Re: Watermarking [Re: Campaign for position of chair and mandate to close this community group]

On Mon, Jan 13, 2014 at 3:49 PM, Danny O'Brien <danny@eff.org> wrote:

> On Fri, Jan 10, 2014 at 10:06:50AM -0800, Mark Watson wrote:
> >    On Fri, Jan 10, 2014 at 9:06 AM, Olivier Thereaux
> >    <olivier.thereaux@bbc.co.uk> wrote:
> >
> >      David, thanks for starting this discussion.
> >
> >      On 10 Jan 2014, at 16:56, David Singer <singer@apple.com> wrote:
> >      >
> >      > I am only aware of two uses of watermarking in copy protection:
> >      >
> >      > 1) Define the system such that all players must detect watermarks
> and
> >      refuse to play unlicensed content. [snip] So, this approach neither
> >      meets your goals nor works well.
> >      [*]
> >      > 2) Use watermarks for forensics. Every copy delivered is
> watermarked
> >      in some way.
> >
> >      Generally used for online music sales these days if I*m not
> mistaken. I
> >      also recall seeing it recommended by professional photographers
> >      associations (such as the ASMP).
> >
> >      I seem to remember earlier discussion in here about this type of
> >      watermarking, and there was significant push back on the idea (from
> >      MarkW?) based on the assumption that such mechanism would defeat any
> >      attempt at caching, and thus render an already hairy bandwidth issue
> >      even hairier.
> >
> >    What I said back then was two things:
> >    - in practice, because of the economics of content distribution,
> per-user
> >    or per-device forensic watermarking would probably be done at the
> client
> >    device and this approach does not avoid the need for
> non-user-modifiable
> >    client components
>
> I wonder if that's just a temporary constraint in a narrow set of
> markets. For instance, Apple uses (or used) simple email identification
> tagging on their MP3 downloads (does that still exist, David?), but they
> have the benefit of controlling their entire delivery system. I can
> imagine that third-party CDN services would be hard-put to place unique
> identifiers on current content and perhaps would have no current
> economic reason for doing so, but what technically is standing in their
> way?
>

CDN edge servers - whether a commercial CDN or single-provider one - are
optimized for massive storage and high throughput. They're generally
limited to the processing power needed to ship data from hard disk or SSD
to the network interface. You can see the design of the servers we use,
for example, here: https://signup.netflix.com/openconnect/hardware

Applying a watermark is a CPU intensive operation, so you're talking about
a wholesale change in the physical equipment deployed at the network edge:
you still need the same storage and network capacity, so the additional CPU
requirements will imply more power, cooling and probably footprint as well.

...Mark




>
> >    - using forensic watermarking to identify individuals brings with it a
> >    risk of mis-identification as a result of stolen credentials,
> identifiers
> >    that persist when a device is sold etc. The consequences of such
> >    mis-identification ought to be carefully considered.
>
> This is true, but there's definitely been some thought put into this. I
> think one of the elements to raise here is the dangers of any such
> identifiers being used to prevent the transit of material; you don't
> want to create fingerprints, and then find yourself mandated to monitor
> those fingerprints deep in the network. (This is David's first category
> of watermarking and the only one I've encountered when dealing with
> video works).
>
> d.
>
>