[Bloat] How about a topical LWN article on demonstrating the real-world goodness of CAKE?

[Tom Henderson]

On 8/10/20 8:57 AM, Jonathan Morton wrote:
>> The current best practice seems to be to instantiate cake/SQM on a reasonably fixed rate wan link and select WiFi cards/socs that offer decent airtime fairness.
>> Works pretty well in practice...
> Yes, AQL does essentially the right thing here, again along the lines of limiting the influence of one machine's load on another's performance, and completely automatically since it has faurly direct information and control over the relevant hardware.  Cake is designed to deal with wired links where the capacity doesn't change much, but the true bottleneck is typically not at the device exerting control.
>
> On that note, there is a common wrinkle whereby the bottleneck may shift between the private last mile link and some shared backhaul in the ISP at different times of day and/or days of week.  Locally I've seen it vary between 20M (small hours, weekday) and 1Mbps (weekend evening).  When Cake is configured for one case but the situation is different, the results are obviously suboptimal.  I'm actually now trying a different ISP to see if they do better in the evenings.
>
> Evenroute's product includes automatic detection of and scheduling for this case, assuming that it follows a consistent pattern over a weekly period.  Once set up, it is essentially a cronjob adjusting Cake's parameters dynamically, so providing a manual setup for the general OpenWRT community should be feasible.  On “tc qdisc change”, Cake usually doesn't drop any packets, so parameters can be changed frequently if you have a reason for it.

Thanks for your insights, they have been helpful.  I also just found 
these references:

http://flent-newark.bufferbloat.net/~d/Airtime%20based%20queue%20limit%20for%20FQ_CoDel%20in%20wireless%20interface.pdf

https://blog.tohojo.dk/slides/llc19-linux-wifi-bloat.pdf

- Tom