DNS Reliability

Eric Brunner-Williams brunner at nic-naa.net
Fri Sep 13 02:32:44 UTC 2013

On 9/12/13 1:39 PM, Rubens Kuhl wrote:
> ICANN new gTLD agreements specified 100% availability for the service,
> meaning at least 2 DNS IP addresses answered 95% of requests within 500 ms
> (UDP) or 1500 ms (TCP) for 51+% of the probes, or 99% availability for a
> single name server, defined as 1 DNS IP address.

unless phil happens to be building out (or spec'ing out $provider's
offered sla) for one of the happy thousand or so celebrants of 2014, a
surprisingly large fraction of which are tenant plays on existing
infrastructure, the bogie above, uninterpreted, is not a controlling

additionally, was phil asking for a metric for an authoritative
server, serving a zone delegated directly from the iana root? was he
asking for a metric for a caching server?

and if the metric is "queries completed vs. queries lost", from where
to where? (that is the "uninterpreted" bit from the bogie rubens
quotes, as we did have to correct some assumptions of the requirement
author -- where is the measurement being preformed?

i'm with randy on this, dns is a service, the better question is what
fails as query response degrades, in the presence of hierarchical
caching and the protocol being used as designed under best effort of
infrastructure and application.


More information about the NANOG mailing list