I use a Raspberry Pi 4 as my router and a TP-Link Archer C7v2 as a dumb switch and access point, connected via a LAN port. I used to run OpenWrt on both, but about a year or so ago I did some iperf3 testing and noticed that when the Archer C7 was running OpenWrt, IIRC it only got something like 200-300 Mbps over Ethernet between it and the Pi. With the TP-Link firmware, it got like 900 Mbps, so I switched to that.
There were some benefits when the Archer C7 ran OpenWrt: access to more 5 GHz channels here in the USA (helpful in my apartment with crowded Wi-Fi), WPA3 support, and probably better OS security.
If I were to put the Archer C7 back on OpenWrt and enable software or hardware flow offloading, would that improve Ethernet performance for dumb switch + AP mode, or is that strictly for use as a NAT router? I notice that in LuCI, flow offloading is in the firewall section, and as I recall, dumb switch + AP mode doesn't use the firewall.
Switching operates at how fast the switch can go. Which should be line rate full duplex?
I'm less sure about wireless AP but pretty sure it's at whatever the CPU can do on a software bridge through the CPU port to the switch. It would require some sort of offload/ acceleration in the wireless to cpu port of which I'm not familiar to what extent different platforms have for that.
IDK how fast QCA9558 can do on a CPU bridge nor if it's slower than what you can achieve on a 3x3 802.11ac AP.
No firewall/routing as you're doing layer 2 bridging. You still want to use the firewall to protect the AP itself but it's not being used in the data path.
Was this wired? sounds like you were doing nat? i.e. you had the wan port hooked up and it was routing / NAT? Was that using iperf3 onto the router itself, was that with it through the switch? AR8327 should be line rate.
It was wired, with the Pi connected to one of the Archer C7's LAN ports.
When I had OpenWrt on both, I ran iperf3 on the Pi and the Archer C7 and tested in both directions. Result: 200 or 300 Mbps.
When I had OpenWrt on the Pi and TP-Link firmware on the Archer C7, I ran iperf3 on the Pi and on my 14" MacBook Pro (M1 Pro) and tested in both directions. The laptop was connected to the Archer C7 via wired Ethernet. Result: About 900 Mbps.
Practical rate will be something like 1.3Gbps in input+output over CPU with all offloads, i.e not full gigabit 1Gbps+1Gbps.
iperf on archer c7 (or any other router for that matter) is not representative in any way of forwarding performance. You have to put iperf server on a wired gigabit machine, then you will get 1.3 if your c7 has eth1 and eth2 or 1.0 if only eth0 and about 700Mbps over wireless in+out , all transfers together not wildly exceeding 1.few gbps.
When you ran tplink firmware with discontinued 32bit NSS hardware offload it was better, no doubt.
iperf3 running on the Archer C7's CPU to the pi can't be compared to wired rate between pi and a macbook pro. CPU on the archer C7 isn't fast enough.
200-300Mbps iperf3 for a mips24kc @ 720MHz sounds about right.
If you're doing bridged AP the switch itself will do gigabit no?
So you're saying wireless in/out for QCA9558 is faster than what you are likely to achieve over the wireless interfaces which is good?
We're not talknig forwarding performance here though? OP says they want to use this thing as a bridged AP now?
I guess tired me was thinking forwarding as in routing, not layer 2 packet forwarding? =P
I didn't think we were talking NAT which you now mention. But I guess you mean flow offloading worked for nat, so it should work for wireless<-> wired offload too?
Never mind. It's not worth the bother trying to remove those packages. When I tried to remove firewall4 with firmware selector (by deleting it or putting a minus sign in front of it), it added it anyway, but I'm not low on space, so it doesn't matter. I'll just follow the bridged AP guide on the wiki as it says, but add the bridger package.