I'm turning up a server. I've been meaning to get an R720 for one reason or another for ages, and I finally had a reason - a new NAS I want to build for video editing and archival - and RePC just happened to get one in for a good price.

It has a pair of Core i-series Xeons from 2012 that I'm going to upgrade to some low-voltage models with more cores from a year later, 32GB of RAM I can upgrade for cheap (but probably won't need to), 12 drive bays, and a premium remote management package.

It's a great machine, which I will probably be underusing, but I just fired it up for the first time since I got it home and got this bone-chilling error message.

Hats off to the firmware dev who actually thought to check for this scenario and handle it explicitly. If someone like this was working at Genband or Cisco, we wouldn't have as many widespread network outages when someone puts in a config directive that null-routes an entire country or class-A network.

"Warning: Committing this config will drop 1,853,980 active sessions. Continue?"

"Warning: Committing this config will change the route for 95% of call paths. Continue?"

But wow. What a frightening message to receive if you weren't expecting it. It actually sent shivers down my spine, and I did expect it.

There's a joke in one of my friend groups, folks who deal a lot with the more uncommon vintage computing stuff (VAXen, late-80s Sun, etc.) than I do.

See, in the 80s and 90s, very few home users had LANs, but minicomputers (and similar hardware) were all networked. Networking is very old technology, and has arguably only gotten less usable over time, but a network in 1990 might not have looked all that different from a modern corporate one. Bootp has changed names and hats, but DNS is still DNS, NTP is still NTP.

As a result, virtually all 1980s minicomputers were dependent on a sprawling mess of remote resources, and on boot, they would expect to be able to reach them, because they were useless without them. Networks used to be much slower and less reliable, too, so timeouts were set to absurdly long values. The end result is that firing up a Sun 2 often looks something like:

Contacting BOOTP server    . . . . . . . . FAILED
Verifying hostname         . . . . . . . . FAILED
Mounting shares:
196.3.55.5                 . . . . . . . . FAILED
196.3.55.12                . . . . . . . . FAILED
90.1.9.15                  . . . . . . . . FAILED
85.85.106.45               . . . . . . . . FAILED
100.33.44.22               . . . . . . . . FAILED
Starting web server        . . . . . . . . FAILED

...and so on, each one taking upwards of 90 seconds, possibly five minutes or longer.

It's still the case now that computers do not sensibly handle this situation. There is no "do I actually have a network?" check built into anything, so when a machine fails to get an IP address, all their services still attempt to fire up. Of course, nowadays many things will get an internal "network unreachable" error, fail fast, and not waste all that much time, but certainly I've seen long boot times on servers when a network was down or misconfigured.

And certainly, in 1989, it was assumed that IT (except they were called "IS" back then) had correctly configured your machine before turning it on, and if not, that they would have brought it up in a low runlevel to fix it first.

So you power up a machine that was built in 1989 and last powered on in 1993, and the system just sits there, refusing to accept that the world it knew, the very last thing it remembered before going into its long slumber, is gone. Disappeared, in the blink of an eye, probably in its entirety, and probably forever.

So many relatives, teammates and acquaintances that don't pick up the phone anymore. The machine just continues down the list, letting each number ring, ring, ring. Surely the old nameserver is still around. How can the planet keep spinning without him?

The computer has to make it through all the stages (bargaining: maybe the secondary NTP server is up?) and reach acceptance before you can do anything with it. Before it's ready to enter the world of the now. It's always, genuinely, sad.

Not just because we're anthropomorphizing it - of course we are, but it's also sad because all those machines shouldn't be gone. We have the space in this world, and the enthusiastic curators, to have kept all. of. it. The sum total volume of every single Unix workstation, every minicomputer, every mainframe ever made probably amounts to less than a single decent size warehouse that is currently sitting empty. These machines were made in the thousands, hundreds, or even dozens. Shredding them for gold and steel scrap was pointless. Made someone a few bucks in the short term, "recovered" materials that were not rare (assuming they were recovered at all.)

It's tragic. A failure of our culture to safeguard its own history, simply because computers are not taken seriously as instruments or tools. People keep objectively bad cars from centuries ago, of which 20 were made, and meticulously restore them. Meanwhile, most computers ever made have been reduced to powder, and when we get some rare specimen from decades ago, every timeout as it boots is a reminder of another machine that we'll never get to meet.

Of course, we are still anthropomorphizing them. It's hard not to imagine the panic rising as it slowly realizes that nothing is out there anymore. But I've never seen a computer just come out and say it like this one.