I have RB1100AHx2 that routes traffic from my customers to internet, preforms NAT and traffic shaping. I have three uplinks. Traffic shaping performs within queue tree and simple queues. In both cases I use pcq as queuing discipline. Also I must say that I have just 14 simple queues. Main traffic passing through queue tree. But recently I started to notice strange CPU utilization at peak hours.
As you see CPU usage for some reason couldn't get more than 70%. Summary traffic from internet to customers was nearly 110-120 Mbps and nearly 20-30 Mbps in reverse direction. I've noticed this statistic post factum so I couldn't say something about how it really worked at that time. But the next day gave me more info about this. Almost at the same time as day before I've noticed that I cannot reach internet from my office. Ping showed me total packets lost to any internet address but my uplinks were ok and my network also was ok. Then I tried to disable simple queues (as my office's traffic shaped there) and I've got things back to work. Ping became ok and summary traffic from internet to customers rose to 140-150 Mbps against 110-120 Mbps before. And even more:
That spike at CPU usage statistic I've got after disabling those queues. And one more thing, when I tried then to enable just one simple queue that even not actually was relevant to my office's traffic I've also got problem with connection to internet and summary reduction of traffic.
At now all traffic shaping performs within queue tree but I'm considered about bad performance issues with simple queues. Queue tree is good but not always convenient. I've tested it with ROS 5.21 and ROS 5.22. I cannot test it with latest ROS 6.X at production system.
Had someone the same issues or something relevant to this? Sorry for veeeery long post and thank you for your attention )
