> SCO Unix on a 25 MHz 386sx with 8 MB and another 6 MB on an add-in card
OK, SCO Xenix not SCO Unix, but I deployed production Xenix boxes with 4MB of RAM on 80386DX processors.
It ran well and supported 4-6 users no problem. As I recall, one customer had 8 users and they needed a RAM upgrade. I think, dimly now as this wasn't my department, it was eye-wateringly expensive. Something in the region of £5,000 to £6,000 (at the time, $10K+).
We did deploy one box with just 2 MB of RAM, but that didn't work well once a few users on terminals logged in. They had to upgrade the RAM.
FWIW these were 32-bit 386 machines, but with no CPU cache – the lower-end IBM PS/2 Model 80 variants of the time, with 16 MHz and 20 MHz CPUs. (The top-end 25MHz Model 70 had a small SRAM cache for the CPU. It cost over £10,000 with no screen, keyboard or mouse, and my Acorn Archimedes A310, which cost me £800, absolutely smoked it: it was about 4x as fast.)
That's when I knew Arm would eventually eat x86: in 1989. It's finally happening now.
But I was running PC DOS and DOS software in a window on my ARM desktop in 1989, in pure software emulation, and it was sluggish but usable. CPU was equivalent to a 2-2.5 MHz 8086, but disk performance was better than a screaming fast SCSI disk, so it balanced out.
Ironically Bill Gates was big into UNIX, see his Xenix interview, and had they not gotten lucky with the whole MS-DOS deal, maybe they would have kept Xenix and who knows how that would have turned out.
Xenix was also my introduction to UNIX.
However due to our school resources, there was a single PC tower running it, we had to prepare our examples in MS-DOS using Turbo C 2.0, and API mocks, and take 15m turns at the Xenix PC.
> had they not gotten lucky with the whole MS-DOS deal, maybe they would have kept Xenix and who knows how that would have turned out.
Oh, absolutely, yes. It's one of the historical inflection points that's visible.
My favourites...
• MS wanted to go with Xenix but DOS proved a hit so it changed course.
• DR had multitasking Concurrent DOS on the 80286 in 1985, but Intel's final released chip removed the feature CDOS needed, so it pivoted to FlexOS and RTOSes, leaving the way open to MS and OS/2 and Windows.
• MS wanted OS/2 1.x to be 386-specific but IBM said no. As a result, OS/2 1.x was cripped by being a 286 OS, it flopped, and IBM lost the x86 market.
• Quarterdeck nearly had DESQview/X out before Windows 3: a TCP/IP enabled X11-based multitasking DOS extended that bridged DOS to Unix and open systems... but it was delayed and so when it appeared it was too late.
* GNU discussed and evaluated adopting the BSD kernel for the GNU OS, but decided to go with Mach. Had it gone for the BSD kernel, there would have been a complete working FOSS Unix for 386 at the end of the 1980s, Linux would never have happened, and Windows 3 might not have been such a hit that it led to NT.
I got whole series of articles out of this, titled in honour of Douglas Adam's fake trilogy about god...
Never saw one of those. Tandy computers did exist in the UK, and even here on the Isle of Man there was a single Tandy's store. (They weren't called "Radio Shack" here.) But while they sold lots of spares and components and toys, they didn't sell that many computers.
> I had kind of the reverse feeling: when the 486 came out, I knew those expensive SPARC and MIPS workstations were all doomed.
Well, yes. Flipside of the same coin.
Expensive RISC computers were doomed. Arm computers weren't expensive back then: they were considerably cheaper than PCs of the same spec. So for a while, they thrived, then when they couldn't compete on performance they moved into markets where they could compete on power consumption... which they then ruled for 30 years.
OK, SCO Xenix not SCO Unix, but I deployed production Xenix boxes with 4MB of RAM on 80386DX processors.
It ran well and supported 4-6 users no problem. As I recall, one customer had 8 users and they needed a RAM upgrade. I think, dimly now as this wasn't my department, it was eye-wateringly expensive. Something in the region of £5,000 to £6,000 (at the time, $10K+).
We did deploy one box with just 2 MB of RAM, but that didn't work well once a few users on terminals logged in. They had to upgrade the RAM.
FWIW these were 32-bit 386 machines, but with no CPU cache – the lower-end IBM PS/2 Model 80 variants of the time, with 16 MHz and 20 MHz CPUs. (The top-end 25MHz Model 70 had a small SRAM cache for the CPU. It cost over £10,000 with no screen, keyboard or mouse, and my Acorn Archimedes A310, which cost me £800, absolutely smoked it: it was about 4x as fast.)
That's when I knew Arm would eventually eat x86: in 1989. It's finally happening now.
But I was running PC DOS and DOS software in a window on my ARM desktop in 1989, in pure software emulation, and it was sluggish but usable. CPU was equivalent to a 2-2.5 MHz 8086, but disk performance was better than a screaming fast SCSI disk, so it balanced out.