Re: [DNSOP] Anycast and DNS questions

[joel jaeggli <[e-mail]>]

On 9/3/14 10:01 AM, David Conrad wrote:
> Hi,
> 
> On Sep 3, 2014, at 8:42 AM, Guangqing Deng <[e-mail]>
> wrote:
>> From RFC1034 section 4.1, it seems that the way used for improving
>> the redundancy and resilience of DNS system is to increase DNS
>> servers. I agree that for the performance of the DNS system, the
>> redundancy and resilience are the first goal and low latency is the
>> second goal. Usually, the first goal mainly depends on the DNS
>> server deployment policy (such as the total number and geographical
>> distribution of DNS severs) and the second goal relates to not only
>> the DNS server deployment policy but also the method used for DNS
>> clients selecting the best DNS server like any cast.

anycast is not a selection mechanism employed by a client. it is the
network that determines the catchment area served by a given anycast
instance.

> Careful here.
> 
> Anycast improves redundancy/resiliency for the system as a whole.  As
> typically deployed, it may not improve redundancy/resiliency for a
> single client.  For example, if a DNS server instance in an anycast
> cloud is no longer responding to DNS queries due to a DoS but the
> routing announcement of that instance is not pulled down, the clients
> topologically nearest that instance will not see improved
> redundancy/resiliency — they’ll not see any responses.
> 
> Anycast may or may not improve latency — it depends on the routing
> system. If the “nearest” instance network topologically to a set of
> clients happens to be on the other planet, latency will not be
> improved for those clients.
> 
> Anycast is a very blunt tool. It can help improve
> redundancy/resiliency and latency if properly deployed, constantly
> monitored, and maintained, but it is very important to understand its
> limitations and implications.
> 
> Regards, -drc
> 
> 
> 
> 
> _______________________________________________ DNSOP mailing list 
> [e-mail] https://www.ietf.org/mailman/listinfo/dnsop
>