[awful]

In retrospect it would surprise some at how advanced the B5000 arch. and implementations were, virtual memory and concurrent, parallel, quorum processing and so on. I learned the stack based, superscalar B7800 CPU to the gate, bit level; single clocking and chasing the op codes and small code through the entire machine including the IO and stacks in memory. I was supporting 4x4 (cpuxio) B7800 and a 2x2 B6700 and all peripherals including hundreds of disk spindles at a 24x7 site. As well as the IOM and then new 4k chip memory infrastructure. Unfortunately I was not able to spend any time at the OS level though, running an early C(?) and Cobol compilers maybe. Spent months taught by Jerry Jackson at the Paoli plant. And sadly no longer have any tapes or docs. My last experience was confguring host devices allowing high speed direct parallel communications between disparate running systems. I cannot describe how powerful the exposure to these technologies were for my future understanding.

[zxcvbn4038]

You can almost always go back to mainframe for inspiration - back in the day they pretty much solved every problem you can think of. It has just taken 40-60 years to be able to do the same thing on consumer class hardware.

Last I checked there was still no mechanism where I could bill back resource usage in a multi-tenant kubernetes cluster so I think there is still work to do.

[pjmlp]

We aren't fully there yet, some of them like Burroughs, were already using quite safe systems programming languages.

ESPOL/NEWP were most likely the first ones to have unsafe blocks, as I think JOVIAL lacked them.

[krylon]

> bill back resource usage in a multi-tenant kubernetes cluster

I feel a bit dumb for asking, but what about process accounting? I'm not sure if it can deal with RAM and I/O, but it's better than nothing. (https://linux.die.net/man/8/sa)

[rbanffy]

> bill back resource usage

If you are not oversubscribed, you can just count the allocation.

Maybe performance counters can help. How long does it take to read op counts prior to a context switch?

[B1FF_PSUVM]

Computer architecture books of the 1980s mostly featured the B5000 and the CDC 6600 as poster children for what could be done with hardware. Some IBMs too, of course.