On killing IPv6 transition mechanisms

Erik Kline ek at google.com
Tue Mar 16 03:33:26 CET 2010

> The google folks can tell you better what their motives are for not
> enabling IPv6 without the whitelist, but there

So the problem is actually two-fold (at least :).

[1] First, and most obvious, is the impact of users not being able to
reach us at all.

There are many things that can go wrong between a web browser and a
web server/service, it's true: DNS, weird CPE issues, misconfigured
firewalls, et alia.  /Most/ of these things, though, tend to affect
all or most of the Internet traffic for affected user/host/network.
As such, diagnosis (at some skill level) begins and may escalate to
the ISP and so on.  But IPv6-related breakage is not of the
"all-or-nothing-so-I'd-better-get-it-figured-out" variety.  It only
affects the sites who advertise AAAAs.  Maybe accessing the
AAAA-advertising site will be important enough to folks that it will
get resolved, and maybe it won't.  0.08% is still a lot folks who
might needlessly have a bad experience trying to reach us, and we can
do better than that.

[2] Second, and perhaps less obvious, is the long-term adverse impact
of latency.

As you probably know right now we have this "Make the Web Faster"
effort.  If you search for comments made by Marissa Mayer in
connection with latency you should find several stories about various
latency experiments.  A few miscellaneous links (since it obviously
won't do any good to link to internal documents here):



The thing to note is that during experiments which added extra
latency, searches were down, but restoring the performance of the site
does not automatically restore the previous user behaviour.  From the
second link, "Even after turning off the [extra] latency, users
continued to search less. That’s right: The perception of slowness had
impacted the users’ attitudes about the usefulness of the site."

The long-term effect of extra latency is that people "learn" that your
site is slow, and don't necessarily or automatically "unlearn" this
when it gets faster.  So clearly we're constantly working to make
things faster.  IPv6 is important for a variety of reasons, but it
definitely shouldn't mean the web gets slower.
Asymmetric/unmanaged/whatever-you-want-to-call-them transition
mechanisms like 6to4 and Teredo tend to add latency which, for us
dualstack sites, is of the "just plain unnecessary" variety.

Hopefully this makes it a little more clear.

More information about the ipv6-ops mailing list