Can business users really justify Linux on economic grounds? Microsoft has an amazing public web page arguing that Linux simply doesn't stack up against Windows NT, now called Windows 2000. Although the page is now more than two year's old, this still appears to be Redmond's main intellectual ammunition to counter the Linux competition in the server space.

Microsoft's Linux Myths page is a piece of purple propaganda that might make even Stalin's commissars shudder; nevertheless it does pose a valid question. While we all know it costs less to buy a single copy of Linux than to purchase a Windows 2000 licence, is Linux cheaper over the long haul?

Amongst the arguments wheeled out to support Microsoft's case is something known as the Total Cost of Ownership or TCO. As the name suggests, TCO identifies the true, underlying cost of owning and operating a server. This includes both budgeted costs and  'hidden' costs. Hidden costs can be viewed as subsidies. Not surprisingly (after all they are hidden) these cost are often overlooked. For example, a hidden cost might be staff training, which doesn't always fall under IT budgets. And there are the costs of business lost during a particularly disruptive change over.

In the Linux Myths page Microsoft claims that Windows NT has a 37% lower TCO than Unix and that, by logical extension this also means Linux. I'll leave you to decide for yourself the overall value of the analysis which was carried out in May 1998 and compares Windows NT Server running on Compaq hardware against Solaris running on SPARC hardware. Here I just want to focus on one aspect.

Microsoft's comparison found that overall costs were US$1.25 million for the Microsoft/Compaq option and US$2.14 million for the Sun alternative. But, and it is a very big but indeed, the analysis puts the initial systems price at US$367K for the Microsoft system and US$791K for the Sun system. If we substitute Linux for Solaris and swap SPARC for Compaq hardware, then factor in the steep Windows NT price rises that have taken place since 1998 this initial system price comparison would now overwhelmingly favour a Linux set up.

Perhaps we could go further. I'm not fully qualified to say whether a Linux system with comparable performance would require more or less hardware expenditure. And, frankly, we'd need to set up some pretty complex benchmarking to test my theory. But, despite what the ?Microsoft Myths' edition of Pravda says, I suspect you'd need more hardware to run the NT system, than you'd need for a Linux server.

The Gartner Group originally developed the whole concept of Total Cost of Ownership analysis as a way of helping business managers make decisions about technology investments. It has been used to great effect. For example, in 1997 Gartner estimated the TCO of a Windows95 PC in a North American corporation to be $US9784. This triggered a lot of soul-searching about whether companies were getting value for money.

One upshot of Gartner's 1997 report was the Microsoft rushed out it's zero administration kit (ZAK) a set of tools designed to reduce the cost of operating networks of Windows-based PCs.

As far as I know, no local TCO data has even been published, but a consultant at IDC told me the corresponding amount for Australia would be 'at least' the same as in the US.

So, with the Australian dollar trading at around US$0.53, we can assume the annual cost of operating a single networked desktop Windows PC in Australian conditions to be around $18,000. That figure assumes an NT server; I'd love to have a handle on comparable figures for a Linux server. Methinks the number would be lower - but by how much? If you can shed light on this question why not add a comment below?

More recently, Gartner produced important TCO figures, which showed major corporations would be better off not upgrading from Windows NT 4.0 to Windows 2000. In other words, we can extrapolate this to deduce that operating Windows 2000 is more expensive than operating Windows NT 4.0. Which, by extension shifts the balance further away from Microsoft's spurious ?37% lower TCO? figure.

Indeed my guess is that Linux provides a better TCO right across the board. I don't have the resources for a full TCO analysis, so I accept my example calculations are far from accurate, but I want to think about those up front costs again.

Just about everybody reading this column will know that Linux costs next to nothing compared with other server operating systems. You'd be hard pressed to find any Linux distribution costing more than $150. Most cost less.

If you've got bandwidth and spare storage you can download it for next to nothing. In many cases people pay no more than a handful of dollars for a homemade CD Rom and the price of a decent bottle of wine for a manual or two. To keep this discussion simple, lets assume the cost to a business user of a single copy of Linux is a nice round $100.

Microsoft's Windows 2000 is a little more expensive. There's actually a complex pricing structure for Windows that depends on a number of factors including how many users you intend to server, the nature of your application, the type of hardware you operate and how desperately Microsoft wants you business. To keep our investigation simple, we'll steer clear of direct sales, arm-twisting and mates' rates and head for the widely publicised Australian Windows 2000 price list displayed online by Harris Technology.

Let's assume we have servers with 5, 25 and 100 seats. For the sake of argument we'll assume the cost of the server hardware is likely to set you back around $5000 and that more users simply means less performance. So the total cost for a five user Linux system is $5100 or $1020 per seat. A five-user version of Windows 2000 Server costs $2051.47. Ignoring small change that's $1410 per seat.

For 25 users, the Linux system will set you back $204 per seat. The 25 user Windows 2000 Server licence is $3523, which gives a per seat cost of $341. For a 100 seat system (which admittedly would operate at a snail's pace but might cope with some basic applications) the Linux cost per seat is $51. Although Microsoft will sell you a 100-user licence version, I've used the $3523 25 user licence plus four 20 user licence packs at $1535 each plus the server cost to arrive at a total cost per seat of $291.

Sure, it's back-of-an-envelope stuff and perhaps I've missed something significant here, but it seems that the up front cost per seat of a Linux server is such a small fraction of the cost of a Windows 2000 that the hidden costs would need to be really major to tilt this playing field in Microsoft's favour.