Hi I'm just wondering if this could still be fixed by software update or it's just a hardware limitation already?
I currently have a Newifi D2 with MT7621 chipset with latest stable build OpenWrt 22.03.2. I have a 250Mbps connection both download and upload and no matter how I tweak the sqm setting I could only achieve stability when I set both the rate to 145Mbps, more than that then ping spikes. I know using cake is CPU intensive and I can only achieved 100Mbps atmost but for the fq_codel + simplest_tbf + software offloading enabled, ping is stable when rate is set atmost to 145Mbps. I also noticed that when I set my sqm past 150Mbps like 185Mbps, then using speedtest while using top command, top command freezes and only goes to normal after the test. While speedtest and using the htop, I could see that there's still a headroom on a cpu so this is where I am confused, why I could not go past the 145Mbps if there's still a headroom in both core?
What makes you believe there would be a "performance issue" here?
With sqm/cake enabled, any kind of offloading is effectively lost - so the real performance of the SOC matters. mt7621 however is a 2*880 mips 1004KEc SOC (network tasks are inherently single threaded, so the second core can't help directly) - which isn't that fast on the CPU side, but profits massively from very well supported offloading- and packet engines (which, as we found out before, lose their mojo with sqm).
Traffic shaping 250 MBit/s is a quite considerable task for classic router platforms, single core ath79 (mips 74Kc) running at 720 MHz maxes out around 70 MBit/s (+/- 10-15 MBit/s, depending on the details), ipq8065 with 2*1.7 GHz ARMv7/ KRAIT300/ ~cortex A15 around 170 MBit/s, so mt7621a's capabilities seem to be within expectations (actually, I wouldn't expect it to cope with 145 MBit/s in the first place). You need considerably faster hardware for your requirements (mvebu[0], mt7622 or x86_64 should play ball, for that and more).
--
[0] beware of mwlwifi's issue, which more or less disqualify this target, if wireless is part of the requirements.
I see, that makes sense. So I better upgrade for a better hardware then. By the way I'm currently using fq_codel + simplest_tbf + software offloading and rate set to 145Mbps, so far it's stable and could attain somewhere 121Mbps when I do some speedtest. As for using cake, I can only attain less than 100Mbps. Anway, Thank you very much for the clarifications.
As mentioned before, 145 MBit/s felt (too-) high to me, compared to the abilities of the relatively similar ath79 SOCs running between 560 and 750 MHz (mt7621 has better offloading, but without that (sqm active), they're rather comparable); fq_codel + simplest_tbf obviously changes the situation somewhat (software offloading less so).
