tested:
BPI-r4 is a dumb AP.
PC1 and PC2 both have a 10gbe adapter.
Tests are done via NFS file transfer by mounting PC2/share on PC1 and dd if=share1/file1 of=/dev/null
The BPI-R4 in between has 2 RJ45 2.5G SFP adapters
from PC1 @ sfp-lan (eth2) to PC2 @ sfp-wan (eth1): 100mbit
from PC1 @ lan1 to PC2 @ sfp-wan (eth1): 100mbit
from PC1 - PC2 (direct link): 5gbit (limited by SATA SSD)
from PC1 @ lan1 to PC2 @ lan2: 1gbit
so the only conclusion here could be that eth1 and eth2 are running at 100mbit speed, even though they claim to be running at 2500mbit.
How could I invesitgate more?
You should test only 1 connection first (pc1 or pc2 to r4),e.g. acessing a file on bpi-r4 nvme,or better using iperf(3) to avoid bandwidth limitation through protocol overhead. I’m not sure how much overhead is added for nfs and how performant it is configured in your case (packet size,latency,…).then adding complexity/protocols step by step. E.g. when using encryption like on sftp/scp maybe this is the bottleneck.
looks like you’re right: when running this on BPI-R4 itself:
mount -t nfs4 PC1/ /tmp/desktopjp
date; dd if=/tmp/desktopjp/largefile.iso of=/dev/null; date
umount /tmp/desktopjp
the results are unpredictable at best (even when swtiching between sfp-lan and lan1), but nowhere at a limited size of close to 100mbit (more like 800-2.2gbit, it depends). so it’s definitely not the SFP adapter. Could it be a bridging problem when a non-dsa port is involved?
I’m not exactly sure what you mean …
In the testing setup:
the bridge interface is the only intrface with an IP.
PC1 is connected via SFP-LAN and LAN1 to a internal bridge on the PC itself and to the bridge of the BPI-R4.
it used to be a bonding link BPI-R4<=>PC1 and bondPC1 was connected to br-lan, but I broke it and performed testing with sfp-lan and lan1 connected to the internal bridge.
I performed testing with sfp-lan (xor) lan1 up and did the command unmounting and remounting every time to work around the file caching of linux. on both interfaces, the speed was way above 100mbit
So, I’m not sure I get what you’d like me to try…
edit: or did you mean: both PCs on a lan1 / lan2 interface? that is 1gbit.
I tested, based on your suggestion, how fast the BPI would copy files from PC1 to internal /dev/null, so traffic speed is measured without actually switching between ethernet ports. Traffic speed from PC1 to PC2 when both are connected to DSA is 1gbit