I posted to the Voxel blog a little while back alluding to our plans for the deployment of the Internet Protocol version 6, “IPv6″ or “v6″ for short. Since then, a lot has changed! The Voxel backbone is fully deployed for support for both Internet Protocol versions 4 and 6 on the same physical circuits and interconnects, commonly known as “dual-stack”. We’ve now extended this capability down to the “edge” of our network, where customers connect.  Existing VoxSTRUCTURE (physical and virtual server) users  familiar with IPv6 and interested in kicking the tires around are encouraged to mail (noc at voxel.net) for a “technology preview” of this exciting new capability. Additionally, we are committed to supporting IPv6 throughout our Content Delivery Network (CDN), and hope to have additional updates to share several months in. Some notes and ramblings from the field below:

  • With the pool of free IPv4 addresses nearing depletion, now’s a good time to begin enabling networks and web applications for v6. This is not a call for panic, but rather a suggestion to become familiar with the technology in play. Nodding to competitors in the vanguard, Limelight Networks, SoftLayer, and of course, Hurricane Electric are doing precisely that, excellently, and on a grand scale.
  • As with the service providers, content adaptation is also key. Google and Netflix get kudos for successfully v6-enabling their sites and living to tell about it [1][2], and viewing www.kame.net from a v6 connection will present you the famous animated turtle; deployment is otherwise sparse, however we can expect some announcements from more top websites shortly.
  • Backbone operators and industry analysts dismissing v6 as a “fad” very clearly have their heads in the sand, and should not be taken seriously. I’ve had my fair share of run-ins there, and have had to vote with our wallet on several IP transit renewals. Competing for the stupidity prize are the backbone provider who wanted to charge a price premium for v6 traffic on a per-megabit basis, and the provider presenting on v6 at nearly every trade show, yet blissfully unaware of how to actually provision a live customer for it.
  • Vendor support for IPv6 has been surprisingly solid of late, and devoid of any major “gotchas”. Foundry and Cisco, the main vendors on Voxel’s IP core, are shipping code that’s thoroughly tested and free of v6-specific bugs.
  • From a tools prospective, our IPv6 roll-out has been smooth sailing as well. Ubersmith, our technology partner there, has been ahead of the curve, and responsive to my feedback and (admittedly few) bug reports — their IP management system was built from the ground up with support for v6, and can churn out customer address assignments quickly and accurately. (In fact, v4 addresses are databased internally as v6 too, with some top-level padding.) Likewise, our DNS tools and infrastructure have supported ‘ip6.arpa’ and ‘AAAA’ records for years, and when tested, worked without a hitch.
  • The landscape for IPv6 peering is interesting. Many network operators are maintaining a similarly stringent set of interconnection requirements for IPv4 and IPv6. Some more progressive access providers have relaxed their peering requirements for v6, opening the door to interoperability testing with a wide variety of participants, though I’d imagine this door will close again following v6-activation on their access networks. Additionally, age-old peering disputes and tensions have succeeded in migrating from v4 to v6 [3][4]!
  • The IPv6 internet is very much in “beta” mode, and absent much of the interconnection and industry knowledge we see with v4. Sub-optimal routing is fairly common, stemming from global providers only meeting at one or few locations, or people routing connections over bandwidth-limited tunnels given a lack of support for dual-stack on their routers. This situation is quickly showing signs of improvement, and will likely shape up for the better as adaptation evolves over the next few years.