Re: [DNSOP] Request for Adoption (draft-moura-dnsop-authoritative-recommendations)

[Joe Abley <[e-mail]>]

Hey Giovane,

On 28 Nov 2018, at 04:55, Giovane Moura <[e-mail]> wrote:

> We have a new draft and we'd like to ask the WG to adopt it:
> 
> https://datatracker.ietf.org/doc/draft-moura-dnsop-authoritative-recommendations/
> 
> This is an informational draft that presents recommendations for
> authoritative DNS operators, based on research works we have been
> conducting over the last few years.

I have read this document. This is not a detailed review, just a more general reaction to the adoption question.

This document reads a bit like it wants to give operational advice but in fact it seems much more like a literature review that has been culturally adapted to an IETF audience. This is not intended to be a negative comment; I think the references and the context are useful and I think it's a great idea to find a way to push pointers to operational research into the RFC series, but I think it would be as well to make it clear that the operational advice contained within is not comprehensive and in many cases is quite superficial. I can give examples if you want more detail.

I also think that the document is not clear on what problem it is solving; in particular, it doesn't define "performance". It's inferred that the performance metric that matters is round-trip query latency for some nominal client at an average distance; what if the operator's priority is availability? That makes R1 potentially bad advice; no doubt there are other motivations that would also not fit the recommendations convincingly.

Much of the document is specifically concerned with the implementation of anycast strategies for DNS, not with authoritative DNS in particular. The advice would hold for any stateless (or short-duration transaction, where short is relative to routing stability). One direction the authors might consider taking this is to update the advice in RFC 4786 with reference to the repeatable experiments that you cite. Ironically 4786 was criticised for being a document about DNS that was pretending to be more general; perhaps when welded to your document the result would be more pleasantly neutral :-)


Joe