Re: [Resolverless-dns] Paper on Resolver-less DNS

[Joe Abley <[e-mail]>]

On 4 Sep 2019, at 10:50, Viktor Dukhovni <[e-mail]> wrote:

> That said, browsers do DNS lookups in parallel, and the user's
> resolver will often already have the answers in its cache.  Is the
> resulting reduction in latency really worth the added complexity
> and potential risks?

I think the context here should usefully be those sophisticated web pages that operate using large numbers of unique names whose only purpose relates to a particular (user, application) pair, and in some cases only have a single-time use (for which caching is just a waste of RAM).

Many single-use names have low or zero TTLs in order to signal that they should not be cached, but many caches impose local policy on minimum TTLs and hence they are cached anyway. The impact on caches can be significant [Dagon]. The queries have been observed to continue to circulate through the DNS globally, regardless of their intended single-use [Huston].

I think it's straightforward to imagine that many, perhaps most of the world's web content would not load more quickly/reliably/pleasingly just because of the introduction of resolverless DNS. However, scale matters and in a world of elephants and mice, the effect on the small number of elephants might matter more than the lack of effect on the rodents.


Joe