So, yet another survey has attempted to answer the question of whether running a system on Linux or Windows is cheaper.

On the face of it buying Linux software is cheaper because it is open-source and Microsoft has a famously pricey licensing system for its products. However, it is the Total Cost of Ownership (TCO) that is the important factor. How much does it cost all in - maintenance, development, training - to run both systems?

We've had numerous surveys, reports and studies in the past two years that have pointed first way one and then the other. In fact, the only thing the reports have had in common is their shameless bias and careful twisting of scenarios to come out in favour of their preferred operating system.

Among such worthy competition, however, last week's study by Giga Research nevertheless manages to amaze with its sheer bravado. Even considering it was sponsored by Microsoft.

The authors point this out early on and begin their executive summary: "Will Linux inevitably dethrone Windows as the leading operating system? Is Linux the better choice?... Giga Research [examined] the relative benefits... by comparing the cost incurred and benefits achieved by two sets of organisations: those using Linux as the basis for their applications and those using Microsoft Windows." Sadly this is as balanced as it gets.

Its conclusion incidentally - just to get it out the way - was that it is cheaper to develop certain programs in Windows rather than Linux. By up to 28 percent.

Discovering how it got there reveals a litany of bizarre unexplained assumptions and selective blindness. In short, this is an entire study geared to producing a single headline. And it has worked. "Windows cheaper than Linux," says CNet; "Microsoft Has Better TCO Than J2EE/Linux," says InternetNews; "Linux Cheaper Than Windows? Think Again," says E-commerce Times; "Windows can be cheaper to use than Linux," says Reuters - and so on and so forth.

Maybe Windows is cheaper than Linux in certain situations but when the evidence is so flawed, you have to question the entire premise.

The study's "primary conclusion" is that there is a "substantial cost saving" of Windows over JTEE/Linux "as a development platform for the applications considered". What applications in particular? We never find out, only what "business goals" the applications would be used for.

We find out that the survey spoke to seven companies running Windows and five running Linux. The report then "extrapolates from these users' experiences" to come up with a brand-new form of TCO that Giga has called Total Economic Impact (TEI) - a method so rich in its extrapolating potential that the company has even trademarked the name.

Why is TEI so great? Because it covers not only all those things that are "typically accounted for by IT" but also "benefits, risks and flexibility". Yes, TEI "weighs the enabling value of a technology in increasing the effectiveness of overall business processes".

Once you've weighed the enabling value of effectiveness in overall procedures, you come up with some interesting figures. For example, it takes 12 months to complete a project on Linux but only nine months with Microsoft. Why? Who knows? Also you need 10 days of training for people on Linux and just seven for Windows. Why? Dummy - everyone knows that Microsoft-trained staff are more readily available and more productive.

Not only this but for a large organisation Linux may cost half in terms of "computer platform", but it is four times more expensive than Windows in terms of "software platform", four times more expensive for "software maintenance" and a third more expensive in terms of skills training.

For a medium-sized organisation, computing platform is somehow double, the software platform an incredible seven times more expensive, software maintenance two-and-a-half times larger and skills double.

How? Well, first of all, it is assumed that anyone using Linux will automatically go for BEA WebLogic and Oracle's database and cough up the surprisingly high licences for each. Why this would be the case when the companies could use highly-rated open source databases and scripting languages like MySQL and PHP for almost no cost is somewhat of a mystery.

Software maintenance is also calculated in a very strange fashion. It is assumed that each year it will cost 25 percent of the list price for the licences. This assumption more than anything shows how skewed by Microsoft this report has been.

Under Microsoft's new licensing rules there is a compulsory annual maintenance fee for ownership of its software which is calculated at 25 percent of the licence. Microsoft says it helps keep customers up to date. However the report assumes that exactly the same is true for all other companies, when it isn't. And, of course, the licences that the report decided everyone was buying were vastly more expensive than Microsoft's in the first place, so the gap just gets bigger.

The report gives two figures for the cost, over four years, of a system on both Linux and Windows: $2,289,000 compared to $1,643,000. Assume for a second though that a Linux user decides to use other open-source software in conjunction with Linux and you find that figure rapidly drops to below that of the Windows figure. Remove the assumption that it takes a third longer in development or training on Linux and that figure falls even further.

If you then take into account the stated assumption that the new Windows Server 2003 is just as stable as Red Hat 9 (something that seems as likely as an iced tea in hell) and so ignore all the extra associated maintenance costs, then Windows really does start losing the case in this carefully constructed scenario.

In a study done last year by the Robert Frances Group, which was just as fervently pro-Linux, though far more ethical with its figures, great play was made of the fact that in the Microsoft set up, when something went wrong, external maintenance help was often needed. This meant, it estimated, that one sysadmin could look after four times as many Linux servers than Windows servers. And yet, Giga has software maintenance running four times higher for Linux.

And while we're on the subject of old surveys. In December last year, IDC ran an equally suspicious survey that argued almost exactly the same point as Giga - and, would you believe, that of Microsoft executives - that Windows (2000 this time) was cheaper than Linux "in the long run" and "for some tasks". That survey considered five years of ownership, Giga's was four, maybe by the middle of next year, we'll have a survey that proves Windows is cheaper in certain cases over three years.

Both studies also argued that software and hardware only account for a small percentage of the TCO. Interestingly, at the time, one analyst had a major problem with this assumption - it was Stacey Quandt from Giga.

The Robert Frances Group survey also had a variety of points that it said made Windows's TCO higher than you would expect. For one, the cost of carrying out an audit on licences (something Microsoft has got stroppy about in recent years). Then, there's the cost of downtime on servers that had been compromised by security holes. And finally, the fact that Linux can be run on a host of hardware architectures whereas Windows cannot.

There is also the issue of migration. If you are running a Unix system, even the Giga report agrees that moving to Windows would be massively expensive "in new systems infrastructure, new skills and perhaps replacement applications". A shift to Linux would be far cheaper. But then Microsoft's aim in all this is not to stop a move from Unix to Linux but prevent people from deciding to jump from Windows to Linux.

If it wants to achieve that, it will find that reducing its licence fees would be far more effective than commissioning dubious research.