[dns-operations] anycasting axfr

[Jim Reid]

> On 18 Apr 2017, at 12:33, Tony Finch <dot at dotat.at> wrote:
> 
> Jim Reid <jim at rfc1035.com> wrote:
>> 
>> Besides, addition or removal of anycast nodes can change the underlying
>> routing topology while the xfer is in progress.
> 
> Anycast works well enough for HTTP, so this isn't a very convincing
> argument against using it for AXFR. E.g.
> https://www.nanog.org/meetings/nanog37/presentations/matt.levine.pdf

You’re comparing apples and oranges. The characteristics of a TCP session for AXFR and HTTP are not the same. The slides you reference say anycast works OK for web sessions: fine. But there’s no equivalent analysis for DNS AXFRs. At least not yet AFAIK. 

BTW that presentation seems to rely on analysis of a small anycast setup consisting of one node in three continents. Anycast offerings for the root and major TLDs are much larger and “denser” than that. [Rhetorical question: how many paths are there between your net and the anycast nodes for F or E or D or... in the UK?] So end clients may be exposed to localised route changes and outages that wouldn’t be as visible for a small anycast network containing just one node in say Europe, another in N. America and another in Asia.