I see that the default option for Packet Steering on my MT6000 is "Disabled" and I want to check for this specific device, is it recommended to leave this disabled?
Read the instructional text next to the checkbox?
1 Like
You may enable it if you are fine with the disadvantages.
https://openwrt.org/toh/gl.inet/gl-mt6000#hardware_acceleration
Packet Steering has been discussed at length in other threads. When Google submittted RPS to Linux many years ago it had some significant gains to performance. There are currently 3 configuration scenarios in OpenWrt, and are all independant of each other:
-
Software flow offloading (SFO) - does packet steering benefit here? Probably yes.
-
Hardware flow offloading (HFO) - (supported by MediaTek SoC devices), fully offloads packets so QoS/SQM can't be used. Does packet steering benefit here? Not really.
-
SQM - used to mitigate bufferbloat, but incompatible with SFO/HFO. Does packet steering benefit here? Well, maybe.
All 3 of these will have different RPS benefits, plus there are single core vs. multicore packet steering options added to OpenWrt. More options are good, but I personally have not observed any measurable benefit for it has a HFO user on this device.
@ trendy RPS is independent of HFO/WED from your link.
5 Likes
So, if I aim to get maximum throughput WHILE using SQM, I should enable packet steering as that would distribute the load to all CPU cores?
I remember reading somewhere in these forum posts that CAKE is single threaded, so I am wondering how packet steering would benefit such a case
Most QoS algorithms are single-core. exceptions are hfsc, codel, ?fifo
Probably start with irqbalance (enable after install) which will run different device interrupts on different cores.
2 Likes
I suggest running test with it on and off, it won't do much with SQM cake but it may help distribute other loads like wifi or whatever else. Just run htop and waveform speed test with it both ways.
1 Like