% fortune -ae paul murphy

Costs: Wintel Vs. OpenSource

One of the true wonders of our age is the ability lots of otherwise reasonable people have to tell you with every evidence of believing it themselves that PCs cost a few hundred bucks and last for years - and then go off to spend well over a thousand replacing last year's product.

That's not, mind you, the only reality defying miracle in sight: the realities of both mass production and retail distribution mean that fully assembled products always retail for less than the sum of their parts - except among Wintel enthusiasts. Thus a $90 lawn mower contains over $600 in retail parts, but lots of Wintel people really believe they can assemble a $400 PC from retail parts - including a Windows license and a processor wholesaling for $939 in lots of 10,000 - for a lot less than $400.

These delusions would be of interest only to social historians and psychologists if they didn't have some nasty consequences - including their effect on the politics of big company decision making. It's a lot easier, and personally less risky, for example, to get a dumb Wintel decision approved than a smart SPARC or open source one.

Imagine, for example, going to senior management for out of budget approval on a new machine intended to handle an exploding web/OLTP workload. In my experience, proposals to put in another Wintel box like an HP 580G5 will slide right through - but proposing an UltraSPARC machine running open source applications will not only expose you to pointed questions and outright antagonism from people in Finance and other management roles, but leave you personally responsible for any problems that ever come up with it if it does get approved.

To get some very rough impression of what this costs companies today, check out this cost and performance comparison between a Sun T2 running an all open source stack and an HP DL580 running roughly the same job using Microsoft software.

Both machines had 64GB of RAM and four of the same 12 x 146GB external storage arrays. The SPARC server had one T2 CPU at 1.4Ghz, the HP four quad core Xeons at 2.93Ghx, one ran an entirely open source stack, the other only Microsoft software.

The HP system cost $204K, the Sun $131 (64% of the PC cost). More importantly, the T2 produced more than twice the Xeon's throughput on the iGEN OLTP 1.6 benchmark and beat the HP by a factor of eight on SWaP (= Perf /[ Space (RU) x Watts ] )

The savings, about $70,000 up front and a few hundred a month in operations, don't amount to much in terms of big company IT budgets - but most companies big enough to use this stuff at all use quite a lot of it. Look at this as an open source dividend over five years and compare the cost of 50 of the HPs to 25 Suns and you're looking at nearly seven million bucks up front and another ten thousand or so per month - enough for another engineering group, a new sales office, even a new production line: whatever makes the money for the business.

This example focuses on comparing Wintel costs to SPARC/open source costs for typical OLTP and web processing but a comparable cost case exists for Cell/Linux versus Wintel on science side processing - and there's really no contest in either case. On both price and performance the the non Wintel opensource combination beats Wintel hands down on jobs large enough to matter.

So why does Wintel continue? because of mistaken beliefs - including most importantly the belief that other people believe in it - and that's bad for you, bad for the economy, and bad for your employer because when people base decisions on delusions they forgo the opportunity to generate competitive advantage from IT, and foreclose the option of spending an available open source dividend on more productive activities.

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.