A Universal Computing Truth

The IT Crowd turned “Have you tried turning it off and on again ?” into one of the best-known jokes in IT.

The annoying thing is that it’s also good advice.

Rebooting fixes a surprising number of problems, from misbehaving applications and stuck services to machines that have simply accumulated too much state since they last started.

But computers don’t get tired, and they don’t need a good night’s sleep.

So why does turning them off and on again actually work?

Computers Accumulate History

Computers don’t have countdown clocks to instability. They don’t get tired simply because they’ve been running for a long time.

But since the last reboot, things have happened:

• Programs have started and stopped.
• Services have been restarted.
• Updates have been installed.
• Drivers have loaded and unloaded.
• Files have been opened.
• Network connections have come and gone.

Ideally, everything cleans up after itself. In reality, resources remain allocated, files stay locked, connections hang around, and applications don’t always return the system to exactly the state they found it in.

Over time, that history accumulates.

A reboot doesn’t need to understand any of it. It simply throws away the temporary state and starts again from somewhere predictable.ystem performance, making a reboot seem refreshingly simple.

“Reboot has one enormous ‘it doesn’t care’. Just stop everything, throw away all that temporary state, start the operating system again from somewhere predictable.”

Resources Don’t Always Come Back

Applications need resources: memory, CPU time, file handles, network connections, threads, and whatever else the operating system makes available to them.

Ideally, an application gives those resources back when it’s finished with them.

Ideally.

In reality, software has bugs. Resources leak, processes get stuck, handles remain open, and old code keeps doing things nobody quite remembers anymore.

The operating system can clean up a lot of this, but not everything always returns to a nice, predictable state.

And that’s where rebooting has one major advantage: it doesn’t care.

Servers, Mainframes, and Uptime

Servers have a different relationship with rebooting. Uptime can become a badge of honour, particularly in the Unix world, where machines running uninterrupted for years have traditionally earned bragging rights.

Mainframes handle things differently. Mainframes can run workloads in separate partitions and restart those partitions without necessarily taking down the entire machine.

And, of course, mainframes don’t reboot.

They IPL.

That’s an Initial Program Load, which is basically a reboot with a much more impressive name.

Why Not Just Restart the Service?

Obviously, rebooting the entire machine shouldn’t always be necessary.

If the application is any good, you stop the service, start it again, and carry on.

But sometimes a service doesn’t stop cleanly. The service may leave resources allocated, dependencies may not recover properly, or something gets stuck between running and stopped.

Windows has a particularly reassuring description for this state:

Stopping.

Which it can apparently continue doing until the heat death of the universe.

The Reboot SLA

So should computers be rebooted regularly?

There isn’t one schedule that makes sense for everything. A laptop, an application server, a Unix system, and a mainframe partition all have different workloads and availability requirements.

But perhaps systems should have something resembling a reboot SLA.

Not a rule saying reboot every Tuesday at 3 AM, but an agreement that maintenance windows actually exist and that systems are occasionally allowed to use them.

Because five nines may not count scheduled downtime.

That doesn’t help much when nobody is ever allowed to schedule the downtime.

On and On Again

Rebooting isn’t magic. It’s just a very effective way of getting rid of accumulated state and returning a system to somewhere predictable.

Sometimes you restart an application. Sometimes you restart the service. Sometimes you reboot the operating system.

And sometimes you IPL the partition and give it a much more impressive name.

Different systems. Different schedules. Same principle.

Computers accumulate history.

Nothing in a server room should be left running unattended for years.

That’s how you end up with a goth in the closet.