Re: [Resolverless-dns] Load-balancing concerns

[manu tman <[e-mail]>]

On Thu, Nov 8, 2018 at 2:04 PM Dave Lawrence <[e-mail]> wrote:

> manu tman writes:
> > Except that it will be based on the IP of the DoH server and not the
> > resolver of the client,
>
> Right, as I said (or thought I said).
>

It was not clear to me, you mentioned it would be resolver-based but this
resolver is now unrelated to the original resolver. I may have missed the

>
> > which in most cases will be more representative of the actual
> > network location (ISP or subnets of ISP) of the client.
>
> I question the validity of this assertion.  It seems to be making a
> huge assumptive leap about where the server providing the records is
> and what its network relationship to any given client is, and
> completely discounting typical resolver locality.  "in most cases"
> will need some data to support it, and it's way too early in the brave
> new DoH world to have good data on that.
>

I am referring to the recursive resolver of the original client, not the
DoH server that would now return records. Before resolverless, you would
return answers based on client recursive resolver IP (which I assume is
somewhat "close" to the end user, but don't have data to back this up
with), with resolverless dns, it would be based on the IP of the DoH server
(or the recursive resolver used by that one).

>
> Personally I expect DoH to be used in a variety of environments, not
> just by the biggest providers with worldwide footprints.
>
> > The DoH server will most likely be mapped to the best location for
> > the majority of the clients using the service
>
> Who said the DoH server is getting mapped at all, or has any locality
> to the additional resources it is trying to provide access to?
>
> Imagine a pretty reasonable set up where some website which only has a
> handful of deployments, say a couple in North America and a couple in
> Europe, wants to point records for one of its partners which actually
> has a much broader footprint.  An Asian client connected to one of the
> North American data centers and getting American-mapped answers when
> the partner site would much rather have preferred you be using one of
> their Asian installations if only you'd asked through normal DNS.
>

This is what I wanted to say and failed at apparently. I did not mean to
contradict your statement, but rather support it.
That being said, chances are that if the majority of the users that use the
first North American server are actually Asian users, chances are the
effect will be that North American users will end up being sent to the
Asian servers of the second site instead of the North American one. As you
said, this is network performance mapping, not geo locality.


> That's just the simple case of network-performance mapping anyway, and
> completely ignores other mapping
>
> --
> Resolverless-dns mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/resolverless-dns
>