problem with with reduced networkinginternet speed

[robdavies]

I have three laptops at home, connected via ethernet wired through the house.

• Advent running Xubuntu 20.04
• Acer running Xubuntu 18.04
• Dell 830 running Xubuntu 20.04

I have broadband supplied by Virgin Media at 200Mb down and 20Mb up


I was getting 200Mb download (on laptop 3 - Dell which was running Xubuntu 18.04 at the time) until around March 2020 (beginning of COVID 19 lockdown), which it then dropped to 60Mb download. Repairing it at the time wasn't a priority, especially as I could still get a usable speed.

Recently, I brought all of the laptops together (to where the modem is) to do a proper speed test. I use a USB 3 gigabit ethernet adapter (TP Link UE300) for one of my laptops, so I used that on all laptops for this test, together with the same cat 5e cable.


Only one device was connected to the modem at one time during the test. These speed tests were performed on speedtest.net, the results were :

• Advent : 60Mb down 20Mb up
• Acer : 200Mb down 20Mb up
• Dell : 60Mb down 20Mb up

So with the same cable and ethernet adapter I was getting different results of different laptops. Even one I used to get 200Mb download on with the same adapter.


I am getting the correct speed from my broadband provider, as I get 200Mb with one of the laptops. I have no ideas how to diagnose/repair this. Can someone help?

Thanks for any help

Update :
I have booted my advent laptop with Xubuntu 16.04.4 live image and was able to get 135Mb download and 20Mb upload. So its not a hardware issue, though don't know how to repair my Xubuntu 20.04 installation

[TheFu]

Good work on isolating the issue. Initially, was considering that the USB2/3 ports in each laptop just weren't connected to the same hardware and some internal limit of the USB internal bus was the problem. Could run an 18.04 Live Ubuntu environment (booted from USB) and completely rule that out? Test that on all three systems for the full apples-to-apples comparison. Same CAT5e and same USB3-ethernet adapter still, please.

Then do the same tests with a 20.04 Live Ubuntu environment (booted from USB) too.

Once that is done AND reported, some other ideas will become clear.

BTW, most websites limit how much bandwidth any single client can use. I limit my users to 1 Mbps, so everyone gets the same.

Within your LAN, transferring files between these systems, what is the performance like? I usually see around 40-75 Mbps, depending on the systems involved. My network is all GigE here. Really, that's where network performance should be measured. Leave the internet out of it. Use iperf3 between systems. Run the server on 1 and the client on the other 2, connecting into the server. That will show you how bad wifi networking is compared to wired AND that some wired ethernet chips are just bad. Some only give 200 Mbps. Others only 650 Mbps. There are good ones that do 920+ Mbps with minimal OS or CPU impact. Those are the ones we want.

[robdavies]

Thanks for your reply and for your help.

I will run the tests you have mentioned and will post the results when they are completed.

Thank you.

[robdavies]

This is what I have so far.

I tried to get Xubuntu 18.04 , 18,04,1 up to 18.04.5 , they weren't available. So, instead I downloaded these versions for Ubuntu. I also downloaded Xubuntu 16.04.6, 18.04.5, 20.04.1 and 20.10

I remember noticing the problem first occurring around March 2020 (I think) , this would fall between the releases of Xubuntu 18.04.4 (release February 2020) and Xubuntu 18.04.5 (release August 2020). It seems to me that an update occurred during 18.04.4 that caused this. Currently, the speed tests I've performed are showing the same conclusion.

I have run all of the Ubuntu versions on all of the laptops. I had a problem with the Dell D830 with the Nvidia nouveau driver, I don't use this driver normally as it freezes my computer, it froze during bootup on the Dell laptop, so I wasn't able to perform the speedtest on all editions on he Dell laptop.

I used the command line speed test for all of the tests, I got more consistent results https://www.speedtest.net/apps/cli

I have been able to get 200Mb download on all tests using Xubuntu/Ubuntu upto and including 18.04.4, after that it slows down to 60Mb max.

The * on the Xubuntu Dell tests, means that I initially got 6Mb downloads, I then had to upplug the USB ethernet and plug it in again. I then got the results as in the PDF. Come to think of it, in normal use, there were times during bootup I couldn't get any ethernet connection (no connection to router) had to unplug it then plug it in again, or reboot. I don't know if this is related.

I will update this post when I have completed the rest of the speed tests.

Thank you,
Rob

[robdavies]

I tested the USB separately using the following

1) cat xubuntu.iso | ps > /media/rob/usbdrive/xubuntu.iso

I got 9.2MiB/s write speed

2) cat /media/rob/usbdrive/xubuntu.iso | ps > /dev/null

I got 17.8 MiB/s

This was run on the Advent laptop when running Xubuntu 20.04.1 that is already installed on it. With the ethernet it get 60Mb download. I wanted to test another device in case it was the USB bus effecting all devices rather than it affecting just this ethernet device.

[robdavies]

I tried iperf3 today, got 260Mb/s over my network between the Advent laptop and the Acer laptop. They are, however, connected via usb2 as neither of them have usb3.

I read that usb2 will realisticly get between 30MB and 40MB per second transfer rate due to overheads, 260Mb/s (32.5 MB/s) is between this rate.

I did get a breakthrough in that I tried another older kernel (5.0.0) from ubuntu.kernel.com and I ran the speed test from my dell laptop (running 20.04.1). It was getting 60Mb/s, this time it got 220Mb/s which is the full and correct speed.

So it seems like its a kernel problem.

Any ideas on who I can contact about this? Or what the next step I should take?

I hope I won't have to install lots of older kernels to narrow down which kernel started the problem for me, there are loads and loads of them.

Thank you,
Rob

[robdavies]

I tried to narrow down kernels that work :

5.3.0 - 200Mb
5.3.18 - 200Mb
5.4 rc1 - 60Mb
5.4.0 - 60Mb

Going to see if I download long term support kernels from kernels.org and see what results I get from that.

[TheFu]

What are the LAN -to- LAN tests using iperf3 showing? For good GigE connections, over 920 Mbps is expected. Forget internet. There's too much out of our control there. Test only LAN-to-LAN performance.

Posting numbers without saying the test setup isn't too useful. I won't guess.

And please don't change between Mbps and MBps units. Use 1 and stick with it!

[Doug S]

I tried to narrow down kernels that work :

5.3.0 - 200Mb
5.3.18 - 200Mb
5.4 rc1 - 60Mb
5.4.0 - 60Mb

Going to see if I download long term support kernels from kernels.org and see what results I get from that.

You have already narrowed things down as much as possible, establishing the start (5.3.0) and end points (5.4-rc1) for a kernel bisection.
Anything in 5.3.18 would only be as backported from mainline, so don't deviate from the master tree.
Always before ever doing a kernel bisection, try the latest Ubuntu mainline PPA kernel (no need to get stuff from kernel.org as ready .deb files already here), which as of writing this is 5.11-rc4. Kernel bisection is a lot of tedious work, involving typically less than 20 kernel compiles. In my experience, 4 times out of 5, once the problem commit is known and used as search term it turns out to be a known problem with a fix in progress (i.e check the mainline kernel weekly).

EDIT: Also have a detailed look at the Ubuntu kernel configuration differences between kernels 5.3.0 and 5.4-rc1. Sometimes, but not often, the issue is actually a configuration change. See the /boot/config* files.

[Doug S]

Forget internet. There's too much out of our control there. Test only LAN-to-LAN performance.

Agreed. However, if it is the only clear indicator type test the OP has, and as long as all tests are repeated several times, then I would continue the investigation. Yes, a more robust, controlled environment, test is ultimately needed for either a launchpad or upstream bugzilla bug report. The detailed method for such a test might be revealed if this thread is tugged at long enough.
if indeed there is some kernel issue, it would be good to isolate it.