On Mon, Feb 4, 2019 at 2:34 AM Mikael Abrahamsson <[email protected]> wrote: > > On Mon, 4 Feb 2019, Dave Taht wrote: > > >> The problem with CAKE/FQ and background traffic is that it can't tell if > >> there is congestion or not, and things like LEDBAT can't backoff and try > >> to avoid causing congestion. So your previous email about allowing some > >> congestion to take place on LE would be good as then protocols that try to > >> avoid causing congestion would have a way to do so. > > > > I do not like that the standard allows for total starvation. I would > > prefer it had a minimum of 5%. > > I fully agree, and that would work even if all LE traffic was put into > single queue with 5% of total bw, and even if that had a 1 second FIFO > with tail drop.
We don't quite agree. "Every application has the right to one packet in the network" - John Nagle Cake adheres to that as closely as possible, yet still keeps latency low for all applications background, priority or best effort. fq_codel doesn't, which leads to shorter queue lengths when ECN is in heavy use. I wish cake would more aggressively drop ECN packets. Sadly cake's uptake is a bit slow, as it is a bit slow. > > -- > Mikael Abrahamsson email: [email protected] -- Dave Täht CTO, TekLibre, LLC http://www.teklibre.com Tel: 1-831-205-9740 _______________________________________________ Cake mailing list [email protected] https://lists.bufferbloat.net/listinfo/cake
