Part 3 This is the third part of The Register FOSS desk's roundup of some of the more memorable missteps and could-have-beens from the beginnings of the microcomputer industry until today.

In part 1 we looked at some of the bad decisions of the eight-bit era, and in part 2 the more expensive mistakes of the 16-bit era.

The 32-bit era was a little different. With few notable exceptions, most contenders from earlier eras sank beneath the PC tide. In the 1990s, the hitherto dull and businesslike Intel computers, which had been office equipment up until then, about as exciting as a staple remover, caught up. They gradually gained the fancy graphics and multimedia features that had previously been limited to proprietary systems.

Much of this was driven by the switch from the reign of DOS to that of Microsoft Windows. How that happened is itself one of the most catastrophic business errors of all time. But other things happened first…

The Apple transition that never happened

In the last installment, we talked about the Apple IIGS and the excellent Megahertz Myth article on its rise and fall. In the section titled "Arm, Möbius, and The Road Not Taken," this history also mentions an even more obscure cancelled Apple computer: its 1985 Möbius prototype.

Apple's Advanced Technology Group was working on an Arm2-based computer that used the RISC chip's raw performance to emulate the 6502-based Apple II, and the 65C816-based IIGS, and even the 68000-based Macintosh, all in software. Co-developer Tom Pittard mentions his work on it. So does Mobile Unleashed [PDF], a 272-page history of ARM, in chapter 3.

Mobile Unleashed refers to another researcher on the project, Art Sobel, who said:

This is how Arm ended up in the first Apple handheld, the Newton.

If Apple had pressed ahead with Möbius, it could have transitioned from the 68000-based Macs directly to Arm processors in the 1980s, a quarter of a century before it did so in our timeline. The chance was right there. It could have bypassed both PowerPC and Intel generations, and two very difficult transitions.

What might have Be(en)

More than a decade later, unable to sell enough of its proprietary Arm desktop computers to keep its unique platform alive, Acorn cancelled the Risc PC 2. One of the rarely mentioned things about the machine codenamed "Phoebe" was that it supported multiple CPUs.

A year earlier, Apple turned down Be Inc's $200 million asking price and instead spent $427 million on Next Computer. This left Be frantically casting about for a future. At the time, The Register said: "BeOS is a very portable OS. The port from PowerMac to Intel x86 took only a couple of weeks."

Perhaps x86, its OS market dominated by wealthy corporate behemoths, was the wrong choice. One of the defining aspects of the radical BeOS was its support for multiple processors. The original BeBox hardware had an unusual RISC chip, the AT&T Hobbit. By the time it launched, it had twin 66 MHz PowerPC 603 processors. If only a transatlantic partnership had happened, by 1998 BeOS could have been running on cheap, quiet, cool-running SMP Arm workstations from Acorn… at a price point (and power draw, and thermal envelope) that no dual Intel or AMD box would reach for decades.

Meanwhile, in beige-box-land…

Of course, the PC industry saw more than its share of big missteps too. In fact, many of them are defining pivot points.

In the 1970s, Digital Research (DR) dominated the eight-bit business micro industry with the CP/M, which turned 50 this year. It missed out on the deal to provide the OS for IBM's new Personal Computer, but its first attempt at a comeback was with the multitasking Concurrent CP/M.

Like Sinclair with the QL, DR misjudged what the market wanted. Sinclair recognized multitasking as a killer feature, and DR had it, but the company chose to pursue the declining multiuser market. Concurrent CP/M [PDF] became Concurrent DOS, which became Multiuser DOS.

Back in 1985, the 286 version of Concurrent DOS could even multitask DOS apps, as The Register noted when OS/2 turned 30. DR was thwarted again, though, when Intel removed the CPU feature that CDOS/286 needed before launching the complete version of the 80286.

In 1986, Compaq launched the first 32-bit PC, the Deskpro 386. The clue is in the name: It was a professional desktop machine, not a server. It catalyzed a market in add-on programs to bring multitasking to DOS, especially Quarterdeck's DESQview.

DR was already selling multitasking PC OSes before the first Amiga launched. If it had offered a cheaper single-user version, ideally with text-mode windowing, it could have sold Concurrent DOS/386 to high-end PC users… but it ignored the growing market for multitasking for DOS power users, and instead chased the blind alley of multiuser PCs with dumb terminals.

DESQview itself also led to what could have been a sea change in the PC industry, if Quarterdeck had moved faster. For years, the company talked about its future product, DESQview/X, which turned DOS into a networked, X11-driven GUI OS. DV/X moved PCs in the direction of UNIX workstations. It could run UNIX apps in a DOS window, but also let UNIX machines on the same network run DOS in a window - hosted, remotely, on real x86-32 hardware. Sadly, it shipped only after Microsoft released Windows 3, and moved the PC market in a different direction.

IBM snatches defeat right out of Digital Research's jaws

As we explained when looking at a pre-release version of OS/2 2, it was the 80286 chip that killed OS/2's chances of being a successor to DOS (as well as those of Concurrent DOS).

Although as far as we're aware, neither IBM nor Microsoft has ever officially confirmed this, the general belief is that Microsoft wanted to target the new OS at the new Intel 80386DX, which had hardware-assisted multitasking for real-mode 8086 apps - that is, DOS apps. Indeed, a Microsoft 386-mode multitasking prototype, codenamed "Football," leaked, and the OS/2 Museum got it running.

IBM, however, was selling lots of 80286-based PS/2 computers and had promised purchasers that they could run its new operating system. Virtually Fun specifically accuses the IBM PS/2 Model 60 as being such a lucrative success that IBM insisted on keeping its promise to its 286 customers - that their machines would run the new OS.

Had OS/2 1 targeted the 386 from the start, it could have matched its billing as "a better DOS than DOS" in 1987, rather than in 1992 when it was all too late. If the new OS had been as successful as both companies hoped, IBM could have shipped every Model 60 owner who wanted OS/2 a free motherboard upgrade. Instead, it lost control of the PC industry it created.

This expensive flop led to the split between IBM and Microsoft. If they had continued to work together, Windows 3 might never have happened. Its huge success in turn led to Windows NT in 1993, which redefined the PC software industry.

Aside from being a usable desktop, Windows NT made a competent enough server operating system, and right from the start, NT 3.1 included TCP/IP support as standard. (For OS/2 Warp, it was a premium edition, called Warp Connect.)

That sailed over the head of Novell, the undisputed ruler of PC networking since the 1980s. It failed to spot that TCP/IP was the next big thing. Instead, it bet the business on network directories - a decade before they became important.

With both OS/2 and Netware of waning importance, Novell and IBM had big problems. Netware 4 was more complicated than previous versions to get running, while NT was easier than ever. NT also made a useful application server - this vulture was installing internet gateways and email servers on NT in Netware sites in the 1990s. The two could have partnered up: Netware for OS/2 combined Novell's high-performance file-serving with OS/2 Warp's GUI and abilities as an applications server. It was a capable combination. Instead, with Netware 5, Novell added a GUI, SMP, memory protection, and other features into Netware itself. ®

https://www.theregister.com//2024/09/04/where_computing_went_wrong_feature_part_3/