Re: [v6ops] IPv6-ness

[Philip Homburg <[e-mail]>]

In your letter dated Tue, 11 Oct 2011 06:38:09 -0400 you wrote:
>The main thing =
>that comes to mind is that some applications are more =
>bandwidth-sensitive whereas others are more delay-sensitive, so the =
>choice made by HE might be good for some apps and bad for others.  

Yes, that is true. If the IPv6 connection works well, but has significant lower
bandwidth than IPv4 then HE will be wrong.

This is not something that is easy to solve in an application. The application
would have to measure the bandwidth over both IPv4 and IPv6 and then take a 
decision. I guess that would be doable in torrent but torrent should not need
HE anyhow.

>(No, it's not reasonable to assume that the application falls back to =
>RFC 3484.  RFC 3484 is also not applicable to all apps.  But neither =
>does it really get in the way of apps, because no app is forced to try =
>addresses in the order returned by getaddrinfo)

I don't think it is so much the issue whether an application is forced to 
use HE or not, but what generic algorithms are available to applications.

>The -04 draft of HE seems to be written on the assumption that a either =
>host only has one IPv4 address and one IPv6 address, or that all of a =
>host's connections are of approximately equivalent capability.  That's =
>not a reasonable assumption in practice.

I don't think HE can solve how to deal with multiple interfaces. Even for just
IPv4 that is mostly an unsolved problem.

One thing is that if getaddrinfo knows what to do, then HE may also be okay,
because HE can be made to prefer whatever getaddrinfo prefers.