At university in the 90s some friends and I ran our own Linux server. It was a 486 or early Pentium and we hooked it up to the university network in a post grad student's office who was happy to just keep it running under his desk.
We even got the campus sysadmins to give us a proper edu domain name. It was a more open and different time and ethernet still meant coax cables with T connectors and terminators.
We were running pre v1 kernel on slackware and it was all installed from floppies. We used it as a web server, coded and played muds, read newsgroups and mail etc. I think tin and pine etc. we easily had 20 users using it from the computer labs.
Anyways the computer kept dying or freezing occasionally. Still early Linux. And the office where it was kept wasn't always open and we didn't have a key.
Being electronic engineering students we built a whole circuit with a PIC controller which plugged into the parallel port. We wrote a watchdog daemon which would keep pinging this dongle. And the firmware on the PIC would check for these pings.
If the server died the pings would stop and the dead man's switch dongle was wired directly into the hardware reset button of the PC.
Worked like a charm for 4 years. And apparently worked for another 5 or 6 after I left.