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

["David P. Reed" <dpreed@deepplum.com>]

[-- Attachment #1: Type: text/plain, Size: 2873 bytes --]


It's even worse. The FCC got focused on max speeds back in the day as its only way to think about Internet Access service. And I was serving on the FCC Technological Advisory Committee and also in its Spectrum Policy Task Force, then later involved in the rather confused discussions of Network Neutrality, where again "speed" in the "up-to" sense was the sole framing of the discussion.
 
Because it was mostly lawyers and lobbyists (not network engineers), this focus on max speed as the sole measure of quality ended up with a huge distortion of the discussion, strongly encouraged by the lobbyists who love confusion.
 
That said, max speed plays a role at all time scales in minimizing response time, but queuing delay has no constituency, even though its impact is FAR worse in real situations.
 
If the FCC and regulators (or even the DoD communications management layers)  ever start talking about queueing delay in shared network services, I will die of shock.
 
But we did have one HUGE temporary success. The speed test at DSL Reports measures lag under load, and calls it bufferbloat, and gives a reasonably scaled score.
 
When I talk to people who are interested in quality of Internet service, I point them at DSL Reports' speed test.
 
That is a big win.  However, marketers of Internet access services don't compete to get good scores at DSL Reports. Even "business" providers provide crappy scores.
 
Comcast Business in the South Bay does very poorly on bufferbloat for its high speed business services, for example. This is based on my measurements at my company.  I know some very high executives there, and Comcast is the only real game in town for us, so I tried to get the folks in Philadelphia to talk to the local managers. Turns out the local managers just refused to listen to the headquarters execs. They saw no monetary benefit in fixing anything (going from DOCSIS 2 to DOCSIS 3.1 which had already been on the market for several years would have fixed it, probably).
 
On Thursday, June 27, 2019 4:33pm, "Brian E Carpenter" <brian.e.carpenter@gmail.com> said:



> On 27-Jun-19 19:49, Sebastian Moeller wrote:
> ...
> > a considerable fraction of home-users seem obsessed in maxing out their
> access links and compare achievable rates; whether such behaviour shoud be
> encouraged is a different question.
> 
> I think this is encouraged by, or is even a direct result of, so called "speed
> tests" for use by consumers (such as https://www.speedtest.net/), and the way
> connectivity "speed" has been used as a marketing tool. At least where I live,
> "speed" is the main marketing tool for switching users to fibre instead of copper.
> No doubt it will be used as the main marketing tool for 5G too.
> 
> It's almost as if those marketing people don't understand queueing theory.
> 
> Brian
> 
> 
> 

[-- Attachment #2: Type: text/html, Size: 4903 bytes --]