StarWhiz / NanoPi-R6S-CPU-Optimization-for-Gigabit-SQM

NanoPi R6S CPU Optimization and Other Notes! Go from 800 Mbps cake SQM to 1400+ Mbps cake SQM!
13 stars 1 forks source link

Asymmetrical Version? #2

Closed chase-miller closed 6 months ago

chase-miller commented 8 months ago

Thank you for sharing!

Why is there a different version for asymmetrical connections? My very rudimentary understanding of this is that we’re essentially instructing something (cake/sqm/some other service) to use the faster cores. What role does the symmetry of the up/downlink have to do with this?

A related question: we would still specify the link speed (via the GUI) for sqm after making these modifications, right?

Thanks again!

chase-miller commented 8 months ago

I just noticed https://forum.openwrt.org/t/nanopi-r6s-with-openwrt/167611/91. Is the idea to give more horsepower (faster cores) to the downlink and less to the uplink?

Now I think to optimize for my assymetrical connection I should focus the big cores on ingress queues and put slow cores on all egress queues.

StarWhiz commented 6 months ago

Hi @chase-miller , I'm not that very smart. I just played around with the affinities and queues and found it worked best for my assymetrical connection.

But to answer your question, Yes. As you said in the second comment that was what I was thinking, to give more horsepower to downlink and less to the uplink.

As for your other question, yes you would still specify the link speed (via the GUI) for sqm after making these modifications.

I think multi-core design is in the works in the SQM space https://docs.google.com/document/d/1HxIU_TEBI6xG9jRHlr8rzyyxFEN43zMcJXUFlRuhiUI/edit?pli=1

But since that isn't available playing around with the affinity and queues is all I can do.