I worked on Tandem platforms (using the Guardian O/S)
from 1986 until about 1994, primarily for International
Funds Transfer banks.
In all the time I worked on Tandem there was no experience of the system
being "down". Not ever. That includes development environments with neanderthal
developers doing everything they could to bring it to its knees.
Perhaps more important than their non-stop ability, was their implementation
of transactions (Transaction Monitoring Facility). Even shitty developers
couldn't get it wrong....largely due to their mature model of requester/server pairs. In the 90's Tandem started to empasise their "fault-tolerant" qualities
more than their "non-stop" - partly because the nightmares experienced by other
manufacturers were starting to get cleaned up.
The amount of money which runs over Tandem hardware is quite staggering.
Anecdote.
I fixed a problem for a bank early one morning in Manhattan.
They came to me and said that they put the fix in production and that they were now "at 6 million dollars" of communications with CHIPS (Clearing House Interbank Payments System). I looked at my watch and said "hey....that's great and its not even noon yet". The bank's representative smiled at me and said
"that's 6 million a second". (Even that figure is probably out of date now).
It was highly scalable and rock solid. One of things going against it
was that it was expensive and "niche". The availability of Tandem
developers tended to swing between extremes. Sometimes there would
be none available....other times the market would get (relatively) flooded
when (say) a large credit card company decided to cancel a big project.
Its easy to think that the Tandem appraoch is/was dying but that's not so.
One of the best web/application servers is Bea Weblogic. This drew heavily from
the Tuxedo product. Tuxedo drew heavily from Tandem.
-Mike