[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Beowulf] Redmond is at it, again


Eugen Leitl wrote:

Yeah, think of all those power Excel hydro code users who're going to
switch...

http://zdnet.com.com/2102-1103_2-5219282.html?tag=printthis

<snip>

For now, Linux has the upper hand, owing to its affinity with Unix--the OS
environment the high-performance crowd is most comfortable with--and the
open-source model, which lets users turn directly to source code for answers
to problems. But a Microsoft product would theoretically integrate better
with Windows desktop machines, and if the company can serve up an impressive
offering, Linux could be in for a tussle.

Gates will have to explain that personally to the graduate students and postdocs who actually do the work, I would think. I wouldn't put it beyond him to try to do that, actually, and it could conceivably be worth his trouble.

<snip>


In a recent interview, Bob Muglia, a Microsoft senior vice president who
leads the development of Windows Server, said the company is interested in
two particular areas: building high-performance computing clusters and
harvesting the unused processing power of PCs.

Although Microsoft is a comparative newcomer to the market, the company could
bring several advantages:

. Machines running Windows HPC Edition could seamlessly connect to desktop
computers, providing instant power for someone such as a financial analyst
performing calculations on an Excel spreadsheet, said David Lifka, chief
technology officer for the Cornell Theory Center, Microsoft's premier
high-performance computing partner.

<snip>


. And Microsoft could build software into its desktop version of Windows to
harness the power of PCs, letting companies get more value from their
computers. It's a technology that's applicable to tasks such as drug
discovery and microchip design.

If there is a way to take this seriously, I thought, it is in "harvesting the unused processing power of PC's." The annual market for HPC is about $5 billion and for PC's about $200 billion. Taking the market price of an "HPC CPU" to be ten times that of a "PC CPU" (starting from an estimated $100 million for 10000 Red Storm processors), that would put 400 times as many "PC CPU's" entering service each year as "HPC CPU's".

From the point of view of HPC, 400 is not a very interesting number, especially if you account for even the tiniest bit of the difficulty you'd encounter in trying to harness any serious fraction of so many more available CPU's.

Even if the price disparity between a "PC CPU" and an "HPC CPU" were more like a hundred, roughly the difference between a mainframe CPU and a PC CPU, that still only makes 4000 times as many available CPU's--still not a very interesting number.

Were I Microsoft, I'd be aiming to tell the world that Microsoft is not making money, it is solving the problems of science with the untapped power of PC's. We will probably see something along those lines from Microsoft. If it stood up to scrutiny, it might be in the interests of science, even if not in the interests of those who have built their careers around linux clusters. It's a great marketing line, and, if anyone would have the resources to make such a thing happen, it would be Microsoft. Thinking again rather cynically in the interests of science, having Microsoft adopt such a marketing posture and making it stick to any extent at all might actually be a good thing, since it would be difficult for Redmond to be a fairly obvious market predator at the same time as it is being the salvation of mankind.

Had Microsoft the wisdom to see it though, it could probably work around the graduate student/postdoc problem by artful use of Windows Services for Unix. A _really_ smooth Microsoft would even give the appearance of supporting the use of Linux for science while in reality using that support as a way of keeping Linux contained.

RM



_______________________________________________
Beowulf mailing list, Beowulf@xxxxxxxxxxx
To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf


This mailing list archive is a service of Copilot Consulting.