If memory serves, the load average calculation was the only use of floating point in the BSD 4.2 kernel. Reworking the algorithm to get 4.2 running on the National 32016 before the floating point co-processor was available was my first time inside the kernel. Fun times.

I'm one week older than the Unix load average!

Or perhaps: my reported uptime is guaranteed to be longer than any Unix system's. :-)

Really, you have never been rebooted?

no. he just sleeps :)

Barring some kind of medical emergency requiring CPR, it's safe to say quite a lot of people have never been rebooted.

Also of interest is "Understanding the (original) meaning of Unix load average"

https://utcc.utoronto.ca/~cks/space/blog/unix/LoadAverageMea...

Given how hard it is to determine what is actually causing high load sometimes I'm surprised we haven't evolved more easily actionable metrics.

You mean like per process time spent active on a CPU?

You could poll that ever so often and log the results to a database if you actually cared; but the act of measuring changes the results...

Or as Bryan Cantrill put it:

  When you're in top, what's the top process?
  top!
  "Hey man, good news, I've found the problem!
  I'm the problem!!
  Who would've thought that?!
  Don't run me!"

taskman.exe is the one exception to this that I've seen, which I've wondered about ever since I switched over to non-Windows platforms. Is there something about NT that makes performance metrics unusually cheap, or is it just "lying" and not reporting that time anywhere?

That's curious. Running Win8.1 and Task Manager lists Task Manager with subprocess Task Manager, anyway, using variably 0 to ~1.0% CPU, 0.3% memory. How "truthful" these stats are, well I guess that's another question altogether...

Taskman sorts differently. Top sorts by the most recent active process, while taskman sorts by the process that has used the most CPU on average lately.

You can choose what to sort by in both.

I mean finding the actual cause of a high load number. There's no way to break it down by process, or even find out if it's because of IO or network, etc. without using other tools.

Fellow poor soul, over the last 10 years whenever a server has high load I've got angry about the fact that load avg is not actionable, and exactly that "or even find out if it's because of IO or network"

ps r -A shows processes in the run queue.

And don't forget that load average means something different between each OS