[dns-operations] Signaling client protocol to authority

[Joe Abley]

On 2011-01-16, at 21:29, Patrick W. Gilmore wrote:

> BTW, I do agree the problem isn't the name server, it is definitely the client.  However, Grandma ain't gonna figure out her OSX is too old to do v6 right (hell, the latest one STILL does broken shit), so perhaps we can help her out in other ways?

While I seem to think I have seen some attempts to mitigate v6 transit brokenness (e.g. google only returning AAAA records, over v4 or v6, when they come from nameservers in an AS they peer with directly) I also have not heard a suggestion for how to deal with pervasive v6 problems in customer networks or client stacks.

The engineering-correct answer would seem to be "make things break and distribute the cost of fixing it to ISP/Microsoft/Apple/Linksys/whatever helpdesks". I think it's not unreasonable for an ISP who rolls out v6 to its end users to bear some of the cost of the implications of doing so, however unpleasant that thought might be to ISPs who are already looking at pretty slim margins on residential DSL. Presumably that cost will be mitigated by the clueful by staging the roll-out of IPv6 across a customer base, and perhaps doing some automated testing to identify users who appear to have some kind of problem.

I can't see how signalling the end user address (or available transports) from a stub resolver through a cache to an authority server is going to work for reasons mentioned elsewhere in this thread (e.g. the legitimate use of an IPv4 resolver on a dual-stack client) but I agree the general problem space, not just in the DNS but more generally, deserves some thought and perhaps some experimentation.

Perhaps a way to start is for content providers to start serving up one-pixel PNGs on both IPv6 and IPv4 to one in every N clients and measure the success in retrieving them in the cases where the client's AAAA DNS requests are deliberately responded to with a name error, or with a real v6 address. This would at least allow the scope of the problem to be understood (e.g. is it 80% of content consumers who are about to break if you serve an AAAA, or is it 0.002%?)


Joe