[SOLVED] Ethernet mostly cuts out after heavy network use

JosSamLoh · 2026-10-07 19:41:50

What?

This started happening a month or two ago while downloading a large game via steam. At first I thought my motherboard was dying, but after testing on a new PC (and narrowing down the issue) I can now confirm that there seem to be some issues with libvirt and dnsmasq.

The issue

With libvirtd.{service/socket} and virtlogd.{service/socket} enabled and/or dnsmasq installed, after heavy, continuous, full gigabit network traffic, my ethernet cuts out (In steam it slowly drops down, probably due to smoothing, while in the KDE network applet speed graph it drops near instantly) and prevents any further downloading of the steam game and any web pages from loading in steam or in my browser. Steam will also say Content Servers Unreachable, and I will get a notification in KDE from Network Management stating: "Limited Connectivity This device appears to be connected to a network but is unable to reach the Internet.". However, strangely, I can still see messages in discord loading, but no media like images, profile pictures or gifs, etc

Table of tested variations:

┏━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━┓
┃ Services ┃ DNSMASQ   ┃ Ethernet ┃
┃ Enabled  ┃ Installed ┃ Dies?    ┃
┠──────────╂───────────╂──────────┨
┃ Yes      ┃ Yes       ┃ Yes      ┃
┃ No       ┃ Yes       ┃ Yes      ┃
┃ Yes      ┃ No        ┃ Yes      ┃
┃ No       ┃ No        ┃ No       ┃
┗━━━━━━━━━━┻━━━━━━━━━━━┻━━━━━━━━━━┛

For each tested variation I started a download of Grand Theft Auto V Enhanced, where:

IIRC the download stopped around 7-11GB
The download stopped at around 13-17GB, waited a bit until it started again then died again at around 23GB
IIRC the download stopped at around 11-23GB
Ethernet never died for the final variation, with full speed consistently for the entire duration of the 96GB download

For the most part, where and when the network dies seems to be a bit random, but usually within a 5-25GB range.

Errors

The only relevant info I can find in the journal is this:

░░ The unit NetworkManager-dispatcher.service has successfully entered the 'dead' state.

Though I am unsure whether this is even relevant.

PC info

Motherboard NIC

Network:
  Device-1: Realtek RTL8111/8168/8211/8411 PCI Express Gigabit Ethernet
    vendor: Gigabyte driver: r8169 v: kernel port: f000 bus-ID: 05:00.0
  IF: enp5s0 state: up speed: 1000 Mbps duplex: full mac: 30:56:0f:ae:b9:47

I can provide any other information to diagnose the issue, just ask and provide me with the necessary commands.

Sorta fixed for now

Thankfully, I am still able to use my virtual machines even with the services disabled, and due to using the user session with usermode networking, I don't need dnsmasq either. Although dnsmasq was recommended as a fix for slow downloads on the steam troubleshooting page, which is strange as it seems to be killing my network rather than speeding it up (though without it I am able to reach max speed anyway).

Hopefully this helps anyone else running into this issue, and hopefully can be fixed at some point too if leaving the services disabled eventually causes some issues, but so far there have been none.

Last edited by JosSamLoh (2026-10-10 23:03:13)

seth · 2026-10-07 20:27:58

When the ethernet "dies" can you still

ping _gateway

or

ping 8.8.8.8

?

dnsmasq was recommended as a fix for slow downloads on the steam troubleshooting page

that would be to avoid slow DNS resolution.

I can provide any other information

Let's start w/

find /etc/systemd -type l -exec test -f {} \; -print | awk -F'/' '{ printf ("%-40s | %s\n", $(NF-0), $(NF-1)) }' | sort -f

Chances are you're concurrently running resolved and dnsmasq and then your DNS resolution trips at some point.

But ideally also post your complete system journal for the boot:

sudo journalctl -b | nc termbin.com 9999

*after* the etherned "died"

JosSamLoh · 2026-10-07 23:00:24

A bit of an update

I was actually able to download just fine even with everything enabled and installed, without ethernet cutting out, and it seemed to be the download cache in steam, as as soon as I cleared the download cache, the issue showed up once more. So likely due to me downloading GTAV:E over and over it cached a lot of it so I didn't run into whatever is causing these loss of connections. Hopefully this info will narrow it down even more.

Diagnostics

seth wrote:

When the ethernet "dies" can you still
ping _gateway
or
ping 8.8.8.8
?

Can confirm I can ping _gateway and 8.8.8.8 fine but not gnu.org, which I believe reported "Temporary failure in name resolution". Nor do any websites load through the browser, though discord messages kept being received without any media like avatars, etc, as per my main post. I will also add that my I'm using a pi 400 with pi-hole, with unbound, as my dns instead of my routers pre-configured dns, if that could potentially change anything.

seth wrote:

Let's start w/
find /etc/systemd -type l -exec test -f {} \; -print | awk -F'/' '{ printf ("%-40s | %s\n", $(NF-0), $(NF-1)) }' | sort -f
Chances are you're concurrently running resolved and dnsmasq and then your DNS resolution trips at some point.

% find /etc/systemd -type l -exec test -f {} \; -print | awk -F'/' '{ printf ("%-40s | %s\n", $(NF-0), $(NF-1)) }' | sort -f
autovt@.service                          | system
dbus-org.freedesktop.nm-dispatcher.service | system
display-manager.service                  | system
getty@tty1.service                       | getty.target.wants
libvirtd-admin.socket                    | sockets.target.wants
libvirtd-ro.socket                       | sockets.target.wants
libvirtd.service                         | multi-user.target.wants
libvirtd.socket                          | sockets.target.wants
NetworkManager.service                   | multi-user.target.wants
NetworkManager-wait-online.service       | network-online.target.wants
openrgb.service                          | multi-user.target.wants
p11-kit-server.socket                    | sockets.target.wants
pipewire-pulse.socket                    | sockets.target.wants
pipewire-session-manager.service         | user
pipewire.socket                          | sockets.target.wants
remote-fs.target                         | multi-user.target.wants
snapper-cleanup.timer                    | timers.target.wants
snapper-timeline.timer                   | timers.target.wants
systemd-userdbd.socket                   | sockets.target.wants
tuned-ppd.service                        | graphical.target.wants
tuned.service                            | multi-user.target.wants
virtlockd-admin.socket                   | sockets.target.wants
virtlockd.socket                         | sockets.target.wants
virtlogd-admin.socket                    | sockets.target.wants
virtlogd.service                         | multi-user.target.wants
virtlogd.socket                          | sockets.target.wants
wireplumber.service                      | pipewire.service.wants
xdg-user-dirs.service                    | graphical-session-pre.target.wants

seth wrote:

But ideally also post your complete system journal for the boot:
sudo journalctl -b | nc termbin.com 9999
*after* the etherned "died"

https://termbin.com/z8ef0

Last edited by JosSamLoh (2026-10-07 23:20:03)

seth · 2026-10-08 07:23:38

Can confirm I can ping _gateway and 8.8.8.8 fine but not gnu.org

So your ethernet is still "alive" but your DNS resolution fails.

Neither dnsmasq nor resolved is in journal or enabled services.
There're also no other indicators or network related issues until

Oct 07 23:42:57 cupdessy steam[2830]: [2026-10-07 23:42:57] Background update loop checking for update. . .
Oct 07 23:42:57 cupdessy steam[2830]: [2026-10-07 23:42:57] Checking for available updates...
Oct 07 23:42:57 cupdessy steam[2830]: [2026-10-07 23:42:57] Downloading manifest: https://client-update.akamai.steamstatic.com/steam_client_ubuntu12
Oct 07 23:42:57 cupdessy steam[2830]: [2026-10-07 23:42:57] Manifest download: send request
Oct 07 23:42:57 cupdessy steam[2830]: [2026-10-07 23:42:57] Manifest download: waiting for download to finish
Oct 07 23:42:57 cupdessy steam[2830]: [2026-10-07 23:42:57] Manifest download: finished
Oct 07 23:42:57 cupdessy steam[2830]: [2026-10-07 23:42:57] Download failed: http error 0 (client-update.akamai.steamstatic.com/steam_client_ubuntu12)
Oct 07 23:42:57 cupdessy steam[2830]: [2026-10-07 23:42:57] Downloading manifest: https://client-update.fastly.steamstatic.com/steam_client_ubuntu12
Oct 07 23:42:57 cupdessy steam[2830]: [2026-10-07 23:42:57] Manifest download: send request
Oct 07 23:42:58 cupdessy steam[2830]: [2026-10-07 23:42:58] Manifest download: waiting for download to finish
Oct 07 23:42:58 cupdessy steam[2830]: [2026-10-07 23:42:58] Manifest download: finished
Oct 07 23:42:58 cupdessy steam[2830]: [2026-10-07 23:42:58] Download failed: http error 0 (client-update.fastly.steamstatic.com/steam_client_ubuntu12)

I'm using a pi 400 with pi-hole, with unbound, as my dns instead of my routers pre-configured dns

Pretty much, yes - have you tried cutting that one out since apparently dnsmasq is not the culprit either and

Oct 07 23:38:02 cupdessy systemd[1]: libvirtd.service: Deactivated successfully.

libvirt was also deactivated?

JosSamLoh · 2026-10-08 15:50:06

seth wrote:

 Oct 07 23:38:02 cupdessy systemd[1]: libvirtd.service: Deactivated successfully.

libvirt was also deactivated?

Libvirt seems to automatically just deactivate after a period of inactivity, searches indicate around 2 minutes of inactivity.

seth wrote:

Pretty much, yes - have you tried cutting that one out since apparently dnsmasq is not the culprit either

After setting my DNS back to the default routers DNS, my download speed was heavily nerfed and wouldn't go above around 250Mbps, however that is likely due to other people in the house using it (or steam hammering the DNS as stated further down). I can test the speeds again later on in the day when there is more bandwidth available.

What is more interesting is that in the pi-hole log steam is absolutely hammering my pi-hole with queries that seemingly never end (hopefully that is what the steam download cache alleviates), and in the "Pi-hole diagnosis tab" I have two entries stating that "Client [my ip] has been rate-limited for at least {number} seconds (current limit: 1000 queries per 60 seconds)" so that could potentially be the issue...

Issue solved?

...and after setting the rate limit very low I immediately get the issue on demand lol. Of course, as always, it's a DNS problem. With dnsmasq and libvirt services enabled and the pi-hole rate limit completely disabled I was able to download the entirety of GTAV:E without my ethernet cutting out. So that confirms it.

I still am quite intrigued whether steam absolutely hammering the DNS is normal operation, I'd imagine it shouldn't need so many IPs to download a game from a CDN... But this guess is definitely about a realm outside of my expertise.

My guess as to why I never got this issue before is probably some kind of network architecture change on steams side, and for why I never seemed to get it when disabling the libvirt services and uninstalling dnsmasq was either them somehow pushing the DNS queries slightly over the edge to be rate limited by pi-hole, or pure dumb luck that it simply randomly worked at the right time lol.

Either way, definitely going to be looking at the DNS for network issues to START WITH in the future. Thanks for all the help!

seth · 2026-10-08 19:13:06

default routers DNS, my download speed was heavily nerfed and wouldn't go above around 250Mbps

Slow ISP DNS?
In that case using dnsmasq or resolved as local DNS cache actually should™ help.

steam is absolutely hammering my pi-hole with queries that seemingly never end

Is this also the case when dnsmasq is actually properly setup and configured?
https://wiki.archlinux.org/title/NetworkManager#dnsmasq

whether steam absolutely hammering the DNS is normal operation

It's not very efficient but also not "abnormal".

JosSamLoh · 2026-10-08 21:04:04

seth wrote:

Slow ISP DNS?

Seems like it.

seth wrote:

In that case using dnsmasq or resolved as local DNS cache actually should™ help.
...
Is this also the case when dnsmasq is actually properly setup and configured?

When connected to my ISPs DNS, it did indeed help, even with nothing else going on in the network I can only reach 400Mbps, when enabling dnsmasq in NetworkManager it shoots straight up to 900~Mbps.

And all also working well when connected to pi-hole, MUCH less queries from steam with dnsmasq setup. Thanks.

seth · 2026-10-09 07:12:49

\o/
Please always remember to mark resolved threads by editing your initial posts subject - so others will know that there's no task left, but maybe a solution to find.
Thanks.

Arch Linux

#1 2026-10-07 19:41:50