Speeddrop after enabling SQM on Netgear R7800


didn't want to open a new thread. I read the whole thing while trying to use SQM and optimizing my speed on my Netgear R7800 and DOCSIS 3.0 cable modem.

Without SQM I get about 130Mbit/s down and 6.3Mbit/s up speed. After that I enabled SQM I played around with the settings described in the wiki and this thread. However, I'm not getting over 90Mbit/s download and 5Mbit/s upload. Any ideas?
SQM_off SQM_on

Try to log into the router while running a speedtest and concurrently run:
top -d 1

and closely watch the %idle in the first or second row, if this reaches zero (or close to zero, say 5%) often enough it would be an indicator that your router runs out of CPU. Note that for for dualcore routers the critical number is 50% as %idle covers all CPUs...

Which version of openwrt are you running, master or something older?

Thank for the hint. %idle drops to about avg 35% download and >90% for upload. So seems to be ok?

I'm using the lastest stable release OpenWrt 18.06.1 r7258.

This might indicate problems, as the r7800 uses a dual-core SoC.

Hmmm I don't understand. Didn't you say, the critical value is <5%?

Or what do you mean by "Note that for for dualcore routers the critical number is 50% as %idle covers all CPUs..."?

Exactly,bsince the r7800 has two CPUs anything below 55% might indicate overload. I say might as with 35% idle each CPU might run with say 17.5% idle, which would still be okay, or rather not a smoking gun. But I consider it to he more likely that one CPU runs at close to 0% and the other close to 70%...

Thanks for the explanaition. With htop I see, that one core is at 98% and the other one at about 30%. So that means, that SQM is not usable with speeds over 90MBit/s for me? Or can I optimize somehow the CPU scheduler or something?

Look for discussion on r7800 and IRQ balancing. By default pretty much everything gets handled by core0 and core1 is idle.

There is discussion about irqbalance and splitting eth0 and eth1 to different cores, and the same on wlans.

I believe that:
might be relevant.