Re: [dispatch] [Sframe] Dispatch of SFrame

On Tue, Jun 16, 2020 at 7:25 PM Victor Vasiliev <vasilvv@google.com> wrote:

> I imagine that for high-bitrate traffic, the difference between encrypting
> every packet and encrypting an entire frame at once is substantial enough
> to make using asymmetric crypto feasible in practice.
>

A few points:

1. The current SFrame draft does not in fact sign every frame. Rather it
signs the past N frames (
https://tools.ietf.org/html/draft-omara-sframe-00#section-4.4) "Adding a
digital signature to each encrypted frame will be an overkill, instead we
propose adding signature over multiple frames."
2. This is maybe sorta true for video in which the frames are large (though
see below) but remember that audio frames are small and have to have low
latency.
3. For video, it's true that I-Frames are enormous but the bulk of the
frames are P-Frames and those would generally be on the order of 1 packet,
at which point you're not getting much improvement.

IOW, it's complicated, but if you're not a fan of the "sign every packet"
strategy, you probably are going to find that "sign every frame" doesn't
work super well either.

-Ekr

> On Tue, Jun 16, 2020 at 10:53 AM Eric Rescorla <ekr@rtfm.com> wrote:
>
>>
>>
>> On Tue, Jun 16, 2020 at 7:27 AM Bernard Aboba <bernard.aboba@gmail..com
>> <bernard.aboba@gmail.com>> wrote:
>>
>>> On Jun 16, 2020, at 6:53 AM, Eric Rescorla <ekr@rtfm.com> wrote:
>>> >
>>> > Yes, I understand that the wire encoding supports signatures, but in
>>> the discussions I've had (including with Emac) I don't think that people
>>> believe that the latency/bandwidth/computation tradeoff is viable.
>>>
>>> [BA] Depends on the scenario.  We are in a pandemic where conferences
>>> are being used for all kinds of things that we haven’t seen before. For
>>> example, consider a situation in which participants are answering a binding
>>> poll and the responses (voice or data) need to be authenticated. SFrame can
>>> handle that. PERC cannot.
>>
>>
>> I don't really think framing this as "PERC vs. SFrame" is that helpful,
>> but there's no in principle reason PERC couldn't have signatures, though of
>> course one would need to define new algorithms.
>>
>> -Ekr
>>
>> _______________________________________________
>> dispatch mailing list
>> dispatch@ietf.org
>> https://www.ietf.org/mailman/listinfo/dispatch
>>
>