There has been some controversy recently over a Microsoft-commissioned TCO (total cost of ownership) study claiming that the long-term cost of Windows is lower than that of Linux. It's time to clear up the confusion.

TCO Primer

Let's start with the basics of TCO. When we talk about measuring the total cost of ownership, it is important to understand what that measurement will actually tell us. TCO is defined as the total cost of a particular item over its useful life. It includes the cost of acquisition, maintenance, support, and disposal. In short, it includes everything you will ever spend on the item, and is useful for understanding future costs that may not be apparent at the time of deployment.

In theory, calculating TCO is easy. Simply add up all the expected costs over the life of the item you are measuring. In practice, however, this can be a little tricky, which is why most organizations don't actually track TCO. A number of factors are typically included in TCO calculations, including hardware costs, software licensing, initial costs to deploy, purchased support contracts, staffing costs, and additional overhead. The following is an overview of the TCO?process:

• Plan to calculate for each class of system (file server, Internet-related server, desktop, etc.)
• Determine hardware costs
• Determine software licensing costs
• Determine deployment costs (outsourced deployment, for example)
• Determine support costs, extend to expected life of system
• Determine staffing costs to maintain system
  -Determine number of staff supporting system
  -Determine hourly costs based on salaries
  -Determine percentage of time spent supporting system
  -Determine total number of staffing hours over expected life of system
  -Calculate dollar value of total staffing hours
• Determine any additional costs for overhead
• Add it all up

Some items on the list are fairly straightforward, such as hardware, software, and support contract costs. Staffing costs can be trickier, especially if your company doesn't have experience with the new technologies. If it were easy, of course, everyone would routinely calculate TCO figures. But it's important to at least have an idea of what's involved.

The Analysts' TCO Comparisons

There have been just a few studies of the TCO for Linux, all looking at costs for servers. An independent study, "Total Cost of Ownership for Linux in the Enterprise," conducted in 2002 by the Robert Francis Group, compared the total costs of enterprise Web servers on Linux, Windows, and Solaris. Data was drawn from actual experiences of interview participants. They found that the three-year total cost of Linux was two and half times cheaper than Windows, and seven and a half times cheaper than Solaris. One reason Linux was cheaper was its reliability. Respondents reported that security patches and accompanying server reboots took longer to administer on Windows, requiring more staff. Another reason was a reduced effort in responding to viruses and Internet worms. According to the survey, the ratio of administrator to server was 44 systems per Linux administrator, and 10 systems per Windows administrator. Average salaries for Linux administrators were about 4% higher than for Windows administrators, and about 20% lower than for Solaris administrators. Taking these factors into account, the staffing costs of Linux were less than one fourth the staffing costs for Windows. Again, this data came from asking interview participants about their already-deployed systems, and so involved real-world experiences rather than speculation.

IDC released a similar study in 2003 comparing Linux and Unix. Sponsored by Red Hat, "Linux and Intel-Based Servers: A Powerful Combination to Reduce the Cost of Enterprise Computing" was based on an earlier report that surveyed 142 companies about costs associated with Linux and RISC/Unix servers. No comparison was made with Windows in this study. IDC found that the TCO of Linux was a little more than half that of Unix for Internet/intranet workloads, and about one fifth that of Unix for collaborative workloads. IDC defined "collaborative workloads" as e-mail, group calendaring/scheduling, shared folders/databases, threaded discussions, and custom application development. The calculations did not include support contracts for hardware or software, but these costs are likely to be higher for Unix.

Microsoft commissioned a report, also from IDC, on the TCO of Windows versus Linux. This report, "Windows 2000 Versus Linux in Enterprise Computing," is currently being promoted in their "Get the Facts" marketing campaign against Linux. This study examined costs in five workloads: network infrastructure, file serving, print serving, Web serving, and security applications. They found the TCO of Windows to be comparable, or superior (meaning cheaper), to every workload except Web serving. The costs of software were calculated as being higher for Linux than for Windows. Staffing costs were also calculated as being higher for Linux.

The Yankee Group claimed in a recent press release that they have completed a study showing that Linux is "not a low-cost alternative to Unix and Windows for large enterprises." Their analysis is that small firms may benefit from a limited migration to Linux, but that Linux did not bring a measurably improved TCO to large organizations. The Yankee Group did not respond to requests to view the report by the deadline for this article, so the analysis here is based on information provided in the press release.

How Can These Reports Be So Different?

How did these studies arrive at vastly different conclusions? The short answer is that if you are abstract enough with your goals and methodology, are selective with the costs that you include, and ask the right questions, then you can arrive at any conclusion you want. I'll focus on the Linux versus Windows studies, since those are the most controversial.

Abstract Assumptions Don't Translate Well to the Real World

The Yankee Group report claimed that some interviewees reported that "a significant Linux deployment or total switch from Windows to Linux would be three to four times more expensive" than the usual Windows upgrade. Well, duh! There should be little question at this point that the total cost of owning a Linux system can be lower than that of a Windows system, assuming that you do nothing with the machine besides run an operating system. But in the real world, software is required to do work, and there is a great variety of software in use. If your organization requires a collection of disparate Windows-specific applications, for example, then you will have to make a number of changes in order to use Linux instead. You may need to rewrite custom applications, switch vendors of critical systems, retrain users on new systems that are quite different from the old ones, and suffer lower productivity because of awkward choices of replacement open source software. If you were to force these changes, then it is certainly possible that the total cost of those systems over their life cycles would be higher than the status quo given the overall costs to make it work. It is certainly possible to imagine scenarios in which this type of forced conversion would lead to a higher TCO for Linux systems.

The pragmatic approach most organizations take in adopting Linux and open source software is to find those areas where they can use it to gain the most benefit with the least cost and disruption. If you were to blindly convert your entire operation to Linux and open source software, you would save money in some places, but potentially at a great cost and disruption in other places. But let's be realistic - no one indiscriminately replaces one technology with another one wholesale. It was an unrealistic question that produced a meaningless and misleading result.

Real-World Data Is More Valuable than Opinion

It also makes a big difference whether the data was collected from real-world experiences of interviewees or from opinion. In the Robert Frances Group study, interviewees supplied all the data from actual field experiences. In the Microsoft-sponsored IDC study, the methodology is less clear. After several readings of the document, it appears to this author that 100 different North American companies were interviewed to help establish the assumptions that would drive the TCO calculations. In other words, the actual data appears to be at least partially generated, not gathered from field experience.

It should also be noted that this study claimed (in the front page "IDC Opinion" abstract) that "the cost advantages are driven primarily by Windows' significantly lower costs for IT staffing." It is unclear from the methodology description how the staffing costs were calculated. In any case, the results are in dramatic opposition to the data gathered by the Robert Frances Group from practitioners with active deployments.

Omissions Can Make the Data Unrealistic in Practice

In the IDC study that Microsoft is promoting in its "Get the Facts" campaign, suspicious omissions make the data unrealistic for most organizations. When the five-year costs are calculated for each workload, a load of 100 end users is used. The software costs are not broken down, but it is clear that Client Access Licenses (CALs) cannot possibly be included. Perhaps IDC considered them to be a part of the client operating system costs, even though they are required for Windows but not for Linux. This is a very important point because at about $800 per 25 users, the CALs really add up. This means that if you have 400 users instead of the calculated 100, you are adding about $9,600 to the software costs, clearly a significant figure. The apparent omission of the required Client Access Licenses is misleading.

What to Look for in a Study

How do you know who to believe? If you're going to follow the advice in an analyst report, then read the document and ask yourself these questions:

• Were the interviews based on experience or guessing?
• Was there actual data from real-world settings?
• Are cost estimates realistic and inclusive?
• Do the authors sufficiently describe their methodology so that I can understand how they arrived at the data?

If you're going to follow advice from analysts then read the reports and look for signs of solid methodology and independence in the research. Think about the advice you are being offered, and decide for yourself if it's helpful or not. As you can see, there is a wide discrepancy in the available analyses.

References

The Robert Frances Group, "Total Cost of Ownership for Linux in the Enterprise," 2002.

IDC, "Linux and Intel-Based Servers: A Powerful Combination to Reduce the Cost of Enterprise Computing," 2003.

IDC, "Windows 2000 Versus Linux in Enterprise Computing," 2002.

The Yankee Group, "Linux Not A Low-Cost Alternative to Unix and Windows for Large Enterprises, says the Yankee Group," 4/5/04 press release.

Adapted from Manager's Guide to Open Source Manning Publications, summer 2004.