It's really fun to recover a system that's designed to never go down. Some years ago I was working with Tandem/Compaq and they had a web server on the NonStop Himalaya platform. Long story short, said server had its first experience with real world web traffic at scale when we were delivering coverage of an international sporting event, and Things Went Poorly.

The onsite Tandem engineers had never been faced with a NonStop setup that just...stopped (actually it was worse than that). Bear in mind this was the company that used to do demos by randomly unplugging hardware and the transactions kept chugging along. We ended up punting coverage to a smaller flat version of the site served off Compaqs running IIS while we cleaned up the mess.

Now, Tandem shouldn't feel bad. After all, Cisco had guys in the hosting facility trying to keep the then-new PIX series up and running.

> It's really fun to recover a system that's designed to never go down

The thing with systems designed to never go down is that you still need to be sure the recovery process works well. Never going down is just half if your nines.

> served off Compaqs running IIS

Yikes!

> Some software was written for Mainframes that effectively NEVER reboots. So you have to modify it to support reboots easily

Run it in a VM instance that can be snapshotted/checkpointed, so that any reboot turns into "suspend and resume from checkpoint"? Or are there any pitfalls to just doing that?

Yes that can help, but VMs are still either Linux or OSX or Windows, and they all require security or other updates that will eventually need to reboot the system. Also the processors are different.