[Make-wifi-fast] a thought about acks

[Make-wifi-fast] a thought about acks

[Dave Taht]

http://www0.cs.ucl.ac.uk/staff/m.handley/papers/hack2014.pdf

-- 

Dave Täht
CTO, TekLibre, LLC
http://www.teklibre.com
Tel: 1-831-205-9740

Re: [Make-wifi-fast] a thought about acks

[David P. Reed]

[-- Attachment #1: Type: text/plain, Size: 1482 bytes --]


Interesting ideas, but these ideas don't belong in TCP.
 
It's perfectly OK to have acknowledgements at the MAC layer. After all, it affects a single contention domain, which is like a shared link, not an end-to-end connection.
 
TCP's flow and congestion control mechanisms are conceived entirely around the idea of a heterogeneous dynamically changing overlay graph.  The expectations of the layer below IP are simple: no buildup of latency, reasonably high reliability (this is what "best efforts" has always meant).
 
The idea of making TCP the "physical layer" protocol was NOT in the original design, on purpose! I was part of the design team working directly under Vint Cerf and Jon Postel when this was decided.
 
Instead, the idea was that the physical layer properties (including "medium acquisition time", better called "arbitration overhead") would NOT be visible at the end-to-end level.
 
So, I would just suggest that these experiments demonstrate how to think about making *WiFi* fast, and not how to make TCP special over WiFi.
 
On Friday, May 31, 2019 11:38am, "Dave Taht" <dave.taht@gmail.com> said:



> http://www0.cs.ucl.ac.uk/staff/m.handley/papers/hack2014.pdf
> 
> --
> 
> Dave Täht
> CTO, TekLibre, LLC
> http://www.teklibre.com
> Tel: 1-831-205-9740
> _______________________________________________
> Make-wifi-fast mailing list
> Make-wifi-fast@lists.bufferbloat.net
> https://lists.bufferbloat.net/listinfo/make-wifi-fast

[-- Attachment #2: Type: text/html, Size: 3055 bytes --]

Re: [Make-wifi-fast] a thought about acks

[Dave Taht]

On Fri, May 31, 2019 at 10:15 AM David P. Reed <dpreed@deepplum.com> wrote:
>
> Interesting ideas, but these ideas don't belong in TCP.
>
>
>
> It's perfectly OK to have acknowledgements at the MAC layer. After all, it affects a single contention domain, which is like a shared link, not an end-to-end connection.
>
>
>
> TCP's flow and congestion control mechanisms are conceived entirely around the idea of a heterogeneous dynamically changing overlay graph.  The expectations of the layer below IP are simple: no buildup of latency, reasonably high reliability (this is what "best efforts" has always meant).
>
>
>
> The idea of making TCP the "physical layer" protocol was NOT in the original design, on purpose! I was part of the design team working directly under Vint Cerf and Jon Postel when this was decided.
>
>
>
> Instead, the idea was that the physical layer properties (including "medium acquisition time", better called "arbitration overhead") would NOT be visible at the end-to-end level.
>
>
>
> So, I would just suggest that these experiments demonstrate how to think about making *WiFi* fast, and not how to make TCP special over WiFi.

Concur. This is just an outgrowth of thinking extremely hard about how
starlink might work (in the total lack of public
information) - for the last week or so, over here:
https://www.reddit.com/r/Starlink/comments/brn6gg/will_starlink_have_bufferbloat/

I'd run across much of mark's work before and he's been quite busy
trying to reverse engineer it - which is *quite fun* and
far more interesting than what I should be doing.

>
>
>
> On Friday, May 31, 2019 11:38am, "Dave Taht" <dave.taht@gmail.com> said:
>
> > http://www0.cs.ucl.ac.uk/staff/m.handley/papers/hack2014.pdf
> >
> > --
> >
> > Dave Täht
> > CTO, TekLibre, LLC
> > http://www.teklibre.com
> > Tel: 1-831-205-9740
> > _______________________________________________
> > Make-wifi-fast mailing list
> > Make-wifi-fast@lists.bufferbloat.net
> > https://lists.bufferbloat.net/listinfo/make-wifi-fast



-- 

Dave Täht
CTO, TekLibre, LLC
http://www.teklibre.com
Tel: 1-831-205-9740