We just upgraded one of our busier ER-POE to v1.9.0 and added a single IP for QoS management using the Basic Queue. Set customer to 15m d/l and 5m u/l. I've tried both FQ_CODEL and SFQ queueing methods with identical results.
The results are such that as traffic picked up through the day, the latency went increasingly higher until this afternoon when we were pass 250-300mbs throught he router, the CPU load was stuck over 4 primarily consumed but the ksoftirqd/0 (but not /1) and the ubnt-util processes. Latency through the router to any host was in the 700ms range.
Deleting the queue had no effect and the only solution was to delete the Traffic Control from the Config Tree.
Is this just a case of not enough horsepower in this router to do traffic control at high throughput or is there a bug here somwhere. I will be implementing this on another router that carries a bit less traffic tomorrow and will compare the results.