Testing limits of home Ethernet wiring

[unmesh]

I will pursue this in the background since the primary goal was not deployment but rather testing.

Thanks for your help.

 

[unmesh]

just chucked usb adaptors on a couple of PCs at work and tested under windows (one is an i7-4770 based desktop, other is a surface pro with an i5-7300U) - out of the box absolutely solid sustained 2.37gbits/sec in either direction using iperf3 ( tested all the way from just a single stream up to 16)

I can also see load on multiple cpus in resource monitor during testing - on the 7300u it'll hit 100% of one cpu and 50% on another - on the 4770 it's about 70% on one cpu and 30% on three other cpus


so yes - something very odd going on with the network stack on windows your end

So I got my USB-C female to USB-A male adapters today and swapped one laptop with another that did not have any USB-C connectors but only USB 3.0 Type A ones. The iperf3 results are still sub 2Gbps.

Could you share your driver version? I'm on the latest from the Realtek website.

Another project for me is to try my desktop at one end of the link.

 

[John Davis]

So I got my USB-C female to USB-A male adapters today and swapped one laptop with another that did not have any USB-C connectors but only USB 3.0 Type A ones. The iperf3 results are still sub 2Gbps.

Could you share your driver version? I'm on the latest from the Realtek website.

Another project for me is to try my desktop at one end of the link.

same as you - the latest off the realtek site (downloaded and clean installed them yesterday morning to test)

I was hoping using something as low powered as the surface pro would have let me recreate your issue - but even that could sustain 2.4gb/s (just).

 

[unmesh]

I will try combinations of Linux and Windows tomorrow in the interest of science but it is very aggravating that there is something different in my Windows laptops that is turning out to be difficult to pinpoint.

Windows Defender Firewall?

 

[John Davis]

I will try combinations of Linux and Windows tomorrow in the interest of science but it is very aggravating that there is something different in my Windows laptops that is turning out to be difficult to pinpoint.

Windows Defender Firewall?

if it was the defender firewall you'd get zero traffic (been there done that) - you should have got the popup from the firewall asking to allow the connection when you first ran iperf, that should be it

try plugging BOTH usb adaptors into your desktop and then run both the iperf client and server processes on the desktop, that will eliminate the laptops totally and hopefully let you pin down the issue

 

[unmesh]

I plugged both into my desktop and configured one as 192.168.2.100 and the other as 192.168.2.101. I confirmed that both are showing up in Network Status

I then opened up two cmd windows and ran the server and the client iperf3 as follows:

iperf3 -B 192.168.2.100 -s
iperf3 -B 192.168.2.101 -c 192.168.2.100

The client times out with
iperf3: error - unable to send control message: Bad file descriptor

Do I need to fiddle with routing tables or something?

 

[John Davis]

I plugged both into my desktop and configured one as 192.168.2.100 and the other as 192.168.2.101. I confirmed that both are showing up in Network Status

I then opened up two cmd windows and ran the server and the client iperf3 as follows:

iperf3 -B 192.168.2.100 -s
iperf3 -B 192.168.2.101 -c 192.168.2.100

The client times out with
iperf3: error - unable to send control message: Bad file descriptor

Do I need to fiddle with routing tables or something?

nope - it'll be windows defender firewall blocking things - I had the same and it seems defender won't pop up and unblock the service 'til it sees the server process on that IP - so for BOTH interfaces start in server mode (and you'll get the pop-up to unblock), you can then kill them and restart in normal client/server setup like you tried

 

[thiggins]

There's no need for the -B. Just run iperf3 -s on the server, then iperf3 -c [server_ip_address] on the client.

And unblock from Windows firewall/defender

 

[unmesh]

I ran iperf3 - s bound to both IP addresses (but did not get a second Defender popup notification)

When I run without binding, I get 7+ Gbps but Task Manager shows no activity on the wires so I'm assuming the loopback interface equivalent of Windows is being used.

If I use -B, I get the timeout as before.

 

[unmesh]

I ran the Firewall Troubleshooter with iperf3 running and it found something to fix after which I can get performance results.

The bad news is that it is clocking just shy of a gigabit. I wonder if this is because the ports hang off an ASM1042 on this motherboard since the CPU utilization stays below 50%

 

[John Davis]

There's no need for the -B. Just run iperf3 -s on the server, then iperf3 -c [server_ip_address] on the client.

And unblock from Windows firewall/defender

except when BOTH interfaces are on the same desktop PC (and that pc also probably has a 3rd interface on the motherboard) - if you don't explicitly bind to a given interface in that case both instances will bind to the first interface they find and you end up just testing the loopback device not the actual physical interfaces

 

[John Davis]

I ran iperf3 - s bound to both IP addresses (but did not get a second Defender popup notification)

When I run without binding, I get 7+ Gbps but Task Manager shows no activity on the wires so I'm assuming the loopback interface equivalent of Windows is being used.

If I use -B, I get the timeout as before.

see my reply above - if you don't bind explicitly to the interfaces you end up testing the loopback device (hence the 7gbps and zero load)

 

[John Davis]

I ran the Firewall Troubleshooter with iperf3 running and it found something to fix after which I can get performance results.

The bad news is that it is clocking just shy of a gigabit. I wonder if this is because the ports hang off an ASM1042 on this motherboard since the CPU utilization stays below 50%

there's a reason ASmedia controllers have a bad rep - so I too would suspect that. The machines I tested on are all using Intel controllers

 

[unmesh]

Will have to wait on a maintenance window on my ESXi Server for that.

Which begs the original question of why the poor performance between laptops using Intel controllers and running Windows rather than Linux. Firmware revision in the USB dongles?

Is there another manufacturer of USB to 2.5 GbE controller chips?

 

[John Davis]

Is there another manufacturer of USB to 2.5 GbE controller chips?

pretty sure realtek is it in 2.5gbe - but Aquantia/Marvell do the AQC111U 5/2.5/1gbe controller - you can find them in things like the qnap qna-uc5g1t and sonnettech solo5g

 

[unmesh]

The reviews say that the 5G adapters cap out at 3.x because of USB 3.2 Gen 1 overhead. I should try out some PCIe 2.5G adapters to see if I have the same issues that I'm having with USB.

 

[follower]

Will have to wait on a maintenance window on my ESXi Server for that.

Which begs the original question of why the poor performance between laptops using Intel controllers and running Windows rather than Linux. Firmware revision in the USB dongles?

Is there another manufacturer of USB to 2.5 GbE controller chips?

windows device manager>remove network adapter>reboot>windows will install a default driver(2015)>retest

 

[unmesh]

windows device manager>remove network adapter>reboot>windows will install a default driver(2015)>retest

I will try that though John on this thread is able to get full performance with the latest drivers.

 

[John Davis]

I will try that though John on this thread is able to get full performance with the latest drivers.

and if you do just let windows update provision the driver you get 10.39 from may 2020 - so only a small amount behind the most current version on realtek's site

 

[BreakingDad]

As a data point, I am using a 100' pre-built Cat5e cable as the backhaul between two RT-AX86U's on their 2.5GbE Ports with no issues.

Ok Boomer