Overview
Hardware: Framework Desktop Max+ 395 (128 GB RAM)
OS: Ubuntu Server 24.04.3
The system exhibited intermittent Ethernet instability. While the network interface appeared operational at first, sustained traffic caused severe degradation and eventual connection failure. The issue manifested as kernel-level transmit buffer exhaustion rather than a simple connectivity or configuration problem.
Symptoms
- Ethernet link came up normally (carrier, DHCP, routing all appeared correct)
- Short tests (single pings, brief downloads) worked
- Sustained traffic caused instability:
- Download speeds fluctuated sharply (e.g., ~5 MB/s → ~17 MB/s → 0 MB/s)
- Long-lived TCP connections were terminated by the remote endpoint
- Large downloads repeatedly failed
- Kernel-level errors appeared during testing
Observed Failure Output (Sanitized)
ping -c 100 gatway_ip
64 bytes from gatway_ip: icmp_seq=1 ttl=64 time=1.59 ms
64 bytes from gatway_ip: icmp_seq=2 ttl=64 time=1.11 ms
64 bytes from gatway_ip: icmp_seq=3 ttl=64 time=1.13 ms
64 bytes from gatway_ip: icmp_seq=4 ttl=64 time=1.12 ms
64 bytes from gatway_ip: icmp_seq=5 ttl=64 time=1.13 ms
64 bytes from gatway_ip: icmp_seq=6 ttl=64 time=1.13 ms
64 bytes from gatway_ip: icmp_seq=7 ttl=64 time=1.04 ms
64 bytes from gatway_ip: icmp_seq=8 ttl=64 time=1.07 ms
64 bytes from gatway_ip: icmp_seq=9 ttl=64 time=1.13 ms
64 bytes from gatway_ip: icmp_seq=10 ttl=64 time=1.12 ms
ping: sendmsg: No buffer space available
ping: sendmsg: No buffer space available
ping: sendmsg: No buffer space available
ping: sendmsg: No buffer space available
14 packets transmitted, 10 received, 28.5% packet loss
Initial Attempted Fix (HWE Kernel)
The system was updated following common guidance:
sudo apt install --install-recommends linux-generic-hwe-24.04
This ensured the latest hardware enablement stack was present. While Ethernet functioned at a basic level, the instability under sustained load persisted. The issue was not that Ethernet failed entirely, but that the kernel exhausted transmit buffers during continuous traffic.
MTU Investigation (False Lead)
MTU mismatches were investigated due to failures occurring during WAN transfers. MTU tuning did not resolve the issue. Failures occurred even during local gateway traffic and manifested as kernel send failures rather than fragmentation problems.
Root Cause
The system was operating with:
- Ubuntu HWE generic kernel
- Realtek out-of-tree DKMS Ethernet driver (r8125)
This combination caused instability under sustained load, leading to transmit queue starvation and kernel buffer exhaustion.
Kernel Verification
To verify the running kernel during troubleshooting, the following commands were used:
uname -r
uname -a
cat /proc/version
ls -1 /boot/vmlinuz-*
During the failure state, the system was running an HWE generic kernel.
After remediation, the system was confirmed to be running:
6.14.0-1017-oem
Corrective Actions
-
Migrated to the OEM kernel:
sudo apt install linux-oem-24.04
sudo reboot -
Removed the out-of-tree DKMS driver:
sudo dkms remove r8125/9.011.00 --all -
Enabled the in-tree driver:
sudo rm /etc/modprobe.d/blacklist-r8169.conf
sudo update-initramfs -u
sudo reboot -
Prevented regression:
echo “blacklist r8125” | sudo tee /etc/modprobe.d/blacklist-r8125.conf
sudo update-initramfs -u
echo “r8169” | sudo tee /etc/modules-load.d/r8169.conf -
Disabled Energy Efficient Ethernet (EEE):
sudo ethtool --set-eee enp191s0 eee off ← change enp191s0 to your Eth name -
Preferred Ethernet over wireless using route metrics:
Ethernet metric: 100
Wireless metric: 600
Result
After migrating to the OEM kernel (6.14.0-1017-oem) and using the in-tree Ethernet driver, sustained transfers stabilized, kernel buffer exhaustion ceased, and connections were no longer terminated. Wireless networking now functions strictly as a fallback.
Key Takeaway
This issue was caused by a kernel and out-of-tree driver interaction under sustained load, not by MTU configuration or missing drivers. The correct fix was switching to the OEM kernel and the in-tree Ethernet driver.