January 2009 Archives

I wanted to post a few IPv6 updates since my last post.

The NTP pool project has been working to add IPv6 support for some time. This month, they've added partial IPv6 support. Hopefully by year's end they'll have full IPv6 support. Several OSes and NTP appliances already have IPv6 support, so there's a fair bit of infrastructure that could support this now.

Speaking of NTP, the IETF's NTP working group is working to define IPv6 NTP options for DHCPv6. The current draft of that spec has received some criticism, claiming that its redundant with the current SNTP option. I hope that is resolved at the next IETF meeting in a few months.

Speaking of DHCPv6, Microsoft has fixed several DHCPv6 bugs in the beta of Windows Server 2008 R2. This should significantly improve interoperability with Unix/Linux clients.

Several North American universities and R&E networks obtained IPv6 allocations in 2008. I pulled the list of allocations from ARIN and noticed some highlights:

* Arkansas Research and Education Optical Network
* Bryant University
* Bucknell University
* CENIC
* California Institute of Technology
* California Polytechnic State University
* Florida LambdaRail
* Furman University
* George Mason University
* Louisiana State University
* Lousiana Optical Network Initiative (LONI)
* National Center for Supercomputing Applications
* New York University
* Oregon State University
* Rutgers University
* South Dakota Board of Regents
* Stanford University
* Tufts University
* University of Alaska
* University of California, Irvine
* University of California, Office of the President
* University of California, San Diego
* University of California, Santa Barbara
* University of California, Santa Cruz
* University of Illinois
* University of Iowa
* University of Nebraska-Lincoln
* University of Pennsylvania
* University of Pittsburgh
* University of South Florida
* University of Wisconsin Madison
* Wheaton College
* Worcester Polytechnic Institute

Several of these allocations are already in use.

2008 was a milestone year for IPv6.

The beginning of the year saw IPv6 added to the DNS root, closing a major hole for IPv6-only communication. In mid-year, the US federal government's IPv6 mandate came into effect, requiring all federal IP backbones to support IPv6. While the mandate didn't have anywhere near the effect that many had hoped for, it did spur many vendors to add IPv6 support to their products. The amount of observed IPv6 traffic increased considerably, but we still lack good data for how much IPv6 is being used.

So, where are we at the end of 2008? I'd like to present a few metrics.

Approximately 150 million IPv4 addresses were assigned in 2008. Only 13% (570 million) remain in the global free pool.

How much of the Internet can use IPv6?

IPv6 continued its slow, steady growth in 2008. One way to measure this is to look at the IPv6 entries in the Internet's routing table:

Active BGP entries

Despite this growth, only 4% of the Internet supports IPv6.

The Internet is composed of a series of interconnected networks (hence the name Inter net). 4% of these component networks (called Autonomous Systems) support IPv6. This is up from 2.4% at the beginning of 2008. That's good to see, but clearly a tremendous amount of work remains to be done. Let's look a little bit closer at the 4% of the Internet that's running IPv6.

The Internet is a large mesh of networks. There are networks that originate traffic, networks that carry traffic for others, and networks that receive traffic. Imagine that I'm sitting in my office at Penn State, and I want to load Wikipedia's home page. Penn State's network and Wikipedia's network aren't directly connected; my packets must flow through several intermediate networks to reach Wikipedia. The component networks that make up the Internet can be divided into two types: edge networks (like Penn State and Wikipedia) and transit networks (like Sprint, Internet2 and Global Crossing). If you add up all the component networks, 4% of them support IPv6. But what if you look at edge -vs- transit networks? 15% of transit networks support IPv6, and just 2% of edge networks do.

Unsurprisingly, that 2% of edge networks isn't generating a lot of IPv6 traffic. There aren't good measures for this, but here's what we do see:

  • At the largest Internet Exchange Point in the world, native IPv6 makes up just 0.2% of traffic, although this is growing steadily.
  • At various DNS root servers, native IPv6 makes up at most 1.5% of traffic, but this varies widely from region to region.
  • The ratio of native to tunneled IPv6 is highly variable from continent-to-continent. Europe has the most native IPv6, and by far, the most IPv6 traffic.

We need more data.

This brings me to my next point: There is an enormous dearth of good statistics on IPv6. Only one large Internet Exchange Point makes real-time IPv6 data publicly available. Only two of the DNS root servers do. There's just not enough data to answer the question: How much IPv6 is out there?

There was a study that attempted to answer that question in 2008, but it had severe methodological flaws. The Internet community has to fix this problem in 2009.

Speaking of measurement, we need to do a much better job of measuring and monitoring IPv6 performance. The IPv6 routing system can change often, and connections show high variability. You can see this in DNS performance.

Hurricane Electric provides daily snapshops of IPv6 performance by measuring latency for IPv6 DNS servers. Every night, they measure if it's faster to use IPv4 or IPv6 to query the DNS. The ratio of which is faster has shown a lot of variability this year:

(Apologies for the small text. Click on the image for a larger version.)

Since fall 2008, the day-to-day variability has started to stabilize. But for nearly 40% of servers, IPv6 represents a performance hit.

IPv6 in US Higher Education

This is still a large grey area. There is a dearth of good data for how thoroughly IPv6 is deployed here. I've been using IPv6-reachable DNS servers as one metric, but this is fairly crude. During 2008, the number of Internet2 members with IPv6-enabled DNS roughly doubled, with much of that happening by early fall. During the fall semester, schools that already have IPv6 added more IPv6. Right now, about 12% of Internet2 members have an IPv6-reachable DNS server.

But DNS is just one protocol. At a minimum, universities should IPv6-enable the DNS, web and email systems. I can count on one hand the number of schools that have done all three. And I won't use all of my fingers. These early adopters seem to be spurring others along.

Even this misses the bigger picture. It doesn't address (pardon the pun) questions such as: How often are there IPv6 outages (a lot!); is IPv6 support part of purchasing requirements; do internal systems (authentication, file/print, directory, etc) support IPv6; has home-grown code been audited for IPv6 support? Many of these questions get dismissed as mere "campus issues," but they're vitally important, and I don't see much effort being put into answering them.

Predictions for 2009:

Let me wrap up by making some predictions for IPv6 metrics in December, 2009:

  1. 90% of top-level domains will have IPv6 glue in the root (right now, 75% do).
  2. 50% of the DNS root servers will support IPv6 (right now, 25% do)

  3. At AMS-IX, 1% of traffic will be native IPv6.
  4. 1400 ASes will have IPv6 prefixes.
  5. Europe will continue to have the most allocated and deployed IPv6.