IPv6 Source Address Selection on Mac OS X Lion

[Bjoern A. Zeeb]

On 16. Dec 2011, at 18:13 , Gert Doering wrote:

> Hi,
> 
> On Fri, Dec 16, 2011 at 06:11:11PM +0000, Mark Blackman wrote:
>>> On Thu, Dec 15, 2011 at 09:31:42AM -0800, Dan Wing wrote:
>>>> It depends on your definition of "better".  If your definition is
>>>> "prefer IPv6", you are right that OSX Lion's algorithm fails.  If
>>>> your definition is "connect to whichever is fastest", OSX Lion's
>>>> algorithm wins.
>>> 
>>> "consistent behaviour".  And this is where Lion fails, because it
>>> flip-flops back and forth between protocols even if nothing changes
>>> in the network.

You want consistent behaviour?  Do what I do.

>> 
>> Isn't this behaviour what Chrome does too?
>> 
>> http://src.chromium.org/viewvc/chrome?view=rev&revision=85934
> 
> on a network where IPv4 and IPv6 have the same (or close) RTTs, chrome
> will *consistently* pick IPv6.  Not "sometimes IPv6, sometimes IPv4".
> 
> Big difference.
> 
> The general idea of HE is good.  HE-in-Lion a step backward.

The missing knob to turn it off sucks badly in either however and while I stop
caring myself I am seeing the problem you describe in dual-stacked networks for
others unfortunately and would love to have a switch to flip to get v6 first v4
then back (at least on demand).

Here, there is no IPv4 address anymore usually, a 127.1 on Lion.
The v4 link-local on XP doesn't seem to matter much currently for IE luckily
and on FreeBSD/PC-BSD the no-IPv4 stack does the job as well:)  No decisions to
make for complex algorithms.  Deterministic behaviour.  User really happy.
Debugging is simpler.  Problem solved.  Good night.

/bz

-- 
Bjoern A. Zeeb                                 You have to have visions!
   It does not matter how good you are. It matters what good you do!