Is Unix Dead?
As Unix faces the stiffest competition of its long life--Windows NT--can it survive?
Tom Yager and Ben Smith
Among operating systems, Unix is the grand old man. Over its long life, it has garnered respect and ridicule, awe and animosity. Never before has computer code managed to so passionately galvanize and divide, anger and excite otherwise level-headed technologists. Unix stories transcend reality. Wizards and demons abound. Gatherings of the faithful generate the kind of excitement usually reserved for evangelists' tents.
Today, however, Unix's seniority and mystique hold little sway. At least 25 different Unixes exist now, and committees dispassionately choose from among various Unix versions, but six providers dominate: Sun Microsystems, The Santa Cruz Operation, Hewlett-Packard, Unix Software Laboratories, DEC, and IBM. The entry of Next, Univel (the joint venture between USL and Novell), Solaris (from SunSoft, Sun Microsystems' software subsidiary), and other new multiplatform Unix players will, odd as it may seem, help consolidate a confusing field of choices. But in the midst of that consolidation, Unix will face what may be its most powerful adversary to date: Microsoft Windows NT.
At this crossroads, it is appropriate to examine some key aspects of Unix: What makes it unique? How has it changed over the years? What are the threats to its survival, and how might it overcome them? And, most important, will it survive?
Reports Greatly Exaggerated
Despite its problems, Unix is not dead; in fact, it's surprisingly healthy. Unix remains compelling because it is the only operating system to offer multitasking, graphics, and cross-platform compatibility in one package. Market researcher Dataquest estimates that, for 1991, Unix sales (including both systems and operating systems) totaled 1.2 million units, for $18.2 billion in revenue. Unix revenue will climb to $44.7 billion in 1996 on unit sales of 4.1 million, according to Rikki Kirzner, a senior industry analyst in Dataquest's software group (see figure 1).
Other analysts believe that Windows NT will slow Unix's growth. NT is a legitimate threat to Unix's domination of the workstation market and to its chances for success on the desktop. NT may be the strongest stimulus to date for inducing Unix providers to stop arguing and start cooperating in an endeavor to give software applications developers a standard software base. The Unix industry has answered that call with new versions aimed at the Intel- based desktop: USL's Unix System V release 4.2 (also known as Destiny), Next's NextStep, and SunSoft's Solaris.
Understanding the Unix Community
There are two distinct branches of the Unix user community. The first group could just as well be running VM, VMS, or any other operating system; it just so happens that its chosen applications run on Unix. These unwitting people may be the fastest-growing segment of Unix users. Participants in many information services--AT&T Easy Link, for instance--are Unix users and not even aware of it. Many retailers run their businesses on Unix. What these users have in common is that some technically knowledgeable person or firm set up and maintains their systems. For this group, Unix is an invisible tool.
The other, better known, Unix community comprises technically adept users, for the most part. They may be using workstations now, but they are the people who were running serial terminals not long ago. In contrast to the first group, these users understand the inner workings of the operating system. This uniquely dedicated group forms the core of the "Unix community," a near-cult with a strict hierarchy based on technical knowledge (see the text box "Unix: A Child of a Thousand Parents" on page 137). The people involved in the creation of Unix built it for people like these. But now that Unix is falling into the hands of the uninitiated, its role--and its shape--are beginning to change.
Those blissfully unaware users of the first group are Unix's future. In fact, they're the future for all multitasking operating systems. Ask today's user what he or she needs, and you'll hear multitasking, graphics, and cross- platform compatibility. These needs are keeping Unix alive, because it's the only operating system that delivers them all.
Unix users have come to take seamless file sharing, network printer services, remote application execution, client/server program support, and multiuser access for granted. MIT's efforts with the X Window System have made graphics part of the mix, providing a network-integrated graphical application environment that's still unmatched by any other popular system. These and other services make Unix a perfect choice for networked applications, and Unix is presently the only operating system that integrates these elements.
Thomas Giammo, assistant commissioner of information systems at the U.S. Department of Commerce's Patent and Trademark Office, agrees: "In the next year or two, there's no other player that can give us what we need," he says.
In spite of that commitment, Giammo echoes the sentiments of both large- and small-scale users who feel stuck with Unix: It solves their short-term problems, but it creates other problems that make it an unlikely long-term solution. Giammo predicts that, as a result, Unix will be effectively wiped out in the mid-1990s; his money's on Windows NT. And he's not alone.
Once Unix became a commercial property, it lost its unity of purpose. In the hands of vendors who knowingly built cross-version incompatibility into their offerings (partly to lock customers into a particular flavor), Unix has generally failed to track user demands adequately. While it meets needs like multitasking and networking well enough, it leaves lesser issues unaddressed. For example, it's still too hard to use and administer, not compatible enough across implementations, and, for the most part, too married to its old character-based, command-line, serial-terminal traditions. "[Unix] is a hacker's version of an operating system that has grown respectable over time," says Giammo.
Overcoming the Problems
Improving Unix is much on the minds of Unix vendors. "If you have an X-based desktop with Mac-like features, the end user won't care that Unix is underneath," says Ken Arnold, an engineer at HP's Distributed Object Computing Program. As base-level machines get more powerful, they can better run the larger Unix operating systems. Then, to the end user, it is simply a matter of what off-the-shelf applications are available.
Avadis Tevanian, director of System Software at Next, agrees. He envisions a GUI that can run productivity applications side-by-side with user-made custom applications. "To get up to millions of units, you have to get rid of [the Unix shell]," he says.
But a GUI has not been an original, fully integrated part of Unix. Instead, it is usually a distinct layer poured over Unix's command line. In contrast, everything on the Mac is handled graphically, and the operating system is tuned for it. Even when the system crashes, a graphical dialog box tells you what went wrong. Most Unix systems boot and shut down in full-screen text mode, and when you need to jump into single-user mode for maintenance, you don't get graphical pull-down menus and radio buttonsyou get a prompt. For graphics to contribute adequately to ease of use, it must be an integrated part of the system.
The Next machine provides a shining example of Unix with highly integrated graphics. Unless something goes terribly wrong, or unless you request otherwise, you'll always have graphics insulating you from the operating system (see screen a). Even the most experienced users--the programmers--work inside a cozy graphical environment with visual tools that make it easy to develop NextStep applications. There's almost nothing the command line offers that the graphical environment does not.
The Next computer, though not a best-seller, has proved that the concept of invisible Unix is a viable one and that graphical workstations can be safely placed on the desks of general users. Jerry Marger, president of the law firm Marger, Johnson, McCollom & Stolowitz of Portland, Oregon, is a believer. He converted his firm's DOS network to a Next system because of the latter's ease of use and performance. "NextStep is awesome. Operating a Next is like operating a Mac, except that it's got a truck engine instead of a four- banger," Marger says.
PC Unix: Waiting for Standards
Plagued by bugs, incompatibilities (particularly in graphical and networking applications), and a lack of cohesive standards, PC Unix never made serious inroads on the desktop. Developers who were once excited about the prospect of a stable multitasking PC operating system quickly discovered the volume of problems to be worked around. Unlike DOS and Windows, which seem to run on every PC, PC Unix was fussy: You had to havethe right processor, bus, drive controller, tape drive, and display card. Software developers also found that they had to create a separate version of their applications for each flavor of PC Unix. While one version of a program can run on every SPARC-based system in the world, the same was never true for PC Unix.
Having so many different PC Unixes is hurting the Unix market. The solution, oddly enough, is to fill the market with a new batch of incompatible PC Unixes from the likes of Univel, Next, and SunSoft. There's some hope that, with the new high-profile players, some de facto standards can be struck. They'll be largely proprietary standards, but the success of Apple and Microsoft proves that those are the standards that stick.
Public standards can't easily be enforced. When every Unix provider can alter the source code to suit its individual needs, how do you keep providers from creating incompatible versions? By making binary licenses easier and less expensive to obtain, the keepers of Unix ensure that a broad set of functions known to be stable, including graphics and networking, will be part of most copies of the operating system. USL's deal with Novell, for example, gives USL a distributor it can keep a close eye on. There will, no doubt, be those who feel compelled to tinker with this stable code, but USL hopes they'll be so overwhelmed by competitors doing fully compatible releases that they'll get back in line.
USL spokesperson Ry Schwark says maintaining compatibility is not the only reason to leave the source code alone. "One of the reasons we have driven development down to the binary level is to off-load as much development responsibility as possible from the OEMs," he notes. Those OEMs will have to pass on the cost of customizing Unix to their customers--a painful choice in a market of shrinking margins.
Of the PC Unix players, SCO and Next are in the minority. Most of the others are lining up behind USL's SVR4.2 (see screen b). Unlike its predecessor release 3.2, SVR4.2 makes networking and graphics more or less standard elements of the distribution. Some vendors will probably choose to unbundle portions of it, but graphical applications built for SunSoft's PC Unix should recompile for Univel's version without changes.
It's not important that these vendors will ship different sets of standard applications. What matters is that developers will have a predictable standard software base on which to rest their code. There will be no more rewriting applications to make them run on variants of PC-based SVR4.2, if it lives up to its promise.
Next doesn't mind striking out on its own. It's bringing NextStep, based on Carnegie Mellon University's Mach kernel, to Intel-based systems in an attempt to capitalize more on the fervor over Windows than on any renewed interest in Unix. Obviously, it can't deliver binary application compatibility with its own Motorola-based workstations, but if Next can reduce porting of applications to a simple recompile, it has a chance.
Pencom (Austin, TX) has ported its X server Co-Xist from NextStep 2.1 to NextStep 486 2.1, a version that Next never released as a product. "We probably had more problems than some developers," says Chris Chauvin, product manager and principal developer of Co-Xist. He adds that the difficulties had more to do with the fact that Co-Xist has a dependence on the CPU. Pencom succeeded in doing the port by rewriting specific chunks of code that were optimized for the 68020 microprocessor. The rewritten code was optimized for the 486 within a week; the actual code rewrite took only a couple of days, Chauvin adds.
"Other people were having one- and two-day turnarounds" in porting Next applications over to the 486 platform, says Chauvin. This lends support to Next's claim of simple recompilation. Pencom will need to port the application again under NextStep 3.0 when Next makes it available in late summer. "Nobody's port will be considered complete until then," Chauvin notes.
Solaris 2.0, a derivative of SVR4, is going to be the acid test for Sun spinoff SunSoft. It remains to be seen whether the software arm of a hardware vendor is truly willing to create a level playing field. Sun is trying to set itself up with a virtual monopoly on SPARC operating systems and, through SunSoft and Solaris 2.0, is planning to extend its reach into the realm of high-end PCs.
The dicey part of this deal is that Sun's hardware side will still be developing its own software. Along with countless SPARC and Intel vendors, it will be licensing its own operating system from its own subsidiary, SunSoft. But Sun will have a unique advantage in the added-value department. As new technologies are born, managers at Sun get to choose whether to keep the jewels locked in Sun's safe or share them with the world through SunSoft. In any case, Sun freely confesses its intention of using proprietary enhancements to keep Sun-brand SPARC systems leading the market.
Through all this, SCO is claiming the high ground for its decision to pretty much stay put. Open Desktop 2.0, SCO's flagship Unix bundle, offers little more than previous versions and is still based on SVR3.2. SCO's approach seems to depend on SVR4's failing to catch on. "The conventional wisdom says that Sun, Univel, and so on are going to eat [SCO's] lunch," says Maureen O'Gara, editor of the Unigram-X newsletter.
Nina Lytton, president of consulting group Open Systems Advisors (Boston), thinks that developers will be attracted to Open Desktop 2.0 because it is proven and stable. "That strategy sounds like a yawn, but [SCO's] is the low- worry approach for Unix," Lytton says. She also notes that Open Desktop 2.0 will run applications written for SVR4.
The NT Factor
Windows NT is the wild card in Unix's future. Microsoft has a talent for whipping up public support for its products while they're still on the drawing board. The U.S. Patent and Trademark Office, under Giammo's guidance, is pre- sold, but he expects to wait a year or two after NT is available before adopting it, to ensure that it lives up to its promise. He's not interested in Windows: "From what I've seen," he says, "Windows is a kludge. It's even worse than Unix." He blames the "DOS baggage" Windows carries.
Microsoft claims NT will run DOS, Windows, new 32-bit applications, OS/2 applications, and X applications within an NT window (see screen c). Dwayne Walker, Microsoft's director of Windows NT and networking products, says, "If you had to summarize NT, you'd say it is powerful, scalable, easy to use, and compatible. And it offers multitasking and multiprocessing."
In addition, Microsoft says that NT will have built-in networking, including TCP/IP, and C2-level security. NT is portable from 386 and 486 PCs to RISC environments, supporting the Mips R4000 as well as DEC's Alpha RISC chips. Microsoft claims an advantage for developers, too: The 32-bit Intel-based applications they develop for NT will run native under Microsoft's Win32S extension for 16-bit Windows. With all that, Windows NT is clearly targeted at Unix, becoming what Walker calls "a high-end operating system for desktops and servers just as Unix is a high-end operating system for desktops and servers" (see "Unix vs. Windows NT" on page 140).
Microsoft clout alone could make NT a success. Next's Tevanian concedes that Microsoft's market presence is a big advantage, and according to O'Gara, at least one other Unix provider is worried: "Even the street fighters over at Sun are concerned when they look at what's happening with Windows. They see the way the market works: The users are a bunch of sheep, and Billy [Gates, Microsoft CEO] is the shepherd," says O'Gara.
For Giammo and others, Windows NT sounds like a dream come true, solving all compatibility headaches in one swoop. But with the end-user version at least three months away from shipping, NT is still a question mark. O'Gara calls NT "the unhardened operating system. You can't tell me that this thing is going to be ready to run mission-critical applications anytime soon." And the strength of Windows 3.1 on the desktop might not help Microsoft in promoting NT on workstations.
Open Systems Advisors' Lytton says Unix users are not impressed by the quality of Windows 3.1. "People thought that Windows would be the path of least resistance, but now that they've had a chance to really assess 3.1, the vision isn't as rosy as it was," she says.
Dataquest's Kirzner sees a bright future for NT: "We believe Windows NT will be the winner, because you can't stop the power of Microsoft, but Unix will do well," she says. Today, Unix has about 86 percent of the workstation market. By 1996, that figure will drop to 47 percent, due largely to NT. Dataquest projects that NT will claim 37 percent of the desktop market by 1996; DOS will retain 40 percent, and Unix will be a distant third with 7 percent (see figure 2).
Promises, Promises, Promises
While NextStep will be one of the contenders for the high-end multitasking desktop, it appears that the fiercest salvo fired at NT will come from an unlikely alliance: Univel. USL, looking to get serious about marketing and distribution, and Novell, hoping to shed some of its proprietary image in the newly competitive climate, have joined forces to offer a new shrink-wrapped Unix operating system that may be available as early as this fall.
Sold as SVR4.2 by USL and as UnixWare by Univel, it has a list of promises at least as long as NT's. What's surprising is that USL doesn't plan to stake out only the high end Microsoft has targeted for NT: SVR4.2's minimum configuration is a 16-MHz 386SX with 4 MB of memory. USL's literature talks about running the operating system comfortably on systems with 60-MB hard drives and even dares to suggest that SVR4.2 would run nicely on a notebook computer.
For standards, SVR4.2 hits a grand slam. The U.S. government and other large corporate customers will appreciate the Posix, FIPS, X/Open, B1/B2 security, OSF AES (Application Environment Specification), and IBCS (Intel Binary Compatibility Specification) compliance. Users of the Intel port will be able to run SVR3.2 and earlier SVR4 applications without hassle. And non-U.S. users get their share of attention, thanks to multibyte characters, alternative date and time formats, message tables, and an internationalized version of the Curses text-interface library.
Difficulty with hard disk file systems has caused a lot of people to turn away from previous PC Unix implementations. SVR4.2 proposes to solve this by introducing a journaling file system. Journaling protects data by first writing it quickly to a reserved area on disk. Then, when the system has time, the journal is flushed and the real file updating is done. At that time, the various Unix file-system data structures on disk are updated as well. Once it has written the data and updated the structures, the operating system deletes that journal entry and moves to the next one.
In the past, data was transferred from the memory buffers directly to its final destination on disk. If you turned off the power or reset a running system, chances were there would be several files in various states of incompleteness, where either the data or the file-system structure would be out of sync.
Modern Unix file systems write disk information in a predictable order that lets a utility called fsck comb through the file system and make sense of it again. But with journaling, that, presumably, won't be necessary. To bring the system up to date, you just play back the entries remaining in the journal. Eliminating the need for frequent file-system cleaning also greatly reduces the time required to reboot a system.
Univel is slimming SVR4.2's size and price by unbundling, breaking with the "everything in one box" tradition. But SVR4.2 is hardly stripped down; both the client and server versions include X and graphical shell support, source- level compatibility with seven common Unix implementations, DOS and Windows support, and networking. USL's role in this is to break with its own tradition of leaving most of the porting job up to licensees. USL will develop targeted binary versions of SVR4.2 (Intel, SPARC, Mips, RISC System/6000, and HP PA RISC versions are promised) and license them to packagers ready-to-run.
Those packagers, with Univel among them, can focus on adding value rather than on doing the basic port. The Intel version will ship from USL with drivers for a variety of display, disk, and SCSI devices. One frequent gripe about X--its dependence on bit-mapped fonts--will be answered by Adobe Type Manager, which will be a standard part of SVR4.2. And SVR4.2's desktop manager will include productivity applications and point-and-click shells for common system administration tasks.
Novell's involvement will get SVR4.2 into the company's authorized dealers, who will provide the added value of NetWare integration. Novell's dealers are an important channel, since it's largely the same one Microsoft is targeting with both LAN Manager and Windows NT. Novell is using SVR4.2 to get its share of the "open systems" pie and keep Microsoft out of Novell's stronghold.
Dataquest's Kirzner believes SVR4.2 will allay the fears of some Unix vendors. "It's a pure, unadulterated version of B1-level secure Unix, but for the PC," she says. "You buy as much or as little as you need." Kirzner says SVR4.2 will be priced to compete against Windows and OS/2, ostensibly in the $100-to-$150 range. USL declines to give a price for SVR4.2, preferring to leave it up to its licensees.
Not everyone is convinced that SVR4.2 will be successful on Windows' and MS- DOS's turf. "It would be stupid to put Unix on people's desktops," says Andrew Toller, a principal with DMR Group, a Toronto market research and consulting firm. He notes that DMR consultants usually use Unix as a back-end system fronted by Windows on the desktop in all but the most technical environments, particularly in software development applications.
Open Systems Advisors' Lytton expects a rough beginning for SVR4.2. "[SVR4.2] is having the same honeymoon [as NT] right now with the press," she says. "But any major new release always has problems that need to be worked out." She cites MOOLIT (the Motif/Open Look Interface Toolkit) as an example, claiming that it won't support any application that uses superset features of Motif.
In spite of the doubts, Toller says that many DMR clients have begun asking about SVR4.2. "These are people that until a year ago didn't even know what Unix was," he says.
When SVR4.2 hits the streets, if it's everything that's promised, it will mark a new era in PC Unix. USL's doing the binary work will help ensure cross- platform application compatibility. USL is straddling the Open Look/Motif fence by offering both GUIs, and it's also committing to technologies like ANDF (Architecture-Neutral Distribution Format) and DCE (Distributed Computing Environment). DCE adds functionality to existing networks by allowing you to share applications as well as data. You can transparently allocate CPU power to networked applications, running pieces of a program on several machines at once.
The goal is to develop these and other new technologies at USL and then get them to packagers in a ready-to-run state, complete with documentation. That's a major shift for USL, and one that more closely parallels Microsoft's strategy than the arcane Unix channels of old.
It's Just Starting
Unix isn't dead, but certainly the next few years will mark a period of intense change. Users and developers will be faced with important decisions. As the Unix market consolidates, the query "Which Unix?" may change to a less confusing question: "Unix or NT?"
To answer that question, Unix vendors, previously content to focus purely on technical issues, will have to spend more time solving ease-of-use concerns. And they will have to counter Microsoft's marketing might head-on.
Unix has a rich history. If the new pack of Unix vendors can keep from bickering among themselves, Microsoft will face a strong competitor of NT.
Editor's note: West Coast bureau chief Andy Reinhardt, news editor Patrick Waurzyniak, and senior technical editor at large Jon Udell also contributed to this story.
Unix vs. Windows NT
Technical issues that are a toss-up between Unix and Windows NT
-- Advanced hierarchical file system with long descriptive names
-- 32-bit linear memory model
-- Preemptive multitasking
-- Memory-mapped files
-- Shared libraries
-- Integral peer networking
-- Distributed global naming (assuming LAN Manager NT)
Areas in which NT looks better than Unix
-- Threaded SMP (symmetric multiprocessing). Some flavors of Unix (e.g., NextStep) are threaded but lack SMP, while others (e.g., SVR4) are SMP-capable but not threaded. Solaris 2.0 claims threaded SMP.
-- Layered device-driver architecture. NT will provide generic disk, display, network, and printer drivers that will allow driver writers to specialize relatively easily to create specific device drivers. Unix drivers are more monolithic and, hence, more difficult to create and maintain.
-- Interapplication communication. NT provides well-defined services (e.g., OLE, or Object Linking and Embedding);Unix does not.
-- A well-integrated scalable-font subsystem. While SVR4.2 will offer Adobe Type Manager and NextStep has Display PostScript, the dominant Unix GUI, X Window System, still relies heavily on bit-mapped fonts.
-- DOS/16-bit Windows support (unproven, but likely to equal or exceed that of OS/2 2.0 and far exceed that of any Unix).
-- Source codecompatible applications, with a hugely successful Windows 3.x installed base.
Areas in which Unix looks better than NT
-- Multiuser capabilities.
-- Distributable GUI (X)graphical applications are intrinsically client/server.
-- Well-standardized basic network services (TCP/IP, sockets, RPC [remote procedure calls]) that work reliably across wildly different hardware platforms and operating-system versions. NT offers a lot of options (named pipes, mail slots, NetBIOS, sockets, RPC, and even, potentially, DDE/OLE, if and when they are made network-aware), but the danger is that there may be no clear path for developers to write distributed NT applications.
-- Well-standardized wide-area E-mail.
-- Availability for more platforms than are likely, at least in the short term, for NT.
Unix's growth in dollar sales will actually outpace that of the market as a whole through 1996, according to market research firm Dataquest. The compound annual growth rate for Unix during this period is 27.5 percent, but only 5.9 percent for all platforms.
Although Dataquest expects Unix to have 47 percent of the workstation market by 1996, it will not fare as well in the desktop arena. There, Unix is a distant third at 7 percent, with Windows NT and MS-DOS holding a combined 77 percent market share.
NextStep (top), SVR4.2 (center), and Windows NT (bottom) all have unique visual traits. NextStep uses heavy shading, floating menus, and animated icons to achieve its glossy twenty-first-century look. SVR4.2's early allegiance was to Open Look, a capable but not popular X interface layer. USL's commercial release will include OSF/Motif as well. Windows NT offers a bridge to Windows users by providing no surprises: It looks, feels, and acts like Windows.
Tom Yager is a BYTE technical editor and author of Unix Program Design and Development for IBM PCs (Addison-Wesley, 1991). Ben Smith is a testing editor in the BYTE Lab and author of UNIX Step-by-Step (Howard W. Sams, 1990). You can reach them on BIX as "tyager" and "bensmith," respectively, and on the Internet at email@example.com and firstname.lastname@example.org, respectively.
Copyright © 1994-1997 BYTE