What ever happened to Tandem and NonStop OS 1973

i've frequently claimed that the 801-risc project in the 70s

was an adverse reaction to the future system project ... and was looking at doing the exact opposite.

future system was going to be a complete replacement for 360 with an extremely complex hardware architecture. it was end before ever being announced ... but while it was in progress, consumed an enormous amount of corporate resources ... in addition to diverting-eliminating most work on possibly alternatives during the period.

at the time, there was a long playing cult film down in central sq (had been playing every night for something like 13 years at the time). i made some unflattering comparisons between the film and the FS project regarding the inmates being in control of the asylum

801 was supposedly going to make an enormous amount of software-hardware trade-offs, drastically simplifying hardware features and implementations. A drastically simplified hardware would also help an extremely complex compliler and software infrastructure do things like maximize performance optimization (i.e. enormous strides in software state-of-the-art and tools would more than compensate for simplified hardware implementation).

there was an advance technology conference in POK in the 70s, where a group of us presented a design for 16-way smp 370 ... and the 801 group presented 801, cp.r, pl.8, etc. one of the 801 people objected to the 16-way smp presentation ... claiming to have looked at the standard vm370 product kernel code and found that there were no control blocks that had fields that provided support for smp operation. the counter argument was that it was smop ... you write the code for it ... various past posts about smp support

and some specific posts about a 5-way smp project involving heavy microcode changes

so when the 801 people presented ... I observed that most of the cp.r and pl.8 code hadn't even been written yet (aka the amount of proposed code to be written for 801 was possibly a couple orders of magnitude larger than the additions to the 16-way smp code that needed to be added to the 370 kernel)

--