Google and IPv6

Erik Kline ek at google.com
Mon Mar 17 04:46:32 CET 2008


2008/3/16 Kevin Day <toasty at dragondata.com>:
>
>  On Mar 16, 2008, at 9:34 PM, Erik Kline wrote:
>  > Speaking personally, I have seen analysis of neither client behaviour
>  > nor connectivity in the IPv6 Internet today.  Are things really as bad
>  > as folks say or is it partly a kind of "urban legend of horribleness"
>  > that persists from earlier tests with less-mature operating systems
>  > and less reliable connectivity?  I  just have 6to4 at home and my Mac,
>  > Linux, and XP boxes all seem to work just fine.
>
>  Every few months I've performed this test on a rather high traffic
>  (but probably tiny compared to what you're used to) website that's
>  comprised mostly of completely non-technically savvy users. The reason
>  I feel the "non-technical users" part is important is because if one
>  of us realizes that we can't reach www.ripe.net, we go fix it - most
>  users think "oh, the internet is broken" and move on to something
>  else. This results in problems going unresolved. :)
>
>  On one page of this site, we put two 1x1.gif images.
>
>  ipv4.gif is on a hostname that has only a v4 A record. This gets a
>  baseline of how many people loaded the images on the site at all.
>  4or6.gif is on a hostname that has both AAAA and A records.
>
>  The order the two images appeared on the page was randomized on every
>  page load to try to reduce any bias there. I measured how many times
>  each image was loaded after a few hours. Results from the last time I
>  attempted this, after removing duplicate IPs:
>
>  ipv4.gif was loaded 278821 times.
>  4or6.gif was loaded 278704 times. (191 hits were on the v6 IP, 278513
>  on the v4 IP)
>
>  The good news is that 0.069% of our viewership was able to load an
>  image on the v6 IP of an AAAA record. Nearly all of these were 6to4
>  addresses, with a few Teredo. I only saw 2 IPs that looked like native
>  v6.
>
>  The bad news is that the image on the AAAA record was loaded 0.042%
>  less than the one on the A record. This meant that by publishing an
>  AAAA record, your site would appear broken to almost as many users as
>  would benefit from having the v6 connectivity.
>
>  I did some troubleshooting with as many users as I could that were
>  willing to look at why they weren't able to access the host on the
>  AAAA record, but didn't get very far with most of them before they
>  lost interest with troubleshooting. In most cases it seemed like users
>  had ended up with v6 activated/configured when they had no v6
>  connectivity. None of the problems appeared to be broken v6
>  connectivity exclusive to our site - if the user couldn't reach our
>  site on an AAAA record, they couldn't reach any AAAA hostnames.
>
>  -- Kevin
>
>

That's actually fairly promising, it would seem, as a measure of
successful connectivity and good client behaviour.  Thanks!

An additional concern would, of course, be performance (i.e. latency).
 If a client can cope appropriately with a AAAA record and it takes
advantage of its IPv6 connectivity but suddenly finds itself going
through tunnels or across continents to the "nearset" 192.88.99.1 6to4
relay the overall user experience could, theoretically, be much, much
poorer.  And poor performance obvious has a direct impact on user
happiness.


More information about the ipv6-ops mailing list