Re: [Resolverless-dns] Paper on Resolver-less DNS

[Erik Sy <[e-mail]>]

On 8/18/19 18:24, John Levine wrote:
> In article <[e-mail]> you write:
>> The privacy problem is that a significant share of DNS resolvers monitor
>> the users' online activities, aggregate these data in user profiles and
>> use these profiles within behavioral advertising or share these user
>> profiles with other parties. Here [1], you can find a comparative
>> analysis of public DNS resolver privacy policies substantiating my claims.
> I took a look at that paper, written by two grad students.  
I think the term grad student is not an appropriate description for a
professor at the American University in Washington, DC. From my
perspective, Samantha Bradshaw [1] and Laura DeNardis [2] have a good 
academic track record in the filed.
> If I were
> their advisor I would have handed it back to them and told them to
> talk to some people who understand the topic so they could fix the
> many painful factual errors.  They got the whole WHOIS GDPR fight
> backward, and made the rather large leap assuming that every word in,
> say, Google's generic privacy policy applies to queries to 8.8.8.8.

The authors describe this limitation in their paper. In detail, some
privacy policies are too generic and lack specific information about
what DNS query data was collected, used and stored. Nonetheless, the
authors concluded that the privacy policies of several DNS resolvers
allow using DNS queries to aggregate user data for advertising.

I'm not surprised by the authors' classification of Google's DNS
resolver. In my view, it is plausible that an advertising company
running a large number of DNS resolvers as a free service is using the
collected user data in the context of behavioral advertising.

> What does it mean to have an account with a DNS resolver?
> I'm not saying that there are no issues, but this paper is not a
> useful discussion of them.

I agree that an empirical measurement substantiating the use of prior
DNS queries within behavioral advertising is certainly a stronger
indication for the mentioned privacy issue than a description in the
privacy policy. However, I'm not aware of prior empirical research work
in that direction.

In total, resolver-less DNS does still send a possibly large number of
fallback DNS queries to the preconfigured DNS resolver. Following this,
the main argument to support resolver-less DNS is a faster connection
establishment on the web as it saves the delay caused by the traditional
DNS lookups. The described privacy enhancement via resolver-less DNS is
just a positive side-effect.

Erik

[1] https://scholar.google.de/citations?user=emHBI1MAAAAJ
[2] https://scholar.google.de/citations?user=PROAnvEAAAAJ