Re: [DNSOP] Solicit feedback on the problems of DNS for Cloud Resources described by the draft-ietf-rtgwg-net2cloud-problem-statement

[Paul Vixie <[e-mail]>]

On Tuesday, 11 February 2020 22:21:05 UTC Linda Dunbar wrote:
> ...
> 
> https://datatracker.ietf.org/doc/draft-ietf-rtgwg-net2cloud-problem-statement/
> 
> During IETF 106, we received comments that the document should cover the
> problems associated with DNS service by different Cloud Operators for
> Enterprise to utilize Cloud Resources even though DNS is not within the
> scope of IETF routing area.  We greatly appreciate DNS experts to provide
> comments to our description.

you've addressed this e-mail to two mailing lists (dnsop@ and rtgwg@) which 
you are a member of, and both will accept and publish your e-mail. however, 
some of us here are members of only one of those mailing lists (me, dnsop@), 
and won't be able to participate in whatever threads may occur on the other 
mailing list (me, rtgwg@). so, i am removing rtgwg@ from my reply here.

> 3.4    DNS for Cloud Resources
> DNS name resolution is essential for on-premises and cloud-based resources.
> For customers with hybrid workloads, which include on-premises and
> cloud-based resources, extra steps are necessary to configure DNS to work
> seamlessly across both environments. ...

you may not realize it, but that is an astonishing statement. i'll explain 
below.

> ... Cloud operators have their own DNS to
> resolve resources within their Cloud DCs and to well-known public domains.
> Cloud's DNS can be configured to forward queries to customer managed
> authoritative DNS servers hosted on-premises, and to respond to DNS queries
> forwarded by on-premises DNS servers. ...

while this is an obvious approach for each and every cloud service operator, 
it depends on lock-in, is not multi-cloud ready, and cannot be made so within 
the DNS paradigm or using any of the common layers added to that paradigm.

DNS is currently viral, if you can look up anything at all global, then you 
can look up everything global. cutouts for non-global names are quite common 
especially when accompanied by NAT. however, collisions cannot be managed this 
way. you can have some names (like .cloud or .corp or .internal) visible 
within your network as long as they aren't also used globally, because there 
is no way to discriminate which collision-name is wanted, other than by 
client-ip address and even that is nonstandard, relying on DNS vendor 
extensions.

similarly, if you use an internal name like .cloud and then want your services 
to be available via or within some other cloud provider which also uses 
.cloud, then there can be no discriminator, and so it can't work. this is why 
most names are global -- there can be no collisions that way. even with large 
NAT buildouts, it's become common to use the global domain name both 
internally and externally, and to simply provide different views of that 
global domain name internally vs. externally.

this was explored 25 years ago: http://family.redbarn.org/~vixie/proxynet.pdf

can you explain why the simple answer (use unique global names for each cloud 
resource) isn't reachable by your proposal?

> ... For enterprises utilizing Cloud
> services by different cloud operators, it is necessary to establish
> policies and rules on how/where to forward DNS queries to. ...

if the names are local, and never collide, this has been possible in one 
vendor's DNS implementation for about 20 years, and is variously possible in 
others. however, this became much more difficult in the DNSSEC era.

if the names are local and may collide, no coherent set of policies or rules 
about how/where to forward DNS queries will be possible.

if the names are global then they will be unique and DNS itself will handle 
the decision of how to route questions to the right authority servers.

> ... When
> applications in one Cloud need to communication with applications hosted in
> another Cloud, there could be DNS queries from one Cloud DC being forwarded
> to the enterprise's on premise DNS, which in turn be forwarded to the DNS
> service in another Cloud. Needless to say, configuration can be complex
> depending on the application communication patterns.

it is that complexity that prohibits scale. are you suggesting that the DNS 
paradigm be expanded to include automated context-dependent naming? i imagine 
it would look something like BGP4, but for names rather than addresses. but 
first i hope you can explain why the simpler and existing viral DNS paradigm 
(all names are global and unique) is unacceptable for your purpose.

thanks for bringing your question here, and thanks to otherpaul and suz for 
what appears to have been vital outreach.

-- 
Paul