OK, I'll admit I'm pretty lost here.

I have a Supermicro X11DSF-E motherboard which has 2 Intel X550 10g NICs in. Both eth0 and eth1 are bonded using 802.2ad (option 4) and appear to be up:

When I look at the info for eth0, it only shows 10g:

Thinking that could just be for eth0, I ran an iperf test on a windows machine with the same motherboard with bonding NICs showing 20g:

As you can see, I'm only getting 10g.

Switch wise, I'm using a Ubiquiti US XG 16 which again I believe I have set correctly (ports 15-6):

I'm still looking into the switch side of things, but wondered if I've missed something on the Unraid side?

Diagnostics attached

Thanks guys!

themightynas-diagnostics-20250606-2115.zip

Edited June 7, 20251 yr by Olly_K
adding diagnostics

A bond will only accelerate multiple simultaneous file transfers, or streams, so it won't show any change with a regular iperf test.

shift your expectations! you are fighting against windmills!

A bond will never add up the possible bandwidths, every single connection just uses one card.

The only thing you might get improved by it, is when your >1000 clients simultaniously try to access the server. The connections of them will be split (equally/randomly/by last bits of MAC or with other means) among all cards that participate in the bond. (but even then, with 10G your hard drives unlikely can even keep up with a single connection, so >1000 won't be sped up too)

But still, none of them can use more than a single card can handle.

For typical home installations with just a few dozens of clients a bond does not make sense at all.

Its mostly a marketing gag.

Save your effords and stress factors, deinstall it and forget that you have ever heard the word bond/LCAPS/port aggragation...

(Addon: the Microsoft page shows you a lie too. Remember: MARKETING GAG!!!)

Edited June 7, 20251 yr by MAM59

Thanks folks, I used additional threads in iperf (-P <num_thread>) to get a higher number which is more in line with what I expect.

I use FreeFileSync a lot which has the option of parallel threads so hoping performance there once I next use it