NOTE: this is a first draft of an article for Please do not copy or distribute it; but please do send me comments and ideas for making it better.

Series Summary: Total Cost of Ownership Revisited


This is the last in a series of eight articles by Paul Murphy comparing the costs and benefits of the Windows and Unix architectures. In this article he summarizes the conclusions and case backgrounds of the earlier articles, so the user can easily select specific cases that model their own systems decisions. The body of the current article is a summary of the conclusions drawn from individual articles earlier in the series, as well as a discussion of what issues are raised by an attempt to quantify the total cost of ownership for a systems architecture and then apply the results to decision making.


Minimizing total cost of ownership isn't so much about technology as it is about management. Picking the right technology helps - but knowing what to do with it is even more important. Is Unix cheaper than Windows? Detailed analysis seems to show that it is, but does that make it the better business decision? That was the real question addressed in this series, not "is Unix cheaper?" but "is Unix Smarter?"

Defining the TCO question

To most people the TCO, or total cost of ownership, concept is about checks written. It relates to the total cost of the decision to own or use something and implies that the decision between two choices, A and B, can be largely determined by comparing the total costs associated with each choice. Implicitly this assumes that A and B produce similar value, or returns to ownership. Since this is rarely the case, most uses of the TCO concept in decision making extend the idea beyond a simple "sum of checks written" to incorporate some acknowledgment of the differences in benefits available from each choice even when those differences are largely unquantifiable.

In the specific context of the Unix versus Windows decision the most obvious problem with our understanding of the cost components going into the TCO idea is that we cannot determine whether Unix is cheaper than Windows without first defining the terms of the comparison and then the basis for making judgments about the results of the comparison.

Defining what we're comparing isn't that difficult. Although "Windows" is a brand label, not a product or set of ideas, while "Unix" is a name for a set of ideas encapsulated in a wide variety of products, we can define comparisons in terms of real products embedded in specific business and technology contexts. For this series those definitions were couched in terms of the system architectures implied by the labels. Thus "Windows" means Windows client-server with a PC on every desktop and applications on shared servers accessed via local area networks [LANs]; while "Unix" means Unix servers with applications delivered via desktop smart displays [defined below] and single level networking.

Setting the basis for evaluation and ranking is much more difficult. In what sense is one approach better than the other? what conflicting valuations must be resolved for a ranking to make sense? Base things only on dollars, and we could conclude that Unix is many times more expensive than Windows because the data center's HP-K580 cost more, in 1995, then the price listed for a Celeron with Windows ME in this morning's drugstore flyer. Attach a time dimension to the dollars without expanding our comparison further and we'll conclude that the best route to lower TCO for a PC is to return the box, unopened, to the vendor.

Clearly, we'd like to recognize and include all costs of the decision to buy and use product group A rather than product group B but cannot because many of those costs are unquantifiable. It is relatively easy to add up checks written, benefits foregone, delays accepted; and so on, but quantifying the value of effects like the organizational design and motivational consequences of each choice is more problematic.

To get this right requires us to consider issues that go beyond dollars and technology to include personal issues and their effect on judgment and decision making. Systems decisions are not made in a purely financial context; there are organizational and personal aspects that matter. Does engraving a scarlet "U" on the decision maker's forehead impose social costs that should be considered? Does a Unix decision made today have "fitting costs" if the organizational computing framework has been established by thirty or forty years of mainframe history? What non dollar costs are associated with the decision to write the first check in a TCO series if the decision falls outside the realm sanctified by local majority opinion?

As a simple example, consider that we know from organizational design work that desktop stability both enhances user productivity in terms of using known software and encourages experimentation, learning, and process adaptation to new software. We know, furthermore, that increased stability is associated with Unix and decreased stability is associated with Windows. Clearly, therefore, there's an organizational productivity benefit to Unix here, but we can't include this in a simple TCO calculation because there's no sensible basis for quantifying the value of the effect.

As a practical matter what this means is that the basic "which is cheaper" question with respect to Windows and Unix can only be partially answered and is therefore not decisive as a guide to making the choice between them. It was in an attempt to improve on this without giving up on the TCO idea that we decided to ask a broader question: transmogrifying the original TCO question to ask: "which is smarter?" rather than "which is cheaper?" That works because the wider perspective the new question gives us enables a more balanced, strategic, view and lets us deal head on with the non-quantitative aspects of the decision.

Defining the Basis for Comparison

Before we can try to decide whether decision A is smarter than decision B we need to define what we're comparing. In this case we're comparing two complete systems architectures consisting of hardware, software, and management methods; one built around Unix and the other built around the Windows product line.

Architecture Server OS Type Desktop Networking Style Systems Management Strategies
Unix with Smart Displays

A typical smart display has a powerful graphics engine and runs Java/OS, Windows, or X clients, often concurrently. There are no moving, or user accessible, parts and typical MTBF ratings are in the 300,000-hour range.

Examples are made by companies including Sun, NCD, thinknic, and IBM.
The 1998 IBM Redbook: IBM Network Station - RS/6000 notebook" by Laurent Kahn and Akihiko Tanishita. provides an excellent, if now somewhat dated, introduction to the technology covering set-up, operations, benefits, and typical business deployments.
Unix Variants:
  1. AIX
  2. BSD
  3. HP-UX
  4. Linux
  5. Solaris
  6. Tru64
  7. Unixware
Primary Desktop:
  • Smart Displays
    • 21 inch NCD900
    • 17 inch SunRay
    • 15 inch Thinknics
Secondary Desktop
  • No restrictions; can mix and match to best suit user needs; PC, MAC, Linux/BSD, etc
Home Desktop
  • No restrictions: PC, Linux/BSD or MAC
All TCP/IP; VPN based access
  1. Leadership oriented
  2. Processes in place to guarantee uptime and performance
  3. Staff have specific personal responsibilities
  4. Staff free to contribute anywhere once responsibilities met
  5. Control pushed out to users
  6. Application support is a user function
  7. Users interact with systems people to make real decisions
  8. People who do technical work make the technical decisions
Windows Client-Server 9 Currently Supported Windows Server Variants
  1. 3 Windows NT 4.0 products
  2. 4 Windows 2000 products
  3. 2 Windows XP products

Effective use requires standardization on one server product.

12 Currently Supported Windows Desktop Variants
  1. Windows 95/98/ME
  2. 3 Windows NT 4.0 products
  3. 4 Windows 2000 products
  4. 2 Windows XP products

Effective use requires standardization on one client product.

Home use systems must be compatible with Office use systems for file exchange and remote access to work most effectively.

Multiple technologies, overlapped services, complex routing and security
  1. Requires top down control and monitoring
  2. Staff are interchangeable
  3. No one person is ultimately responsible for anything
  4. Staff do not fraternize with users
  5. Control is highly centralized
  6. Application support devolves to PC help desk
  7. All decisions made by senior committees

The two architectures have a range of significant differences in how they are installed, maintained, operated, and supported. From a senior management perspective the most important issues all relate to the role of systems in the organization:

Is Windows evolving toward the Unix/Smart Display Model?
A particularly good example of the use of the TCO idea to compare deployment strategies independently of the acquisition decision is provided by a recent article: Managing Your IT Total Cost of Ownership by Julie Smith David, David Schuff, and Robert St. Louis in Communications of the ACM. (January 2002, pp101ff) In this study TCO is defined almost exclusively in terms of infrastructure spending but applied, not to a choice between two architectures, but to a comparison among a large number of implementations of the Windows client-server architecture and so to the discovery of methods to reduce TCO within the PC architecture.

The TCO reducing method the researchers discovered, centralization of everything, is consistent with my views and this no doubt explains why I liked the article; their results, after all, really just demonstrate that the best way to reduce the cost of owning Windows PCs is to upgrade the servers and spend lots of effort turning the desktop PCs into not so smart displays.

  1. the complete reliability of the smart display means that the systems department does not operate a help desk facility. There is no desktop OS to go wrong and no confusion between OS and application related issues is possible.
    As a result the demarcation between the systems role and the user role is very clear and no control conflicts occur. That yields productivity benefits as application training and assistance are handled by domain experts within the user community and Unix technical staff are free to work directly with users in originating, evaluating, and making service changes;
  2. both architectures require shared central servers to function, but Unix unifies an organization while Windows fragments it.
    Although, from a purely practical perspective, Unix simply makes collaboration easier and cheaper, the real "drivers" here are perceptional. A Unix/Smart display system is obviously a corporate resource intended for business use and is therefore consistent with the underlying reason the organization exists - to align the activities of a number of people toward the achievement of common goals. Windows, on the other hand, is presented as a personal power tool and is therefore antithetical to the underlying reason the organization exists.
    Add in the basic unreliability and complexity of the Windows environment and the resulting conflicts are exacerbated by the need for the Windows CIO to, in effect, live a lie - giving users the illusion of control that comes with physical possession of the desktop computer but having to hire someone with a title like a chief firefighter, CTO, or deputy, to act as the vice-principal for user discipline and ensure that they do not try to exercise that control in any meaningful way.
    The Unix CIO, in contrast, relies on the inherent stability of the system to create a trust relationship with the user community in which both groups work toward common goals instead of against each other.
    For more information on these issues and how systems managers can help their companies transition from the Windows environment to Unix, please see the author's The Unix Guide to Defenestration
  3. in the typical Windows environment most of the systems staff spend their time battling the disaster du jour; whether that's a new virus, a software upgrade, or a network issue doesn't matter. It is the effect that matters because it means that the CIO and other management staff live from one crisis to the next, always in firefighting mode; never able to concentrate on long term revenue or strategic issues. The Unix CIO, in contrast, can count on things working and focus on revenue generation and longer term strategic issues. Thus the Unix CIO can turn the Systems group into an organizational asset while his, or her, Windows counterpart is condemned to the role of cost sink, always spending monies earned by others.

The Series Summarized

The primary purpose of the seven case studies in this series was to wrap these considerations in real business contexts to see how they express themselves in reality. For that reason the cases are as realistic as possible and the solutions proposed generally go beyond simple TCO to the more fundamental question of which architecture represents the smarter choice.

In each case some attempt is made at quantifying the TCO impact of each decision but this is only possible for costs that can be easily estimated and attributed. In most cases there is no sensible basis for either quantification or attribution of major management and related organizational costs. It may be obvious that one solution frees the CEO to concentrate on running the business while the other choices redirect senior management's attention to ephemeral issues like getting the e-commerce server up again or apologizing to customers for making their credit card numbers available on the web; but it isn't obvious what this difference might be worth in dollars attributable to the systems decision.

As a result the TCO part of the discussions generally show Unix to be cheaper, but the business elements discussed do something more important: they illustrate that Unix is usually the smarter choice too.

The table below summarizes the individual articles in the series. A different kind of summary, one listing only conclusions and comparisons between Windows and Unix is available here.

Article Comparisons Conclusions
Strategic Comparison: Unix with Smart Displays versus Windows Client Server

Two groups:

  • a 500 user school
  • and a 5000 user manufacturing company
face new technology choices and find Unix about a third cheaper than Windows on cash cost alone.

This story defines the terms and sets the stage by doing a straight up "sum of checks written" TCO analysis and then moving into consideration of unquantifiable productivity and organizational design consequences of the systems decision.

  • Five year capital and operating costs for both cases.
  • Productivity under differing assumptions about systems reliability, organizational focus, systems change, and application support.
  • Organizational design and control implications arising from systems decisions.
  • Effect on the home user of a business decision between Unix and Windows.
  • Effects on business adaptability to external change.

For large organizations Unix savings can amount to 60% of the Windows cost over five years.

Cheaper is good, but consideration of other issues shows something more important than just "cheaper": for these two groups the Unix decision is Smarter.

For the school that happens because Unix lets the school focus on Education, and for the business it happens because Unix lets management transform the Systems Department from cost sink to revenue generating asset.

Making the case for Unix at Whackabilly Toy Inc.

A small manufacturing company on the verge of being bankrupted by an ill considered move to Windows based E-commerce finds a solution in switching to Unix.

This story is all about failure and the excesses that led to many of them.

  • Cost of Ecommerce under both approaches.
  • Effect of the options on executive time and focus.
  • Effect on the company's ability to meet customer needs.
  • Effect of removing systems failure from the picture.
  • Effect of increased performance and screen sizes.
  • Effect of providing access to more, and better, software.
  • Effect of providing faster, more secure, home access.

An average 40% cost savings from Unix is shown to be insignificant relative to gains in user productivity and management's ability to focus on running the business, instead of systems.

Again, it is the gain in business focus, not the cash savings, that make Unix the smarter choice.

Impress Recruiting - Freeing developers from Windows Costs

A small software developer finds renewed business opportunity by adopting open source ideas and dropping the Windows cost burden.

Many software developers try to sell into the Windows market because it seems so large. This story is about the enormous costs this idea imposes on the developer and its customers.

  • The proportion of the total sale which represents income to Impress.
  • The barriers the technology places in the way of making the sale.
  • The customer's perception of Impress Recruiting
  • Development and infrastructure costs.
  • Value of customer support skills
  • Development flexibility and competitive opportunities

When Impress sold its software only on Windows, about two thirds of each bid reflected other people's products and most bids required customers wishing to adopt the Impress package to do other things they did not want to do - like licensing the latest release of Microsoft Word and paying for the operating system and hardware upgrades to run it. By dropping this Windows cost burden, Impress reduced its average bid price by one third without affecting its revenue per sale and became able to concentrate on selling its own product.

What really made Unix Smarter for Impress wasn't the cost savings, it was their transformation in customer eyes from their previous role as Microsoft enforcers to their new and more appropriate role as people the customer can count on to make things work.

SafetyJet International - Things only Unix can do

Development considerations for an application designed to fully meet the automation needs for a major airline demonstrate three unique Unix strengths:

  • scalability
  • reliability
  • secureability

In this case, there are no other workable choices; the problem is simply out of range for other architectures.

Primary comparisons are stated in terms of the new airline's posture relative to its existing competitors and include:
  • vastly improved customer service opportunities
  • the smoothing and predictability effects logical customer pricing has on earnings
  • vastly improved on-time performance
  • 80% IT staffing level reductions
  • 3-5% operating savings per seat mile flown
  • Only Unix can scale to solving very large mixed integer program problem in near real time.

    If we tried to print the full problem matrix using 6 columns and four rows per inch we would need a sheet of paper slightly larger than West Virginia.

  • Only Unix can meet the need for 100% uptime
  • Only Unix can meet the need for absolute certainty in regard to data integrity on security applications.
The Happy Valley Tax Authority- Where Unix is a four letter word

Social considerations drive out business and cost considerations when Windows bigotry marks the Unix expert as the common enemy.

The comparisons here are all implicit and include:
  • the difference in risk exposure to the audit firm generated by a competent Unix reviewer is compared to that generated by a technically naive auditor who signs off on absurd, but buzz word intensive, client activities.
  • the "abusability" of a Unix system relative to a Windows one.
  • Unix stability versus getting trapped into a pointless upgrade cycle
  • Unix security versus PC wish-walling

The Windows people in this story do and believe things that are so obviously absurd that they become hard to comprehend - but reality trumps rationality every time.

The issue here is the social cost of knowing what others refuse to accept -being right is no substitute for being listened too when your goal is to help the client. The consultant in this story keeps demonstrating expertise - and every time he does, the effort further distances him the people he's supposed to be helping.

The most serious mistakes here are his -and that is what makes Unix smart in this story.

Ann Andersen Printing - Betting on Linux in the face of industry-wide change

A small printing company faces sweeping change in its markets and is urged to bet on Linux as the key to business survival through to about 2005.

The story within this story shows how the use of a small Linux machine and the web to distribute reports can replace a large printing and distribution job to save millions of dollars.

This case is the dual of the Impress Recruiting case with the customer view replacing the developer view of the open source movement in general and Linux in particular.

The primary comparison here is between earning a margin as a return on investment -the combination of people with physical infrastructure- and earning that margin as a return to organization - the combination of people and information.

The management comparison lies between being driven by the technology and using technology to drive the business. That's the point of the story within, it's the impact on business operations, not the direct savings, that counts.

Here we peer ahead five years from mid 2001 -or at least three new Microsoft product waves- and see that Linux offers stability, control, and opportunities to make better use of software dollars. That's what makes Unix the smarter choice for Andersen.

From this client's perspective Linux doesn't just offer "cheap", it offers the only chance now visible for gaining real competitive advantage through software as the twin impacts of large scale consolidation and better PC software in client hands, decimates its industry.

Martin Cutter Mills - Unix and the politics of generational conflict in a mid range business

Executives at a large manufacturing business have lost control of its information systems, allowing themselves to be manipulated by Systems in exchange for not being brought to task. Now push has come to shove and the options are taking action or letting the company slide into bankruptcy.

The action required is re-assertion of control and exercise of their managerial responsibilities. That action is independent of technologies and doesn't require them to cross the social boundary into nerddom.

  • Multiple technical solutions (Unix, As/400, Windows)
  • Supply chain and ERP solutions
  • IT Leadership versus IT management
  • Corporate versus local views of systems problems
  • Appeasement versus taking responsibility
The comparison here is behavioral. Left to themselves, systems organizations tend to be parasitical. In this case they've developed a plan that meets the career goals and needs of each little fiefdom within the systems organization - but completely ignores the reality of the company's needs.

The smarter technical solution is Unix, but this case is Whackabilly writ large, and demonstrates that technology and the cost of technology are minor issues relative to management and the cost of control.


The extended vision of what comprises total cost of ownership, or TCO, embedded in this series asks whether Unix is usually the smarter choice when the alternative is Windows client-server. On the numbers, Unix is always cheaper but, more importantly, it consistently wins out as the better choice when the decision is embedded in wider organizational or personal contexts.

That makes Unix the smarter choice; not only is it unambiguously cheaper, it is usually also better from an organizational design or senior management viewpoint.

But this raises an obvious question: if Unix is so much smarter, why isn't it everywhere?

There are two rather different answers to that. First of all, Unix is everywhere. Make a phone call, you're using Unix; connect to the internet, and you're using Unix; pay bills on-line, and you're probably using Unix; go out for exercise to clear your head, and the chances are good that your shoes, your bicycle, your canoe, or your ski gear were designed on Unix workstations.

Is there a bulge on the left side of the normal curve?
In the spring of 1984, people could buy Unix, Mac/OS, or PC-DOS based machines.
  • About 50,000 (2%) people picked the multi-user Tandy 6000HD. At $4,499 this offered 512K of RAM, a 15MB hard disk, a 1.2MB eight inch floppy, an integrated 24 x 80 screen, and two CPUs - a Z80C for backwards compatibility with TRS-DOS and a 7.54MHz MC68000 [with 32 bit data registers and 16/24 bit addressing] that ran Xenix off the hard disk.
  • About 30,000 people (1%) picked the MacXL. At a list price of $5,495 this offered 1MB of RAM on a MC68000 at 7.54Mhz (5 on early models) with a 10MB disk, a 720K 3.5 inch floppy, the MacOS GUI and a full suite of graphical applications on a high contrast, black on white, 640 by 480 screen.
  • just over two million people (97%) picked the IBM PC/AT. For only five dollars more than the Mac, this offered a maximum of 256KB of ram on a 16 x 16 bit chip running at 5.77Mhz with a 10MB disk, a 360K 5.25 inch floppy, PC-DOS, a BASIC interpreter, and a 24 x 80 green screen.
Note: The PC sales number is a guess. I have not found a credible source this. Please email me if you can help.

But if we restrict ourselves to just desktop and small server computing this isn't true. It's Windows everywhere, not Unix. So how did that come about?

The short answer is that nobody really knows. Longer answers start with the mindset of the typical IBM data center manager in the early eighties and the combination of his control over corporate systems expenditures with the absolute resistance to change mandated by the gear, people, and software he worked with. Add in the incredible amounts of easy money to be made by helping users fight his budgetary and processing control, and you have a recipe for conflict explaining why the national accounts have never shown a GDP benefit offsetting expenditures on computing.

Although we don't know how things got this way, we can ask a somewhat different question: why do they stay this way? Users now have real power in corporate systems decision making, why aren't they demanding Unix?

I believe the answer defines the challenge facing the Unix community -AIX, BSD, HP-UX, Linux, Solaris, Tru64, Unixware- all of us. Most users are not demanding Unix because most of what they know about it is negative -and dead wrong.

Consider that:

  1. very few user executives have ever seen a smart display. Most associate Unix, not with powerful graphics software under frontends like KDE, GNOME, or even CDE, but with the flickering and unstable yellow on black command line seen in a Windows telnet session;
  2. most user executives have seen Unix used in a data center - and, because Unix is generally mismanaged in those data centers, associate the high cost and complexity they encountered with the technology rather than the incompetents running it; and,
  3. people don't update information they've internalized unless there's an external reason to do it. Very few business people understand the cost picture for Unix today, most unconsciously compare heavily advertised home PC pricing to Unix costs remembered from years earlier. Tell the average business executive that you can buy a (rather small :-)) Sun server with Solaris for less than the cost of renting a Windows 2000 Advanced Server license for a year and you'll find your sanity being questioned. Go beyond that to point out that any hardware capable of running Windows will also run Unix - something you'd think would prove that there is no PC hardware advantage to Windows - and you'll generally meet complete incomprehension.

So how do we fix this? Teach, communicate, show people what works - build the Unix community one user at a time.