By Daniel P. Dern
Network World, 2/24/97

If you believe what some research firms say, network computers (NC) will not only be more inexpensive to buy than desktop PCs, they'll be significantly more inexpensive to own and operate. By at least a couple of estimates, the reduction in total cost of ownership (TCO) should hover around $3,000 per year - a significant chunk of change by any measure.

This assumes, of course, you're prepared to believe the numbers analysts are coming up with. We, being skeptical reporter types, were not. So we dug around behind the numbers and came to the firm conclusion that nobody can say with a reasonable degree of accuracy or confidence how much, if anything, an NC might save you. There are too many variables that have yet to be explored and lots of assumptions, omissions and gotchas in the data being bandied about.

Nobody, for example, has yet delved into questions addressing factors such as the productivity benefits to be achieved with one device vs. another, or the integration and migration costs associated with NCs. Assumptions are rampant, simply because there is no NC knowledge base to work from; instead, analysts extrapolate from user experiences with X terminals or diskless workstations.

Analysts freely admit this is not an exact science.

''We're not fussy about what the exact numbers are,'' says Bill Kirwin, vice president and research director at Gartner Group, Inc. in Stamford, Conn. ''People report anywhere from half to three times what we estimate - it's more important for our numbers to be in the ball park.''

This is not to say that NCs flat-out won't save you money. They might. On the other hand, they may only transfer administrative costs to a different place, such as from the client to the server. The point is you can't take analyst estimates at face value. You've got to look behind the numbers, understand how they were arrived at, and think about what they don't tell you.

''Each analyst's studies reflect different types of user bases, as well as different choices of what costs to include and exclude,'' says Greg Blatnik, vice president at Zona Research, Inc. in Red wood City, Calif. Zona, for example, focused on workgroups of about 15 users and a single server. Gartner, on the other hand, looked more at large companies with revenue of $1 billion or more.

''These sorts of studies definitely don't present the entire picture,'' Blatnik says. ''They usually don't take into account factors such as productivity, or new application deployment costs.'' Additionally, he notes, there's a wide variation in costs for the same thing, such as the salary for a system administrator - they simply cost more in most large metropolitan areas.

Proper use of these studies also requires that you understand what you sacrifice by replacing PCs with NCs. If you were Federal Express Corp., would you replace your trucks and planes with bicycles because the latter cost less to buy and maintain? Answer: Probably not, except in some very specific locations.

What's an NC?

• Client-oriented network computer. This is typically a diskless device running a thin-client operating system. Code such as Java modules execute locally but reside on the server, so server connectivity is essential. Some types may not inherently support Windows applications, so you may also need a server-based emulator such as NTrigue, from Insignia Solutions, Inc. Example: Sun Microsystems, Inc. JavaStation; Wyse Technology, Inc. Winterm 4000; and Network Computer Standard from IBM, Netscape, Oracle Corp. and Sun.

• Server-oriented network computer. This is basically a diskless smart terminal with the application executing on a server. Example: Citrix Systems, Inc. WinFrame used with Wyse terminals.

• NetPC: Windows-based diskless or disked PC that has no data stored locally; its ''identity,'' meaning user profile and configurations, are server-based. Examples: Microsoft Corp.-Intel Corp. NetPC specifications.

The standard against which NCs are compared and contrasted in most analyst studies is a PC running Windows 3.X or Windows 95, attached to a LAN for file and print services, with typical quality of management. ''Typical,'' it appears, translates to anywhere from average to poor; few corporate desktops are being managed anywhere near as well as they could be.

Crunching the numbers

According to Gartner's figures, Sun JavaStations, along with similar offerings on the way from other vendors, are likely to be the most cost-effective of the NC choices, saving up to 40% in purchase and support costs as compared to desktop PCs. That might be more than you'll save by simply managing your PCs better, but the numbers don't take into account the cost of migrating to NCs in the first place.

Putting several of the leading market research firms head-to-head-to-head in what is admittedly not an apples-to-apples comparison, due to the varying methods research firms use to reach their conclusions, it appears annual TCO savings of $2,000 to $3,000 for each user station, or 30% to 45%, wouldn't be suprising. Are these figures reasonable? A National Aeronautics and Space Administration facility at Cape Canaveral has 1,000 hybrid desk top/NC systems from Advanced Modular Solutions, Inc., devices that can be configured as either disked or diskless PCs with data and identity saved on the server. The facility reports savings of 80% of its user administration budget.

Indeed, judging by analyst projections shown here, any NC strategy should offer savings of 20% to 30%, with the Java-class solution doing best. But so can implementing ''best practices'' with existing systems, according to Gartner.

''We believe you can get 25% to30% cost reduction by simply managing your current PC environment better,'' Gartner's Kirwin says. Better management includes providing adequate tech support resources, applying state-of-the-art system management including remote diagnostics and control, doing electronic software distribution, and imposing greater levels of standardization.

With NCs, standardization is pretty much enforced, Kirwin says, because you have a standard client configuration, built-in system management technology and centralized management. "Network computers are the embodiment of best practices," he says.

It also can include simpler things, as Ryder System, Inc.has shown. Wilbert Williams, group project manager for Ryder, based in Miami, says his company has centralized the ordering of PCs and reduced the number of vendors and configurations it deals with. ''Up-front capital costs of PC hardware and software have been cut by at least 20%,'' Williams says.

Further, he anticipates reductions in support costs because ''with this smaller number of system types, we can now test 100% of our applications on each PC configuration before shipping them to the field.'' Equally important, the value of the PC systems to the user and user satifaction have both gone up because the company can ship bug-free applications sooner.

At present, about 6,000 Ryder PCs are involved; Williams anticipates 10,000 to 15,000 PCs will ultimately be deployed under these cost-saving measures. So it's clear that NCs have serious competition in the dollar-savings arena.

And there is the sheer range of numbers given by some analysts. Depending on your exact situation, there could be cases where it actually costs more to have and hold an NC as compared to a PC.

No wonder there's a lot of cynicism out there regarding these cost-of-ownership studies.

''Most of the comparisons are made on the basis of clearly insufficient information and insufficiently precise assumptions, or they are applicable to a range of buyers and apps, but not the whole market,'' says Thomas Nolle, president of CIMI Corp., a Voorhees, N.J., consultancy.

Among Nolle's questions are whether licenses for Java-based code will cost less than existing software and how existing applications might perform on diskess machines. ''What will Microsoft Office do when the code says, 'Write current keyboard buffer to disk,' but your system doesn't have a disk?''

Art Hutchinson, senior consultant at Northeast Consulting Resources, Inc. (NCRI), a Boston-based consultancy, agrees it's too early to accuratey assess NC ownership costs. ''Nobody really knows yet - there isn't enough data or experience to say,'' he says. ''The products are just emerging, and a lot of the costs such as support or management may be hard to identify early in the game.''

Research firms admit, and even caution, that there is potential for variation in the numbers they present and what your experience might be.

Bob Tasker, senior vice president of computing research at The Yankee Group, sees his firm's figure as a ''two sigma number,'' meaning ''66% of the people impacted will be within plus or minus $500 of it, and 95% within $750.'' He also cautions hat the TCO figures do not include the cost of migrating applications to the new environment.

The wide range in International Data Corp.'s (IDC) numbers reflect the variety of sites in its survey, according to IDC analyst Chris Christianson. In our chart, those spending $4,000 annually per user represent large centralized sites running production environments; those shelling out $12,000 each year per user are ''large distributed environments with hundreds of remote locations and with business processing workloads,'' Christianson says. IDC surveyed several hundred PC and terminal-oriented shops, he says, with the NC data extrapolated from diskless PC user information.

Rather than replacing PCs, as a rule Christianson sees NCs going to new turf, especially in terminal-based environments. In that scenario, NCs are likely to cost more - 25% per year for each user over five years - but they will bring improved functionality.

Making the choice

Assume for a moment that a given NC can do the tasks currently being handled by PCs, and that there will be cost reductions as claimed. You've still got to get from here to there, meaning migrating applications, buying new equipment and staging everything over.

To determine whether it's worthwhile, Kirwin says you've got to compare your estimated migration costs to the anticipated savings over time. For example, if it will cost $5,000 per system to migrate and you only expect $1,500 per system per year in reductions, it's a raw deal. ''If you can't recover your migration costs within two years, skip it,'' he says.

Theo Forbath, a consultant at NCRI, says you should pick your spots for NCs. ''If you live in a world where dumb terminals play a vital role, you will do well with an NC,'' he says. Similarly, for supporting workers who rely on a relatively fixed desktop, such as a receptionist, support person or administrator, he says companies may see lower TCO with NCs. On the other hand, ''a lot of folks will still need notebook computers, which can't currently be replaced by NCs.''

You've also got to consider whether TCO estimates measure enough of the problem to be a useful guideline.

''The NC folks would like to have you believe support costs of existing PCs is the only thing that matters,'' argues NCRI's Hutchinson. ''I haven't seen discussions with clearheaded assessment of the 'benefit of ownership' for each approach, such as worker empowerment and work teams, and whether employees are still able to be as productive, mobile, flexible and creative for the business.'' If the value of the employee goes down because, say, they can't plug in while on the road, he says any savings may be negated.

''What we tell our clients is that this is a real horse race,'' Gartner's Kirwin says. He recommends a three-step course of action. First, clean up your existing PC installed base, implementing best practices. Then look to places such as branch offices that can benefit from server-based offerings that deliver Windows-based applications, like Citrix's Winframe. Finally, target specific-function NCs such as JavaStations for highly structured environments. ''There's a place for all these technologies, and none will be a clean sweap of the market,'' he says.

CIMI's Nolle remains cynical. ''A network computer will never be a substitute for a desktop system for any user with a significant degree of computer literacy. The storage and flexibility limits of NCs mean that knowledgeable users will always be dissatisfied with performance.''

What about NCs as a terminal replacement? ''Network computers will not be a substitute for 'dumb terminals' in any applications where the connection would have to be rearchitected, or where there is no economic benefit to doing the conversion,'' Nolle says.

There will, he believes, be some cases where NCs will replace 'dumb display functions' during the next five to 10 years. ''But the network computer will not displace desktop computing significantly.''

Instead, sensible planners will consider the various flavors of NCs where they have to deploy new hardware or replace old hardware anyway. Meanwhile, be on the lookout for ways to manage those existing networked desktop PCs more efficiently.