RoboGeek

CNET: Power could cost more than servers, Google warns

Here we go: Power could cost more than servers, Google warns

The CNET article is referring to an article published in ACM's Queue: The Price of Performance (ACM Queue vol. 3, no. 7 - September 2005, by Luiz André Barroso, Google) An Economic Case for Chip Multiprocessing

His point is that as system performance is going up, power consumption increases to match.  Hence as system performance increases lets you cram more and more into your data center, the power needs for the data center will increase dramatically.  Eventually, his numbers say, the cost of powering your computers will cost more than the computers themselves.

As I wrote before I'm just happy that the new systems we're selling have lower power needs than previous systems.  I am very interested to have the world I live in be clean, and I know that the more power we humans use the more polluted our world becomes (because of the way we get the power).

I know that one can often do the same work (e.g. light your room) while using less power (e.g. using compact flourescents or LED lightbulbs), which makes me itch for the rest of the humans around me to catch on that they don't need to use as much power as they're using today.

It's about efficiency and I like the way Luiz puts it in his paper.  Performance per watt.

It's also about coming up with the right measurement to capture the desired end goal.  See, the results one gets are always based on the question you ask.  If you ask simply for "give me more processing power" then the easy answer is to make the CPU run faster and faster.  But we've seen with Intel's CPU approach how the faster you make the CPU go, the more power it consumes, the more heat it dissipates, the more you have to spend on cooling systems and the more you spend on power.

But if you ask for a broader picture of "more power at lower cost of ownership" that changes how you approach the problem.  And if you toss in "oh, and it would be nice if it saved the planet" the approach is changed again.

I'm a software guy and if I say anything more I'll probably get in trouble.

(2005-12-09 09:45:07.0) Permalink Comments [0]