As far as I understand, Css326 and crs326 switches have the same hardware, the only difference is in the installed software. I wonder if queue trees (especially mangled by dscp/priority tagging) are hardware accelerated in crs326 or not. (Or maybe it is not an issue, because CPU requirements for queues are low?)
The Wiki says: feature is not available, meaning that L3 Hardware offloading MUST be disabled for these features to work Maybe it will become avaiable in later versions of ROS 7.
Source: https://wiki.mikrotik.com/wiki/Manual:CRS3xx_series_switches#L3_Hardware_Offloading
Currently: https://wiki.mikrotik.com/wiki/Manual:CRS3xx_series_switches#Traffic_Shaping
This is interesting. So there is no stable CRS switch that could handle QoS (other than global egress/ingress rate limiting) by the hardware. I think that without hw offloading, the relatively slow CPU is not able to handle queues when it matters (e.g. when there is congestion on a 1 gig link). When there is no congestion, queues are not needed anyway. Sooooo if I really need to prioritize traffic, I guess I should not use CRS.
Well except CRS317-1G-16S+ 