Quote: "stability that makes it fail (statistically speaking) once in a billion years"
...I got clients that can kill a Midrange machine dead in 24 hours, I had one last week.
They had just one query running, which was destined to write just over 10 billion records, because they failed to add a JOIN for one file. 24 hours later, it was still running, no-one noticed, and the disk was at 98.5%. The System Operator message queue was screaming at them...no-one noticed. The files were set to NOMAX, which basically means they'll keep expanding forever. There was no facility for the server to IPL itself. They were literally less than an hour away from bringing the entire business to a grinding halt.
A couple of weeks ago somebody lost their machine to a power-cut...they had no UPS!
When IBM say their machines are stable, they assume their customers are competent. Put the two together and they are stable. But this country still suffers from the legacy of accounts departments doubling as an IT department.
[EDIT] For those that don't play with Midrange machines, getting them back up after a failure is like trying to get a beached whale back in the water.
BatVink