by Mike Loukides
The rise of UNIX in the early 80s has almost entirely to do with economics and hardware -- and the two are inextricably linked. It has almost nothing to do with "open systems" or any of the ex-post-facto justifications or explanations that have circulated.
Quite simply, beginning with Sun Microsystems' first workstations, UNIX was the software that ran on the best low-cost available hardware. For a given price, you could get much more performance than was possible before. The proprietary vendors (DEC, etc.) got behind the performance curve, and stayed behind it. Pretty soon (by 1985, if not 1986), there was absolutely no reason to buy DEC hardware, except to run VAX/VMS -- and, if VAX/VMS was making you suffer sub-par performance, that reason wore pretty thin. UNIX, in whatever flavor, became the leader because you could buy faster hardware, cheaper. You could put a workstation on everyone's desk, and still have spent less than a mainframe would have cost.
But why did all those aggressive little start-up vendors go after UNIX? Well, in fact, they didn't: Apollo is a notable exception. But most start-ups relied on UNIX ports. Why? This doesn't have anything to do with any intrinsic virtue in UNIX -- rather, it's purely economic. For a relatively small piece of change ($20K or so) you could buy a complete working operating system, with source. That's nothing compared to the expense and time of developing an operating system from scratch. To make it concrete: Multiflow Computer (an ill-fated supercomputer startup for which I worked) had their first complete (hardware) board-set sometime around August 1986. By October, we had a machine you could rlogin to. By November, UNIX was running stably. By January, we had a product on the market. (And serial numbers 2 and 3 are, I believe, still in service). Of course, before any hardware existed, the software people were working busily with instruction set simulators, etc. But that kind of quick development just would not have been possible if Multiflow had to develop a proprietary OS. Nor could Multiflow conceivably have had the money to develop a proprietary OS. We burnt over $50M in our tragic corporate history, as it was; developing our own system would probably have doubled that.
The story was repeated again and again: Sun, Convex, Multiflow, Masscomp, SGI, Pyramid, Stardent, etc., etc. NeXT is at the end of this list. In short, by 1985 if you were going to have "modern" computers, you were going to have a computer that ran UNIX. Small startups built better hardware than the mainstream vendors, like DEC and IBM; and the small startups couldn't afford the time or the money to develop a proprietary operating system. Soon, the companies with proprietary operating systems even started moving to UNIX, with DEC and IBM being the big exceptions.
Education was another factor in UNIX's growth. Ultimately, this is also an economic factor. Schools were turning out loads of very competent computer users (and systems programmers) who already knew UNIX. You could therefore "buy" a ready-made programming staff. You didn't have to train them on the intricacies of some unknown operating system. It takes a long time to become a guru; with UNIX, you could buy them ready-made. A friend of mine at IBM research said "You may not believe it, but IBM really is committed to UNIX. They've realized that they can't hire people out of school who know anything about their mainframe system. They can't even hire system administrators." For IBM, this was the writing on the wall.
A third factor was evolution. UNIX provided an evolutionary path -- the UNIX of 1994 is significantly different from the UNIX of 1978; sophisticated networking, many utilities and tools, a windowing system, new shells. However, that evolutionary path worked because Berkeley (primarily) and AT&T put a lot of effort into integrating new developments into the basic system. Most other operating systems have remained more-or-less the same from their invention until their obsolescence; they've added incremental improvements, but haven't been able to transcend their origins. And again, this is an economic factor. Users want new features; vendors want to sell new features; developing new features in-house is expensive. Again, Multiflow provides an interesting example: their machine was probably the first standalone computer to have a network interface before it had working serial ports! That couldn't have happened without the BSD networking code.
I haven't mentioned "open systems" or anything like that because, frankly, "open systems" as a concept was invented sometime around 1988 -- as I said before, after the fact. Fact is, running UNIX was a decided disadvantage until sometime around 1988. You could guarantee that any interesting third-party application software wasn't ported to UNIX; you could guarantee that the vendors of these software packages weren't happy about having to port to dozens of slightly different UNIX platforms. Given that people really buy computers to run applications, the world wasn't all that rosy. Eventually, the hardware advantage forced the software vendors to play ball -- particularly in the scientific arena, where customers could exert a lot of pressure on vendors to port software to faster platforms. To some extent, though, UNIX still labors under the "application software" disadvantage (though now, the comparison is to Windows, rather than VMS or VM/CMS). I'd say that the "advantages of open systems" (whatever open systems might be, and that's another essay -- open systems is itself a largely meaningless term) are real, but they were only appreciated afterwards. Corporate users realized "Gee, we have all these UNIX systems, and they're more or less alike, and more or less compatible with each other, and we can hire people straight out of school and put them to work without training them on PrimeOS -- there are some real advantages here."
Where does that lead us, then? My big point is that hardware and applications drive the market; users are going to buy the best hardware at the lowest price on which they can run their important applications. I don't think there's any inherent loyalty to UNIX, much less to "openness," or to anything aside from getting more done for less cost.
What I've been aiming at is to figure out what the historical factors are that allowed UNIX to succeed, and see if we can use them to figure where the market is now, and what's happening next. Some observations:
That's where the market is now. Where's it going? Well, the most important new development is the introduction of Microsoft NT into the picture. To me, that's important because, for the first time since 1983 or so, users will have an alternative to UNIX on the sexiest hardware platform around. As I said, I don't think there's deep loyalty to UNIX in itself; users will buy the best hardware they can, for the best price, and it looks like that will be Alpha. And it isn't a single-platform OS: NT runs on the Intel series, of course, but is also slated for DEC's Alpha, the MIPS series, and other high-performance architectures. Whatever you may like or dislike about it, NT is a fully functional operating system (unlike DOS) that can cover the entire spectrum from low-end to high-end (like UNIX), providing a uniform environment from the desktop to mainframes.
I am assuming that Microsoft understands the importance of high-end applications (numerical analysis, structural engineering, quantum chemistry, the whole works, not just fancy word-processors). It's possible that they don't, but they're a smart company; I don't think that will happen. I would assume that they're out there wooing the developers now.
So, where are we headed? NT is a viable operating system for high-end platforms, which was one of the key factors in UNIX's favor. Even if NT is incredibly expensive to license, the startup hardware vendors that were sensitive to price have largely been killed off; DEC and HP can afford large licensing fees. If anyone releases a cheap, source-licensable version of UNIX, or NT, or anything, you could see a new generation of startups. That would throw the whole ball-game into question. However, there's a good chance that Jobs is right, and that it costs just too much to develop a significantly different hardware platform now.
The educational factor was another important (and ultimately, economic) element in the ascendency of UNIX. That's a little hard to think about. On one hand, NT has an advantage because it's new and interesting; hackers who like new and interesting things will want to hack on it. On the other hand, source code will be hard (or impossible) to come by. Microsoft has made some noises about low-cost source licenses to universities. This would be critical to their success -- in fact, if they do it well, Microsoft would be well positioned to win across the board: high-end, low-end, and middle. Which means that in four years, there will be armies of pre-trained NT hackers. Vendors tied to UNIX will be swimming upstream. The big question, though, is "Will Microsoft do it?" I can't answer that. I don't know if they have the slightest idea about how to make this kind of deal, through there are all sorts of rumors floating around. My guess would be that they'll make a deal, but it will be too restrictive. They won't want to give up sales to university students, so they'll price it weirdly, or have strange strings attached.
Finally, there's the evolutionary factor -- in which, of course, education played a very important role. If UNIX is going to continue to compete, it's critical for it to retain the ability to evolve. NT may evolve, but frankly DOS really hasn't changed much since it was released, and I'd expect NT to follow the same pattern. However, if the UNIX industry degenerates into a bunch of squabbling vendors who talk about cooperation but never cooperate, UNIX won't change much over the next decade, either.
UNIX grew because it ran on great hardware, it had a central role in universities, and it had an evolutionary path. With the demise of small computer vendors, the hardware factor has disappeared; if Microsoft plays its educational cards right, the University factor will disappear. So the big issue for the future is evolution: will a new evolutionary path emerge that allows UNIX to integrate new developments in a more-or-less uniform way? That's what I'm waiting to see.
Mike Loukides is an editor at O'Reilly & Associates.
This document is maintained by John Loomis, last updated 10 August 2006