By Fred Baker, Cisco Fellow
During the week that World IPv6 day happened in, I was curious to see how the various networks involved were doing over time. I set up a test from my home, using a Hurricane Electric tunnel, IPv6 on my Mac (10.6.7), and my 871 router (15.1(3)T). I put together a simple script that would accept as input a set of web sites like http://www.cisco.com – the web sites that ISOC said were going to be IPv6-accessible on the 8th of June – and spidered them (e.g., read the web page using the unix ‘curl’ utility and scanned for href specifications). I then added the URLs I had learned to my list and continued to try them, gathering statistics on success rate for those that had AAAA records. As a result, I was doing about four page loads an hour from each domain in the list from June 5 through June 9 – all times GMT.
One observation I made was at once gratifying and “as expected”. The various sites were coming up in advance of the magic day and, for the most part, serving IPv6 data successfully. One observation that surprised me a little at the time but in retrospect makes sense – the download rate increased over time as well. Why? Well, it takes some time to attempt to download and discover that the AAAA record is not up or is up and the service isn’t quite there yet. As my probability of a successful download increased, my download rate increased.
The one that had me a little surprised and very pleased was the number of sites I actually visited. There were about 400 participants in the event, but I actually successfully accessed 751 separate URLs at 554 domains using IPv6; the extra ~150 were sites that participants pointed to that were also IPv6-capable.
The failure levels were very gratifying. I of course had failures from sites that didn’t have their AAAA records up, and initially (on Sunday and Monday) some sites that had their AAAA records up but whose IPv6 didn’t work. I found myself, on Tuesday, corresponding with one participant whose service was just fine when he measured it, but from my perspective simply hung. After a few tcpdumps, it became clear that my traceroute to him differed from his traceroute to me, and I was receiving every other TCP segment he was sending. He changed his routing to work around a network that appeared to be having problems, and voila, it worked.
So, from my perspective, I saw two important things. IPv6 worked, cold stop, and not just with the magic few but with their business partners as well. Where there was a problem, the kind of fairly straightforward interactions that our industry is known for got them straightened out.
Which brings me to a question. I expect that broadband service providers and other networks will be deploying IPv6 services to their customers in the relatively near future – very likely adding IPv6 services to existing IPv4 services and degrading those IPv4 services with some form of address multiplexing. Testing an “IPv6 Day”, from my measurements, in fact created an “IPv6 Week”. What would happen if, as residential and other edge IPv6 deployments occur, we held an “IPv6 Week”?