[workaround] nfoce forcedeth slow network-speed

Vrenn · Guru Joined: 15 Dec 2004 Posts: 327

Hello!

I have a strange problem. On my old very relyable machine (build 2007) I used to backup over a 1GB-Ethernet.
Now I saw that the netspeed is only 100Mbit max (sftp/rsync ~9MB)

NIC is nforce motherboard.

bbgermany · Posted: Mon Jan 29, 2018 8:09 am Post subject:

Hi,

can you check via ethtool the advertised modes etc?

greets, bb
_________________
Desktop: Ryzen 5 5600G, 32GB, 2TB, RX7600
Notebook: Dell XPS 13 9370, 16GB, 1TB
Server #1: Ryzen 5 Pro 4650G, 64GB, 16.5TB
Server #2: Ryzen 4800H, 32GB, 22TB

Vrenn · Guru Joined: 15 Dec 2004 Posts: 327

bbgermany · Posted: Mon Jan 29, 2018 3:29 pm Post subject:

Hi,

np about the english, im no native as well. Usually issues on gbit belong to bad cable, bad connections on patch panels or even the wall outlets. Sometimes it is a bad nic or even a misconfigured switch port (if possible).

greets, bb
_________________
Desktop: Ryzen 5 5600G, 32GB, 2TB, RX7600
Notebook: Dell XPS 13 9370, 16GB, 1TB
Server #1: Ryzen 5 Pro 4650G, 64GB, 16.5TB
Server #2: Ryzen 4800H, 32GB, 22TB

NeddySeagoon · Posted: Mon Jan 29, 2018 3:35 pm Post subject:

Vrenn,

1Gb uses all 8 wires in the network cable.
10/100Mb only uses wires 1, 2, 3 and 6.

It matters to 1Gb about kinks and squashed cables too.
Watch out for 4 wire patch cables.
_________________
Regards,

NeddySeagoon

Computer users fall into two groups:-
those that do backups
those that have never had a hard drive fail.

Vrenn · Guru Joined: 15 Dec 2004 Posts: 327

Both cable I used are cat6. I cleaned all components, and installed the cable again. If it happens again I'll buy/replace the cable, but as the backup-cable (cat6 too) did show the same, just under Linux for two days I though it was a driver-issue. At least ethtool tells me what my diver is thinking at the moment :-)

Thank you for the help!
_________________
With nice greetings
Vrenn

Vrenn · Guru Joined: 15 Dec 2004 Posts: 327

When the machine is turned off a while I now get 100 mode per default.

Vrenn · Guru Joined: 15 Dec 2004 Posts: 327

Just tried a NIC with Realtec-Chip "RTL8139C"
Didn't know if driver 8139cp or 8139too neede, so I configured both as module.
Both loaded (lspci -v) and openrc told me it got an IP.
But then local network access was not possible. I loaded just one of them, no success. After reboot with ubuntu-dvd, modules were not loaded at all. Back in my genoo ifconfig/lspci -v don't show eth1 either, it's like its dead...

nforce,D-link, Realtek, third card burned?
Now I'm really desperated.
_________________
With nice greetings
Vrenn

bbgermany · Posted: Wed Jan 31, 2018 10:12 am Post subject:

Hi,

if you have installed a new card, it could be named as enp3s0 or something similiar. eth1 or even ethX is not a must anymore since the new naming scheme of udev.

Did you shut off eth0 (your nforce card) when you tried the 8139C? Accordind to the usb lan adapter, you wont get 35-40MB/s if you buy a USB2.0 card, since these usually connect with 100Mbit/s only and if you buy a USB3.0 based card, im not sure how well they perform on USB2.0

If you can, you should try to get a PCI Ethernet adapter from ebay or even amazon, based on an Intel card if possible.

oh and btw, the 3m cable is direct attached to the switch, or to another computer or is it connected to the wall outlet?

greets, bb
_________________
Desktop: Ryzen 5 5600G, 32GB, 2TB, RX7600
Notebook: Dell XPS 13 9370, 16GB, 1TB
Server #1: Ryzen 5 Pro 4650G, 64GB, 16.5TB
Server #2: Ryzen 4800H, 32GB, 22TB

Vrenn · Guru Joined: 15 Dec 2004 Posts: 327

Hello.

ifconfig first showed the new card, but it disappeared after some minutes of use. First both modules loaded and bind on the card by themselves. Then after restart modules were not loaded anymore, modprobe them didn't made any change.
Seems like the cars have not been found by linux anymore, even ubuntu-livedvd did not get them up.

Current setup: problemPC-3mcable-NetgearGBswith-router-anotherNetgearSwitch-Server/laptop

What about a powerunit/motherboard-fault? CPU run stable compiling kdeplasma or libreoffice. But it might harm the cards I installed in the last slot? Or why they are visible as eth1 first with loaded modules and IP, and after a short time not even found any more by the gentoo or ubunu-kernel?

An USB-NIC might be more motherboard/powerunit-independent. I already tried two Realtec/D-link PCI cards on it.

Thanks for the intel-tip. I might check the cable-connetions on the board first.
_________________
With nice greetings
Vrenn

NeddySeagoon · Posted: Wed Jan 31, 2018 7:54 pm Post subject:

Vrenn,

Post your lspci output and put your dmesg onto a pastebin site.
We need dmesg from the very start, so you may need to reboot to get that.

USB2 provides at best 440Mbit/sec but its only half duplex.
You won't get 40MB/sec of user data over that, even if the USB root hub is not driving anything else.
There is USB overhead, TCP overhead, half duplex ... so you lose over 10% of the theoretical 440Mbit/sec that USB2 provides.

A 33MHz PCI bus won't quite do 1Gbit/sec either.
_________________
Regards,

NeddySeagoon

Computer users fall into two groups:-
those that do backups
those that have never had a hard drive fail.

bbgermany · Posted: Wed Jan 31, 2018 7:56 pm Post subject:

Can you post the output of "lspci -v" and maybe the output of dmesg?

greets, bb
_________________
Desktop: Ryzen 5 5600G, 32GB, 2TB, RX7600
Notebook: Dell XPS 13 9370, 16GB, 1TB
Server #1: Ryzen 5 Pro 4650G, 64GB, 16.5TB
Server #2: Ryzen 4800H, 32GB, 22TB

Vrenn · Guru Joined: 15 Dec 2004 Posts: 327

I now hat time to test:
(rarely used pastebin so put all test in one text file and marked them with original filenames where protocol started, and yes, please look at lsCPi..., files are wrong named by me...)
https://pastebin.com/HGAkXbmh

1. I recorded dmesg and lspci of the buildin nforce-nic showing up (sometimes) with 1GB: Search for dmesg_only_nforceNIC.txt lscpi_only_nforceNIC.txt
2. I recorded dmesg and lspci of the buildin nforce-nic after a cold stard, seems to always be at 100Mbit: Search for dmesg_only_nforceNIC_coldstart.txt lscpi_only_nforceNIC_coldstart.txt
3. Applied the 1GB D-Link card to only free PCI slot.
4. I recorded dmesg of the 1Gbit DGE-528T D-Link card, not appearing: dmesg_DGE-528T_1GB_asETH1_firsttry.txt
5. Removed the 1GB D-Link card from the only free PCI slot for the other card.
6. I remembered forgetting lspci, ramming it back in its slot, notice that eth1 comeing up with the r8169 driver and in 1GB mode: dmesg_DGE-528T_1GB_asETH1_secondtry.txt lscpi_DGE-528T_1GB_asETH1_secondtry.txt

No call me getting insane, I would believe you.
I now think its a hardware-problem with the nforce NIC, not knowing anything about it except its behaving bad (Speed/instable).
The two other NICs, at least the D-Link is now showing up after being rammed in the dam slot at about 5 times. Might the time of some years not using the slot may lead to connection-problems insulting the kernel-drivers in some way?
It works today and I'll look at it the next days. I copied several GB-big files, no disconnet and coming always up with 1GB mode.

(ps: I controlled the right fit of the PCI-cards each time in good light before powering up, its frustrating somehow)
The machine was rockstable for so many years (with gentoo) but nvidia/kde/nouveau-fight and now the NICs are forcing their tribute.
Anyway I am learning from you, thanks for that.
_________________
With nice greetings
Vrenn

Vrenn · Guru Joined: 15 Dec 2004 Posts: 327

I now tested the "new" card for one week. Seems I have to blame the old PCI-slot that it didn't work in the first tries.
No Idea why the nforce-NIC doesnt use 1GB-mobe at cold startup and is unstable at replugging.

The 1GB D-Link card now works. Sometimes the speed goes down even to 7 GB after some minutes, but NIC-mode still stays in 1GBmode.
This, so seems to be a different thing, but at least it stays stable.

Thank you for your help.
_________________
With nice greetings
Vrenn