[AusNOG] Stupid web browser behaviour Was: My Predictions for the ISP Industry

[Geoff Huston]

Hi Karl,

> On Thu, 2012-03-15 at 13:20 +1000, Craig Askings wrote:
>> For example on my dual stack connection at home when I go to 
>> (http://labs.apnic.net/measureipv6/) it works first time with Firefox 
>> but fails with Chrome because it has yet again decided to stop asking 
>> for AAAA records.
> 
> I may be just displaying the depth of my ignorance regarding Chrome
> here, but doesn't Chrome implement "happy eyeballs"?


not really - it has its own interpretation of that draft, as do the other "happy eyeball" browser implementations.

What appears to be happening in Chrome is that whichever DNS response arrives first, Chrome appears to select as the preferred protocol and attempts to connect by sending TCP SYN packets. Unlike Safari, Chrome attempts to connect using 2 ports in parallel in the preferred protocol, and after a 250ms delay, if there has been no SYN ACK in response, it will send a further SYN on a third port using the other protocol.

regards,

Geoff