On 5 Mar 2026, at 00:24, George Michaelson <[email protected]> wrote:

> I have run local root. I don't see any problems with my service, running 
> local root.

I am not doubting your operational diligence George, but for many people I can 
imagine persistent and enduring failures that will go unnoticed and the result 
will be new TLDs that are not universal or retired TLDs that persist. If you 
don't use something and you don't measure it, how would you expect to see 
failures?

I appreciate that DNSSEC is there to save us from such problems, but software 
has bugs and humans make mistakes and our goal ought to be to protect the 
namespace expecting that those things are true, not trying to legislate that 
they must be false. We have certainly seen fixes to DNSSEC validation failures 
of the form "turn off validation". Hope is not a strategy. 

> I don't see the relevance of fetch mechanisms to success or failure here, or 
> the rate of churn in the root as a significant issue for a local root copy 
> mechanism.

Me neither. The data transfers are minuscule with or without incremental 
transfers and I do not understand some of the strong opinions about data 
distribution mechanisms unless they are anchored in not-invented-here.

However, to avoid the risk of sounding too much like I am in favour of all of 
this, let us continue...

> I do accept that there are a cohort of people who have downside consequences 
> of reduced traffic to the roots.

We have a root server system that is already quite hard to measure, an easy 
example of which was the extensive fear and loathing around the first KSK 
rollover, but we manage to come up with plausible numbers for availability and 
system health that are sufficient to convince us that the system is stable and 
secure. 

I don't know how we convince ourselves of such things if local-root becomes 
prevalent. Some are saying that this is a rare and niche local optimisation and 
prevalence is not expected, but I also hear that this is all fine because we 
can trust the major implementations to do this safely and well. But if all the 
major implementations are doing this then it no longer seems rare and niche; it 
seems like one default setting away from mainstream. 

This all sounds like a solution looking for a problem to me. Unless the problem 
is actually "let's make the root server system unnecessary" I don't really know 
what this is all for; I don't see arguments for increased security, 
observability or stability. 

I am not arguing against this work. I have concerns but I don't think it's 
actively harmful. Just because I think it smells funny doesn't mean others 
shouldn't enjoy its delicious and heady flavours.


Joe

_______________________________________________
DNSOP mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to