How the Linux-vs-BSD culture clash looked in the 1980s/1990s

[liam_on_linux]

(Repurposed HN comment.)

The BSD/Linux thing was there right from the start, but it was more complicated than a simple us-vs-them. The thing is that there were a whole bunch of competing commercial Unix-like OSes in the 1980s.

But there were other prejudices as well.

In Proper Grown-Up Unix terms, PCs were toys, poorly-made weird little things that were no more than office equipment. So nothing worth using ran on the 386.

There was no local bus yet, no IDE or EIDE, slow AT expansion bus, no processor cache, and so on -- meaning a forest of proprietary or semi-proprietary extensions and buses and special slots. This opened up a market for a vendor to port to Brand X PCs and Brand X's own weird storage and display.

Enter Interactive Corp, which tried to combat this, and worked on Unix ports for various vendors' hardware. Expensive OS for expensive machines.

And there was SCO which wasn't proud, wasn't fancy, ran on commodity kit, and didn't try to be a general purpose OS like that white lab-coat brigade expected. So SCO Xenix worked, and you could run apps on it, but in the box there was no C compiler, no networking, no X11, nothing. It was a runtime-only OS and it was still expensive.

Everyone sneered at it but it did the job. I put in a lot of it.

Then if you weren't paying, someone else was who would never see the word "Unix", there were all the vastly expensive RISC boxes with their vastly expensive expansions and vastly expensive -- well, everything. Sun, HP, DEC, IBM, SGI, loads of company would sell you rooms full of workstations, single-user minicomputers with big screens. They cost as much as a house.

Actual BSD ran on actual minicomputers that cost as much as a small street of houses and those dudes wouldn't even look at PCs.

Which left a market for enterprising vendors squeezing Unix-like things onto low end kit.

Various flavours of BSD, including BSD/OS; SCO Xenix in both 286 and 386 versions; Interactive 386ix; several vendors' own-brand licensed Unixes, including Dell, later, an official Intel one that mainly ran on Intel's own pizza-box workstations.

And all the proprietary computer vendors entered the game too. Commodore did Unix for high-end Amigas; Atari did Unix for high-end STs; Acorn did Unix for high-end Archimedes; Apple did Unix for high-end Macs, allegedly originally just to get a US military deal; etc. etc.

All these are still $1000 per instance OSes though.

Then, universally scorned, MWC Coherent, a real Unix-like OS for $99... and QNX, which was apparently good but mainly focused on real-time stuff, and cost more than the casual could afford.

(As a European I never saw this but it was in all the ads in all the US mags. There was a lot of "cheap" American stuff we didn't get over here, like paid-for shareware. We had metered phone calls so no BBS scene. Only rich Americans got that stuff.)

Coherent was so good that AT&T accused them of theft and sent Dennis Ritchie around to check. He came back and said, no, it's legit.

And Andy Tanenbaum's Minix, a toy for students, not for real work, but essentially free with a book.

These latter indirectly showed that you _could_ copy AT&T's holy grail and make it work, so while Richard Stallman was building all the tools but choosing the wrong kernel and sabotaging the whole thing, along came this Finnish kid with his learning exercise, and excited beardies on Usenet said that it actually worked and it was at least as good as Minix and was getting to Coherent levels.

So the point is, there was a spectrum, from legendary machines made from purest unobtainium, to ludicrously expensive x86 stuff for very specific (and ludicrously expensive models) of PC kit, to the still ludicrously expensive SCO that got no respect, to "cheap" stuff that nobody had in Europe because it had no business purpose. There was legendary free stuff in America but it only ran on room sized computers that cost as much as a lottery win, so I never saw it. "Free" as in "it's free if you're so rich it doesn't matter."

And "free" shareware that was "free" as in "the phone bill to get it will cost more than just buying a commercial version in a shiny box".

But there _was_ a spectrum, from vastly expensive to "a small business will pay for this", down to theoretical stuff in America that you could dream about... which paved the way until the point where an ordinary PC was a 32-bit machine with a memory management unit and hundreds of megs of disk and several megs of RAM, and suddenly, this Lin-Min-Gnu-ix thing was doable, if you had a beard and a checked shirt with black jeans and wore hiking boots every day. 

Comments

[autopope]

A footnote:

Xenix was developed by Microsoft in the 80s. For a while Microsoft had more UNIX licenses out there than every other VAR/OEM combined! But around '86 BillG decided to focus on OS/2 and hunted around for a buyer. SCO, previously just a device driver shop, wangled a deal in return for Microsoft taking out a 19% stake in the company. What SCO found hard to swallow was that Xenix relied on the Microsoft C compiler, so the development system required them to pay royalties. Xenix also relied on AT&T code, so: royalties. System 7 UNIX hadn't originally come with TCP/IP, so Microsoft licensed an IP stack from another vendor -- more royalties. So basically if SCO wanted to sell a complete system they had to pay out a lot of money each time, hence the lack of a bundled compiler and Xenix coming by default with UUCP rather than TCP/IP, and so on.

In 1988/89 Larry and Doug realized this was no way to run a business, so they paid AT&T for a perpetual license to System V 3.2, which became SCO UNIX. But to maintain binary compatability with Xenix (all the installed base, remember), they needed to keep it running with MS C.

Circa 1993, every copy of Open Desktop that SCO sold -- the all-in, X11 plus Motif plus IXI Desktop on top of SCO UNIX bumper pack -- incurred £500 in royalty payments. So old-SCO had a hard floor beneath which they couldn't sell Open Desktop (or just plain SCO UNIX) profitably, and that floor was already in three digits. Trivial if you were in the minicomputer business, but by then SCO was targeting 486s and the then-new Pentium

From about 1990-1993 SCO joined the ACE Initiative to define a new RISC-based PC architecture -- partners included HP, I think, and it was intended to run the not-yet-there Windows NT as well as Open Desktop as a rival for OS/2, IBM's seemingly invulnerable corporate behemoth. But ballooning development costs broke ACE apart, and SCO bailed (with a 10% downsizing across the global development team).

They also made a bad call in 1992 with the design of Open Server (which shipped late, in 1995). They expected thin clients to prevail once the 1990-92 RAM famine receded, as hard drives of sufficient size to hold a full workstation UNIX were fiendishly expensive back then. So they hit "pause" 8 months into development and re-targeted Open Server as a symlink farm that could be pointed at a local installed filesystem or at a remote file server. But by the time it was ready to ship, PCs were coming with 500Mb drives by default and most folks didn't want to run it as a thin client over 10mbps ethernet. So they kneecapped the design by mistake.

The also had an ageing code base built atop SVR3.2, which nobody wanted: the new hotness was SRV4. But AT&T had wised up and wanted $200M up front for an unlimited license: Sun coughed for it (the result was Solaris) but that was an entire year's revenue for SCO. So SCO engineering in Watford essentually turned their SVR3.2 codebase into a white room clone of SVR4 -- only the AT&T copyrights in the header files remained! -- but couldn't use "SVR4" in marketing, so everyone thought they were increasingly behind the curve. Oops.

In 1994-95 they began negotiating with AT&T for a license to SVR4.2, which they eventually landed (after I left in 95 to join a web startup). But by then it was too late: they'd gone public then promoted an accountant to be CEO, and he … did not make good decisions. Culminating in selling most of the business to Borland, and the trademarks and sublicense agreements to a Linux distributor called Caldera …

And the rest is history.

(Source: first-hand memory, I worked for SCO from 1991-95.)