[dns-operations] Comments regarding RFC 3484

[Florian Weimer]

* Peter Koch:

> On Tue, Sep 18, 2007 at 07:57:54PM +0200, Florian Weimer wrote:
>> Any more comments on this topic?
>
> Thanks for spotting the issue.

Well, at Debian, we've received bug reports, and it appears that it's
impacted our CDN to some extent.  IOW, we've got something used to
work in pre-RFC 3484 days, and no longer does due to fully
deterministic address sorting (which was introduced with an libc
earlier this year).

> The proximity heuristics provided by CommonPrefixLen() do not make
> sense for IPv4, and only limited sense for IPv6.

I agree in the most strongest terms.  I didn't want to bias the
discussion from the onset.

> The basic question is what the actual state of deployment of
> RFC 3484 is.

Test results are very inconclusive.  For instance, according to my
tests, FreeBSD 6.2 performs round-robin across all addresses.  But
others have reported that FreeBSD implements Rule 9.  Someone else
could not reliably demonstrate that Fedora implements Rule 9 (even
though the libc maintainer, a Red Hat employee, claims that RFC 3484
support is a must or something like that).  Some of that could be
attributed to different caching resolvers used in different tests, but
some results are clearly bizarre (such as Windows XP not choosing the
Rule 9 address in all cases). *sigh*