LAN disconnects after heavy network traffic

[Lincee]

I have an issue with some VM's where the LAN disconnects after some heavy network traffic.

For instance, if i download some linux iso's after a while the connections drop. when i go into the vm and try to ping anything (IP based or DNS based, doesnt matter) it fails. all i can do is reboot the vm and then it works again. I tried switching cards, but that doesnt seem to work.

I had it on both an Alpine VM and a Debian VM, strangely enough the proxmox host has no network issues.

 

[Kingneutron]

I had some issues with DHCP in-vm, try setting it to static IP

 

[Lincee]

All my vm's have static IP's

 

[pct86]

@Lincee having the same exact issue with Nextcloud syncing data to client.
https://forum.proxmox.com/threads/proxmox-vms-lose-network-connection-with-high-traffic.145549/

 

[Lincee]

@pct86 indeed, thank you for mentioning! I'll watch that topic as well

 

[Kingneutron]

What speed are these connections? Granted I've had no trouble after getting rid of DHCP in a couple of instances and haven't tried transferring ~1TB of data in/out of a VM yet

I have 3 separately-wired networks - 1Gbit, 2.5Gbit and 10Gbit (MTU 9000, point-to-point, no "switch" except for Proxmox host) and no trouble with VM net disconnections.

Does ' systemctl restart networking ' have any effect in-guest, or is it only a reboot that fixes it? If it was at the physical level I would suspect NIC overheating.

Any messages in journalctl when it cuts out?

 

[Kingneutron]

Could y'all who are having this networking issue post the configs of your VMs somewhere? Not necessarily in the thread, pastebin would do. Never know what commonality might matter for this

I'm firing up a SuSE VM to test iperf3 continuous 10Gbit transfer, will see if it triggers the issue

 

[Lincee]

ok, these are things i'll have to check tonight!

Its a selfbuild NAS that used to run Alpine Linux, and i had no issues there. So its somehow connected to Proxmox. Also when i limit the download speed to 15MB/s it seems to hold out longer, so far

 

[Kingneutron]

Well, my iperf3 test ran for 136 minutes and change until I finally killed it:

Received ~13TiB
Sent ~5.3TiB

No RX/TX errors, dropped or overruns
DHCP4 lease refreshed (4) times, apparently no issues in journalctl

‍ I'll keep checking the thread, but no issues on my end

 

[Lincee]

There's no specific messages when the connections stop. restart networking works.

iperf runs normally within the VM as a client, as well with speeds between 98MB/s & 135MB/s. When i use iperf as a client (again, sinde the VM), it cut out at 101GB with a bitrate of 210 Mbit/sec and i need to restart networking again

 

[Kingneutron]

Could be a kernel module driver issue, could be overheating, could be the cable, maybe something else. Could be your cat or some RF interference, IDK

What manufacturer is the NIC, what kernel module driver and have you tried replacing the cable (try CAT6) / putting it in a different port on the switch?

Post results of ' lspci |grep -i net '

https://askubuntu.com/questions/46942/how-do-i-stop-my-ethernet-network-connection-from-dropping
(older thread)

https://www.redhat.com/sysadmin/beginners-guide-network-troubleshooting-linux

You may want to consider wiresharking / sniffing the comms, but not sure if that applies here

 

[Lincee]

A kernel driver issue i can accept All others i can not. Since this only happens with Proxmox

Your second link leads me to this: https://forum.proxmox.com/threads/r...eze-performance-with-promox-7-4-and-8.134212/ which has some more information, same problems So i'll see what that does! And lspci:

root@pve:~# lspci | grep -i eth
04:00.0 Ethernet controller: Realtek Semiconductor Co., Ltd. RTL8111/8168/8411 PCI Express Gigabit Ethernet Controller (rev 15)

So i think the kernel driver issue is pretty likely the issue

 

[Kingneutron]

You could get around it pretty quick by throwing an Intel pcie NIC on the motherboard, or possibly going with USB3 adapter.

 

[Lincee]

yeah i ordered myself a small nic see what that does

 

[Lincee]

in the meantime i tried getting it to work with dkms-8168

now i have no network at all anymore im beginning to think going back to Alpine is the better road with this realtek mess

 

[Lincee]

You could get around it pretty quick by throwing an Intel pcie NIC on the motherboard, or possibly going with USB3 adapter.

So, i got a USB3 adapter (https://www.tp-link.com/us/home-networking/usb-converter/ue300/ ), only to discover that this also has a Realtek chip
Same problems, both an Alpine VM and a Debian VM. so i guess i really have to wait untill i have my intel nic. Ive changed the VM nic from Virtio to Intel E1000 but that didnt help. The Proxmox host itself remains unaffected, network keeps working even though the VM looses connection.

 

[Kingneutron]

Yah, you gotta read those Amazon pages 1st and find out what chipset they have ;-)

https://www.amazon.com/gp/product/B00MYTSN18
^ This 1Gbit has an AX88179 chip, I would see about returning the Realtek