To:
kre@munnari.OZ.AU
Cc:
ipng@sunroof.eng.sun.com, dnsop@cafax.se
From:
Bill Manning <bmanning@ISI.EDU>
Date:
Wed, 15 Aug 2001 13:30:04 -0700 (PDT)
Sender:
owner-dnsop@cafax.se
Subject:
why A6 might be harder to deploy
cc list trimed: A couple of other considerations: After IETF49, we held a ws that ran DNS on v6 transport. We tried A6 & AAAA records in the context of the DNS root. Dales notes reflect that FreeBSD's resolver was unable to recognise an A6 record when it was "glue" for a root server. It was able to recognise a AAAA record. This has not been rechecked to my knowledge. This is also applicable between the servers for a zone. In a mixed environment, it is possible to have servers running a variety of code, from Bind4, Bind8, Bind9 and even some of the other genetic varients (Microsoft, Lucent, PowerDNS, Akami, et.al.) Indeed, this is one of the strengths of the DNS. And in such clusters, RR 28 is well known and is passed around. If an NS records "glue" is type 28, zone transfers will occur. However nearly none of these versions/varients knows or supports type 38. And given the strict checking, for coherancy, in most of the recent code, if the glue for an NS is type 38, the zone transfer will -fail- That seems to indicate that if type 38 is to be used, all the servers for the zone in which they reside must be running code from a single branch. (the unknown RR type problem) And there are indications that that particular branch is not all that speedy. Some numbers I have seen indicate the QPS rate is about 50% less than the previous branch. Are we really ready to take that performance hit to gain A6 support? Random idea... Presume 10,000 v6 users today. If we give them AAAA now, and agressively explore/test A6, on the experimental track for 12-18 months, we might presume there would be 1,000,000 v6 users by then. That is still statistically "small" as an installed base and we would have a much stronger case that A6 performance issues are addressed, A4synth could be better understood, and operational tools are available. If we put forth the effort, it should be a no brainer to push it back on stds track and consider moving AAAA to historic (sort of like RIPv1 is historic... people still build it and use it. :) The installed base is still small enough that we can make the change... I think. --bill