Hector Martin

headless systems not coming up after reboot because lennart decided that eth0 is not good enough for an interface name

As much as I love blaming systemd, the problem goes back much further to kernel devs making device numbering non-deterministic (asynchronous probe). But until Lennart that only affected rare use case of >1 eth port.

Nope. udev used to have rules making device numbering deterministic (record-on-first-use). then systemd removed that in favor of their insane numbering scheme.

I wouldn't knowingly argue in favor of nondeterminism.

It's still nondeterministic on first install though. Not great if you're doing reinstalls or deploying across hosts. TBH I'm not a systemd fan but I think the new interface naming scheme makes more sense in the long run. And I don't even use systemd on most machines.

This is why legacy IDE had jumpers. Hardware topology conf should always be physical on the hardware to facilitate (re)install, deployment across many machines, moving disk to replacement hw, etc. ID by mac/guid/etc is bogus.

It's eno1, eno2, etc. for internal ethernet when the motherboard firmware is decent and falls back to bus location. The default of using the USB topology instead of a MAC might be what ends up annoying people most since it changes if you switch the port. https://www.freedesktop.org/wiki/Software/systemd/PredictableNetworkInterfaceNames/ …

I turn it off on my laptop, since it almost always only has eth0/wlan0 on boot and anything else is hotplugged. I also sometimes turn it off on trivial systems with a single port and no hotplugging (e.g. embedded). But it's good for anything with 2+ ports, particularly servers.