I am using RouterOS as a transparent bandwidth manager (via PCQ) over two bridged nics on a x86-based system. At peak load, the CPU runs around 10% pushing a few thousand PPS at around 60Mbps.
We are seeing dropped packets and periods of high latency, even for traffic the bandwidth manager should be prioritizing. Then there is this very strange issue . . . sometimes, when users try to download a file, the download gets halted part-way through and they cannot finish. However web pages, streaming media, etc . . . seem to work fine.
The first thing I did was disable all of the queues, there was no change. Next, I rebooted the system and the problems went away . . . for a while, and then they returned. Then I replaced the NICs, no change. So I replaced the motherboard, no change.
Has anyone ever run into anything like this? I have no idea where to go next. The strange thing is that even disabling the queues doesn't help. I have to reboot the system, it works for a while, and then the same exact problems return.