Norwegian IPv6 brokenness experiment, October results

[Tore Anderson]

Hi list,

I mentioned about a week ago that I was conducting an experiment here in
Norway, trying to figure out more about any eventual breakage that would
be caused by dualstacking web sites.  This generated quite a few
off-list messages from people wanting more information, so I thought
that I'd rather just post the October results here instead so that
anyone interested can take a look.  Feedback or questions are of course
very welcome.

The test was done with the help of my customer A-Pressen Interaktiv
(http://www.api.no), which is the publisher of about 55 online
newspapers (Norwegian-language and mostly regional).  The readers are
not particularly techical or anything like that.  The site is in the
Norwegian top five (when adding all the publications together).

API have added an IFRAME tag in their HTML templates for most of their
pages, which includes a linkgen.php script on my test server (loaded
from a IPv4-only hostname).  This scripts spits out two IMG tags both
pointing to a 1x1 pixel large PNG, one loaded from a dualstacked
hostname, and the other from an ipv4-only hostname.  The ordering of
these tags is random.  In addition, both of the URLs pointing to the
PNGs include an ID which is generated randomly each time the linkgen
script runs.

All the traffic is going to the same test server, but all the hostnames
are pointing to different IP addresses and the TTL is only five seconds
(in order to prevent any bias introduced by caching).

The executive summary:
- On the dualstack host 0.153% of all hits are lost
- IPv6 penetration (preference over IPv4) is at 0.334%

The Opera web browser (coupled with Windows Vista or newer, which
automatically configure Teredo and 6to4) is causing most of the
problems, because it unconditionally prefers IPv6 over IPv4.  If I
simply exclude all log lines which contains "Opera" in the user-agent
field, I get the following executive summary instead:

- On the dualstack host 0.023% of all hits are lost
- IPv6 penetration (preference over IPv4) is at 0.134%

I've submitted Opera bug DSK-269385 about this issue, but have yet to
receive any reply.  If anyone has any contacts in Opera they can poke,
that would be greatly appreciated.  I doubt I'd be able to convince API
(or any other of my customers) to dualstack their production web sites
until Opera have released a fixed version.

I've attached two tables with numbers, one excluding all hits from Opera
browsers.  A few things to note:

- I've been running a 6to4 relay on the test server
- Data from the 11th, 15th, and 20th are removed (apparently outliers)
- The 11th I started announcing 192.88.99.0/24 to my peers on NIX
- The 28th I implemented a split-horizon DNS scheme that masked the
  AAAA records for the dualstack host from two eyeball networks known to
  filter 6to4 (thanks to Geert Hendrickx for that suggestion)

Many thanks to API for allowing me to experiment with their users and
for letting me post the results, and to Steinar H. Gunderson of Google
for helping me out tremendously along the way.

Regards,
-- 
Tore Anderson
Redpill Linpro AS - http://www.redpill-linpro.com/
Tel: +47 21 54 41 27
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: 200910.txt
Url: http://lists.cluenet.de/pipermail/ipv6-ops/attachments/20091104/87ce7c0c/attachment.txt 
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: 200910-noopera.txt
Url: http://lists.cluenet.de/pipermail/ipv6-ops/attachments/20091104/87ce7c0c/attachment-0001.txt