Re: [Ohttp] Centralization and deployment models

[Martin Thomson <[e-mail]>]

On Tue, Jun 15, 2021, at 22:35, Vittorio Bertola wrote:
> I think that we should consider the possible implications on Internet 
> centralization from the start, exactly like we need to consider 
> security and privacy impacts.

This I agree with.  We are collectively still grappling with just what it means to consider these issues, so maybe a little more deliberate effort is needed in new efforts to engage with the topic.  I just don't know how to turn that into something actionable.

It's not through lack of trying either.  The IAB spent a lot of time on the subject in the 4 years I was there and have done since.  We have no shared understanding of the subject and so no concrete advice to give.  We just have a general sense that consolidation is bad.  Maybe that is limited to when centralization leads to single points of failure or control.

We don't know what leads to centralization and - in particular - what protocol designs might to do influence it.  I've heard it said that DoH has worsened centralization, but I don't think that question is settled either.  What is natural forces of market pressure and what the protocol enabled are hard to disentangle.

For DoH, I don't know if we could have done anything different then, even with this foreknowledge, except to the extent that we would have tried harder to engage with those who might be affected.

And just to make it clear, I don't think that the parallels with DoH and this work that are being drawn are quite right either.  Part of the reason I am engaging with this work is that it helps a major privacy problem with DNS centralization problem: the aggregation of profiles at DNS resolvers.

All that said, I'm no richer in terms of knowing what to write in a charter on the subject.

> Again, this depends on deployment models. Would OHTTP proxies be 
> anonymously available to the general public, or would they require user 
> authentication before use, or be restricted to traffic from specific 
> networks, or specific devices, etc? 

No deployment model that has been suggested to me involves any anonymous, general purpose, and open proxy.  Indeed, as I mentioned to Eliot, the protocol as proposed cannot be used for general purpose use cases.  You might define some sort of extension for that purpose, of course.

> [...] a new standard should include deployment recommendations to 
> prevent the worst scenarios from happening, or explanations of why 
> those scenarios are unlikely or unattractive for abusers.

This is a reasonable request; FWIW, I've opened an issue on this.  We have some of this text already, but maybe not as comprehensive as you might like.