Don’t know why you start with multiple instances instead of testing a single one.
You run 4 servers with 4 streams at the same time in your test.
To simulate multiple users, but if I launch a single instance I get the same result (4Gbps).
This looks like a PCIe link (bandwidth) limitation of the computer. The load is balanced right so the PCIe switch does it’s job correctly.
List the output of lspci here
Already mentioned above, you’re not reading
And
This is not the output of lspci and you’re getting rude again!
OK, thanks for your comments, but you haven’t provided a solution to the problem, so I won’t reply anymore.
Once again, IPFire is perfect; it’s the users who are “idiots”.
Good luck in your live with that behaviour! 
indeed it is kind of a challenge to keep up here
despite the fact that phil scar captured ned bourton's topic
the issue is that with phil's hardware
vyos delivers ~10GB/s and ipf ~4GB/s
and phil assures that this happens on the
very same machine and that is some kind of
similar to ned bourton's issue ...
unfortunately ned seems busy 
b.t.w.
neither!
I don’t believe that at all.
PCIe 3.0 has 8GT/s per lane, or 0,97 GB/s (7,76 Gb/s), without protocol overhead. Since PCIe works in full-duplex per lane, you can take this throughput for the max. of the NIC. With protocol overhead and a single lane 4Gb/s is too less. Should be about 6,5Gb/s. With a single PCIe 3.0 lane there will be never ever 10Gb/s.
But the Nxxx CPU series has only 9 PCIe 3.0 lanes. This computer has no PCIe switch, see lspci.
They should have checked the amount of PCIe lanes connected, used and the link speed to the NICs. But they now it better and are rude. So they can find their own way 
.
@xperimental
indeed the pcie-lanes-question and i [have to] quote myself:
it seems that this was not done by @pscar13
and your provided infos from the datasheet
kind of confirms that.
so where do these network speed values come
from phil and bertel posted?
unfortunatly bertel and phil jumped in here
and ned was not seen again untill now 
chances are still good that ned has not
a nine-lanes-only setup 
maybe due to the lack of moderation
we have a not so helpful mixup here
now with different scenarios ...
I don’t see that there and don’t know where this graphic comes from. Or do you have 0.5 lanes connected to the i226?
lspci -s <ID> -vvv
Look for LnkCap and LnkSta!
@pscar13 afaik the image you posted is ServeTheHome sourced.
And I found “interesting” the braketed numbers for 10gbit adapters and intel network adapter.
It doest not seem “addressing” to me… But I don’t know if there are the lane numbers.
N355 anyway is PCIe 3.0 declared and (as posted) 9 lanes.
So.
2+2+1+0.5+0.5+0.5+0.5 = 7. So only 7 out of 9 lanes used in this mainboard?
You can not assign 0.5 lanes.
I’d agree. But anyway I can’t find the meaning for that numbers…
Who cares about that graphic. To me, it says nothing. Run the command, look for LnkSta and you got the answer.
When I write
I’m not theorizing
I’ve done a test
That is still possible. The OS controls the hw resources, so different OS may manage the hardware resouces different.
Is that exactly what I’m asking?