I have few CCR112 which is equipped with the QCA-8511 and I’m trying to determine whether or not VLAN translations can be/are performed in hardware.
According to the wiki page at https://wiki.mikrotik.com/wiki/Manual:CRS_features#Bridge_Hardware_Offloading for the CRS1xx devices the features in the switch menu do not break hardware offloading. I’m not sure how to interpret that. Either it only means using these features still allows packets to travel between post-6.41 bridge ports without hitting the cpu. Or it means that the features from the switch menu are actually handled in hardware.
I’ve noticed that if I use the ‘old’ software-based bridge with VLAN interfaces created on the etherXX ports my CPU is fully taxed before I can saturate the wire speed on my interfaces, not going above 350mbps on a single port.
Changing the config to make use of the new VLAN-aware bridges requires enabling vlan-filtering=yes on said bridge which also breaks hw offloading, again causing full cpu load before wire speed is reached.
When I do my vlan translations using /interface ethernet switch {vlan, egress-vlan-tag, egress-vlan-translation, ingress-vlan-translation} menus, I actually do manage to saturate my wire but while doing so, the CPU is heavily loaded. This either means these switch features are much more efficient that vlan-filtering=yes on the new bridge (while reaching a similar end result), or that there is some hardware offloading involved in this. Which is it?