Re: [DNSOP] New Version Notification for draft-yao-dnsop-root-cache-00.txt

["Joe Abley" <[e-mail]>]

Hi again,

On 29 Sep 2015, at 21:23, Jiankang Yao wrote:

>> What reason do you have to think that response latency from root 
>> servers
>> has any measurable impact on end-user experience?
>>
> I think that there are some papers which explain it.

Then some citations would be useful.

To be clear, I'm not having trouble understanding the concept. I just 
don't believe it.

> One observation:
> due to the complexity of the network environment, the current quality 
> of access to root service is uneven globally.  For example, CNNIC 
> finds through comprehensive monitoring and analysis that in China the 
> time delay of access to the 13 root servers varies greatly from 
> province to province, showing a difference of up to 200ms for most 
> root servers, and in a number of provinces nearly 60% queries fail to 
> hit the root mirrors deployed in China.

I believe that, for sure. And I understand the desire to bring the root 
server infrastructure closer to end-users in China, motivated by 
concerns about availability and the problems that would result from 
prolonged non-availability.

However, I think it's controversial to imply that end-user performance 
is substantially improved by the wider (and closer) deployment of root 
zone data. That assertion needs some justification.

For example, a recursive resolver whose clients mainly ask for names 
that end in (say) COM only need to talk to root servers every 172,800 
seconds (that's the TTL on the NS set in the root zone and at the COM 
apex). Even if it takes a really long time to get a response from a root 
server when I need one (e.g. ten seconds, which seems like a high 
estimate), 10/172800 is less than 0.1%, and even in that case it's 
likely that records for popular COM names are independently cached and 
in fact there would be zero impact on end-users that depend on those 
common names even during such an event.

> BTW,
> this draft is trying to solve the same problem speicified in 
> draft-ietf-dnsop-root-loopback.
> I think that the authors of draft-ietf-dnsop-root-loopback  will have 
> a better explaination than me.

I had much the same feedback to the authors of that draft. In that case, 
however, it was observed that slaving the root zone was common (if 
perhaps not widespread) practice, and it was worth documenting the 
trade-offs and recommending a consistent approach for those who insisted 
on doing it. That's not the case here.

I think I would need to see a convincing problem statement and 
understand how this proposal provided effective solutions before I could 
support it.


Joe