Sun Pats Rump SCO - Tarantella Cashes Out After Lots of Agony

A teeny tiny acquisition announcement brought back a lot of memories today.

Remember Santa Cruz Operation - no, not the SCO you read about fighting IBM and Novell, but the "old SCO"? Bob Greenberg and friends did a very brain-damaged version of Unix for the PC (originally derived from Version 7 and System 3) way back in the dark ages. Bob had done a Version 6 Unix derivative for the RAND Corporation called BobG Unix. The group was spun (thrown? or maybe walked?) out of Microsoft because Microsoft really didn't want a Unix system if it wasn't written in BASIC.

In 1982, Intel offered Symmetric Computer Systems CEO and Founder William Jolitz a great deal on 286 processors when he was deciding on processor bids to use in their new workstation funded by Technology Funding Partners (Symmetric's lead venture firm). There were lots of problems with the 286 and Unix: 1) the instructions were not restartable, so if the operation could not complete (like memory wasn't loaded) you could not reliably reload the instruction (there were steppings that supposedly could, but you never knew what you'd get), 2) the only way the address space was made large was by the reloading of segments - we'd encountered this problem before with the PDP-11 (William Jolitz work as an undergrad was on overlays for the PDP-11, so he was very familiar with this problem) - performance goes to hell when you move data from overlapping 64 kbyte segments to other segments, with all bets off when you hit an exception during that time, and 3) the calls within the segment and intrasegment calls made for variable sized stack frames, and the way Intel, Microsoft and others agreed on stack frame layout required a major rewrite in Unix - ironically, this made it difficult for early Windows programs derived from DOS as well. From these issues, we knew 286 Unix would never be a successful product, because there were too many compromises to move too many software packages from architectures like the VAX to it. SCO went ahead and made a Xenix based on the 286 - took them three years and a lot of work - and it was still a disappointment.

While the technology folks at Intel waved a siren song of processor dreams, intimating the 386 would be "just around the corner", their chip sales people sang a different more croaking tune. They were adament that the 286 would stick around for five years or more, and the 386 would not be available soon. So sadly facing reality (in other words, we couldn't hold out for 3-5 years on a promise and sales wouldn't sell us what we had to have), we opted for the NS32000 processor chipset, which meant a real BSD Unix (good) but also meant we had to work with National Semidestructor, uh, Semiconductor (bad), who couldn't decide if they were in the high-end processor business or the low-end volume chip business. They compromised by selling the chips at a premium while not bothering to support them - the worst of both worlds.

By the time the 386 came out in 1985 and Xenix 286 was released, we were already selling Symmetric 375's running Symmetrix 4.2BSD Unix (see The Symmetric 375 and Symmetrix Owner's Manual). The Symmetric 375 was reviewed in Byte Magazine in January 1988, 4 years after we had begun selling them. SCO trod a much harder road than we did. We sold BSD Unix on hardware, just like another Berkeley spinout called Sun and a lot of other workstation manufacturers. Microsoft basically licensed their software to PC OEM's so you still bought the PC with DOS (later Windows). Apple sold hardware with their own OS. But the Unix market for PC's was an untried market. How do you prove it exists to cynical businessmen if it isn't there yet? (Some would argue we're still not quite there yet).

In 1985, Intel also introduced a 386 version called the 376 that did not support real mode so it could not be used for DOS. It went directly into protected mode. The reason was so they could sell the same chipset brain-damaged as a low-end processor. Symmetric Computer Systems was told it was 1/4 the price of the National Semiconductor processor but twice as fast (we could buy an entire year's inventory from a PC vendor who had misordered for a song). This meant that past the disk drive, the next biggest expense was not the processor but the OS licenses. We really wanted to make this work, especially since the MMU and clock chip was built-in to the 376 (which meant a smaller board footprint, lowering costs). We didn't know this at the time, but it would also have worked with the FPU - Intel kept saying "No, it wouldn't work", but turned out it did - it did work with the 8287 and 8387. It turned out Intel sales wanted us to buy the 386 plus FPU and that carried a very high pricepoint (the Rolls Royce configuration) and preferred no deal to the smaller and more feasible (for us) deal. We would have used the 376, dumping the NS32000 then and there (answering Intel's tech groups dreams) if we had been told the correct information by Intel sales. But the 376 team got sold down the river by their own chip sales (not for the first time).

In essence, the difference between National and Intel was that National wanted to price and then do technology to compete. In the late 1980's, after Intel reorganized (fired) some of their old-time chip sales (none too soon), it wanted to compete, even with itself if necessary, and then consider price. If the chip didn't move, National allowed its market size to drop and prices to rise, and postponed technology upgrades for longer periods of time - releases got fewer and fewer, prices higher and higher, and customers fled. If the chip didn't move, Intel discounted, so sometimes the newer processors were cheaper than the older processors. For a time, the 386 was cheaper than the top end 286 - compilers and parts of the OS had to change and investment had to be made to get that performance, and your costs rose. If you had a DOS app, you went back to 8086 mode, if you had a DOS extender mode in protected mode, you went back to 286 mode. Only if you had everything right could you use 386 mode. The stack frames and alignments would all be different. This made it a support nightmare for the SCO people.

It wasn't until the 486 came out in 1989 that we became convinced that Intel would displace the larger mainframe processors, because the first one was half the speed of supercomputers like the CDC 7600 used at NASA Ames a decade ago, and that the scaling would result in a factor of ten room. That's why we did 386BSD (the 386 was the commodity PC chip and affordable to all) and wrote "386BSD: A Modest Proposal", setting the stage for the modern open source operating system.

It is somewhat ironic that a decade later, in 1992, we released 386BSD to the public. So Intel did eventually get us to write Porting Unix to the 386 - it just took a while until they got their act together.

So, what ever happened to the old SCO? Well, SCO for all their work (and they did go IPO) had a number of embarrassing scandals (of the "keep your hands off me" kind). They also ran smack dab into the open source revolution of the mid-1990's. But they did have a number of business accounts, and kept going, acquiring Unix in 1995 from Novell (who had gotten it from USL after their disasterous Berkeley v. USL lawsuit), went for the "high-end" processor market (which didn't really materialize), and finally sold Unix along with their name (suitably modified by the buyer to the more trendy "SCO Group") right back to Ray Noorda (Novell) and his new startup Caldera in 2001. Renouncing their Unix heritage, they reinventing themselves as Tarantella. So all was fine with the world, right?

Well, they didn't have very much left, but they did do some network management software. So now they are gone, sold to Sun for $25 million in cash. Thus ends one of the first Unix on Intel processors companies, not with a bang but with a small cash whimper. May it rest in peace.