[Cake] [Rpm] [Make-wifi-fast] The most wonderful video ever about bufferbloat

[Sebastian Moeller]

Hi Dave,


> On Oct 26, 2022, at 22:42, Dave Taht <dave.taht at gmail.com> wrote:
> 
> I loved paced chirping.

	[SM] Yes it sounded like a clever idea (however I would prefer a clearer signal from the network about queue-filling). However I have heard precious little about parced-chirping actually working in the real internet, which IMHO means the following questions are still open:
a) does it actually work under any realistic conditions?
b) under which condition will it fail?
c) how likely are these conditions over the existing internet?

IIRC it uses packet spacing to deduce whether capacity has been reached, and since packet spacing is a known unreliable source of information it needs to average and aggregate to make up for operating based on questionable data. IMHO it would be preferable to solve the "questionable data" problem as my gut feeling is with better data would come simpler solutions to the same challenge.
Or I am simply misremembering the whole thing and be barking up the wrong tree ;)

> I also loved packet subwindows.

That was allowing "rates" below one packet per RTT?

> I wish we could all agree to get
> cracking on working on those two things for cubic and reno rather than
> whinging all the time about the stuff we will never agree on.

;). You might have seen some of my code, which should indicate that maybe I am not going to be all that helpful in the "get cracking" department ;)


Regards
	Sebastian