[dns-operations] Anycast vs. unicast NS

[Stephane Bortzmeyer]

On Fri, Mar 18, 2011 at 12:16:40PM +0100,
 Shane Kerr <shane at isc.org> wrote 
 a message of 54 lines which said:

> In principle you're right, although in practice I see quite good
> correspondence between physical location and routing topology.

There was a very good talk by Steve Gibbard (Nominum, ex-PCH) at the
OARC workshop this week
<https://www.dns-oarc.net/oarc/workshop-201103>. His experience and
conclusions were quite opposite. For instance, testing J-root (which
has an instance in San Francisco) from four different points in San
Francisco led to many places...

My own anecdotal evidence supports it: traceroute to our anycast
servers from various places in the world show that routing does not
take place in an Euclidian space... 

Anycast is great for reliability and resiliency, but it does not
optimize the RTT.