[Rpm] [Make-wifi-fast] tack - reducing acks on wlans

[Keith Winstein]

Hi Michael -- glad we are mostly of like mind! Let me try to explain where
I was coming from with the super-delayed E2E ACKs below...

On Thu, Oct 21, 2021 at 1:42 AM Michael Welzl <michawe at ifi.uio.no> wrote:

> Hi Keith !
>
>
>    - Protocol 1: The endpoints can now decide what to do with this extra
>    info. For example, they could decide that the endpoint receiver should
>    switch to super delayed ACKs itself (maybe one every N seconds or M
>    megabytes), and the sender will trust the Protocol 2 ACKs for
>    congestion-control and retransmission purposes, while still using the
>    Protocol 1 ACKs for ultimate reliability. (I.e. the sender won't discard
>    outstanding data from a reliable stream until it's been ACKed by the
>    endpoint.)
>
> Not sure I see the point of super delayed ACKs, but …. either way, I think
> we agree on “The endpoints can now decide what to do with this extra info.”
>  There can be many variations.
>

 My thinking here was that ACKs serve at least three different purposes in
an ARQ-style transport protocol:

   1. congestion control
   2. loss detection and triggering retransmission of previously sent data
   (via timeout, dupacks, sender-side timestamp schemes (Sprout/RACK), etc.)
   3. guaranteeing the abstraction of a reliable byte stream, by making the
   sender retain a copy of the outstanding data (ready for possible
   retransmission) until ACKed by the receiver

I think most endpoints might be willing to delegate #1 and #2 to an
untrusted intermediary that's along the network path and eager to help
(it's no worse than a transparent TCP accelerator today), especially if
there's some ability for the endpoints to double-check the intermediary's
work and say, "you know what, no thanks" if unsatisfied.

But #3 feels different -- if an untrusted intermediary's false ACK causes
the sender to delete its copy of data that never makes it to the receiver,
that threatens the core service model of the protocol. It's not possible to
recover from that without escalating to the application. So the sender
probably shouldn't discard data irreversibly until it hears a
cryptographically authentic ACK from the E2E receiver, but that could come
much later and less frequently.

-Keith
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.bufferbloat.net/pipermail/rpm/attachments/20211021/44679665/attachment.html>