| Are network computers just another false alarm, like the
first personal digital assistants (PDAs)? That's the down-to-earth question you can't help asking when high-profile CEOs such as Oracle's Larry Ellison, Sun Microsystems' Scott McNealy, and IBM's Lou Gerstner describe their utopian vision of a $500 Internet appliance that could replace today's PCs. It's a question that provokes strong reactions from both the visionaries (network computers are coming soon, and they'll change everything) and the naysayers (network computers are impractical, underpowered throwbacks to dumb terminals). This much is certain: The idea is sweeping the computer industry like wildfire. Acorn Computer, Apple Computer, Geoworks, IBM, LSI Logic, Oracle, Silicon Graphics, and Sun Microsystems are just a few of the companies that are either developing network computers or licensing their technologies to other vendors. In late January, Oracle unveiled a prototype of its network computer in Japan (see the photo). Bandai, a Japanese company, licensed Apple's Pippin architecture and plans to ship a low-cost system this month (see the sidebar "Inside the Pippin"). Many sources predict that network computers will be in mass production by the end of the year. Things are moving so fast that people are still inventing new names for the machine. Network computer (NC), Internet appliance, Web PC, Java terminal, Internet access device, browser box, net-top box — they're all minor variations on the same concept. The concept is a dirt-cheap computer that discards today's overweight OSes and bloated, platform-specific applications. Instead, it runs a microkernel OS and platform-independent software written in an interpreted language such as Sun's Java. Monolithic, kitchen-sink applications give way to component-based programs and applets that you download from a network or the Internet only when you need them. In theory, you could do almost anything with an NC that you do with a PC, except it will cost you a lot less for the hardware and software. Connectivity costs, however, may be significant. To transform that theory into reality, NC vendors must capitalize on four key technologies: cheap, speedy microprocessors; lightweight system software and componentware; platform-independent programming languages; and fast, affordable access to networks. But the biggest question is: Who needs one? Not Just for Neophytes NCs aren't just for home users who can't afford or can't comprehend a full-blown PC. In fact, proponents expect the early adopters to be Fortune 500 companies. "The key to [the corporate] market is for people to be able to do what they do today, but with less hassle," says Andrew L. Laursen, vice president of Oracle's Network Computing Division. (Although Oracle says it has no plans to manufacture NCs, it is licensing technology to other vendors and is hoping that NCs will create more demand for server software.) Corporations are the first target for NCs for two reasons: They already have high-speed networks in place, and their administration costs for conventional PCs are soaring. Administration costs have become so significant, the low purchase costs of NCs won't likely be their main attraction for corporations. According to studies by the Gartner Group (Stamford, CT), the five-year cost to a business of owning a Windows PC is about $40,000. By moving software maintenance to servers and limiting the ability of users to tinker with their machines, NCs could be easier for MIS departments to maintain and less susceptible to user errors. So far, no one has predicted the ownership costs of an NC, but X Window System terminals provide a good starting point. The Gartner Group estimates that companies with a 24-user X-terminal and Unix-server network can save more than $200,000 over five years compared to companies with a PC-based LAN of the same size. The X Business Group — which is a more biased source — estimates that network administration of a 40-seat X-terminal network could be $270,000 less expensive over five years than a similar PC LAN. NCs could also save money by slowing the costly spiral of upgrades. Each new release of a behemoth business package, such as Microsoft Office, requires more memory, more hard drive space, and more CPU cycles. More powerful computers, in turn, spur developers to create even bigger programs. If programmers broke up those monolithic applications into smaller, dynamically loaded components, the computer could accomplish (in theory) the same tasks more efficiently. It would be more efficient because you would be loading only the part of the program that you need, when you need it. A simple example: One sample Java applet is a small (27 KB) spreadsheet. It doesn't perform linear regression or offer the extensive graphics of Microsoft Excel. However, the idea of components is that complex mathematics and graphics would be separate modules that you would download when you need them. At 28.8 Kbps, it would take about 7 seconds to download such a spreadsheet — or about the same amount of time it takes to load Excel 7.0 on a 120-MHz Pentium system. Although componentware was invented for conventional PCs, NCs could benefit from it even more. NC proponents envision hardware designed for a distributed environment in which administrators store and maintain the software components on a network, not on the client. That network could be a corporate LAN, an enterprise WAN, the Internet, or a secure intranet. Once the NC downloads the software from the network, however, the components execute locally on the client. NCs are not dumb terminals, nor are they X terminals, where applications (called clients in X) execute on a central server. Some analysts aren't convinced that NCs will save corporations significant money. Improved system management tools could wipe out any cost advantage of NCs, and some corporations are turning conventional PCs into virtual NCs by modifying the hardware and software to prevent user tampering. "This is a race between tools for better desktop manageability and a special appliance that is effectively lobotomized to make manageability easier," says Bill Kerwin, an analyst for the Gartner Group. "While network computers are a much more manageable environment, the bottom-line cost might be about the same." The second major market for NCs is schools. Schools desperately need more computers to prepare their students for a job market in which PCs are becoming as common as telephones. Yet according to a study by the Arrington Research Group (Monterey, CA), the ratio of students to computers in U.S. public and private schools (grades K-12) is nearly 7 to 1. Almost half of those computers are 8-bit Apple IIs, Commodore 64s, Commodore PETs, and TRS-80s. "School districts have told us that $500 is an important entry point, a price point that would allow them to buy computers en masse," says analyst Michael Arrington. "It could move computers from the media lab onto the students' desktops." Finally, there's the consumer market. Nobody knows how many people would be interested in buying a home computer if it cost only $500 and gave them access to the burgeoning services on the Internet. The potential is so lucrative, however, that at least a dozen companies — including Asian conglomerates that specialize in high-volume manufacturing of consumer-electronics products — are rumored to be hotly pursuing the idea. The push for NCs is also a rare chance to outflank Microsoft. As the biggest vendor of monolithic, PC-centric applications and system software, Microsoft seems most vulnerable to a paradigm shift that could make its underlying platform obsolete. Some proponents of NCs can barely conceal their glee over Microsoft's recent defensive moves, such as its licensing of Java and its announcement of VB Script. The danger is that wishful thinking could cloud the vision of network computing and lure its supporters over a cliff. Dream Machines Microsoft CEO Bill Gates is skeptical about inexpensive NCs. "When someone talks about a terminal at a certain low price, ask them what they dropped from a PC," Gates said at a recent Internet Strategy Workshop in Seattle. "Did they drop 3-D browsing? Did they drop motion video? Did they drop local storage? PCs have to drop in price, but they also have to advance in capability." Indeed, some observers scoff at the idea that vendors can build and profitably sell a useful computer for only $500. Yet the industry sold millions of sub-$500 computers in the early 1980s, before the IBM PC standard raised expectations and prices. Perhaps the best-selling computer of all time was the Commodore 64, which made its debut in 1982 at $595 and soon fell to $200. Today, a 486/33 with a small hard drive and minimal RAM could probably sell for about $500. Such a system might represent a low-end NC. Of course, a Commodore 64 isn't powerful enough for the kind of applications users demand today. However, computer technology has come a long way over the past decade, and it's possible to build surprisingly powerful boxes at low prices. Apple designed the Pippin before the current hype over network computing, but the Pippin, which uses the Power Mac architecture, could easily become a $500 NC. The latest home game machines from Sega, Sony, and Nintendo also come close to the design points for NCs, and they're capable of graphics that would put most PCs to shame. They all cost less than $500. The question isn't whether $500 computers are possible; it's whether a $500 computer can meet the expectations of users. Those expectations vary widely, depending on the market and the application. A business user who's juggling spreadsheets has different needs than an eighth-grade student who's researching a report using the Internet, or an adult who's interested in home shopping. Therefore, it is unlikely that a single machine at a single price will fit the needs of the corporate, education, and home markets. "Network computers will not replace PCs on the desktop," says John Robb, an analyst with Forrester Research (Cambridge, MA). "They're more of a supplement to PCs. They'll go into places where there are ease-of-use issues or cost issues." Mobile PCs will prove even more resistant to obsolescence, because it will be a while before wireless networks are ubiquitous and fast enough to make mobile NCs practical. Eventually, however, public NCs could become as widespread as public phones. Instead of toting a notebook and a bag of accessories, all you'll need is a credit card. After you log on to the network from the airport, airliner, or hotel room, the NC will download your working environment and all your files. Network Catalyst The four technologies required to make NCs practical are all important, but networking is the keystone that supports the entire structure. The World Wide Web, which didn't even exist five years ago, is the most remarkable networking development for NC proponents. On the surface, the Web is just a visual, point-and-click user interface on the Internet. Seemingly overnight, however, it has transformed a bleak, text-based landscape into a colorful cyberworld brimming with opportunities, both artistic and commercial. As Netscape's initial public offering proved, the Web may be sparking the last great gold rush of the millennium. The Web and the Internet enable network-centric computing on a global scale. However, only a few million users with high-end PCs currently have access to the Web. And few of those users enjoy the kind of bandwidth that's required to fulfill the grander ambitions of Web visionaries. NCs could get by with the narrowband access provided by analog modems (14.4 to 28.8 Kbps) or basic-rate ISDN (64 to 128 Kbps). However, a device that drinks most or all of its software from a network really needs a bigger straw. That's why the ultimate fate of NCs may depend on the adoption rate of technologies such as asynchronous transfer mode (ATM) and cable data modems. Cable modems could be the primary catalyst; hundreds of millions of homes, schools, and office buildings throughout the world are already wired for cable. Hewlett-Packard and Motorola have announced contracts to sell more than half a million cable modems this year. A proposed industry standard (IEEE 802.14) would codify the protocols for data access over cable at a speed of about 10 Mbps. (Theoretically, 25 Mbps is possible, but line noise is a major problem.) Further out is ATM, which can range from 1.544 to 622 Mbps. If ATM becomes widely available, network-centric computers could become a lot more attractive. "Once people get access to the Web at broadband speeds, there's no going back — especially if it can happen at a price point that's competitive with narrowband access," says Tony Stelliga, vice president of product marketing for MicroUnity Systems, whose innovative Mediaprocessor could be the CPU in tomorrow's broadband NCs. One catch: If cable modems take off, can the networks keep up? The Internet may bog down under the strain of millions of new users, and cable TV companies will have to invest in new equipment. To minimize their investment, cable companies may link anywhere from 500 to 2000 customers to a single data network node. If 10 percent of those customers subscribe to cable data services, as many as 50 to 200 users might be sharing the same slice of bandwidth during peak hours. "It's a nightmare in the making," says Vincent Schmidt, product manager for ADC Telecommunications, which makes T1 modems. "Cable data modems are only the tip of the iceberg. The back end cannot handle that many high-speed users." Schmidt thinks that phone companies — which have more sophisticated switching networks already in place — will intervene with their own high-speed alternatives to cable modems. Some analysts expect these alternatives to be in place in 1997. "Eventually these problems will be solved," Schmidt says. "If there is a path to money, people will find it." Cheap Chips High-speed networking won't save the day if NCs are stripped-down boxes that run like slugs. For NCs to seriously challenge PCs, they must deliver comparable user experience and performance. That's not easy for a product that retails for $500 or even $1000. A 120-MHz Pentium chip alone costs about $350, and some versions of the Pentium Pro cost nearly $2000. That's OK, respond NC proponents, because NCs don't need leading-edge microprocessors. Instead, NCs will ride the trailing edge of the price/performance curve, where they can reap the benefits of plunging prices. The industry focuses so much attention on state-of-the-art CPUs that you can forget how much processing power is available for just a few dollars. For example, Integrated Device Technology (IDT) and NEC Electronics manufacture a wide range of 64-bit RISC processors based on R4000 cores from Mips Technologies. IDT's R4640 executes 175 MIPS (Dhrystone: 2.1 MIPS) — comparable to a 100-MHz Pentium — and costs only $28. NEC's new 133-MHz VR4300 executes 120 MIPS (or 80 SPECint92, 60 SPECfp92) and costs only $35. Versions of the new StrongARM processor, a joint project of Digital Equipment and Advanced RISC Machines (ARM), deliver from 115 to 230 MIPS. The StrongARM is expected to cost about $30 to $50 (see "StrongARM Tactics," January BYTE). This kind of bargain-basement CPU power makes the latest generation of home game machines possible. Nintendo's new Ultra-64, designed in partnership with Silicon Graphics, uses a 100-MHz version of NEC's VR4300. "When we first showed people what the Ultra-64 could do, some Silicon Graphics engineers were stunned," says John McCrea, the manager of Silicon Graphics' WebForce unit, which is making tools for Web developers. "The Ultra-64 has graphics equal to a $100,000 Silicon Graphics workstation of five years ago." The company is putting that same graphics-rendering technology into its Cosmo Motion engine, part of a graphics library for Java. By using modern design tools and standardized modules, today's engineers can integrate fast RISC cores with numerous other components on a single chip. This saves even more money and results in highly integrated parts that are tailored for specific customers. For instance, Sony's PlayStation, which is a $299 game console, has a single ASIC that integrates a Mips R3000A CPU core, a specially designed geometry transform engine, a graphics processing unit that can render 1.5 million polygons per second, a sound processor, a JPEG decompressor, and miscellaneous I/O logic. LSI Logic custom-designed Sony's ASIC in only eight months. LSI has a new fabrication process that can squeeze 49 million transistors on a chip. (By comparison, a Pentium Pro CPU has 5.5 million transistors.) LSI recently announced a superchip that's intended for low-cost NCs. Called the Internet on a Chip, it can include a Mips R4x00 CPU core of the vendor's choice, a graphics processor, a sound processor, a memory controller, and circuitry for a V.34 modem — all for $50. Alternatively, LSI can replace the V.34 modem with an ISDN interface for about the same price. And the CPU cores aren't slouches. For midrange NCs, LSI suggests a 40-MHz R4010 that executes 100 MIPS; for higher-end requirements, LSI offers an R4020 core that delivers 200 MIPS. To make an NC, says LSI, all the vendor needs to add is 4 MB of DRAM ($110 at current wholesale prices), an analog chip for the V.34 modem ($2 to $5), and perhaps some ROM for the OS kernel and Java engine (the price varies according to how much, but mask ROM is inexpensive). "There has been a lot of negative reaction in the press because nobody has been able to show a bill of materials that would make a sub-$500 box possible," says John Daane, who is vice president and general manager of LSI's Communications Products Division. "We designed the Internet on a Chip to prove it could be done at this price point." The Internet on a Chip is part of a wave of highly integrated superchips coming this year from suppliers like Chromatic Research, MicroUnity, Nvidia, Philips Semiconductors' Trimedia Product Group, and Vadem (see "Chip Fashion," November 1995 BYTE). Intel is less likely to be a player in this market. It prefers to sell cutting-edge microprocessors, which are much more profitable. Those profits pay for the billion-dollar foundries that Intel needs to maintain its leads in production capacity and process technology. Another company that's a pioneer in low-cost systems is Acorn, which has been making inexpensive computers for years. In 1995, Acorn introduced the A7000, a machine aimed primarily at the U.K. educational market. The A7000 is based on the ARM 7500 microprocessor, which is comparable to a 486DX2/66. Although the A7000 has only 2 MB of RAM, its RISCOS OS is capable of multitasking several programs in that space, says Peter Bondar, director of Acorn's Applied RISC Technologies Division. Bondar says Acorn is in "advanced stages of discussion" with U.S., Japanese, and Korean companies that want to build NCs with technologies licensed from Acorn and ARM. (Acorn founded ARM and owns 42 percent of the company.) Acorn is offering its efficient RISCOS, its low-cost board manufacturing, and ARM's processors. "The debate over the $500 Internet computer is a moot point as far as we're concerned," says Bondar, who adds that his company's low-cost computers prove the NC concept is viable. Because $500 computers will be a low-margin business, the market probably won't attract many U.S. manufacturers. More likely, Pacific Rim companies that specialize in high-volume consumer electronics will make NCs. Where's the Software? Hardware is only half of the equation — perhaps the easiest half. In a market where many people shun the Mac because it has merely thousands of applications instead of tens of thousands, NCs are at a real disadvantage: They have no software at all. "A dedicated Java machine can't run one line of code that's currently for sale anywhere, and that's a major obstacle to overcome," says Michael Goulde, an analyst for the Patricia Seybold Group (Boston, MA). "Have you ever seen a football game in which the receiver starts running downfield before he finishes catching the pass? He ends up dropping the ball. I think that's what is happening here." Everything seems to depend on Java. Other programming languages may have roles to play — Telescript, VB Script, and TCL come to mind — but Java was largely responsible for igniting the mania over NCs, and Java is the critical element that will make NCs from different vendors compatible with each other. Java is a high-level interpreted language that runs on a virtual machine. In other words, the Java engine is a run-time interpreter that insulates applications from the nitty-gritty details of the OS and hardware. That's why it hardly matters which OS or CPU is inside the NC. Vendors can choose what delivers the best price/performance advantage, and they can even change the OS and CPU when something better comes along — without breaking any applications software. As long as there's a Java engine on the computer, it can run any program written in Java or JavaScript (see "Wired on the Web," January BYTE). This architecture is crucial to the platform independence and low-cost model of NCs. It's also the next logical step in the evolution of software development. In the 1970s and early 1980s, programmers wrote software in assembly language because it was fast and memory-efficient. As PCs grew more powerful and the pressure for new features became more urgent, programmers began working with high-level compiled languages such as Pascal, C, and C++. Compiled code runs slower and requires more memory than hand-tooled assembly language code, but it's easier to write. It runs acceptably fast on today's systems. Interpreted languages such as Java are even easier to work with than compiled languages, and developers don't have to port their code to different platforms. The flip side of interpreted programs is that they're even slower than compiled programs. This could be the Achilles' heel of NCs, whose low prices will preclude the fastest microprocessors. If performance becomes a major problem, full native compilation could be the fallback position. Java programs are compiled into a bytecode format that's interpreted by the run-time engine. It's possible to create a compiler that turns Java bytecode into native machine code. To preserve software portability, the compilation could happen automatically when the NC downloads Java objects from the network. In a business environment, the server could store precompiled objects. Some NCs may cache the compiled binaries of frequently used objects on a local hard drive. (Caching would also help compensate for slow network connections.) To attract developers to Java, Sun and Oracle are sponsoring a Java Cup contest with $1 million in prizes. According to Geoff Baehr, Sun's chief networking officer, there will be no shortage of entries. "People are actually writing this stuff," he says. "This isn't one of those things where people will be saying, `Whatever happened to Java?' in a few years." Proponents claim that Java will unleash a wellspring of entrepreneurial spirit that dominant software companies and a restrictive distribution system currently stifle. NCs could alter that balance of power in two ways. First, a shift toward components and applets could make room for smaller developers. Second, because NCs by definition live on a network, they will encourage new models for merchandising software. Instead of buying an unlimited-use license, as you do now, you may pay a small fee every time you download an object. Or you'll pay a small monthly fee for unlimited use. Or perhaps you'll pay nothing at all — just as most people today don't pay for Netscape Navigator. "We're on the brink of having someone with a good idea being able to deliver that good idea to literally millions of people overnight," says Oracle's Laursen. Microsoft Fights Back The debate over NCs has led to much loose talk about the destruction of Microsoft. Can a company with enough programmers to populate an infantry division compete in this brave new world? At the Internet Strategy Workshop, Gates said he isn't worried. "We've always competed with free software. For about $10, you can buy 10 CD-ROMs with lots of free software, and that software will be available on the Internet, just as it's been everywhere else. "We think most companies would rather pay $100 per module to make sure their employees have the best spreadsheet or the best word processor," Gates continued. "We think they'd rather pay 0.1 percent of the employee's salary to make sure they can use that software all they want instead of leasing it on a per-use basis." Still, Microsoft was compelled to license Java from Sun. This isn't the cataclysmic event some observers think it is, however. Microsoft needed Java to keep its Web browser competitive with Netscape's. There are no indications that Microsoft is shifting emphasis away from Visual C++ and Visual Basic. In fact, Microsoft announced VB Script, a subset of Visual Basic that's intended to compete with JavaScript, the Java-based scripting language in Netscape Navigator. Because VB Script is an interpreted language — and because Microsoft will freely distribute the run-time engine — it, too, has the potential to become a platform-independent standard for Web developers. It's likely that all Web browsers and NCs will support both JavaScript and VB Script. In theory, the regular version of Visual Basic could have been Java. Both are high-level interpreted languages whose programs are compiled into an intermediate bytecode or p-code that in turn executes on top of a run-time engine. But Microsoft has not ported Visual Basic to other platforms, except for the abbreviated version in Microsoft Office. Until now, Microsoft's idea of "cross-platform" has been Windows 3.1, Windows 95, and Windows NT. However, we hear rumors that Java delivered the jolt that Microsoft needed to rethink this strategy. Don't be surprised if Microsoft ports the full Visual Basic run-time engine — though probably not the development environment — to the Mac later this year. Paradoxically, Microsoft's greatest strengths are the very things that opponents claim are its greatest weaknesses: Windows and Office. Although both are behemoths, there's undeniable value in those megabytes of code. Writing a collection of Java components and applets that duplicate the functionality of Word, Excel, PowerPoint, and Access is what's known in the industry as a "nontrivial task." So don't hold your breath. Yet if NCs are to make inroads into the corporate market, the industry must measure their worthiness against the prevailing standard, and the standard is Microsoft Office. Similar reasons motivated Apple to base Pippin on the Power Mac architecture. "Ultimately, the success of these machines will depend on the applications that are available for them," says Mark Orr, Apple's manager of business development for Pippin. "Anytime you launch a new platform, you've got the classic chicken-and-egg problem. You need content to attract an installed base, but you need an installed base to attract developers to create content." Of course, NCs don't necessarily have to battle PCs head-to-head for the same kinds of applications. Some observers think corporations will dedicate NCs to other tasks: presentations, self-guided training, on-line documentation, Internet research, E-mail, database access, and light-duty spreadsheet functions. "There may still be many employees in the world who don't have a computer because their job doesn't revolve around Microsoft Office," says Silicon Graphics' McCrea. In Search of a Kernel One of the most crucial technical challenges will be finding a lightweight OS that will let NCs compete against Microsoft Windows. Although the OS doesn't matter in the sense that Java programs will run on any system that supports the Java engine, the OS must be small and fast enough to run within the reduced resources of a low-cost machine. There are many memory-stingy microkernels around, but they're generally designed for embedded applications and lack such luxuries as high-level graphics libraries. The OS in an NC must satisfy the needs of programmers, who have grown accustomed to the rich APIs of Windows, the Mac OS, and the various flavors of Unix. Also, to make the basic concept of an NC feasible, the OS must have some mechanism for dynamically loading distributed objects over a network. Fortunately, this problem has attracted a great deal of attention in recent years from companies that are working on technologies for interactive TV set-top boxes. The futuristic TV networks appear to be a little further in the future than experts once predicted, so set-top-box designers are eager to adapt their technologies to NCs. (A skeptic might wonder if the whole push for NCs is merely an attempt to amortize the R&D expended on set-top boxes.) Oracle is developing an OS called NCOS for its NC reference design. The company has released little information about this OS, except that it's based on a microkernel that supposedly runs in "considerably less" than 1 MB of memory. One reason NCOS is so small is that it dispenses with the big APIs of conventional OSes. Instead, separate subsystems and class libraries of Java objects will handle high-level functions. For example, Oracle claims it has a subsystem for remote-method access that runs in about 64 KB, and a graphics subsystem that needs only 300 KB. "And it's just as rich as the Mac OS and the graphics subsystems in Windows," says Oracle's Laursen. "There's going to be an ultralightweight version of CORBA [Common Object Request Broker Architecture] that lets me communicate between my Java objects across the network," Laursen says. "The name services are pretty straightforward. They just have to cache the set of objects I've been communicating with, so I don't have to do a name lookup every time I access some remote object. So through a hierarchical caching architecture and a lightweight distributed-object environment, I can do everything I do with PCs." Laursen points out that today's OSes and applications are memory hogs. They load the bulk of their code into RAM at start-up, and that code stays in RAM whether you need it or not. With an NC, the OS and applications could be more object-oriented, and the objects won't load into memory until the OS summons them. Two other OSes that seem particularly well suited for NCs are Acorn's RISCOS and Geoworks' GEOS. Both run well on one-generation-old CPUs with as little as 1 MB of RAM, yet both are rich OSes that offer much to developers and users. Introduced in 1987, RISCOS is a 32-bit OS with cooperative and preemptive multitasking. It can reside and execute in 4 MB of ROM, and multiple applications can run in as little as 2 MB of RAM. RISCOS supports a general-purpose desktop GUI and more than 2000 applications. It also has the advantage of running on inexpensive ARM processors. GEOS traces its ancestry back to the Commodore 64 but currently runs on x86-compatible CPUs. Geoworks says it is porting GEOS to multiple RISC chips, partly because some of its licensees want to introduce NCs later this year. (Geoworks won't name the customers who are planning NCs, but GEOS licensees include Brother, Canon, HP, Nokia, and Toshiba.) GEOS, like RISCOS, can reside and execute in 4 MB of ROM, and that includes some built-in applications. It can get by with as little as 1 MB of RAM. Geoworks CEO Gordon Mayer notes that Canon and Brother already sell GEOS-based dedicated word processors that are really thinly disguised PCs. They include a full suite of applications plus a floppy drive, an ink-jet printer, and a color monitor — all for about $500. "If you add a modem or a cable modem, you've got a network computer," says Mayer. In other words, packaging may define an NC as much as technology. Conceivably, someone could turn a warehouse of old 486-based PCs into trendy new NCs by stripping the systems down to their essentials. Simply replace Windows with a smaller OS, install the Java run-time engine, and bundle them with a cable modem or a network card. Nintendo, Sega, and Sony could likewise turn their home game consoles into consumer browser boxes with little effort. Nintendo has already demonstrated a Web browser on its Ultra-64. Microsoft has shown a Web browser running on a set-top box designed with NEC. Especially for the consumer market, there's a fine line between a game machine, a set-top box, and a scratch-built NC. Reality Check Before hailing the birth of a new era, consider two disturbing facts. First, as we go to press, nobody is willing or able to demonstrate a functional system that was designed from the start to be an NC. That's rare for a technology that's supposedly a few months away from mass production. IBM, which made a big fanfare about NCs at Comdex, refused to discuss the subject. Second, there seems to be an attitude that "If we build it, they will come." The assumption is that a $500 NC will uncork pent-up demand that's currently frustrated by the high prices and intimidating complexity of PCs. But easy to afford doesn't necessarily mean easy to use. Businesspeople who are already familiar with PCs will adapt quickly to NCs, but the holy grail is to lure millions of new users. "I don't want to have to buy a book that's called $500 Network Computers for Dummies," says Apple's Orr. "Too many people are assuming that the biggest barrier to personal computing is price, when in fact there's all kinds of research that shows most people simply see no relevance in personal computing." The corporate applications for computers are well established, but where are the breakthroughs that will make NCs both compelling and accessible to the 65 percent of U.S. households that don't own a PC? Will NCs simply reinvent the wheel at a lower price? Possible answers: The Web will supply the compelling applications, and the point-and-shoot user interfaces developed for set-top boxes will mollify computerphobes. Neither answer is farfetched. The Web is evolving rapidly, and the killer application seems within reach. It could be home shopping, if Web browsers become as easy to use as mail-order catalogs. It could be home banking, especially if cashless debit cards catch on, turning NCs into automatic teller machines that dispense electronic cash. It could be something nobody has thought of yet. Even then, NCs — like Apple's Newton — might be a little too far ahead of the curve. Years may pass before the Web becomes a secure place for commerce. Nathan Myhrvold, group vice president of Microsoft's Application and Content Group, thinks that PCs will drive the Web's growth in the foreseeable future. Myhrvold estimates that 100 million people in the world who have PCs aren't on the Internet, despite so many free browsers. "It's actually more expensive for [these people] to get on the Internet by buying a $500 terminal," he says. "So where are the next 100 million users of the Internet going to come from? The people with PCs, or some small percentage of people who buy $500 terminals?" Oracle's Laursen acknowledges that NCs probably won't be an overnight success, but he predicts they will be a major force by the turn of the century. "I think it's a 10-year play," he says. "If you look at the changes that happened over the last 10 years with PCs, I think these changes are equally fundamental. Except I don't think it will be as hard." It's Inevitable The concept of a $500-or-thereabouts computer is undeniably appealing — just as it was 15 years ago, when low-priced computers were commonplace. But two things have changed since then. First, the Web is creating a giant electronic library that globalizes information. Second, millions of people will need access to that library to keep up with a world that's fueled on information. Computers aren't luxury items anymore; they're survival tools, but they're still priced like luxury items. Without question, the technology exists to produce a low-priced volkscomputer that could serve those millions of workers, students, and consumers. By year's end, we'll have one. Sidebar:Where to FindAcorn Computer
Four Key Technologies
Architecture of an NC Old PC vs. New NC
NCs Could Make Notebooks Obsolete
Oracle's Network Computer Mark Orr, Apple's Manager of Business Development for Pippin Tom R. Halfhill is a BYTE senior editor based in San Mateo, California. You can reach him on the Internet or BIX at thalfhill@bix.com. Inbox / July 1996Trust the Web PC"Inside the Web PC" (March cover story) was a great read and made many good points. But do you think computer users are going to trust their most sensitive data to the Net, where they will have to worry about both security and accessibility? Small hard drives neither cost nor weigh that much anymore. Can it be that we are really just talking about a new breed (with a new name) of cheap computer? What will a network computer really add that a cheap PowerBook can't? Frode Hegland http://www.liquid.org frode@liquid.org I expect that most "network computers" will have hard drives for caching software and storing local copies of user files, but the user won't have to manage the hard drive, any more than PC users today must manage the contents of their RAM-based disk cache.By caching copies of the OS and applications on a small hard drive, the NC can run faster. Most NC users won't store their files on the Internet. In corporations, universities, and even public schools, they'll store files on LAN servers, just like they probably do now. Users at home could store them on a local hard drive or on a secure server at their Internet service provider. This frightens some people, but consider: professionally maintained servers are backed up regularly; locally stored information is only as safe as the physical security of the device it's stored on; all the most sensitive information about your life — bank accounts, medical records, etc. — is already stored on network servers. A PowerBook could indeed be used as an NC if the operating system and applications were replaced. Unfortunately, there's no such thing as a cheap PowerBook. — Tom R. Halfhill Copyright 1994-1998 BYTE |