Network connection speed changes suddenly from 1000BaseT to 100BaseT (TrueNas 12->13)

[broozar]

Hi everyone,
I recently upgraded my homelab server from TrueNas 12 to 13. I am using the server as a backup to sync a SynologyNAS over TrueNas' CloudSync functionality via SFTP. This has worked well in v12. Now in v13, my file transfer starts out great, but after a few minutes, the network connection speed drops from 1000 to 100 BaseT while running. Curiously, this speed reduction persists over multiple reboots. I only get GBit back when I unplug the machine and then power it on again. What could be the reason and how can I fix it?

The common suggestion in this forum seems to be bad cables. Since I did not touch the cables between the upgrade and the other hardware did not change either, I doubt that's it.

Setup: H81M-C board with Realtek RTL8111G going into a Netgear GS116 switch connected to a FritzBox (Router) and a Synology DS 620 (NAS I want to back up).

 

[Samuel Tai]

Sadly, yours isn't the first case of Realtek NICs doing oddball things, and won't be the last. The usual recommendation is to switch to a genuine Intel PRO/1000, which are cheap and plentiful on eBay.

 

[broozar]

Thanks, will do.

 

[broozar]

I have replaced the NIC with an Intel card as you suggested. Initially, the behaviour was as erratic as with the Realtek chip (started as GBit, dropped to MBit after a few minutes). Then, I configured IPv6 in addition to a static IPv4 and it appears to be stable at GBit for hours now. Are these 2 things related (IP protocol version and performance/reliability) or is it a coincidence that the speed has not dropped again?

 

[Samuel Tai]

Probably a coincidence. You may need to disable the on-board Realtek in the BIOS for complete stability. Also check your power supply.

 

[broozar]

I disabled the Realtek chip immediately after installing the Intel card. However, both interfaces were listed in the dashboard (one as not connected), and both had the same IP(v4). So I deleted the Realtek interface in the dashboard. Maybe that was part of the solution too. The server is now going at full speed for several hours now, so I consider this solved - thanks.

 

[broozar]

Sorry for coming back to this. For days now, I am falling back to 100 Mbit again. I replaced all the wirees with new Cat6. The runs are only 5m long. I don't know what to try next. I read somewhere that a power saving feature might be at fault, but I don't know where I should look for that, the Bios certainly didn't show anything suspicious. Any suggestions?

 

[Patrick M. Hausen]

Is the Netgear switch managed? I'd look for any settings claiming to be about power efficiency.

 

[broozar]

No, it's a cheap unmanaged switch. All my traffic from all my computers goes through that switch and I have not observed any other PC having this issue.
Any other suggestions?

 

[Arwen]

One note about 1Gbps copper verses 100Mbps copper, is that 10/100Mbps Ethernet only use 2 of the 4 pairs. One pair for each direction. But, 1Gbps copper uses all 4 pairs, and uses each pair for bi-directional data.

What that means is that you probably have a cable or connection with a loose or bent pin for one of the gigabit specific wires.

I've seen bent pins on the female jack, on a workstation system board. Bending the pin back temporarily solved the problem. (We had vendor support, so that was the permanent solution, replace the system board.)


Try looking at all the female sockets for the Ethernet path. All 8 pins should be regular, straight and springy.

 

[broozar]

Changed the cables again, getting 1GBit. Here's hoping that it holds this time. Thanks everyone for your suggestions.

 

[Arwen]

As I clearly stated, it is NOT always the cables. I have PERSONALLY seen damage to the board side socket that the cables plug into, where their was a bent pin.

I suggest checking all socket sides of the network path. You might need a flashlight and or slight magnification, (depending on how good your eyesight is).

 

[broozar]

Yes, I understand. There are several female connectors involved: Server to patch panel and then to switch. By "replacing the cables" I meant the entire run, including taking another place on the patch panel and choosing another port on the switch. Do you think the connector on the panel is borked?

 

[Arwen]

Okay, great.

As for your Port 21, the 2nd & 3rd from the right do look a little wonky. But, I can not be sure. And research into which those are in relation to your network path would have to be done to figure out if they are part of the 1Gbps copper path.

Actually thinking about it, the 3rd from the left & 3rd from the right form part of the 10/100Mbps path. And it is either the far 2 right or far 2 left that make up the other pair for 10/100Mbps. But, this is dependent on your layout.

 

[broozar]

Thanks for the suggestions. Just a small update, I had more and more PCs on the network acting up in the last few days to the point where I received IPs from my router but got no internet connectivity at all, constant ping timeouts etc. Looks like my switch was to blame after all, since everything is back to normal after turning it off and on again. Very odd.