Re: [v6ops] Lion/Snow Leaopard side-by-side on an IPv6 enabled ADSL2+ home line..

[Keith Moore <[e-mail]>]

On Aug 6, 2011, at 10:44 AM, Rémi Després wrote:

>> Finally, if you provision your networks so that IPv4 and IPv6 exhibit "roughly equivalent" levels of path latency and congestion, then we're going to try very, very hard to use both IPv4 and IPv6 in "roughly equivalent" measure.
> 
> Serious disagreement on this one.
> In this case, v6 MUST be preferred:
> - If both paths are "good" (as they should be!), preference for IPv6 tends to increase the v6/v4 traffic ratio, which will lead to a visible fading out of IPv4 use.
> - It's easier to implement, AFAIK, than trying hard to evenly split the traffic between v4 and V6.
> Any solid argument against this?

As far as I can tell, it's easier to implement if you don't try to bias in favor of IPv6.  Then the difference between v4 and v6 will amount to random noise, and the choice between a v4 path and a v6 path will essentially be random.  I don't know that it's necessary to "try very, very hard" to use both; I think this will happen fairly naturally.

But it might be reasonable for the software to do something like: if there's not a statistically significant difference between the measured performance of the different paths available, fall back to an rfc 3484-like policy table. 

Keith