Re: [Ppm] Interim topic: Shall DAP support heavy hitters?

[Christopher Patton <[e-mail]>]

Hi everyone,

I've prototyped most of the changes for Proposal #1 in Daphne:
https://github.com/cloudflare/daphne/pull/575

Lots of this would be reusable for #2. I hope this gives folks a sense of
the scope of changes required for supporting heavy hitters.

Chris P.

On Mon, Apr 8, 2024 at 4:18 PM Christopher Patton <[e-mail]>
wrote:

> Hi folks,
>
> As noted at IETF 118, there are a number of issues for the DAP spec
> related to supporting heavy hitters via Poplar1:
>
> https://datatracker.ietf.org/meeting/118/materials/slides-118-ppm-dap-open-issues
>
> We'd like to spend a significant chunk of time during the interim on
> resolving these issues. To that end, Tim G. and I have been leading a small
> design team over the last couple of months towards identifying a set
> protocol changes to support this important use case. We need your help
> deciding between three different options:
>
> - Proposal #0: Don't support heavy hitters in DAP (punt to a future
> protocol spec)
> - Proposal #1: Support heavy hitters with minimal changes to DAP
> - Proposal #2: Support heavy hitters, but amend the collection
> sub-protocol to minimize the amount of information revealed to the Collector
>
> Each proposal is detailed in the following google doc:
>
> https://docs.google.com/document/d/1ZjXz-1kGsTDf2Vn2u-fwYqR8BSc3tOYlELVHAYvAfjk/edit?usp=sharing
>
> It would be helpful if folks could spend some time ahead of the interim
> (22 April, two weeks from now) digesting these options, providing feedback,
> and figuring out what information you need, if any, to form an opinion. My
> hope is that we'll be prepared to make a decision during the interim on the
> questions enumerated in Section 6.
>
> Relatedly: If we decide to support heavy hitters, then we'll also need to
> decide how to deal with a class of attacks against DAP/Poplar1 articulated
> by Simon Friedberger and Phillipp Schoppmann. We call these "steering
> attacks", as they exploit the fact that the attacker in our threat model
> has partial control of the protocol execution that is not considered the
> original Poplar paper (https://ia.cr/2021/017). There are a few
> mitigations in the google doc that we'd like folks to mull over.
>
> I'm looking forward to seeing folks online in a couple of weeks.
>
> Thanks,
> Chris P.
>