On 26 Feb 2020, at 14:18, Scott Mutter via mailop wrote:
[⋯] Do any DNS resolvers actually cache
data for the period stated in the TTL these days?
Many do. If you're operating a recursive for any sizable user
population, you want to minimize the response time. Having the response
in your local cache is actually as fast as you can get. Then again, with
long TTLs comes the longevity of errors. This is why public resolvers
have heuristics / buttons to forget data ahead of time or trigger a
refresh.
I've seen some studies that compare large recursive resolver
performance, that left me with the impression that at some sites, the
resolvers are resource-starved. I wouldn't think this is a deliberate
stance, as it degrades the quality perception of customers.
If you look at gmail.com it's TTL is 300 seconds - now... granted that
IP
address is not used to actually connect to mail server to send out
mail,
it's just the IP address for the front facing gmail.com.
Likely, they need to be able to point to a wholly different anycast node
on a whim, or don't want you to carry a cached response when roaming
between networks. I would not consider any large sender as a good
example of the discussion on this context, because with that scale, come
very specific challenges.
Many of the infrastructure elements I manage have sub-1d TTLs in their
DNS records except for things like TLSA records and such. In our case,
this is to ensure that changes can be deployed quickly. This of course
comes with the price that we will disappear much faster from the DNS if
we manage to screw up our geo-diverse name servers.
I definitely would subscribe to the notion that TTL should not matter
for
this. But should and does are two different things.
+1
-lem
_______________________________________________
mailop mailing list
[email protected]
https://chilli.nosignal.org/cgi-bin/mailman/listinfo/mailop
