Can a new type of computer, designed from the outset to work on a network, topple the PC‘s dominance of the workplace?

S here is a war brewing-by the time you read this, we may be in the thick of it. On one side are companies such as Oracle

Systems Corp., Redwood Shores, Calif., and Sun Microsystems Inc., Mountain View, Calif., who view the network-not the personal computer- as the cornerstone of modern computing. They are proposing that a new, simpler type of desktop computing device, called the network computer or NC for short, take over the desktop from the more complex PC, thereby becoming the entry point to the network for the vast majority of peo- ple. A host of these NCs have already been launched into the marketplace.

Opposing them are Microsoft Corp., Red- mond, Wash., and Intel Corp., Santa Clara, Calif., which now dominate desktop technology. As this side sees it, moving away from the PC would be disastrous, not only for them but for industry and commerce as well. While they now concede that the PC could be modified to make it more manageable-a configuration they call the NetPC that is slated to appear later this year- they insist that it still makes sense to retain the WindowslPentium core of today’s PC.

Essentially, this is a war of ideas, about how networks-Inter- or intranets-should be used to distribute and control software and how comput- ers should be built for use in that environment. In Oracle‘s view, the NC is not for business use alone, it is a device that will make computer and network technology pervasive in everyone‘s every- day life. But before that can happen, it seems that the NC must first be tested in the business arena, at which the first NCs to come along have been targeted and where the PC now dominates.

The conflict’s outcome will decide the future direction of the computer industry and the for- tunes of those engaged in the struggle. If any- thing is certain, it is that the computing desktop will never be the same, and that future systems will be easier and cost less to manage.

The first shot No stranger to rhetoric or revolution, Paris got

both two summers ago from Lawrence C. Ellison, the chairman and chief executive officer of Oracle. Speaking at a technology forum there on 4 September 1995, he said, “We believe the world is moving from a desktop point of view to a net- work-centric point of view, and when you have a network-centric point of view, you don‘t need a device as complicated as a PC. You can get a ter- minal for as [little] as us $400 to $500.”

“A PC is a ridiculous device; the idea is so complicated and expensive,” he continued. “What the world really wants is to plug into a wall to get electronic power and plug in to get data.” With these words, Ellison, head of the second-largest software company in the world, in terms of rev- enue, introduced the basic concept of the net- work computer, or NC, which he sees as trans- forming the world of computing.

A fellow participant in the Paris forum was William E. Gates, chairman of Microsoft, the soft- ware company that is at the heart of what Ellison called the ”ridiculous” PC. In response to Ellison, he said that the dominant desktop device would never be Ellison’s “dumb terminal.” And noting Ellison‘s statement that the future system would be less expensive because it had no disk storage, Gates claimed, “You‘ll still need a way of storing

RICHARD COMERFORD, Senior Editor

IEEE SPECTRUM MAY 1997 no1 8 - 9 2 3 s l 9 7 l $ t n on01997 IEEE 21

the applications that you download from the network, and your personal data."

Since then, Ellison has been hard at work fleshing out his basic premise, and Gates has been just as active trying to prove him wrong.

The gathering clouds While those first shots may have been

fired in Paris 2 1 months ago, the seeds of the conflict were sown earlier in this de- cade. During the 1990s, Microsoft tight- ened its hold on computing with its pro- prietary DOS/Windows operating sys- tems, which run mostly on Intel hard- ware In 1995, by dint of its savvy market- ing, it in effect controlled FC software technology development, and seemed likely to dictate desktop computing tech- nology for the toreseeable future.

In addition, by coupling the Windows NT operating system it introduced in 1993 with the latest Intel Pentium proces- sors, Microsoft saw it could also launch an attack on the market for higher-perfor- mance network systems. Then, too, the runaway success of Web software start- ups like Netscape Communica t ions Corp., Mountain View, Calif., awakened Microsoft to the importance of the Web and the Internet

The Internet network server market is broadly shared today, but its No. I play- er is Sun. Sun's systems-based on its own open Sparc microprocessor archi- tecture (which it licenses to others) and its version ot the Unix operating sys- tem-are at the center of many compa- nies' Internet operations.

Systems running Windows NT for Networks on Intel processors are a threat to Sun. Windows NTs share of the server market has been growing rapidly, and Paul Otellini, Intel executive vice president for

worldwide sales, told a meeting of New York City financial analysts last May that "Servers ... represent one of the last large untapped market segments tor our archi- tecture . . . . The Pentium Pro processor is positioned to address this directly."

Oracle is the leading developer of data- base management system (DBMS) soft- ware, which is widely used in network en- vironments. As companies have begun to tie their DBMSs into their intra- and Internet plans, Oracle has made it known that it wants to be a central player in the Net game, and it currently provides many tools for developing Web/database appli- cations. Although Microsoft's database products are no equal to Oracle's today, Microsoft has begun setting standards for database application programming interfaces (APls) that could give it con- siderable clout in Oracle's market, partic- ularly as it ties itself more closely to the Web and the Net.

T h e nature of their business plans makes Intel and Microsoft the natural enemies of Sun and Oracle. At no time have the heads of those companies shied away from disparaging their rivals' prod- ucts. But it was at that Paris technology forum that the competition came to a head. Whereas network technologies such as clienthewer computing had been in development since before the dawn of the decade, it was the ever-spreading deployment of the Internet and its atten- dant technologies (file transfer protocol [ftp], e-mail, and the World Wide Web) that showed the way to implement widely useful computer networks.

With the rise of Netscape and other Internet software companies, Microsoft- which had admittedly underestimated the Internet's impact-refocused itself on developing Web-centr ic software, re

vamping its Windows desktop and server operating systems to be more Net com- patible. Indeed, the next version of its Windows 95 operating system, now in beta testing, is built around Web tech- nologies, including Microsoft's version of Sun's Java technology.

Waking up with Java The Java software technology created

by Sun was developed by people who were totally enmeshed in the Net. Sun was ahead of Microsoft in grasping how impor- tant the Internet could be to the future of corporate computing. After all, Sun had been delivering its workstations to the technical community for years, and in so doing was working with many scientists and engineers around the world who were developing fundamental Internet and Web technologies. It was thus natural that, as technologists started adding more nodes to the Net-not only for research but for commerce, too-they relied on the com- puters they knew, and Sun's business grew.

To continue promoting the use of its systems for Internet applications, Sun set up a separate team to develop advanced network software and hardware technol- ogy. The fruit of that group's work was an object-oriented programming lan- guage called Java, designed to run on any system in a network. The premise underlying Java's development was that all computers can be described generical- ly, that is, in terms of general functions like display and input capability; when a computer is described in this way, it is called a virtual machine.

Java's developers realized that i f all computers can be described in terms of general functions, then programs that are written using only general functions could run on all computers. Turning the

[11 Data from the Gartner Group show that, when it comes to the total cost of owning a PC, its purchase price is just the tip of the iceberg. The biggest part of that cost, which has more than doubled since 1987, results from end-users tinkering with PC hardware and software.

22 IEEE SPECTRUM M A Y 1997

Disk

Diskette

Flash memory

CD ROM

No

No

Optional 4 MB

No

Modem External supported

Boundless Technoloqies Inc.

HDS Network Svstems Inc.

Network Computing Devices (NCD) Inc Comuanv Sun Microwstems Inc. Wyse Technology Inc.

Winterm 4000 series

(226 X 60 X 174)

Product (size in mm: H X W X D )

XL Station (45 X 248X 2290)

XLT Station (72 X 370 X 390)

JavaStation (250X302X117,

Tower; box, 70 X 190 X 310)

@Workstation (-)

Explora" series (41 X 184 X 254)

$749 for 8MB (black-and-white

display)

$695 for 8MB (without display)

$800 for 8MB (without display)

$750 (without display)

Price >us $900

Intel i960 (several speeds)

Intel i960, 66 MHz

PowerPC, 28 and 33 MHz;

66 MHz in future

Sparc; Java chips in future

StrongARM 1 I O , 200 MHz Processor chip

4-64MB; 16MB typical

8-12MB; 16MB typical

8MB+; 16MB advised

8-64MB; 8MB standard

8-16MB; 8MB standard Memory

Token ring or Ethernet MAC Permanent ID Ethernet MAC Ethernet MAC Ethernet MAC Ethernet MAC

No No Changeable ID Possible in future No No

"- Optional <IGB, on internal PC Card

No ----+-- Optional

Optional 2 4 M B on internal PC Card

Optional

N 0;

likely in future No;

yes in future Optional 4MB

B N 0; future release No No No

PC adapters No No No No Yes

Flat panels and many others Display Several offered Options Options Several offered

640 X 480 1024 X 768 1280X 1024

640 X 480 1024 X 768 1280 X 1024

Display resolution, pixels

640 X 480 1024 X 768 1024 X 768 Through

1600 X 1200

Small PC, std. PC, 3270, 5250 options Standard PC Keyboard Standard PC I Standard PC Standard PC

PSI2, two-button

pc, two-button Mouse pc, I two-button

pc, three-button

External supported

PSI2, two-button

Internal option

I None 1 External supported supported

I

NCD. with IBM Corp, dl50 nianufa<tures i h e Network Station to much tlie 5 a n i P 5p?iificdtions the Explora Source D H Browri Associates In i , I E E E Spectr i im

~ O \ l l < l ~ 8 I < l ~ I l l l I : \ l I I i ! I l l< 1 / 1 1 l l l ~ h l i l l ' 23

[2] Working with Oracle Systems Corp., Advanced RlSC Machines Ltd. came up with this design, based on its ARM7500FE processor, that meets the Network Computer (NC) Reference Profile specification. Design flexibility is a hallmark of the NC; many of the items shown in the diagram are optional. For instance, while an NC must have a text input device, it need not be a keyboard; it could be a pen device or touch-sensitive screen. The pointing device could be a mouse, a joystick, or a pen device. The output device could be a monitor or a Tv, or even a virtual-reality headset.

general functions into the specific tasks would be a simple job for the computer on which the program ran.

For a computer to run a Java program, then, it need only have a so-called inter- preter to translate the Java code into in- structions it can execute directly. Unlike current Windows applications, Java pro- grams do not care what the computer‘s underlying operating system software or processor is. By freeing people from a par- ticular type of system software, Java was Sun’s declaration of war on proprietary system software in general and Microsoft‘s Windows in particular.

Reality check Of course, it is possible to add the abil-

ity to interpret Java to a Web browser

24

that runs on a PC. So the Java approach to computer technology might not have had any immediate impact on the PC had it not been for studies of what it costs companies to buy, operate, and maintain PCs that were conducted by a number of market-research firms.

At the Paris forum, OracleS Ellison sug- gested that the annual cost of supporting a PC was US $5000 per year. This figure may have been an understatement, unusu- al for Ellison. Numbers released in 1996 by the Cartner Group Inc., Stamford, Conn., indicate that the total cost of own- ership (TCO) for a PC, over the typical amortization period of three to five years, is more than $40 000, or between $8000 and $13 000 per year [Fig. 11, as Intel’s Otellini pointed out in a May 1996 meet-

ing with New York financial analysts. According to the Gartner Group, the

total cost of ownership for a PC has four main factors. The PCs actual cost makes up about 21 percent of the annual figure. Of the remaining 79 percent, support expenditures contribute 27 percent and administration accounts for 9 percent. But a whopping 43 percent of the cost is at- tributed to “end user operations,” which means the user is fiddling with the PC, wasting time adding special software or hardware and using it for tasks that are not strictly job-related, Gartner says.

What Intel and Microsoft have to con- tend with in continuing to sell PCs-and what Oracle and Sun hope will give pause to would-be corporate customers for those systems-is reducing this total cost

IEEE SPECTRUM MAY 1997

ATM = asynchronous transfer mode IP = Internet protocol ISDN = integrated-services digital network LAN = local-area network SLIPIPPP = serial line interface protocollpoint-to-point protocol

of ownership. As Sun’s chief executive officer, Scott McNealy, stated, “The thing that is going to drive this business is TCO .... We plan to bring that number down to $2500 a year, all up, all in.”

Defining the NC While Ellison‘s lambasting of the PC in

Paris had called attention to his cause, he was now faced with having to make good on his claim that the low-cost system he described could be built. On 26 February 1996, he took the first step, showing a pro- totype of a network computer built around a RISC processor from Advanced RISC Machines (ARM) Ltd., Cambridge, United Kingdom. Of course, showing a prototype is no guarantee that such devices would eventually move into mass production.

Then, on 20 May 1996, Oracle and four other companies-Apple, IBM, Netscape, and Sun-released the first guidelines for constructing an NC. The document was called the Network Computer Reference Profile and it stated its intention ”to pro- vide a common denominator of popular and widely used features and functions across a broad range of scalable network computing devices, including personal com- puters [emphasis added].”

In a spirit of openness, the group was saying that PCs could join. Indeed, it was possible for a PC to have all the attributes of an NC. But providing a bare-bones

COMERFORD -THE BATTLE FOR THE DESKTOP

description of what an N C must have would make it difficult for a typical PC, with its large hard disks and considerable RAM, to compete with leaner, built-to- the-guidelines boxes.

The profile describes a set of four user-interface resources that are required: for output, a screen with a minimum res- olution of 640 by 480 pixels-a video graphics array (VGA) or its equivalent- and audio output; for input, a pointing device such as a mouse, and text input capability. Note that the display require- ment would permit TV monitors to be utilized, and text input, while including keyboards, could also refer to some other means, such as handwriting or selecting text from an on-screen alphabet. The pro- file also states that persistent storage, such as a disk drive, while not forbidden, is not required, either.

Though deliberately vague about re- source requirements such as RAM and ROM or processor performance, the profile is specific about the communication proto- cols that must be supported to make the NC a network-centric device. First of all, an N C must handle the transmission con- trolhternet protocols (TCPAP). It must also support ftp, telenet, the simple net- work management protocol (SNMP) first used in IBM mainframe terminals, the net- work file system (NFS), and the user data- gram protocol (UDP) used by the NFS.

Source: D.H. Brown Associates Inc., IEEE Spectrum

Again, in keeping with the network- centric view, the N C must be designed to boot up on the network. Just as a PC has its basic I/O system (BIOS), which tells it to load its main operating system from a hard disk drive, the N C has a boot pro- gram that tells it how to connect to the network and identify itself. The guideline offers one of two boot configuration op- tions: the dynamic host configuration pro- gram (DHCP) or the bootstrapping proto- col (Bootp). While there are differences between the two protocols (Bootp requires that information in a server database be manually pre-configured while DHCP allows for dynamic configuration), what‘s important is they allow an NC to discover its own IP address, the address of a server, and the name of a file to be loaded into its memory and executed.

In addition, the NC must support the fundamental Web standards: the hyper- text transport protocol (HTTP) for trans- ferring Web pages between a server and a browser, and the hypertext mark-up pro- tocol (HTML), which describes how a browser should display a Web page’s con- tents. There are also four multimedia file formats-JPEG and CIF for images and WAV and AU for sound-widely used on the Web and required for the NC, and the Java virtual machine with its runtime environment and class libraries.

So that the NC can handle e-mail,

25

[3] An ad hoc industry group called the Desktop Management Task Force created a layered scheme to make NetPCs easier to manage. Each piece of its hardware and software is uniquely identified and this information, saved in a database, lets administrators track its status.

three protocols are required: the simple mail transfer protocol (SMTP), t h e Internet message access protocol version 4 (IMAP4), and the Post Office protocol version 3 (POP3). Lastly, there are emerg- ing APIs for security as well as, optionally, Smartcard ( I S 0 786) standards and the Europay/MasterCard/Visa specification for secure financial transactions.

In addition to providing an NC-plat- form example [Fig. 21, Oracle did more- it announced the names of at least a dozen other companies that were also hit- ching their wagons to the NC. The com- panies were categorized as system manu- facturers, technology partners, and dis- tributors andor system integrators.

In the first category were such compa- nies as Acorn Computer Group, Bound- less Technologies (formerly SunRiver Data Systems), Nokia Group, Olivetti Group, and Wyse Technology, which would make NCs. The technology part- ners, who would create the hardware and software components needed to build the systems, included ARM, Cirrus Logic, Corel, Digital Equipment, Mitsubishi Electric, Motorola, and VLSI Technology. The distributors and system integrators

2 6

included Hitachi, Japan Telecom, NEC, NTT Data, Otsuka Shokai, Softbank, and The Japan Research Institute.

With these announcements, the NC acquired credibility and, within a few months of the reference platform an- nouncement, several of the manufacturers introduced NCs.

NCs for real Today, there is a whole host of NCs to

choose from, the specifications for a sam- pling of them are shown in Tables 1-4. The first company out of the gate was HDS Network Systems Inc., King of Prussia, Pa., which introduced its @work- station on 11 June 1996. Like the other NCs to follow, it was based on a RISC processor (a 66-MHz Intel i960) and was much smaller than a PC (about the size of an X terminal). Also, like the others that followed, its $749 price in a usable con- figuration did not quite meet Ellison's "under $500'' prediction.

On 5 September, IBM Corp. led the way for the reference-document partners, unveiling the IBM Network Station. The box, jointly manufactured by IBM and the X-terminal manufacturer Network

Computing Devices (NCD) Inc., Moun- tain View, Calif., was built around a 32- bit PowerPC RISC chip. At 191 by 267 by 32 mm, it was smaller than the HDS system, even "smaller than an algebra textbook," as IBM put it at the time.

Network Computing Devices markets a similar system under the name Explora using its own NCDware as the operating system. The Explora's Wincenter soft- ware gives users access to Windows appli- cations running on a server. Wincenter is based on the WinFrame technology cre- ated by Citrix Systems Inc., Fort Lau- derdale, Fla. Originally created to let workstations and other computers run Windows applications, it was easily ad- opted by other NC manufacturers, in- cluding Wyse and Boundless. Network Computing Devices is also offering the HMX series of NCs, which are based on 64-bit MIPS R4000 RISC processors and have greater memory capacity. None of the systems has a hard-disk drive, but a floppy drive is an option.

Boundless Technologies, Austin, Texas, is offering the XL and XLC series of NCs, which are based on Intel's RISC processor: the i960. The XL is slightly larger than the NCD Explora (45.0 by 248.0 by 229.0 mm) and does not offer a floppy-disk option.

For its part, Wyse Technology Inc., San Jose, Calif., has chosen to base its Winterm 4000 series NCs on the blazing- ly fast (200 MHz) StrongARM 110-a processor designed by Digital Equipment Corp . using the ARM architecture it licensed from that company.

While all these NCs are network ready, what is obvious about them is that their core processors are different from unit to unit and, while the operating system soft- ware may be based on common technolo- gies, the implementation for each proces- sor is also different. Thus NC reference designs based on different processors are now springing up. At the Comdex Show in November, Motorola offered a reference design for NCs, developed together with NCD, in which the core central processing unit (CPU) was an MPC8xx embedded PowerPC RISC processor, while Digital Equipment, working with ARM, recently promulgated a reference design based on the StrongARM processor.

Sun has not yet issued a Sparc or Java chip reference design, but on 29 October 1996, it did introduce its own version of the NC, the Javastation. Its CPU is a microSparc I1 t 00-MHz RISC processor equipped with both Citrix software to run Windows applications and Sun's JavaOS to run Java applications.

Wintel replies In spite of their differences, all these

introductions might have launched the NC on a short road to success had it not

IEEE SPECTRUM MAY 1997

SBCS, DBCS = single-/double-byte character specification

been for an announcement made just one day before Sun introduced the Java- station. On that day, Microsoft and Intel announced an initiative to develop their own reference design, for a new type of desktop computer t o be called the NetPC. And just as the NC initiative months before had announced its sup- porters, so, too, did the NetPC. Included in the ranks were most major US.-based PC manufacturers: Compaq Computer, Digital Equipment, Gateway 2000, Hewlett-Packard, NEC, Packard Bell, and Texas Instruments. Conspicuous by its absence was IBM.

Like the NC, the NetPC would be sim- pler than a PC, it would "not require the flexibility and expandability of the tradi- tional PC." But unlike the NC (and not surprisingly), this new type of system would be built around only one processor architecture, Pentium, and only one oper- ating system, Windows. It would also be based on two initiatives to make PCs more manageable: one that Intel had begun ear- lier, called Wired for Management, and a new one from Microsoft, called Zero Administration for Windows.

In addition to meeting the above re- quirements, and unlike an NC, a NetPC must have an internal hard disk for cach- ing. Further, systems would have to com- ply with Plug and Play requirements, and

have unique IDS for each installed device and add-on. Each system would have to be uniquely identifiable by a machine-read- able ID. Addressing the problem of end- user fiddling around pointed out by the Gartner Group, the PC case would have to lock physically, and there would be no internal expansion slots available to the end-user. Optionally, the system could have a floppy drive, a CD drive, a PC Card slot, a Universal Serial Bus connec- tor, and/or an IEEE 1394, the so-called "Firewire" connector [see last month's issue of IEEE Spectrum, pp. 19-25].

Although the NetPC must be locked, a recent announcement by Intel, Microsoft, and Compaq, called the Device Bay ini- tiative, uses FireWire to let users attach a peripheral box into which devices can be easily plugged. Any devices plugged into the Device Bay must meet the NetPCs ID requirements, so that network administra- tors could know what was being added and still control system configurations.

The point of the NetPC specification is not to reduce the purchase price of a system to under $500. It is to provide a system that would be inherently easier to manage in a corporate environment; i f users cannot get inside the box and if ad- ministrators can see exactly what hard- ware and software the user has, then the box will be inherently more manageable

Source: D.H. Brown Associates Inc., IEEE Spectrum

and its support costs would drop. Some have said that the NetPC is

merely an attempt to blunt the thrust of the NC, and undoubtedly that is the case, in part. But Intel, for one, had been look- ing at the problem of the TCO for some time. It was certainly aware of the problem in May 1996, when Otellini addressed the analysts meeting in New York City and indicated that having a Windows NT/ Pentium Pro system, along with tools for managing distributed systems, would reduce TCO by about $3000 per year.

A month before the NetPC announce- ment, on 24 September 1996, Intel for- mally unveiled the Wired for Management initiative in New York City in a day-long business computing forum. As part of its presentation, Intel unveiled a LANdesk (for local-area network desk) management product aimed at PC help desks, a series of Pentium-Pro-based servers with built-in management capabilities, a hardware- monitoring chip to track PC health, and new guidance for customers on imple- menting more manageable systems.

Behind these introductions was work that had been going on since 1992, when informal meetings on the issue of desktop management blossomed into a coopera- tive strategy for managing systems. It be- gan with Intel, Microsoft, Novell, SunSoft (a Sun Microsystems' business unit), and

27 COMERFORD - THE BATTLE FOR THE DESKTOP

Source: D.H. Brown Associates Inc., IEEE Spectrum

SynOptics Communications, but Hew- lett-Packard, IBM, and Digital Equipment soon joined them in forming the Desktop Management Task Force.

For the next two years, the task force worked at developing a solution in the form of a desktop management interface and the management information format. In 1994, the effort was joined by Apple Computer, AST Research, Compaq, Dell, and Syman- tec, and by September of that year, the task force had developed its initial plan.

Management components The basic design involves three layers.

At the bottom is a group of manageable products, or components. These are div- ided into five basic groups: software ap- plications, operating systems, hardware products (such as system motherboards, network cards, mass storage devices, and displays), peripherals that can attach to the systems, and systems hardware (de- vices such as processors, memory, and other ICs).

The attributes of these products that are needed to manage the system (part type, part number, operating character- istics, and so on) are described in a lan- guage called the Management Interface Format (MIF). MIF has a defined gram- mar and syntax, and the MIF file for each manageable product is added to the MIF database when the product is installed in the system.

Products communicate through a com- ponent interface to the middle, or service, layer. The service layer is actually a pro- gram that resides on the network and col- lects information from products, manages the data in the MIF database, and passes information to the highest layer: the man- agement application layer. The service layer uses a special management interface to communicate with management appli- cations: the management console, the desktop management application, and the LAN management application [Fig. 31.

2 8

This structure, coupled with the four concepts of the Zero Administration ini- tiative, forms the basis for the NetPC. The first of Zero Administration's con- cepts is automating the system software's configuration, which is what the NC does. When a NetPC is first booted, its op- erating system will automatically update itself from a server without user interven- tion. Then an "automatic desktop" feature built into the operating system will pro- vide the user with all the applications that he or she is permitted to have, automati- cally installing them when they are invoked by the user.

The next concept is "automatic system- state storage " All data created or entered by the user can be automatically reflected to a server, so users will have access to all their data no matter what NetPC they use to log onto the network.

The third concept is central adminis- tration and lock-down. All aspects of a NetPC's configuration are to be control- lable by a central administrator through the network, and the system's configura- tion locked. Yet the central administrator can give a user more flexibility, allowing him or her to add various software or hardware options to the NetPC.

This leads directly to the fourth con- cept: applications flexibility. This flexi- bility means that the NetPC user can have thin clients-Web-style applica- tions-as well as full applications for personal productivity and larger client- server operations.

It is Microsoft's intention to support the Zero Administration concepts both in Windows and in the applications it delivers. It will also support the work of other software vendors to develop appli- cations that will run on this operating en- vironment. Indeed, Microsoft's announce- ment that its future operating systems would incorporate these features, virtual- ly identical to those supported by Citrix's WinFrame, sent Citrix's stock into a tail-

spin from which it had not fully recov- ered as of this writing.

The bottom line The key distinctions between the NC

and the NetPC hark back to the original debate between Ellison and Gates in Paris. The NC has no hard disk and hence can not locally store the full applications that Microsoft refers to when discussing appli- cation flexibility for the hard-disked NetPC. But the types of applications that NC supporters see in its future are all thin clients based on Java, and therefore do not require the NetPC's hard disk. The NetPC is tied to one platform architec- ture, Intel's and Microsoft's, while the NC opens the door to other processors and operating systems.

For now, it looks as though either type system could trim business outlays. What remains, then, is to see which side can deliver all the resources needed to deploy their systems quickly, efficiently, and at the lowest lifetime cost. +

To probe further Since the NCs are aimed at the intra- and

internet applications, its not surprising tha t there i s a weal th o f informat ion about them on the World Wide Web. The NC Reference Profile can be found at http://www.nc.ihost.com/, along w i th a list of i t s endorsers. The NC suppliers list- ed in this article's table all have Web sites: Boundless Technologies, http:// www.boundless.com/main.htm; H D S Network Systems, http://www.hds.com/; Network Comput ing Devices, ht tp: / / www.ncd.com/; Sun Microsystems, http://www.sun.comljavastation/; and Wyse Techno I ogy, h t t p://www.wyse .co m/.

More on the NetPC can be found a t http:// www.microsoft.com/windows/netpc/,on Microsoft's Zero Administration initiative at http://www.microsoft.com/windows/zaw/. and on the Wired for Management initia- tive at http://vwwv.intel.com/managedpd.

IEEE SPECTRUM MAY 1997