UBNT-stig wrote:
Lochnair wrote:
dtaht2 wrote:stig: the portion that I'd hoped be handled properly was GRO peeling on the interface with cake on it.
are you saying that enabling qos on one interface disables all offloads on the others?
If I remember correctly enabling QoS on an interface disables offloads only on that interface.
Yes, it does not disable it across the board. If a packet goes through the QoS path then the skb get marked that it's not a candiate for offload. So if you're only shaping on the WAN output then the download should still get offloaded.
Just did some quick test after you broke the news (and thanks). This is not completely true. The CPU utilization tells me some incoming SKB's still go through QoS path but I did see a drop. Could still put to good use by end users though.
Sorry for the little digression.
