[dns-operations] cache and subnet option
kbeevers at ns1.com
Thu Apr 20 01:48:47 UTC 2017
I had this half typed when Mark responded so dropping it here anyway.
In message <m2shl36fq2.wl-randy at psg.com>, Randy Bush writes:
> > dear lazynet,
> > caching resolver R serves clients on subnets S0 and S1. a device on S0
> > queries for wanna.know. R queries for the data using the subnet option.
> > a device on S1 then queries for wanna.know. does R respond with the
> > cached data from the S0 query, or issue a new query with the S1 subnet
> > data?
> > [ to my na=EFve eye, issuing a second query would seem to have, shall we
> > say, interesting scaling properties ]
> > randy
> It depends on the addresses and the answers. Yes, ECS has scaling
> issues and it depends on the authoritative servers returning the
> shortest prefixes possible which in turn depends on ISPs using
> sensible addressing plans which keep topologically close clients
> under the same prefix.
Answer is as usual "it depends" -- but more or less, it's "the latter" --
it issues a new query with S1 as the subnet. The most important caveats
are that it depends on the scope of the subnets and what R considers a
sufficiently specific but not too revealing scope -- usually in practice,
this is /24; and that it depends on the scope returned in the response from
the ECS-enabled authority. For example if we receive a query scoped to
126.96.36.199/24 but we know we will return the same response for all addresses
in 188.8.131.52/16, then we will respond with that scope in the hopes that the
resolver caches with that scope and is smart enough not to ask a different
question for 184.108.40.206/24.
As Mark pointed out, there are indeed scaling challenges here -- especially
on the caching side.
Kristopher Beevers, CEO
kbeevers at ns1.com
desk: +1-855-438-6766 ext. 701
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the dns-operations