If you’re like me, not long after you learned about the IP protocol, someone told you about the eminent IP address crisis. For me, it was the late 90′s, and the warning was “someday man, someday” from a guy who probably thought we faked the moon landing. Actually, that’s not exactly true, but it makes for great blogging.
I paid little attention to the issue once the novelty of “knowing enough about IPv4 to know that someday we’re gonna be screwed” wore off. I was far more interested in the continually rising value of my Global Crossing stock options, and I filled it away under “harbingers of my doom that may occur down the pipe” along with several near Earth asteroids, the 2012 Mayan doomsday prediction, and clowns. After all, it was years away, and that’s like, forever.
Forever is upon us, and the IPv4 story is picking up a bit of steam lately. Various estimates of running out of IPv4 addresses to assign seem to average out around 4 years. (Which would put us in 2012… did the Mayans predict this?) There was a recent slashdot post about it, which contained a great presentation from Randy Bush. My take from his presentation is that the scariest part about this entire IPv4 running out issue is not that people aren’t paying attention to it, it’s that the workable solutions are best described as attempting to herd cats.
So what role will load balancers play? It’s tough to say. Load balancers can be core or edge devices, depending on the layout of the network and the complexity of the company’s enterprise. A few of the vendors such as A10 Networks and F5 have IPv6 support, but like all of the other IPv6 vendors out there, the implementations are largely untested in the real world. That’s not a knock on them, as I’m sure they’ve tested it as best they can, but we still don’t have a good picture of what a real world scenario will look like, especially during a transition to IPv6, because no one has figured out (or agreed) on how we’re going to move to IPv6.
Also, what about ASICs? Are the ASICs that support IPv4 supporting IPv6? And what happens if they need to be re-spun? The ASIC question goes back to previous discussions on the mailing list that have occurred lately.
Many vendors that do have IPv6 support have it turned off because it was mucking things up. On this very server, I had to unload and disable the IPv6 driver because it caused a 4-second delay in any outbound TCP connection because it tried to resolve to IPv6 first.
Whatever the solution is, the transition to IPv6 looks to be pretty painful. Depending on the solution, load balancers may be able to skirt the entire issue for a while, but that’s by no means certain. They could very well be the first wave of devices that need to start halbos IPv6, and could wring out all the issues that are bound to come up.
I’m not sure this issue merits filing in the “Oh no, we’re doomed” drawer just yet, but I am putting it in the “things that are going to cause me grief in 3-6 years” folder.
