The History of XENIX
A Failed Future and a Long Life
In the November 1980 issue of BYTE, the publication reported that Microsoft signed an agreement with Western Electric for the rights to develop and market UNIX from Bell Laboratories. The version of UNIX from Microsoft was to be specifically for the PDP-11, the Intel 8086, the Zilog Z8000, and the Motorola 68000, and it’s name was XENIX. It’s major selling points were that it was supposed to be available for 16 bit microcomputers and that it would have MS BASIC, FORTRAN, and COBOL which were already widespread.
At this point UNIX was quickly gaining market share on minicomputers, and CP/M was already the dominant system for 8 bit microcomputers. By the time the BYTE article was published, Microsoft had already purchased a DEC PDP-11/70 for XENIX development, and the Z8000 port was scheduled for release before the end of 1980 with 8086 and 68000 ports following in the second quarter of 1981. This was particularly ambitious as Microsoft’s first Z8000 target was a Codata machine delivered by Bob Greenberg to the guys working on the port in the middle of 1980. Despite a PDP-11 and a Z8000 both being computers and despite UNIX being written in C, the two machines are very different. This would have been a Herculean task. Pricing was to be $500 (around $1866 in 2023) for a single-user copy, but XENIX was going to be sold exclusively to OEMs. Perhaps the most interesting point from the article in BYTE was their expectation that XENIX would become the dominant OS for 16 bit micros due to CP/M’s 16 bit port being limited to the 8086. The first delivery of XENIX to a customer occurred in January of 1981. The consulting firm Human Computing Resources (HCR) in Toronto was trying to expand into the commercial UNIX market, and rather than develop and release their own version, they licensed XENIX from Microsoft. Their first deployments were on the PDP-11 and Zilog Z8000 with XENIX version 2.1 (version 2.0 was only an internal version on the PDP-11 at Microsoft AFAIK, and there was no 1.0). HCR and Microsoft had a good relationship with HCR contributing enhancements to XENIX and doing some work with hardware ports.
By May of 1981, Robert G. Greenberg (“Bob”) had become the product manager for XENIX at Microsoft. He wrote a very lengthy article for Byte about the product, and spared no details. He was particularly talented in explaining the very complex origin and history of UNIX up to that point in time as well as the state of the UNIX market. His article is of importance for the history of XENIX as he clearly explains Microsoft’s goals with OS.
Microsoft’s clearly stated intention was to bring the uniform and consistent design of UNIX and its multiuser, multitasking, I/O rich capabilities to the (at the time) humble microprocessor. To this end, Microsoft had purchased a source license for Western Electric’s UNIX distribution for the PDP-11 and ported it to 16 bit microprocessors just as the prior announcement had stated. The Z8000 was first, this makes sense as Zilog was the clear leader of the microprocessor market at that time. As of June in 1981, Microsoft was still working on their versions for DEC LSI-11/23 (a PDP-11 using microchips), Intel 8086, and the Motorola 68000. Between February and May (when I presume the BYTE article was written) something had changed as some porting targets had fallen out of favor or at least lost priority within Microsoft (like Texas Instruments). Of course, Microsoft wasn’t just porting; they were doing bug fixes, optimizations, stability improvements, and reliability improvements. Microsoft was actively adding features as well with things like distributed processing, new networking technologies, multi-CPU support, record and file locking, semaphores, shared memory management, and both software and hardware error recovery. A key for XENIX, however, was approach-ability. The computer industry was certainly still young in 1981, but it wasn’t totally new. Millions of amateurs and novices knew Microsoft’s programming languages, and XENIX offered these. If someone were a home user who knew BASIC, BASIC was on XENIX. If someone were an astronomer at a university and knew FORTRAN, XENIX had FORTRAN. If someone working in accounting and new COBOL, XENIX had that. For the odd Pascal user, Pascal was present too. XENIX being UNIX at this point might have been a struggle without these tools. Most computer users and software developers would have been completely unfamiliar with C in 1981. And yet, Microsoft still felt more was needed. The company established a clearinghouse for quality software for XENIX to try and get maximum distribution for that software to XENIX customers.
By November, there were some third party resellers of XENIX. One example was a company called Management Systems Development (MSD) who offered XENIX bundled with their own Uni-Calc spreadsheet on PDP-11 workstations, along with installation, configuration, and support services.
Another early customer of XENIX was the Santa Cruz Operation formed in 1979 by Larry and Doug Michels. SCO was originally a consulting business. The company changed overtime into one focused on porting UNIX and UNIX applications to new hardware. They developed Dynix (UNIX v7) for the PDP-11, but abandoned that effort in 1982 with a joint development agreement between SCO and Microsoft for XENIX. SCO and Microsoft then worked with Logica in the UK and HCR in Canada on XENIX enhancements and porting efforts.
With the IBM PC’s launch and success, a new strategy was developed for XENIX. Microsoft had a three tiered structure of upward compatible operating systems: MS-DOS, XEDOS, and XENIX. For MS-DOS/PC-DOS, a user would be on an 8088 or 8086. MS-DOS was written in 8086 assembly language so ports would be difficult and of little value. XEDOS, however, was written in C. It was essentially a single-user version of XENIX, and it was targeted at the same hardware as XENIX: 8086, 68000, Z8000, and LSI-11. These three systems were demonstrated at COMDEX in a working state with unmodified code being run on all three systems. For graphics compatibility across these three systems, Microsoft used AT&T’s Presentation Level Protocol. The intent here was clear. A user could start on MS-DOS, then migrate to XEDOS where both some MS-DOS and all XENIX software would be usable, and then in the event that he/she needed still more computing power, he/she could migrate to XENIX.
In March of 1982, Western Electric reduced the royalty fees that it charged for UNIX with the introduction of UNIX System III (a combination of UNIX V7 and PWD). This was done by raising the source license cost, but lowering the royalty per user license to $100 (about $318 in 2023) from $250. This meant that Microsoft’s prepayment of $200000 in 1980 (around $700k in 2023) for a discount in volume was voided if they wished to use the newer system. That same month, on the 10th, Paul Allen held a seminar in NYC where he outlined plans for MS-DOS 2.0. The system featured some windowing capability with the ability to select files and programs via cursor movement, on-line help (meaning help available live on the running machine), networking that was compatible with XENIX systems, a print spooler, an enhanced debugger, several direct ports of XENIX utilities, and disk I/O buffering,
In November of 1982, Microsoft and Intel signed an agreement to make XENIX available on iAPX 286 systems marketed by Intel. When the Apple Lisa was announced in 1983, a lot of chat was about its GUI OS, but there were ample words about XENIX as well. The agreement between Microsoft and Apple was most likely inked by February of 1983 as Apple was featured in Microsoft ads of that time frame.
By this point, the ambitions of MS-DOS 2.0 had been scaled down. Despite this reduction in scope for MS-DOS 2.0, it did carry many bits of XENIX. The system adopted I/O redirection via less-than and greater-than symbols, piping, a hierarchical directory tree, file handles, installable device drivers, background tasking, the ability to change the switch operator from
- like XENIX/UNIX, and the ability to change the path separator from
/ along with changing the way devices are accessed in the shell (to things like
/dev/con). These latter two were accomplished via undocumented (at least I can find zero documentation about this today)
From what I can tell, by the time MS-DOS 2.0 was released, Microsoft had effectively abandoned the XEDOS idea in favor of slowly changing MS-DOS into what XEDOS was intended to become. XENIX, however, was not abandoned at all, or at least not yet.
The US government forced the breakup of AT&T on the 8th of January in 1982. In January of 1983, AT&T released UNIX System V for the PDP-11 and VAX and offered it directly as a commercial product rather than solely as licensed source. AT&T was also able to enter the computer market directly. This could be why Microsoft ended the development of XEDOS and tempered their attempts to migrate users from MS-DOS to XENIX, but it could also be that Bill Gates saw Visi-On at COMDEX. Still, XENIX was the most widely installed UNIX on the market by 1983, so Microsoft continued with it. As AT&T entered the UNIX market directly, Microsoft’s license for UNIX source was now from AT&T instead of Western Electric, and until the 1990s, Microsoft held the single highest volume license of UNIX.
By March of 1983, Tandy Radio Shack had made XENIX the default operating system on the TRS-80 Model 16 with up to three concurrent users. They also supplied XENIX to buyers who’d previously bought a Model 16. For TRS-80 Model II users, if they bought a hard disk and an upgrade kit, they could also run XENIX if they had at least 256K RAM.
By late Spring and early Summer of 1983, what would later be referred to as UNIX workstations were becoming far more common. Systems featuring the most powerful 16 bit processors of the day, large amounts of RAM (for the time), and UNIX or a UNIX-like operating system were available from multiple vendors. XENIX was doing well as the OS of choice for those systems. Importantly, XENIX picked up support for ethernet, X.25, and IBM SNA networks more quickly than its competition. This seems shocking at first given that XENIX could run on an 8086, but in workstations employing the 8086 it wasn’t entirely uncommon to see support chips freeing up CPU time. For example, the Altos 586 had two I/O processors and an MMU (sort of) to make the system a bit more capable. Of course, getting a XENIX system was an expensive affair just like later UNIX workstations with pricing starting around $8000 (about $25000 in 2023) and going as high as $15000. While quite steep, this cost gave the purchaser a network capable, multitasking, multiuser system with CPU speeds at or above 10MHz (for some architectures), RAM measured in megabytes (for some architectures), and hard disks of 10MB or more. Consumer machines were typically using CPUs clocked at or below 5 MHz, had less than 1MB of RAM, and usually lacked a hard disk drive.
Amazingly, all of this growth from 1981 and into 1983 was all largely a single version, 2.1. This version of XENIX was successfully ported to all intended targets by 1983. Machines with these CPUs and with XENIX on them were available from Paradyne, Compucorp, Bleasedale, Kontron, Forward, Fortune, Sritek, IBM, Intel, Victor, Seattle Computer Products, Altos, and more. In the case of the PDP-11, XENIX 2.2 became available in early 1983, and toward the end of the year 2.9 was available. All of these ports were still using UNIX v7 as the underlying code base allowing Microsoft to continue leveraging their original licensing deal established with Western Electric.
1983 saw the first releases of XENIX 3.0 based on UNIX System III with enhancements from BSD. This was first made available on the Altos 586. In early 1984, the Intel 310/380, and machines from Acorn received UNIX ports of this newer version. After that, XENIX 3 was made available for the Apple Lisa and many other machines. The port to the Lisa and Lisa 2 were important for SCO. These were sold as boxed and shrink-wrapped software by SCO, and were not only their first binary product, but also proved the viability of UNIX being sold in such a manner. The LISA as a 68000-based workstation with XENIX was essentially a multiuser minicomputer of a few years ago for around $5000 dollars (about $15000 in 2023), and SCO sold thousands of copies. SCO ported XENIX 3 to the Tandy Model 16B on which XENIX was the default operating system once again. SCO had another significant triumph that year. SCO ported XENIX to the IBM PC XT without requiring an MMU or extra processor card. To do this, SCO leveraged the work done by the Altos team (in partnership with Microsoft). Compilers and assemblers had already been made for that effort, and SCO just needed to be certain that their code was binary compatible with the Altos release to make sure applications would be available. This too was a commercially successful port. By the end of 1984, Tandy was the largest UNIX systems vendor of their day, and all of their UNIX machines ran XENIX. Microsoft’s (and SCO’s) continued use of System III saw them lose the contract with Intel who turned instead to Digital Research. There was also one market into which Microsoft had largely failed to gain a foothold, big machines. On mainframes and minicomputers (other than the PDP-11), XENIX was absent. This isn’t too surprising since Microsoft’s focus (as its name implies) was microcomputer software. Most of those machines ran UNIX proper or ran bespoke operating systems written expressly for them. This sudden explosion in XENIX ports was made possible by SCO being granted distribution rights for XENIX in the USA from Microsoft.
XENIX System V was released in 1985 and its first port was for the IBM PC AT (286) and this enabled ATs to address up to 3MB of RAM and access disk drives of more than 20MB. This version of XENIX listed by Microsoft (though most of the work was done by SCO and SCO also sold it) as Xenix 286 was released with support for 286 protected mode, but XENIX System V was made available for all of the previous targets. Microsoft and SCO made a decent amount off of the 286 port. Xenix 286 was a triumph for SCO. Beyond the technical aspects of 286 protected mode and supporting up to ten user terminals with an AT, this version beat IBM’s own PC/IX to market, and more importantly, IBM dropped their own XENIX system in favor of SCO’s. It was, however, the IBM PC AT that would bring about the end of Microsoft’s adventures with UNIX. On the 10th of June in 1985, Microsoft entered into a joint development agreement with IBM to develop OS/2. Microsoft would continue to use XENIX internally to run their mail servers and to develop OS/2, but as OS/2 took shape it would replace XENIX as the system for software development. Likewise, MS-DOS’s XENIX-like flair would end.
Despite Microsoft’s shifting priorities, SCO’s business was built around XENIX almost entirely. Oddly, as they were taking more of the primary role in XENIX development, they picked up none other than AT&T as a customer. AT&T was selling PC 6300 and wanted a UNIX for it. As SCO had already ported XENIX to the PC platform, AT&T had them do a port to their enhanced machine with support for their STARLAN system, Display Enhancement Board, and modem.
The last bit of work on Xenix within Microsoft was to merge many of XENIX’s improvements and ports into UNIX System V SVR3.2 and SVR4. This was a good move on Microsoft’s part as they then received royalties on sales of those systems. Microsoft did continue to use XENIX for the mail system within Microsoft on 68000-based systems until that was taken over by Microsoft Exchange in the 1990s.
By the middle of the 1980s, XENIX was used in many different industries. ISPs had replaced large minicomputers with XTs or ATs running XENIX. Retailers, fast food chains, hotels, universities, smaller research labs, manufacturing centers, and many more companies were buying Lisas and ATs to act as UNIX servers to which they’d then attach a few dumb terminals. This was far cheaper at the time than buying each user a PC. Yet, aside from the unreleased graphics system from the MS-DOS/XEDOS/XENIX triumvirate based around AT&T’s protocol, XENIX did not include a GUI. Some companies then created their own GUIs, and one of those was Siemens’ release of XENIX named SINIX which included a GUI named Collage.
Within SCO, Larry Michels was the businessman and his son Doug Michels was the technical mind. They operated as a team, and they struggled throughout the 1980s. Cash flow was often an issue and they had to use personal funds to keep the business going. Some years would have decent profits and those profits were often used to fund the down years. Things changed a bit as 1986 came around. With SCO being the primary distributor of XENIX in the USA, the profits increased. In November of 1986, Santa Cruz Operation bought Logica in the UK. This established SCO in Europe, and it increased their total addressable market as SCO gained European distribution rights through this acquisition.
In 1987, SCO released the SCO Xenix 386 Toolkit. Developers were now able to start creating both applications and device drivers for the Intel 80386 and when SCO released Xenix 386 (System V Release 2.3.1) later that year software was already available for the new OS. This version sold well and found its way into many small to medium sized businesses partially helped by the fact that it was the first fully 32 bit operating system available for the 386, and also helped by the fact that it could run multiple MS-DOS applications concurrently. Xenix 386 also included SCSI support, TCP/IP, and other more modern technologies After the release of Xenix 386, ads for products listing OS support start to refer to “SCO XENIX” and by 1989 SCO and XENIX were roughly synonymous. Over the course of 1986 and 1987, the company grew immensely from just a handful of people including the Michels to nearly 500. The company then moved from a small shop just outside of downtown Santa Cruz to 400 Encinal Street and started construction of a new building at 425 (across the street). Among the new employees was a board member Jim Harris who had previously led Microsoft’s OEM sales.
Both the success of the 386 release and SCO’s expertise in UNIX on x86 would see that platform become the primary focus of SCO’s efforts. Additionally, within the market for microcomputers, the x86 platform was growing rapidly and defeating most competitors.
1989 was another big year for SCO. In February, Microsoft completely exited the UNIX business, but purchased between sixteen and twenty percent of SCO. This deal provided funds to SCO for expansion, put Microsoft CTO Frank Gaudette on the SCO board, made it certain that Microsoft wouldn’t exercise control over SCO, secured on-going royalty payments for Microsoft, and gave SCO full control over XENIX.
In June, SCO released SCO UNIX System V/386 which brought the system up to SVR3.2. This work was two years in the making and required extensive work. SVR3.2 and Xenix 386 (based on SVR2) were not initially binary compatible, but working with Intel and AT&T, SCO was able to to incorporate the Intel Binary Compatibility Standard. The next major hurdle was getting device drivers to work, which took the majority of the second year of development time on the release. The name change was worked out with AT&T during 1988. SCO UNIX was released as a higher-end product with Xenix 286 and Xenix 386 still being in the product line. With SCO UNIX, the company offered several add-on packages including TCP/IP, X Windows, Motif, NFS support, and the Ingres relational database. In 1990, SCO released SCO Open Desktop. This release essentially bundled all of the add-on packages into a single software release, and added SMP support.
On the 9th of May in 1990, HCR announced that it would be acquired by the Santa Cruz Operation. HCR’s founder Michael Tilson said the deal involved a share swap that had a value measured in millions. At this point HCR was a much smaller company than SCO with around 50 employees. This deal made SCO the sole source for XENIX globally, reduced the royalties they needed to pay to just Microsoft, and gave them control of Microsoft’s C compiler for XENIX as well as the Cfront C++ compiler.
Unfortunately, SCO Open Desktop didn’t sell particularly well, and SCO was once again burning cash. Adding the cost of acquiring HCR, extremely high costs in system documentation (something SCO was extremely good at, producing thirty thousand pages of documentation in an eighteen month period), large head count, and the high cost of application software development to the cost of developing an OS that didn’t sell and SCO’s position was looking quite dark. Venture capitalists owned around a fifth of the company by 1991 as SCO was struggling to survive. SCO’s unfavorable financial position wasn’t helped by the rest of the industry. The Advanced Computing Environment was announced in 1991 with Compaq, Microsoft, MIPS, DEC, and SCO all taking part. The idea was to create a commodity computing platform around the MIPS CPU. Microsoft was porting OS/2, and SCO was porting UNIX. The UNIX port was a merger of SCO Open Desktop, DEC OSF/1, and DEC Ultrix. SCO’s investment was considerable but ACE fell apart just a year after starting. The response within SCO was to cut staff. The company had reached a headcount of around thirteen hundred people, and over one hundred fifty were let go. SCO also brought new managers in from the outside to lead this newly reorganized company. Almost all UNIX applications projects were terminated.
By the middle of 1992, it appeared as though SCO was ready to make an IPO. Yet, SCO’s past company culture prior to the reorg and new management was haunting it. SCO had operated more like a university with the good and the bad that come with that. It was rather normal for the company to have hot tub parties during working hours, to have beer available in the office at all times, and for fun and camaraderie to exist simultaneously with sixty hour weeks. End of day meetings were usually held on the roof. Doug Michels reflected on SCO generally saying “We knew how to code, we knew how to market and sell, and we knew how to party. We were growing like crazy and didn't know quite when to stop.” That kind of culture is certainly fun for many, but there tend to be downsides to large numbers of drunk people being in frequent close proximity in an informal setting. In early December of 1992, three former executive secretaries filed a lawsuit against Larry Michels for sexual harassment. A fourth woman joined on the 15th with a more serious allegation. On the 21st, Larry Michels resigned as CEO of SCO, and Jim Harris filled the role of interim President. Toward the end of January, Lars Turndal who’d been the leader of SCO’s European efforts became CEO, Harris became Chairman of the Board, and Doug Michels remained as Executive VP and CTO. The lawsuit was settled out of court in April of 1993, and criminal charges were never filed.
Despite all of the drama, SCO’s reorganization was successful, and SCO posted a profit for 1992 and the first half of 1993. On the 25th of February in 1993, SCO purchased IXI Limited in the UK which produced X.desktop which was the basis of Open Desktop’s GUI. By this time, SCO held a market share of roughly sixty five percent for UNIX on Intel. The company’s IPO was on the 27th of May in 1993, and its NASDAQ symbol was SCOC. It didn’t exactly have a great time. By the end of the year, the stock price had fallen to half its starting value. The company began killing off unprofitable products and placed its focus back to its UNIX system. I do not normally mention any deaths on ARF, but I would be remiss not to say that both Harris and Gaudette passed away in 1993. Turndal became Chairman of the Board and CEO in December of 1994, and Alok Mohan became President and COO. The loss of Harris and Gaudette so shortly after a shakeup at the company undoubtedly hurt SCO’s performance in the stock market. It would be difficult to make any serious investment in a company whose leadership isn’t certain, and this is especially true when Microsoft had released NT in the middle of the year.
On the 12th of December in 1994, SCO acquired Visionware of the UK which produced PC-Connect, XVision, and SQL-Retriever. PC-Connect was a program that was essentially a terminal emulator for MS Windows, but it allowed things like cut/paste between MS Windows and X Windows. XVision was a X Server for Windows that maintained the visual style of Windows for applications being accessed from X Windows. Finally, SQL-Retriever was an OBDC-compliant database client allowing applications to easily access Oracle, Interbase, or another RDBMS. These two acquisitions gave SCO’s UNIX products a little more value and more differentiation.
SCO Open Desktop had received updates and patches during all of this time. The naming of SCO’s release changed to SCO Open Server with next major release, SCO Open Server 5, in 1995. This release changed the executable format to ELF, added dynamic linking, added a journaling filesystem, added RAID support and disk compression, added APM, expanded SMP for up to thirty two processors, and vastly increased hardware support. This release also added many graphical system administration tools. The system cost over $50 million (roughly $100 million in 2023) to develop and took three years of work. This was SCO’s primary product, and supported many other endeavors. A rather important piece of software for SCO Open Server 5 was Global Access. This was a web client and email client software package as well as web server and email server software. It included POP3, SMTP, NCSA HTTPd, and Mosaic. Another venture was PizzaNet. SCO and PizzaHut built the first ecommerce platform ever, and they did it to sell pizzas. I find this rather fitting considering SCO’s culture. The site was essentially a webform with a menu of the company’s offerings. On form submission, the order would go to a server in Wichita, then get sent to the PizzaHut location. To verify the order, an employee would call the customer. Payment was done in person at the time of delivery.
On the 20th of September in 1995, Novell sold Unix System Laboratories to SCO. Novell had become both the technology holder and trandemark holder of UNIX a few years prior, the technologies went to SCO and the trademark was sold to the Open Group a year later. For SCO, this meant that they’d get some of the talent involved in the creation of UNIX, and that they’d shed the licensing costs of UNIX while gaining the some of the revenues of existing licenses (but not trademark royalties). After some rather complicated negotiations, the system that resulted from this deal, which included HP, was UnixWare 7 based upon a combinations of technologies from Open Server 5, UnixWare 2, and the SVR5 kernel. This product was announced in March of 1998. In March of 1999, SCO introduced UnixWare 7 Data Center Edition which added support for clustering and other high availability features along with support for non-x86 hardware.
By 1997, SCO’s market share had grown and the customers included megacorporations of varying types from retail to finance to transportation and even to governments. The company even had rather high revenues, but their profit margins were rather small, and Open Server 5 was responsible for around eighty percent of their income. The company did another round of layoffs, and in April of 1998, Doug Michels became President and CEO while Alok Mohan became Chairman of the Board. His first goal was to move the company to electronic licensing and electronic distribution of software. This would eliminate inventory costs as well as the costs related to printed manuals, boxes, and pressed discs.
From 1998 to the year 2000, SCO was working with IBM and Intel to produce a UNIX platform under the code name of Project Monterey for the upcoming Intel Itanium architecture. This was publicly announced on the 26th of October in 1998 and one year later 32 bit UnixWare binaries were running natively on IA-64 hardware running an IA-64 UNIX. By 2000, Itanium was late, IBM was more interested in Linux, and the entire endeavor fell through.
SCO saw a brief increase in profits in 1998 and 1999 as it began offering consulting services and sales for Linux distributions from Caldera, SuSE, and TurboLinux, as well as offering Y2K compliant products and Y2K consulting services. The company’s stock likewise saw an increase. By March of 2000, the company was once again losing money with Linux on one side and Microsoft on the other. On the 7th of May in 2001, SCO sold its UNIX operating systems to Caldera Systems and spun off Tarantella (the software from IXI and Visionware) as its own company. Caldera would rename itself to SCO Group in 2002 and subsequently become one of the most reviled companies in the tech world. The rights to UnixWare and Open Server were acquired by Xinuos in 2011. Open Server 5 is still supported making XENIX, perhaps, the longest lived UNIX.
I have readers from many of the companies whose history I cover, and many of you were present for time periods I cover. A few of you are mentioned by name in my articles. All corrections to the record are welcome; feel free to leave a comment.