Ars Technica today published an article titled “Half an operating system: The triumph and tragedy of OS/2“. Although very interesting, unfortunately the article to a significant extent engages in what can best be called fantasy history, which causes the text to contradict itself. And while it attempts to present a coherent view of the history of OS/2, the article is more a hodgepodge of facts/products/trends (PS/2, NT, PowerPC) picked at random to support an argument.
It also presents the old “geniuses at Microsoft vs. dumb IBM” story which is increasingly difficult to justify as time marches on. IBM is the bad guy, overly bureaucratic and incompetent. Yet listening to old IBMers a somewhat different picture emerges, with Microsoft being a disorganized, sloppy company producing poorly documented and buggy software. As always, there’s some truth to both sides of the story.
Let’s start with some of the inaccuracies. When OS/2 was conceived and designed, it wasn’t called OS/2 at all. In fact there is no known evidence that the name OS/2 existed before 1987. The name OS/2 had a lot to do with PS/2, but the OS itself not so much.
Then there’s this gem: “Today we take virtualization for granted […], but in 1985 the concept seemed like it was from the future.” Or at least it would have, if time flowed backwards and the future was 1972—see IBM’s own VM/370.
Then comes the the fantasy history. The 286 sucked—okay, that’s not an unreasonable argument. Therefore, OS/2 should have been designed to run on a 386. Now, hold on there. First off, 386 PCs didn’t even exist before 1986, and only covered a tiny fraction of the PC market when OS/2 was released. OS/2 was meant to be a mass market product, not a niche operating system guaranteed to be ignored by major ISVs.
Then there’s the minor detail that it took Intel a while to build 386s that were capable of running 32-bit code in reality instead of just on paper (the famous double-sigma CPUs).
But the biggest argument against 386-only OS/2 is given in the article itself: Memory. If 16-bit OS/2 1.x was considered to be too memory hungry, exactly how was bigger and slower 32-bit OS/2 supposed to solve that problem?
Yes, it took Microsoft and IBM much longer to complete 32-bit OS/2 than originally planned, but given that even in 1992, OS/2 2.0 was often considered to be too memory hungry, it’s hard to see how releasing 32-bit OS/2 years later would have improved things.
One of the biggest issues with the article is the claim that good DOS and Windows application support prevented native OS/2 applications from being written. That’s probably why Windows 95 didn’t support DOS and why Windows NT didn’t bother with a 16-bit Windows subsystem… right? The truth is that good DOS and Windows support was a necessary evil and without it, OS/2 would have fared just like the other PC operating systems with no or minimal backwards compatibility. NeXTSTEP, Solaris, UnixWare, and the like.
The article several times falls into the trap of assuming that people working for the same company must necessarily share the same goals. That wasn’t true of Microsoft (Windows 9x vs. NT), and it definitely wasn’t true of IBM, where several hardware groups fought each other and didn’t necessarily see eye to eye with software divisions. Once Microsoft quit OS/2, IBM was in an unfamiliar and difficult position of having to support competitors’ hardware, and competing against Microsoft on IBM’s own hardware. The inevitable internal infighting didn’t help OS/2, but that doesn’t mean IBM as a whole behaved irrationally.
Microsoft’s anti-competitive tactics are not mentioned, despite being very well documented. Arguably once Microsoft got a stranglehold on OEMs and forced them to preload DOS and Windows (securing a guaranteed source of cash flow), the game was over and no one was going to change that, not IBM and not anyone else.
Windows NT is presented as some kind of mythical software that magically ships a bug-free version 1.0. Most people don’t remember NT 3.1 as “multiplatform, ridiculously stable and fault-tolerant”, but rather as bloated, slow no-show that hardly anyone ever ran (NT 4.0 was certainly a different story!). And calling the PowerPC version of NT a “consumer operating system” seems to stretch reality more than a little.
Looking back from 2013, the answer to the question of which company was smart and which was stupid is less and less clear. IBM cut its losses with OS/2 and later on, sold its PC Company while it was still worth something, rather than waiting like HP and Dell. The article says that “IBM threw in the towel on OS/2 and then on PCs in general” as if that were a dumb thing to do… but the reality is that getting out of the PC hardware market when IBM did was was a very smart move, and killing OS/2 was more or less the only option IBM had, especially after the spectacular failure of Workplace OS.
IBM may be 100+ year old, but you don’t get to be that old without having learned a trick or two. Microsoft bet the farm on Windows, but Windows is not sexy anymore and making money with operating systems is increasingly difficult. In another decade, we might more clearly see who was really smart.