Re: [Cake] man page bikeshed

[Dave Taht <dave.taht@gmail.com>]

other stuff.

* gso-split no-gso-split not documented (what section do they belong in?)
* ingress/egress keywords not documented

I would like to bring out more that cake is a shaper that can also be
left unconfigured and run at line rate.

I honestly *didn't know* everybody else here wasn't using it that
way... it's my default qdisc on everything with spare cpu! dogfood it
via: sysctl -w net.core.default_qdisc=cake

"       CAKE  (Common  Applications  Kept  Enhanced) is a shaping-capable queue
        discipline which uses both AQM and FQ."

is a bit weak.

"       CAKE  uses  a  deficit-mode  shaper, which does not exhibit the initial
       burst typical of token-bucket shapers.   It  will  automatically  burst
       precisely  as  much  as required to maintain the configured throughput.
       As such, it is very straightforward to configure."

I have fiddled with the metro setting for local lans. at a gbit, the
results are "interesting" as a local qdisc,
at 100 flows, we end up with full throughput, 70k of buffering, 40k of
bql, tons of marks, bunch of drops in the
switch (which is like 192k of buffer there), and short rtts and lots
of retransmits. cwnd is capped too much.

(the sch_fq alternative is like 1.7MB in size and I haven't tried to
compare the two)

Or maybe that was a 500 flow result. Can't remember just now.

On Mon, Jul 30, 2018 at 10:23 AM Dave Taht <dave.taht@gmail.com> wrote:
>
> I would rather like the man page (one day) to include two realistic
> examples - (say a week's uptime), taken while loaded,  one example
> outbound, one example inbound, rather than the unloaded one it
> currently has and the less than massively loaded one...
>
> I loved looking over this result (at 40 users!) today.
>
> https://forum.lede-project.org/t/sqm-cake-and-piece-of-cake-qos-high-cpu-usage/17794/2



-- 

Dave Täht
CEO, TekLibre, LLC
http://www.teklibre.com
Tel: 1-669-226-2619