Electric slide for tech industry?

SANTA CLARA, Calif.--The computing industry is on a power trip, but it doesn't want to be.

Technology providers, customers, government officials and researchers gathered at Sun Microsystems' headquarters here Tuesday to try to tackle some of the problems posed by the runaway consumption of electricity by computing gear.

The problems arise from a confluence of business demands, rising energy prices and technology changes, which have led to chips and computers that consume more electricity. The result is a conflict of priorities. Some people try to pack servers in more densely to use floor space better, while others try to space them out to reduce overheating problems.

"We're at a perfect storm," Ben Williams, vice president of commercial business for chipmaker Advanced Micro Devices, said in a speech at the Global Conference on Energy Efficiency in the Data Center. "Business units are requiring IT departments to do more with less. We have (chief financial officers) question why all the racks aren't filled. The IT guys are saying 'I need another data center.' Then we have the rising cost of energy that's bringing all this to the forefront."

But it's in the interest of anyone consuming power to improve efficiency, argued Andrew Fanara of the EPA's Energy Star program. "Companies have to ask themselves, 'Am I willing to bet the cost of energy is going to go down?' That's the cost of doing nothing," Fanara said.

To get a grip on the problems, the tech industry has come up with solutions ranging from more energy-efficient Xeon processors to liquid cooling of high-end server systems.

Trends aren't promising. Four or five years ago, a 6-foot tall rack full of computing gear would consume 2 to 3 kilowatts, "but now we're talking about 10-, 15- or 20-kilowatt-draw racks," said Sun Chief Technology Officer Greg Papadopoulos. Google has warned that its server electricity costs soon might outpace its server purchase costs.

"Companies have to ask themselves, 'Am I willing to bet the cost of energy is going to go down?'"

--Andrew Fanara, Energy Star program, EPA

The problem is getting worse as data centers packed full of computers proliferate and grow in size. Data centers typically have raised floors with holes in them to direct specially cooled air coming from below straight into server compartments.

"Between now and 2009, we expect 12 million additional square footage of raised floor going into marketplace," IDC analyst Vernon Turner said. By comparison, the Mall of America in Minnesota, the world's largest shopping mall, measures 2.5 million square feet. "Think of that filled to the brim with servers," Turner said.

A large fraction of the energy consumed in data centers goes to waste, said Bob Sullivan, a data center design expert from the Uptime Institute. In a survey of 19 data centers, 1.4 kilowatts of power are wasted for every kilowatt of power consumed in computing activities, the research consultancy found.

Measure the problem
The first step, several at the conference agreed, is to develop a useful common measurement of system performance. The industry could then balance that against power consumption, to judge how bad the power efficiency problem is and how effective solutions might be.

"Most companies agree the priority...is an objective measurement of the service being delivered," Jonathan Koomey, a computer power expert who works at Stanford University and Lawrence Berkeley National Laboratory, said in an interview. He expects that measurement to emerge from discussions he's involved in, which have included server makers, the Environmental Protection Agency and others. "I would hope a year from now, we'll have at least a draft metric," he said.

But most speakers here were reluctant to suggest a performance measurement process, and several agreed it's a thorny issue. For one thing, different companies inevitably try to pick tests that make their own equipment look good. For another, it's never easy to pick tests that represent the performance of everything from processors to storage and networking.

"It's going to be tough to pick something that's broad enough and yet simple enough to be practical on a day-to-day basis," said Peter Bannon, vice president of architecture at P.A. Semi, a start-up that develops low-power processors.

Sun's Papadopoulos said that even something as apparently simple as a vehicle's fuel efficiency isn't straightforward to measure--and compare--in reality. A motorcycle beats out a sport-utility vehicle in raw miles per gallon, but an SUV can be more efficient overall because it can carry more passengers. But then SUVs often aren't always filled to capacity with passengers, so utilization also must be factored in.

[cyn1c]

Power Supply Efficiency

"The EPA lets manufacturers give such supplies, which convert AC power from the wall to DC power used inside the computer, an "80+" label if they're more than 80 percent efficient. That "Energy Star" label means that they lose less than 20 percent of the power they draw in waste heat. But efficiencies of only 70 or 72 percent are typical for power supplies."
Efficiency measurement is going to depend on power supply loading. There is no efficiency constant for the power supply, it is a curve from light load to full load, reaching a peak somewhere in the middle (like 75% full load).
Perhaps if power supplies were better selected for their specific use, they could achieve real improvements in wasted power reduction. I get the feeling that the power supply is usually an afterthought, oh 550 Watts that ought to be enough.

[mendozamanny]

72-75%??? 80+ Power Supplies is the way to go.

I have been pushing 80+ Seasonic Power supplies. PG&E will not only give a rebate of up to 10.00 for each 80+ power supply in a server, you will save money on your electricity bill and have a cooler system.

www.jameco.com for more information

Teach Them To Fish

Most of the press this week has been focused on what IT equipment and component manufacturers can do, should do and are doing. Not once have I read a single statement that puts any responsibility of efficient energy use back on to the owners and operators of data centers. The industry, as a whole, has provided many best practices documents to help customers make the best of what they have. However, complacency has been the order of the day. Take the basics best practices such as hot-isle/cold-isle, using blanking panels and plugging holes in floors. Over 50% of the computer rooms I've been into don't or won't adopt those practices.
LBNL published a study in Feb 2003 that showed the average energy balance of energy consumption of 12 data centers. Cooling cost as a percentage of overall energy use ranged from as low as 22-25% to as high as 50-55%. Those doing a good job obviously spend less on cooling due to more efficient airflow management. Our Advanced Technology Team in Houston has been been publishing White Papers on best practices for more than three years.
There is a saying "while I'm pointing one finger at you (hardware industry) I must realize the there are three fingers pointing back at me".
Before any hardware solutions can begin to make a significant difference, Computer Room operators and owners need to realize their role and responsibility in using what they have in a practical and efficient manner. Most data centers are so screwed up in their layout and management, no hardware fix will help cure their energy consumption problems.
"Teach them to fish and they can feed themselves".

Bob Pereira
Enterprise Infrastructure Technologist
Industry Standard Server Group
Hewlett Packard

[Zymurgist]

Power supplies?

While I appreciate an efficient power supply as
much as the next guy (and I believe that
efficiency for the EPA is measured at peak
load), how about the components? Those things
are getting *HOT*. If you want to talk
inefficient, there you go.

I'd also add that we home consumers would really
like some of that energy efficiency too. Home
servers, media centers, etc. are getting to be
pretty common. Could we get something that
doesn't dim the lights when we turn it on,
please?

[CableDOC]

Forgotten A/C Loads

The article only briefly mentions air conditioning, yet a 20 kw draw rack consumes 6 tons of A/C, or more than a house. Time for demand loading, or else send all rack rooms to the artic and melt icebergs.

Article about efficient computing that does not mention Opteron?

You guys outdo yourself every time!

[MediaRites]

finger pointing

The government doesn't want to focus on the users because they won't agree to wear the e-star mark on the top or front of their heads. ;-)

Efficiency, Usage, and AC/DC conversion

I would be astounded if the average powers supply
gets over 70% efficiency. More likely is the
situation found with "wall warts" in which the
device consumes 1 watt even if there is no load,
plus 20% of whatever is plugged into it.

If a cell phone charger or similar intermittent
usage device is plugged in 24/7, then you might
see 70% efficiency when the device is actually
charging a battery, and 0% efficiency (100% waste)
when the battery is full. Total system efficiency
might average out to 10% or 20% depending, and
that assumes that the power supply is properly sized to its load.

If, as often happens, the power supply is twice
as large as it needs to be, then there will most
likely be additional base power consumption, which
means that the total efficiency will fall down to
the 60% range.

Computer systems which turn off disk drives, monitors,
etc. do not help with this problem: You have to
disconnect the main power transformer from the
wall plug.

The actual situation with regard to power supplies
is that, while many systems, especially in servers,
are able to achieve efficiency over 75%, there
are a large number of power supplies out there
which are under 20%.

[mastmar221]

Factual Errors

Sorry to nit pick, but this is a little bit of home town pride. While the mall of america is the largest mall in the US, it is not the largest in the world. Edmonton, Alberta in Canada has the largest mall in the world named West Edmonton Mall. Sorry guys.