George Michaelson wrote:
What did I mis-understand? The APNIC 1x1 is a random sample over
users, and it sees significantly more than 75% EDNS0. More like 93%.
by query volume, or by rdns ip?
By query volume. And, we're at authority where I suspect you see the edge.
Once the query is covered by a modern competent resolver, and gets
passed, it acquires. EDNS0
most middleboxes who think they know what a dns packet has to look like
and willfully intercept or rewrite or drop them, are between the stub
and the (desired) rdns. i am not sanguine about getting edns0-carried
data to a stub often enough during my lifetime to say, let's solve the
a-vs-aaaa problem using another protocol extension.
if the internet is a territory, then the dns, like bgp, is the map of
that territory. since many commercial and government actors want to
control access to the internet, they mostly do it by controlling access
to the dns. and their incentives are misaligned such that they do not
care what they break or what upgrade paths they deny.
note that solving a+aaaa with an edns0 extension may be good enough, if
we think that getting the aaaa rrsets cached more often will mean that
the stub's aaaa query will less often lead to a cache miss, as long as
the largest share of RTT for a stub query that leads to a cache miss is
still on the rdns-to-auth path and not the stub-to-rdns path.
--
P Vixie
_______________________________________________
DNSOP mailing list
DNSOP@ietf.org
https://www.ietf.org/mailman/listinfo/dnsop