AP1 (connected to the router) and AP2 form a mesh (acting as a wireless bridge) and AP2 and AP3 are connected by cable. The setup works as devices connected over wifi to AP3 can access the internet, but there is a performance loss I do not understand yet. I used iperf3 to analyse the 5G wifi performance between the nodes.
I am using AP1 as an iperf3 server. AP2 connecting to AP1 reaches about 400 Mbit/sec. AP3 to AP1 also gives about 400 Mbit/sec. A Notebook connecting to AP3 reaches 200Mbit/sec but the same Notebook connecting to AP1 just gives 50Mbit/sec.
iperf3 should not be run on a router/ AP, as it needs a lot of CPU performance (more than routing- or forwarding packets), distorting your results
repeater effect, each repeater roughly halves your throughput, as the same radio first needs to receive your data and then (afterwards) retransmit it again - so everything takes twice as long (and that happens with every hop in the chain)
I made a sketch to clarify the setup I am using (see below).
I tested the speed between AP2/AP1, AP3/AP1, AP2/Router, AP3/Router, AP3/AP2 for which the performance consistently shows to be good. Testing Notebook/AP3 and Notebook/AP2 provides good results also. But as soon the traffic passes the wireless bridge the performance drops significantly (Notebook/AP1 and Notebook/Router).
While iperf3 running on the routers could distort the results, this should be the case in all tests.
There should also be not repeater effect as I use a wireless bridge and no repeater.
For testing purposes I just added an additional wireless network on AP2. This network, while of course suffering from the repeater effect, performs better than when connected to AP3.
The 802.11s link shares the physical radio just on AP1. But if this was a problem then the Performance from AP3 to AP1 would be bad - which is not the case.
I tried to visualise where the performance loss occurs:
