I was tempted to undo this change in my kernel, however this change works around a bug in the I219 adapter and I didn't want to end up with an unstable system. Related to this bug I read about TCP segmentation offload and changed tso and gso values with ethtool. Eventually "ethtool -K enp0s31f6 tso" off did the trick for me and network performance is back to normal. You might have to change the name of enp0s31f6 to the name of your Ethernet adapter or play with on/off values for tso and gso.
In order to persist this setting after a reboot, I created a udev rule that calls ethtool, maybe this option works for you as well.
I also encountered network performance degradation on my server, however I'm not running Ubuntu but Debian with a self-compiled kernel. With bisecting the kernel, I could track this issue down to this commit between 4.14.2 and 4.14.3: /git.kernel. org/pub/ scm/linux/ kernel/ git/stable/ linux.git/ commit/ ?id=0f478f25d50 cb6464678284a13 f514fefc16e271
https:/
I was tempted to undo this change in my kernel, however this change works around a bug in the I219 adapter and I didn't want to end up with an unstable system. Related to this bug I read about TCP segmentation offload and changed tso and gso values with ethtool. Eventually "ethtool -K enp0s31f6 tso" off did the trick for me and network performance is back to normal. You might have to change the name of enp0s31f6 to the name of your Ethernet adapter or play with on/off values for tso and gso.
In order to persist this setting after a reboot, I created a udev rule that calls ethtool, maybe this option works for you as well.