"This IPv6 traffic is mimicking classic end-user 
     bandwidth shaping," Levy says. "It's not machine driven; 
     it's human eyeball driven."

I would be very interested in know how they can make this distinction.

I would imagine its probably sort of like the electric grid. In the morning there is a huge surge as people wake up, flattens out during the day while people are working and goes down at night while most (normal) people are asleep.

If it was 'machine driven' traffic I'd imagine it would be pegged constantly, where as bandwidth for a given region of the world would behave similar to that of electricity usage.

> We're talking about a traffic spike that is 30-to-1 type ratios. In other words, 30 times more IPv6 traffic is coming out of Google's data centers than before.

Given that Google enabled IPv6 traffic to a high-bandwidth service like Youtube, is the spike really of any note?

If a week ago 100 people were using Google Search via IPv6, then today one person views a single Youtube video, that would easily account for a "30-to-1" spike

The actual number of users with IPv6 would be interesting, the fact there was a traffic spike seems quite irrelevant..

I wonder if Google has any way of figuring out what proportion of their IPv6 traffic is coming natively over IPv6, versus over one of the various IPv4-to-6 tunneling schemes? If a lot are in the 2nd category, this might not be a great move: a high-bandwidth site like this could overwhelm the tunnels and make new tunnels less attractive to operate.

You can tell pretty accurately by looking at the source address; 6to4, Teredo, and the major tunnel brokers allocate out of known prefixes. Google should be receiving little tunneled traffic since AFAIK they only return AAAA records to clients at ISPs with "good" IPv6 support.

Nice. Now I can watch videos and chat on IRC with IPv6. Move HN and Github over, and I won't need IPv4 anymore!