lynxthecat / cake-autorate

Eliminate the excess latency and jitter terrorizing your 4G, 5G, LTE, Starlink or other variable rate connection!
https://www.bufferbloat.net
GNU General Public License v2.0
268 stars 24 forks source link

Field Report for cake-autorate 3.0.1 on D-Link DIR-878 #238

Open richb-hanover opened 1 year ago

richb-hanover commented 1 year ago

Using D-Link DIR-878 with OpenWrt 22.03.5; generating traffic from Rasperry Pi 4b connected via Ethernet. My ISP connection is 75/75 mbps. I used the idle value from top for most measurements because htop doesn't summarize the CPU usage across multiple cores. (Higher "idle" percent is better.)

a) Run out of box; lots of logging; htop shows CPU runs 15-25% when idle; passing traffic goes to 80-85% (I was probably only observing one of the cores, though. Nonetheless, lots of spare CPU.)

b) I turned off logging (output_processing_stats, output_load_stats, output_reflector_stats, debug set to zero; output_cake_changes set to 1). With logging off, cake-autorate & SQM running, top shows idle percent of about 80% with no traffic; 38-45% with speedtest running. htop showed CPU 12-22% (slightly lower) when idle; 60-75% with traffic. (Again, probably only recording one core, but plenty of spare CPU)

c) Turn off cake-autorate (leaving SQM running); top shows 85% idle percent no traffic; ~50% idle with traffic

d) Turn off SQM; top shows 99% idle percent when no traffic; 79% idle with traffic

TL;DR The multi-core DIR-878 has plenty of CPU to handle a 75/75 mbps ISP connection with cake-autorate installed. I have left cake-autorate running as a service.

rany2 commented 1 year ago

That's great news for a 880Mhz mips mt7621 CPU. Thanks!

lynxthecat commented 1 year ago

@richb-hanover many thanks for this report!

@moeller0 wrote about the MT7621A here.

Actually my outdoor 5G router - Zyxel NR7101 has something called the MT7621AT.

Looks like the MT7621AT is a variant of the MT7621A.

moeller0 commented 1 year ago

. I used the idle value from top for most measurements because htop doesn't summarize the CPU usage across multiple cores. (Higher "idle" percent is better.)

With htop configure it to show all CPUs and then look at the idle percentsge of the CPU with the highest load (or simply pick the lowest idle number reported), once the cake cpu isfully loaded you reached the limit independent of the load on the other CPUs....

richb-hanover commented 1 year ago

With htop configure it to show all CPUs and then look at the idle percentsge of the CPU with the highest load (or simply pick the lowest idle number reported), once the cake cpu isfully loaded you reached the limit independent of the load on the other CPUs....

Cool tip. My htop configuration simply shows (what I believe is) utilization on each of the cores.

Thanks.

image
moeller0 commented 1 year ago

Not strictly better, but I typically enable the detailed CPU graph that shows software interrupt processing separately, I also add an additoonal textual display for each CPU. However currently I am travelling without a real computer and can not test how to configure things 'my way' nor create/post a screenshot I think...

moeller0 commented 1 year ago

Not strictly better, but I typically enable the detailed CPU graph that shows software interrupt processing separately (F2, Display Options , Detailed CPU time), I also add an additonal textual display for each CPU (F2, Meters, -ECANNOTTEST). However currently I am travelling without a real computer and can not test how to configure things 'my way' nor take htop screenshot.