Number of the Day: 11.8 Million
by Michael Graham Richard, Gatineau, Canada on 06.19.08

11.8 million -- The number of computer servers in the USA in 2007, up from 2.6 million in 1997 (source: IDC).
30.3 million -- Total number of computer servers worldwide. Also growing fast.
1/2 -- Half of the energy consumed by a typical data center (the place where the servers are housed) is used to remove the heat they generate: air-conditioners, fans, etc.
15% -- Most computers in the average data centers run at 15% capacity or less, idling the rest of the time but still consuming electricity. Thankfully, the trend is going in the right direction and the whole industry is looking for ways to reduce energy usage data centers.
25 -- By 2011, data center energy use could almost double and require the equivalent of 25 power plants. In the US, they are already (since 2006) using more energy than color televisions. "According to recent study from McKinsey & Company, could well surpass the airline industry as a greenhouse gas polluter by 2020." Source: New York Times


















Wow, more than televisions?
Too bad they're idling. Should run Folding@home or something :P
Higher CPU usage = higher power consumption. Believe it or not, most nerds are very conscious about wasting energy, both for the sake of the environment, and the massive cost of running these facilities. There's not a week that goes by that I don't see a new article on how to make server farms cheaper to run or more eco-friendly.
"Higher CPU usage = higher power consumption. "
Yes, but the CPU is still just one of many things.
Much better to have 50 computers running at 100% than 100 computers at 50%. That's why new energy management systems look at redistributing the load and shutting down as many servers as possible.
IBM has a nice cooling plan in the works. They have a new chip that has tiny pockets for water cooling. So not only do you have a more efficiently cooled chip, you can take that water and recapture some of the lost energy. I don't have a link for the video or anything, but they have a nice demo of how their proposed system would work.
It's always upsetting that we use MORE energy to get rid of the heat generated by computers. Fans are loud and not very efficient. The water cooling and recapturing method seems very promising. Imagine the cost savings such a system would give. The only problem is that business have already invested large amounts of money in current computing and cooling technology. It's going to be a while before these water cooling / energy recapturing systems become ubiquitous. Hell, I don't even think they're ready to go yet.
Here's a link to the IBM stuff (unfortunately the video is not there):
http://www.eetimes.com/showArticle.jhtml?articleID=208402316
http://www.engadget.com/2008/06/05/ibm-cools-stacked-silicon-chips-with-water/
http://arstechnica.com/news.ars/post/20080605-ibm-demonstrates-water-cooling-for-3d-processors.html
Found the video:
http://gizmodo.com/377696/ibm-mixes-water-chips-to-make-supercomputer-and-cook-dinner-too
http://www.youtube.com/watch?v=ioCZojN4A0g
The sad truth is that, unlike waste caused by un-efficient vehicles like SUV's, servers do a job that can't be done by a more efficient machine.
The clients I serve want efficient servers/workstations to cut down on utility bills, and larger companies are moving more into virtualization than simply adding more hardware.
If someone could invent a processor that performs as well as current generations of processors but produces little to no heat, that patent would probably be squashed faster than the patent for that engine that ran off sewage. The manufacturers of the cooling plants make too much money cooling down the datacenters around the world.
I think the best option at the moment is virtualization, and I think watercooling is just about ready for prime time. It's been used in high end personal machines for a while now to keep computers really cold, and I think that it is feasible to recapture some of the heat with the number of systems in a large server farm.
Of course, so long as companies want huge amounts of uptime, web hosting will be really inefficient. Another cool idea I read about was that some botnets now are using infected computers to host their scammy websites. (See a recent column on DansData.com) This could be used in a good way. Imagine opting to host part of your favorite website and receiving less ads in return, or say a monetary reward or a discount if the site is a shopping site. Or perhaps someone could design a web protocol like bit torrent to host popular sites, so that the more users who visit the site, the more available it is.
Can you tell I'm not a computer scientist :( ?
Everyone boycott the internet! discuss:
Virtualization is a good idea. It seems like a better option to the problem would be if computer manufacturers could make an instant-on PC (as in, could boot in a couple seconds or less). Then decisions don't have to be made about how many computers to keep idle to handle the fluctuations in demand. And no more wasted booting up/shutting down cycles.
Tendomentis, that is essentially what CPU manufacturers do every time they step down to a smaller process size (e.g., 90 to 65nm). The power requirements go down, and typically so does the heat. They are more efficient for the work they do. Demand in the market however, has settled on getting a faster processor for the same power requirements, rather than more efficient comparable CPUs. But there is a growing market in the mobile and compact laptop for more energy efficient machines, many of which use a similar processor at a smaller die size. It will be interesting to see if these start appearing more in the desktop and server market.
"Much better to have 50 computers running at 100% than 100 computers at 50%. That's why new energy management systems look at redistributing the load and shutting down as many servers as possible."
Thats the DUMBEST thing I think I've ever heard anyone say. So what happens when 1 or 2 of your servers needs to go down ? You have 48 servers running at 105% ? There are articles everyday on tech blogs about "greening the datacenter" etc etc, processors are getting better at running on less power.
Jack, you don't need to be rude.
I meant is strictly as "it's better for energy use" in the abstract. Not "it's better in a real world situation". Of course you always need some margin for failures, sudden peaks, etc.
But that's still the goal of the energy management systems, to get as close to possible to max CPU us while keeping uptime as high as possible.
15% usage, what we have now, is certainly not the optimal number...
Story on a related topic:
New green supercomputer powers up at Purdue
http://news.uns.purdue.edu/x/2008a/080610McCartneySICortex.html
Shows that there is a huge need for power management solutions to efficiently utilize power resources.
Story on a related topic:
Whitepaper More Power to You: Improving uptime and capacity planning through rack-level power and environmental monitoring
Today, IT and facilities managers are asking, "Am I getting enough power to my blades?" "Are there racks with excess capacity?" and "Am I getting enough cooling to those racks?" This white paper explores how intelligent rack PDUs can address these uptime and capacity planning issues.
http://www.raritan.com/resources/white-papers/power-management/