% fortune -ae paul murphy

When Windows beats Linux: a cautionary tale

Back in March of 2005, when Lufthansa finalised its control of the Swiss International Air Lines operation cobbled together out of the 2001 SwissAir bankruptcy by the Swiss government and several of its larger banking partners, its first order of business was house cleaning.

As part of that, management promptly accepted an EDS sponsored project proposal aimed at replacing a Linux based e-commerce system with an all Microsoft production - and thus authorized a project that has since become a minor star on Microsoft's anti-Linux page.

Here's how Microsoft summarizes the project and its results::

Swiss International Air Lines (SWISS) needed to gain every revenue increase and cost reduction possible from its most profitable sales channel - its e-commerce Web site. Thanks to a migration from Linux technologies to Microsoft Windows Server 2003 with Internet Information Services 6.0, the Microsoft .NET Framework, and the Microsoft server product portfolio, SWISS is meeting that goal. Agility, as measured by the frequency of software updates to meet business requirements, is up 400 percent. Responsiveness on the site is up 300 percent and unplanned downtime is cut 90 percent, giving customers a more natural and enjoyable experience. These changes helped to support a 30 percent increase in online revenue last year - and did so while putting SWISS on the flight path to save U.S.$600,000 per year in reduced operation costs.

That's impressive - and particularly so because a quick google suggests that reliability and performance really did improve afterwards.

On the other hand, it's pretty counter-intuitive - at least to me because I put the problem of proving Windows better and cheaper than Linux in the same category as proving that fifty men using wheelbarrows can more quickly, reliably, and cheaply move 25 cubic meters of ready-mix per day up a mile long mountain road than can one person operating a a Cat 950F wheel loader.

Luckily there's a case study which makes it clear how this miracle was achieved: basically Swiss Air had a four year old running the loader, and starved it for fuel too.

More specifically, what appears to have happened illustrates a general problem and should serve as a clear warning to anyone considering the use of Linux primarily because of its money saving potential.

The general problem is that most people believe that computers are just computers -that skills learned in one environment translate to all others - and that's just not true. Linux is not Windows, and replacing only the OS layer in a Windows architecture with Linux isn't a recipe for saving money, it's a recipe for disaster.

Consider this from the case study:

In 2005, SWISS kicked off this process with a more physical migration of its e-commerce solution: The solution was outsourced, moved from within the SWISS datacenter in Basel, Switzerland, to the facilities of technology services provider Swisscom IT Services, based in Berne. The airline reservations system, with which the booking solution communicates, is hosted at EDS.

"Reliability was one of the reasons we decided to outsource our infrastructure," says Meyer. "We concluded that we didn't have the critical size needed to justify the expense of a high availability infrastructure. We turned to outsourcers who could provide that reliability more cost-effectively. And we manage the reliability through our service delivery organization."

Once the original infrastructure was physically relocated to Swisscom IT Services, the technology provider worked with the airline to implement a series of changes. The Web front-end to the e-commerce solution was rewritten using Microsoft C# technology, introducing object-oriented programming to what had formerly been a script-based solution and enabling the solution to be updated and expanded more easily in response to business requirements. The application was rehosted on a consolidated environment with 4 Windows Server IIS 6.0-based computers replacing the 10 computers that had hosted the former Linux version, for a net consolidation of 60 percent.

The new solution architecture takes advantage of Windows®-and .NET-based technologies to replace the mixed presentation and application layers of the Linux-based version with a true n-tier architecture with fully separate presentation and application logic.

The content management solution migration to SharePoint Server is being accomplished along with a server consolidation similar to the one achieved in the Web-serving portion of the solution. The 10 computers in the content management solution are being consolidated to 4, with SharePoint Server eliminating the need for a pair of search servers, a pair of staging servers, and a pair of content servers. With a similar reduction in the development and test environment, the 60 percent consolidation of the content management solution is eliminating 12 servers. In all, the combined savings is 18 servers.

What they don't go out of their way to tell you here is that the hardware being replaced went in before 2001 bankruptcy and had been on life support ever since - meaning that the real reason they were able to improve performance while reducing the server count by nearly two thirds has nothing to with Linux, and everything to do with the fact that the operation of Moore's law let them go from 30 servers offering a cumulative 27 PIII Ghz to 12 offering a cumulative 76.8 Xeon Ghz.

Similarly the software put in originally had minimum updates since - the mention of Red Hat 4.0 notwithstanding, the case study makes it clear that the software being replaced - Obtree, Mercado, and Oracle - had come in with the original system and barely been updated since.

It's when you look more closely at the claimed savings, however, that you get some clues to what really happened. Consider this bit:

SWISS has also achieved significant cost savings from its migration from Linux to Windows Server. The reduction in Web servers saves the company $150,000 per year, with the outsourcing of the solution contributing another $50,000 in savings, according to Heintel. Replicating the consolidation in the content management portion of the solution should triple the savings to $600,000 per year, he estimates. And the savings will grow. With the migration to SharePoint Server, SWISS eliminates a second platform for content management so it and its outsourcer need only manage a single environment. SWISS expects the SharePoint Server-based system to deliver content twice as quickly as its Linux predecessor. The reduction if not elimination of Linux in its environment similarly reduces the burden of managing multiple operating systems and frees SWISS from having to locate what Meyer sees as relatively scarce and expensive Linux expertise.

Developing the solution for the .NET Framework also resulted in faster and more cost-effective development than SWISS would have seen had it used Java, Heintel estimates. The Web server portion of the solution took three months and $120,000 to develop; had SWISS used Java, Heintel estimates, the solution would have taken 50 percent more time and money.

Read that carefully and you'll see they're claiming to have saved $25,000 per server, per year for each PIII machine eliminated - and then claiming to save another $12,500 per server per year from outsourcing operation of the 12 new machines.

Overall they're reducing their server count by 18, replacing free software with licensed software, out-sourcing development and hosting - and claiming that these actions produced direct savings on the order of $33,300 per server per year for each of those 18 machines - i.e. that each time 2.5 P3s running mostly freeware were replaced by a 3.2Ghz dual Xeon running Windows along with licensed and custom developed applications, the company saved $50,000 (=$6E5/12) per year - in perpetuem.

As far as I know the only way to make sense of this is to assume that it took at least eight more people to run the 30 Linux servers than it does to run the 12 Windows ones -after all the money didn't go for reinvestment, it didn't go for hardware, third party support, or software, so what else is left?

Obviously, however, this makes no sense: at least eight people for thirty small Linux servers? - but look at two things they say about the content management system:

"The content management system, with approximately 1,000 pages of content,"

The 10 computers in the content management solution are being consolidated to 4, with SharePoint Server eliminating the need for a pair of search servers, a pair of staging servers, and a pair of content servers. With a similar reduction in the development and test environment,

That's ten computers for a thousand pages of content -and notice that they needed separate search servers, staging servers, and content servers to do it. That's not a Linux on P3 set-up, it's a structure for NT on 486 or P1 - and once that became clear a quick google for resumes from people who used to work for SwissAir suggested a simple picture of what happened.

That system didn't go in on Linux, it came on NT 3.51 and later got migrated first to 4.0 and then to Linux after the money ran out and IBM's political strategy of selling Linux as the anti-Bush OS for Europeans made it attractive to people then reporting to government over-seers.

That's how they ended up needing at least eight people to run two thirds of a rackmount, and therefore the miracle isn't that replacing an NT architecture with an XP one allowed them to show improvements over treading water for nearly five years; but that the applications more or less worked throughout that Linux interregnum - and that apparently despite the best efforts of all concerned.

And therein, I think, lies the lesson in the story: as the current recession deepens we're all going to be under tremendous pressure to cut costs - and because Linux is cheaper than Windows, many Windows users are going to be tempted to do what these guys did: adopt the cheaper technology on a one for one replacement basis.

As they found out, however, Linux isn't Windows and changing to Linux on a one to one OS replacement basis only because it's cheaper is about as dumb as it gets. Changing to Linux is generally a good idea, but you do it because it's better, not because it's cheaper -you use "cheaper" to sell the change to others, but always remember: it's only cheaper if you use it properly - otherwise? otherwise, you'll become like these guys: failures in full retreat, the stuff of ridicule - like four year olds trying wrestle thirty ton loaders up mountain sides - during rainy season.


Paul Murphy wrote and published The Unix Guide to Defenestration. Murphy is a 25-year veteran of the I.T. consulting industry, specializing in Unix and Unix-related management issues.