Andrew, At 2016-03-23 09:11:44 -0400 Andrew Sullivan <a...@anvilwalrusden.com> wrote:
> On Mon, Mar 21, 2016 at 06:55:00PM -0700, Paul Vixie wrote: > > i think we have to start planning for a world in which EDNS0 never reaches > > 75% penetration due to middleboxes having the high ground. > > Well, you could get most of the way there for this proposal by having > just recursives do it. End users are mostly going to do happy > eyeballs anyway, so if a recursive gets the AAAA with the A its own > load goes down because it can answer more stuff from cache. I'd like to argue that supporting A+AAAA on the stub side is potentially huge. Given that 90%+ of a recursive resolver's time is spent reading stuff out of cache, being able to get the data to clients with fewer queries seems huge to me. This is a big, fat win for resolver operators. It also seems not hugely complicated. On the stub resolver side, it should be relatively simple to discover if this functionality is supported, and use it if it is. We could add a special query for this that a host can use for this, or any of 100 other methods. On the recursive resolver side, it should not impact performance, as no additional searches should be required to get both the A and AAAA data for a recursive resolver. While there is additional logic to handle this approach (because of issues of collecting A and AAAA together into a single reply), it's not hugely complicated (at least not compared to all of the other the crap that resolvers have to do while talking to the authoritative world). Cheers, -- Shane
pgpD8M71tjQp_.pgp
Description: OpenPGP digital signature
_______________________________________________ DNSOP mailing list DNSOP@ietf.org https://www.ietf.org/mailman/listinfo/dnsop