Talk:EDNS Client Subnet

incompatible with caching?

How is this supposed to work together with caching on public recursive resolvers? When the answer (for a query which contains EDNS client subnet information) is personalized, it cannot get cached, and so the reason to use a caching resolver goes away - the client has the same RTT time as if it would asked the authoritative server himself. --93.200.30.76 (talk) 00:43, 3 September 2022 (UTC)

ECS isn't incompatible with caching. But a resolver needs to cache records separately for different subnets. Thus multiple devices within the same /24 network (if that's the prefix used by the resolver) will be able to share the same cache, but devices in a different network will have their own cache. With smaller subnets (larger prefixes) cache hit rates will reduce and RAM usage will increase. The resolver gets to decide on the prefix, so if a resolver wants to have a higher cache hit rate and use less RAM, it can choose a prefix of (say) 16. This will reduce the ability for an authoritative DNS server to be highly granular in returning geographic-appropriate responses, but won't prevent it altogether, and the granularity often doesn't need to be better than country or state.

So it's not quite correct to say that ECS is incompatible with caching. But ECS granularity and cacheability are somewhat opposed, in a balance that can be tuned at the resolver.

The article could do with a section on how ECS affects caching. Jlaidman (talk) 00:26, 30 April 2024 (UTC)