When one of my tests crashed one of those unprotected mainframes, two guys who were then close to my age now stared at an EBCDIC core dump, one of them slowly hitting page down, one Matrix-like screen after another, until they both jabbed at the screen and shouted "THERE!" simultaneously.
(One of them hand delivered the first WATFOR compiler to Yorktown, returning from Waterloo with a car full of tapes. I have thought of him - and this "THERE!" moment - every time I have come across the old saw about the bandwidth of a station wagon.)
Spending hours wrangling sendmail.cf, and finally succeeding, felt like a genuine accomplishment.
Nowadays, things just work, mostly. How boring.
I recently installed Tru64 UNIX on a DEC Alpha I got off eBay. I felt like it was more sluggish than it should be, so I looked around at man-Pages about the VM (virtual memory, not virtual machine) subsystem, and was amazed how cleanly and detailed it was described, and what insights I could get about its state. The sys_attrs_vm man-page alone, which just describes every VM-layer tunable, gave a pretty good description of what the VM subsystem does, how each of those tunables affects it, and why you might want to change it.
Nowadays, things are massively complex, underdocumented (or just undocumented), constantly changing, and often inconsistent between sub-parts. Despite thinking that I have both wide and deep knowledge (I'm a low-level code kernel dev), it often takes me ages to figure out the root cause of sometimes even simple problems.