Re: Getting to consensus on packet number encryption

[Spencer Dawkins at IETF <[e-mail]>]

Speaking as a curious observer with no hat on ...

On Tue, May 1, 2018 at 1:59 PM, Praveen Balasubramanian <
pravb=[e-mail]> wrote:

> Yeah this doesn’t have to be unilateral and is a negotiation. Either side
> is free to terminate the handshake and fall back to TCP if it doesn’t want
> to support the transforms being offered in the negotiation. This works
> perfectly fine for the intra-DC use cases.
>

So, please help me understand whether I'm tracking this.

Most of the time, QUIC connections are expected to traverse something like
the public Internet, so a bunch of participants want PNE as an
anti-ossification thing in this case.

Sometimes, QUIC connections would traverse only a controlled network path
(which I THINK is the point of the conversation about datacenters), for
server-to-server connections. Some participants would like to avoid any
overhead they can avoid in those situations, and don't see any need for
PNE, only additional (single-digit percentage) costs, and would like to
negotiate not using PNE for those connections.

Is that about right, so far? Assuming so ...

Is it correct that those connections are using TCP today, and are seeing
sufficiently good performance with TCP to make additional overhead look
problematic?

Is it correct that if an implementation doesn't want to use QUIC with PNE
overhead, Plan A would be to refuse the QUIC connection and fall back to
TCP?

Thanks in advance for clues ...

Spencer