Hi Paul and others, On 29 Jan 2026, at 17:15, Paul Hoffman <[email protected]> wrote:
> On Jan 28, 2026, at 18:22, Geoff Huston <[email protected]> wrote: > >> This makes the measurement of "DNS latency" to be so extremely variable that >> the concept >> becomes somewhat meaningless, particularly when you might want to use such >> measurements >> as a comparator. > > +1 to everything above. Note that this also affects the definition of "query > latency": the latency for a query from a stub resolver to a recursive (with > the RD bit set to 1) will likely be very different than the latency from a > recursive resolver to an authoritative server (with the RD bit set to 0). > > If you want to try to define latency in the DNS, you have to specify what > types of queries are being sent. This comes up semi-regularly, not necessarily here in this working group but elsewhere. An example is the constant fervour around RTT measurements on DNS service provider ranking sites. I have often wished there was more appreciation for other metrics like availability and route diversity for anycast service addresses; while some DNS queries are often in the hot path for user experience, others are only very rarely in that situation (e.g. referral responses to TLD nameservers are almost always always in the cache, and the ability to reliably fill the very occasional cache miss seems much more important than whether that cache fill takes 2ms or 200ms). Perhaps this is something that is worth writing down. It feels vaguely BCPish. Joe _______________________________________________ DNSOP mailing list -- [email protected] To unsubscribe send an email to [email protected]
