger networks.

For those that are curious and/or looking for an addressing plan, section of the ARIN Number Resource Policy Manual (NRPM) provides useful guidelines for IPv6 end-user assignment that I feel make sense from a common-sense, operational point of view. The manual can be found at:

One of the subjects not covered in the guide are point-to-point (PTP) subnets. There are two schools of thought on this:

  1. Always allocate a /64 subnet on PTP links. This way, you are not breaking the built-in autoconfiguration capabilities of IPv6. Or,

  2. Use something smaller. Personally, I shiver at the thought of allowing auto-configuration on a core (or even important customer access) PTP link. Also, the idea of blowing 2^64 addresses (18,446,744,073,709,551,616) addresses on a PTP link seems awfully wasteful — harking back to the days of default /8 allocations to folks like the Ford Motor Company and MIT. I like to use /124 subnets as a standard. The nibble boundary makes for easy subnetting, and with 16 available hosts (2^4), it covers 99.9999% of PTP scenarios on my network.

The debate about PTP links is just one small part of a larger argument about IPv6 addressing — namely, how much waste is too much? Many people argue that there are so many IPv6 addresses that we will never use them all. While I agree that 2^128 (3.402e+38) is a huge number, I don’t see a problem with conservation where it makes sense. Nobody knows what the future will hold, and the explosion of the Internet came as a real shock to people who originally thought that we could never utilize all of the addresses in the IPv4 space. After all, 2^32 (4,294,967,296) is still a pretty big number.

If you’d like to join the debate (or just enjoy the show), the ipv6-ops mailing list is an excellent place to start. You can join at

There you also find link to the list archives, which make for some interesting late-night reading.

The one constant in IT is change, and I’d be interested in seeing what kind of subnetting/addressing/aggregation plans others have rolled out.

Ken Mix - May 09, 2009

IPv6 living in a IPv4 world

Good article on how improperly implemented 6in4 tunnels are holding back the v6 content providers and giving IPv6 a bad rap.

tags: IPv6
Cody Lerum - December 27, 2008

Linux IPv6 MTU issues

I’m seeing an annoying issue with Ubuntu 8.10 and IPv6. It appears that the Linux OS is sending packets larger than 1500 bytes out of a gigibit Ethernet interfaces which is configured for a 1500 byte MTU.

This is easy to reproduce with a dual stack machine. Just SCP/FTP a file down via IPv4 and one via IPv6. You will only see about 10% of the IPv4 throughput on the v6 transfer. A tcpdump or wireshark will confirm that packets sourced from the Linux machine are larger than 1500 bytes…​.sometimes 10x the size. When the network connecting these two machines only supports 1500 bytes, all these oversized packets get dropped.

I’ve opened a case with the ubuntu guys —

However, I’m seeing similar results on CENTOS 5.2 so I’m thinking it’s less distro specific.

I haven’t really heard of anyone else noticing this which makes we wonder if I’m missing something painfully simple in the configuration, or nobody is really serving any IPv6 content off a server with 2.6 kernel.

Anyone else seeing this?

Cody Lerum - November 30, 2008

IPv6 Nameserver Glue

Looks like we will be moving our domain registrations to another provider shortly. We are currently with Dotster for cost and ease of administration (mostly cost), but when I inquired about adding IPv6 glue for our nameserver records they were without clue. After repeated requests I finally got through to a tech who proclaimed that nobody is using IPv6 and they have no demand. confirms this. Looks like Network Solutions might be getting a little more business.

Now if only a few thousand other people moved their registrations for this reason we might pop up on the radar.

tags: IPv6 DNS
Cody Lerum - November 30, 2008
About Knowledge Bombs
Random bits of knowledge that we don't want to forget and that might help you!
Cody Lerum
Ken Mix