I haven't really tested network speeds lately, as my own connection is too slow, only 100+ Mbit/s
But in late 2016 (when I got R7800) there was discussion about HTB qdisc causing SQM simple.qos to miscalculate some delays for dual-core CPUs (like in R7800). Discussion starts at
But like I mention in the discussion, those were then mitigated in the upstream Linux.
It was also discussed with SQM & cake authors in https://github.com/tohojo/sqm-scripts/issues/48
But at that time, cake performed better than simple/fq_codel.
My input to the discussion is that
- it is quite possible that there is some regression.
- it might be also in the upstream Linux kernel.
- it might also affect just certain devices like single/dual-core CPUs ARM/MIPS/X86, so debugging can be difficult
- it might not be the "main" qdisc like cake or fw_codel, but the helpers like HTB (or TBF)
Hopefully those earlier discussions from 2016-2017 help in debugging.