Re: [Pearg] About hiding in crowds

[Shivan Sahib <[e-mail]>]

Thanks for the note Christian. I remember that a similar point was brought
up in the *"What can you learn from an IP?" *paper presented by Nikita
Borisov at ANRW 2019 last year:
https://dl.acm.org/doi/10.1145/3340301.3341133, video of presentation:
https://youtu.be/pEnT2BJvyv0?t=106
In the discussion after, it was brought up that IPv6 could make things
better (more to choose from, so less stickyness if servers wanted) or worse
(less incentive to have the same IP address for multiple services). There
was also some discussion around what

   - the CDNs can do
   - the websites can do, and
   - clients can do

Sara and I were chatting today and we reckoned this was an architectural
consideration (with research input) for Internet privacy. It seemed to us
that a first step might be to write up the concern and possible solutions
in an Internet Draft.

Thanks also for the blog post
<https://huitema.wordpress.com/2020/08/09/can-internet-services-hide-in-crowds/>;
I tried checking out https://github.com/private-octopus/centraldns but got
a 404.

On Mon, Aug 10, 2020 at 4:25 PM Christian Huitema <[e-mail]>
wrote:

> A lot of the privacy extensions recently developed amount to "hiding in
> crowds". For example, SNI encryption assumes that multiple servers are
> accessible through the same IP address. If the SNI is hidden, outside
> observers won't know which one was accessed. DNS encryption makes the
> same assumption in an indirect way. It assumes that we gain privacy by
> hiding the DNS exchange that maps www.example.com to an IP address. This
> is fine, except for the fact that most servers have their own IP
> address. You can hide the DNS exchange, you can hide the SNI, but
> outside observers will still be able to understand which servers you are
> accessing by simply looking at the address header. If we want real
> privacy, we will need something else!
>
> How do I know? I started with the Majestic Million list of domain names,
> and resolved 25,000 of these names, and found out that on average a
> given IP address was shared by about 1.21 names, as explained in:
>
> https://huitema.wordpress.com/2020/08/09/can-internet-services-hide-in-crowds/
> ).
> And then I resolved the next 25000 names to be more sure of the results.
> The average increased slightly, from 1.21 to 1.22, which does not change
> the results much. 74.6% of domains use an address that is unique to
> them, 8.7% use an address shared by 2 domains, and only 8% use an
> address shared by 10 or more servers. DNS encryption and SNI encryption
> do bring privacy for a minority of connection, for which it may well be
> important. But they do not improve privacy in 75% of the cases.
>
> I understand that privacy-warriors can use VPN, proxies or Tor. But
> these tools are far from perfect -- see the recent Sybil attacks against
> Tor, or the outveiling of shady business practices by many VPNs. In any
> case, these tools at best provide "privacy for a few active users". But
> that leaves aside the bulk of Internet users. Thus my question for this
> program: how would we provide privacy for the masses?
>
> -- Christian Huitema
>
>
> --
> Pearg mailing list
> [e-mail]
> https://www.irtf.org/mailman/listinfo/pearg
>