[Ecn-sane] [tsvwg] per-flow scheduling

[David P. Reed]

Please!
On Monday, June 24, 2019 3:31pm, "Jonathan Morton" <chromatix99 at gmail.com> said:



> > On 24 Jun, 2019, at 9:57 pm, David P. Reed <dpreed at deepplum.com>
> wrote:
> >
> > TCP doesn't have a "natural sawtooth" - that is the response of TCP to a
> particular "queueing discipline" in a particular kind of a router - it would
> respond differently (and does!) if the router were to drop packets randomly on a
> Poisson basis, for example. No sawtooth at all.
> 
> I challenge you to show me a Reno or CUBIC based connection's cwnd evolution that
> *doesn't* resemble a sawtooth, regardless of the congestion signalling employed. 
> And I will show you that it either has severe underutilisation of the link, or is
> using SCE signals. The sawtooth is characteristic of the AIMD congestion control
> algorithm.
 
Of course AIMD responds with a sawtooth to a router dropping occasional packets as congestion signalling. Are you trying to pretend I'm an idiot? And various kinds of SACK cause non-sawtooth responses.
 
My overall point here is that you seem to live in a world of academic-like purity - all TCP connections are essentially huge file transfers, where there are no delays on production or consumption of packets at the endpoint, there is no multiplexing or scheduling of processes in the endpoint operating systems, etc.
 
TCP sources don't work like that in practice.

> 
> - Jonathan Morton
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.bufferbloat.net/pipermail/ecn-sane/attachments/20190624/4f5e99d7/attachment-0001.html>