iperf tests on various scenarios (with SBCs and some x86 machines)
Now here is the cable lengths connecting the machines. The switch is actually required as the USB to Ethernet couldn't handle the lengths between the RK3399 #1 and the two sides.
- USB to Ethernet adapters : r8152 TP-Link model
- RK3328 and Exynos 5420 (+ that router) : CAT 6 cables of 3m.
- RK3328 and 2 unrelated routers : CAT5 of about 10 meters.
- Unrelated routers to RK3399 #1/Switch : CAT5 30m (3399 itself is connected to Switch via a 1m CAT6)
- Switch to RK3399 #2 currently on Grandma's table for now : CAT5 30m (total ~70m to RK3328 with 2 long distance repeaters)
The routers are connected via 1m cables, so they don't count as two.
- RK3399 #2 to the 3 RK3566s : CAT6 1m each
- RK3566 #2 to all the JH7110 and the laptop/router/desktop : 6m + 6m = JH7110 + 3m = Router + 3m + 3m = Desktop/Laptop each
And by the way, this project took a long time to set up and I don't know about the worth of it. After all, many other factors couldn't be accounted for. But for SBCs, I think the limit is about just that. But just remember the RK3328 and the 5420 at close range nearly saturated their Gigabit ports. Could be that NAT'ing and signal degradation is something that is in play, but is difficult to resolve. Althrough it is weird that the RK3399 #2 despite the distance, scored similarly to the first one. But the RK3566s only had 2/3 the bandwidth of the RK3399s. I am sure the PSUs are quality, as I am using a ATX PSU for some SBCs and a PINE64 USB-C power supply for others (65W max)....