[dns-operations] Quick analysis of TLD SOA's

[Joe Abley]

On 2011-07-13, at 12:22, Simon Munton wrote:

> On 13/07/2011 16:54, Phil Regnauld wrote:
>> 	Fair enough, provided NOTIFYs are received
> 
> NOTIFYs have to be ACKd by the Slave (RFC1996 4.7), and are re-sent if not received ... up to a point

Some redundancy of design is possible, though, e.g. if zone data flows

  M -> ( DM1, DM2, ... ) -> ( NS1, NS2, ... )

(M being "master", DM being "distribution master", NS being nameservers that answer queries from the world and hence represent the delegation/apex NS sets for a zone)

then you can configure a a partial mesh of NOTIFYs, e.g.

  M -> DM1
  M -> DM2

  DM1 -> DM2
  DM2 -> DM1

  DM1 -> NS1
  DM2 -> NS1

  DM1 -> NS2
  DM2 -> NS2

  (etc)

By making sure there's at least two NOTIFY paths for each nameserver you reduce the chance that a (network, other) problem in one NOTIFY path leads to a nameserver falling back to (REFRESH, RETRY) semantics for zone transfers.


Joe