Support NeoGAF

[A Black Falcon]

I don't think this has been posted...

That 90% number refers to that apparently the average internet company data center server is only actually using 6 - 12% of its processor capacity at any one time, but it's using 100% of its electricity at all times. And there are huge numbers of data centers now, all over the country.

Plus, as an added bonus for even more wasted power, huge amounts of diesel exhaust from the backup generators all data centers run as backup for if the electricity goes out. They're not all running at all times, but do have to run occasionally, or for testing, and create a lot of pollution. Yeah, not good. I wonder, will anything be done about this sometime? This kind of power waste can't continue...

Here's the article. It's six pages long, so I won't quote the whole thing. It's very interesting, read it.

http://www.nytimes.com/2012/09/23/t...belying-industry-image.html?pagewanted=1&_r=0

Some choice sections from the full article:
(but read the full thing too!)

Most data centers, by design, consume vast amounts of energy in an incongruously wasteful manner, interviews and documents show. Online companies typically run their facilities at maximum capacity around the clock, whatever the demand. As a result, data centers can waste 90 percent or more of the electricity they pull off the grid, The Times found.

To guard against a power failure, they further rely on banks of generators that emit diesel exhaust. The pollution from data centers has increasingly been cited by the authorities for violating clean air regulations, documents show. In Silicon Valley, many data centers appear on the state government’s Toxic Air Contaminant Inventory, a roster of the area’s top stationary diesel polluters.

Worldwide, the digital warehouses use about 30 billion watts of electricity, roughly equivalent to the output of 30 nuclear power plants, according to estimates industry experts compiled for The Times. Data centers in the United States account for one-quarter to one-third of that load, the estimates show.

Even running electricity at full throttle has not been enough to satisfy the industry. In addition to generators, most large data centers contain banks of huge, spinning flywheels or thousands of lead-acid batteries — many of them similar to automobile batteries — to power the computers in case of a grid failure as brief as a few hundredths of a second, an interruption that could crash the servers.

That secrecy often extends to energy use. To further complicate any assessment, no single government agency has the authority to track the industry. In fact, the federal government was unable to determine how much energy its own data centers consume, according to officials involved in a survey completed last year.

The survey did discover that the number of federal data centers grew from 432 in 1998 to 2,094 in 2010.

Today, roughly a million gigabytes are processed and stored in a data center during the creation of a single 3-D animated movie, said Mr. Burton, now at EMC, a company focused on the management and storage of data.

Just one of the company’s clients, the New York Stock Exchange, produces up to 2,000 gigabytes of data per day that must be stored for years, he added.

EMC and the International Data Corporation together estimated that more than 1.8 trillion gigabytes of digital information were created globally last year.

“It is absolutely a race between our ability to create data and our ability to store and manage data,” Mr. Burton said.

About three-quarters of that data, EMC estimated, was created by ordinary consumers.

With no sense that data is physical or that storing it uses up space and energy, those consumers have developed the habit of sending huge data files back and forth, like videos and mass e-mails with photo attachments. Even the seemingly mundane actions like running an app to find an Italian restaurant in Manhattan or a taxi in Dallas requires servers to be turned on and ready to process the information instantaneously.

Nationwide, data centers used about 76 billion kilowatt-hours in 2010, or roughly 2 percent of all electricity used in the country that year, based on an analysis by Jonathan G. Koomey, a research fellow at Stanford University who has been studying data center energy use for more than a decade. DatacenterDynamics, a London-based firm, derived similar figures.

The industry has long argued that computerizing business transactions and everyday tasks like banking and reading library books has the net effect of saving energy and resources. But the paper industry, which some predicted would be replaced by the computer age, consumed 67 billion kilowatt-hours from the grid in 2010, according to Census Bureau figures reviewed by the Electric Power Research Institute for The Times.

Direct comparisons between the industries are difficult: paper uses additional energy by burning pulp waste and transporting products. Data centers likewise involve tens of millions of laptops, personal computers and mobile devices.

The Viridity tests backed up Mr. Stephens’s suspicions: in one sample of 333 servers monitored in 2010, more than half were found to be comatose. All told, nearly three-quarters of the servers in the sample were using less than 10 percent of their computational brainpower, on average, to process data.

The data center’s operator was not some seat-of-the-pants app developer or online gambling company, but LexisNexis, the database giant. And it was hardly unique.

McKinsey & Company, the consulting firm that analyzed utilization figures for The Times, has been monitoring the issue since at least 2008, when it published a report that received little notice outside the field. The figures have remained stubbornly low: the current findings of 6 percent to 12 percent are only slightly better than those in 2008. Because of confidentiality agreements, McKinsey is unable to name the companies that were sampled.

A company called Power Assure, based in Santa Clara, markets a technology that enables commercial data centers to safely power down servers when they are not needed — overnight, for example.

But even with aggressive programs to entice its major customers to save energy, Silicon Valley Power has not been able to persuade a single data center to use the technique in Santa Clara, said Mary Medeiros McEnroe, manager of energy efficiency programs at the utility.

“It’s a nervousness in the I.T. community that something isn’t going to be available when they need it,” Ms. McEnroe said.

Terry Darton, a former manager at Virginia’s environmental agency, said permits had been issued to enough generators for data centers in his 14-county corner of Virginia to nearly match the output of a nuclear power plant.

Some industry experts believe a solution lies in the cloud: centralizing computing among large and well-operated data centers. Those data centers would rely heavily on a technology called virtualization, which in effect allows servers to merge their identities into large, flexible computing resources that can be doled out as needed to users, wherever they are.

Others express deep skepticism of the cloud, saying that the sometimes mystical-sounding belief in its possibilities is belied by the physicality of the infrastructure needed to support it.

Using the cloud “just changes where the applications are running,” said Hank Seader, managing principal for research and education at the Uptime Institute. “It all goes to a data center somewhere.”

Some wonder if the very language of the Internet is a barrier to understanding how physical it is, and is likely to stay. Take, for example, the issue of storing data, said Randall H. Victora, a professor of electrical engineering at the University of Minnesota who does research on magnetic storage devices.

“When somebody says, ‘I’m going to store something in the cloud, we don’t need disk drives anymore’ — the cloud is disk drives,” Mr. Victora said. “We get them one way or another. We just don’t know it.”

 

[grimshawish]

Using the cloud “just changes where the applications are running,” said Hank Seader, managing principal for research and education at the Uptime Institute. “It all goes to a data center somewhere.”

“When somebody says, ‘I’m going to store something in the cloud, we don’t need disk drives anymore’ — the cloud is disk drives,” Mr. Victora said. “We get them one way or another. We just don’t know it.”

I like these people.

 

[goodfella]

http://www.theverge.com/2012/9/23/3377868/cloud-internet-infrastructure-waste-energy-new-york-times

Retort from The Verge

 

[Suairyu]

I'm amazed that companies can justify the expense of wasting energy. You don't even need the environmental argument - they're just throwing away money.

 

[saltywalrus]

Why is CPU usage relevant? Just because CPU isn't being used doesn't mean the disks and memory aren't being used.

 

[Magni]

Why is CPU usage relevant? Just because CPU isn't being used doesn't mean the disks and memory aren't being used.

If the disks get accessed, then the CPU is used.

 

[Ether_Snake]

Free market!

 

[jaxword]

So they switched to streaming instead of hoarding their collection.

 

[goomba]

Servers should start using more energy efficient architecture, like ARM CPU''s.

 

[Dr.Acula]

Do they know how much it costs to build a datacentre? They're not profitable on day one. So why would anyone build a datacentre that's already at maximum capacity on day one?

I don't buy a PentiumII and overclock it to run Win7.

TFA even says that they can virtualise servers to take advantage of idle cycles. Ooooh, so now it's a good thing the existing data servers are undercapacity, isn't it? Cloud needs sky!

 

[iamblades]

If the disks get accessed, then the CPU is used.

What he is saying is that cpu utilization is not the only thing that matters to power usage, and it's not a good measure of how much these servers are being used.

A basic file server can have really high utilization of bandwidth and file system and still be low utilization on CPU.

 

[SRG01]

There are different types of servers for different types of loads. That being said, the article is not off the mark. There are many ways to build reliable/redundant systems without energy usage on this scale.

Interesting fact: much of a data center's energy usage goes towards cooling! In fact, infrastructure is perhaps more of a design challenge than the servers themselves.

 

[danwarb]

They should totally switch that power saving mode on.

 

[The Faceless Master]

print media disses cloud, talks up paper... news at 11...

 

[SquiddyCracker]

Worldwide, the digital warehouses use about 30 billion watts of electricity, roughly equivalent to the output of 30 nuclear power plants, according to estimates industry experts compiled for The Times. Data centers in the United States account for one-quarter to one-third of that load, the estimates show.

Jesus christ.
So much for using resources more efficiently.

Sorry Zaptruder, this is why won't be hitting post-scarcity anytime soon.

 

[lexi]

Servers should start using more energy efficient architecture, like ARM CPU''s.

not sure if serious.

 

[DonasaurusRex]

hey gotta keep all that stuff cool or it goes BOOM. Im kinda wary about how accurate this could be, that had to be an exhaustive inquiry into all the hardware used i mean...damn. I dont necessarily doubt it...because lets face it a data center is mission critical , all other metrics are secondary to the time online and available. I wonder if newer power gating , and downclocking technologies have helped make a dent in this , then again data centers hold onto hardware much longer than smaller customers/end users , so probably not.

 

[ElectricBlue187]

yeah fuck that cloud nonsense

Spoiler

the cloud is actually quite useful in certain instances

 

[teh_pwn]

not sure if serious.

Yeah, I don't know why everyone things ARM is so efficient. I'd like some numbers, but I bet Intel's Haswell beats it in performance/watt. And servers need performance.

Once Xeon designs based on Haswell and SSD are standard/cheap/large, server computing will run on much less power.

SSDs use like .1 watts. So even with some crazy large network RAID for a server we're talking very little power.

 

[Windu]

 

[Gallbaro]

Free market!

This its the free market at work, customers want high availability more than they are concerned about that extra margin of costs and are willing to pay for it.

 

[Technosteve]

servers should run on the matrix human power supply

 

[zou]

What a stupid and confused article. The author makes no attempt to actually explain the different parties and instead lumps everyone together when talking about "the datacenters". He also makes a lot of ridiculous and misleading claims (like calling servers with "only" on average 10% utilization comatose) and never bothers to provide any useful details on those cherry picked numbers of his: Were all the servers owned by one company? Maybe it was a facility providing co-location hosting?

He also seems awfully naive or clueless when talking about those "extravagant" generators that are running on stand-by just in case they lose power for just a few ms. And Amazon still went down, see? And then he includes that one random dude that's saying these people are way too conservative and that there are perfect alternatives.. These being...? Mr. Author? And while he claims everyone is being secretive, he completely ignores and doesn't even mention the publicized energy conservation efforts and hw/dc designs by the likes of Facebook, Google and MS.

And lastly, the premise of the whole article is shit to begin with. Unless you are going to prohibit/restrict customers from purchasing/renting servers based on their usage, there is always going to be plenty of idle servers. Add to that anyone running critical services that will want to have stand-by servers as a fail-safe. Which leaves all the companies that are actually running the majority of these servers and that have plenty to gain from improvements in energy efficiency: And guess what? Every single one of them is working on it and has been for a few years now.

 

[BobTheBub]

Yeah, I don't know why everyone things ARM is so efficient. I'd like some numbers, but I bet Intel's Haswell beats it in performance/watt. And servers need performance.

Once Xeon designs based on Haswell and SSD are standard/cheap/large, server computing will run on much less power.

SSDs use like .1 watts. So even with some crazy large network RAID for a server we're talking very little power.

There are serious efforts to use ARM in the Datacentre, HP announced an initiative with these guys http://www.theregister.co.uk/2011/11/01/hp_redstone_calxeda_servers/

But seem to have abandoned it for the moment and are doing work with Atom based servers.

 

[Inflammable Slinky]

Uhh.


Why are the data centers using tons of power? How can it be wasteful if the center is paying for electricity? I mean, I doubt these companies enjoy paying for electricity for no reason.

Sure, it could be wasteful, but this article makes it sound like there's some sort of conspiracy to waste as much energy as possible just for lulz.

 

[speculawyer]

Yeah this may be very misleading. By their nature, computers spend most of their time doing nothing unless it is a videogame or some complex simulation. Just fire up your task manager and look at the % of CPU use. Most of the time it is nothing.


But there is waste in server systems and a lot of companies in Silicon Valley are actually working hard to address this problem. So to the degree that this is true, there are people working on the issue.

 

[KHarvey16]

Just skimmed, but it seems that by this articles logic, my computer uses the same amount of electricity whether it's sitting idle or I'm playing a graphics intensive game, simply because in both cases it's using 120V. Please tell me this isn't what it's saying.

 

[Zaptruder]

An interesting article and one that is definetly a little biased against technology.

But reading between the lines - understanding that this technology is essential, and that it nonetheless does things that traditional technology isn't capable of, or does it much more efficiently (even with the wastage) - what's important is to change the culture of data centers.

Push data towards virtualization, towards the cloud - move the onus of maintaining data off individual companies, and onto data center and virtualization companies that are able to maintain user and company data as required.

Of course, virtualization is its own very complex construction... all I can say is that it's possible, and can help to dramatically boost the utilization rate of existing hardware.

At some point in the near future, we have to focus on getting that utilization rate up. As we move forward into the future, the demands of technology will continue to grow... rightfully so as it may be, but this is not without impact, and not immune to the considerations of efficacy as with any other facet of society.

 

[midonnay]

1.3 percent versus 25 percent for transport?

sounds more like an argument for more videoconferencing and less zoom zoom :/

 

[Zaptruder]

Just skimmed, but it seems that by this articles logic, my computer uses the same amount of electricity whether it's sitting idle or I'm playing a graphics intensive game, simply because in both cases it's using 120V. Please tell me this isn't what it's saying.

It is been a little simplistic - but the point still holds some degree of validity.

When you use 10 machines to do what 1 can do at full blast - you need space for 10 machines, you need power enough to idle 9 machines and operate 1 at max capacity, you need cooling for those 10 machines, and you need essentially all the back end infrastructure that we don't think about for those 10 machines, even though you really only need the computing power of one machine.

For my personal machine... I believe normal operation is roughly equivalent to idling it (i.e. using web browser, maybe running music) - takes about 100-150 watts of power. Playing a graphics intensive game, the utilization goes up to around 500 watts.

Going off that - you're still using 20-25% of power even when idle. A rough back of the napkin calculation shows that 10 machines are drawing 100% + (20*9), or roughly 3 machines worth of power to do 1 machine's worth of work.

 

[HAL_Laboratory]

We need cycle farms. 100 rows of 2000 cycles, all powered by human legs. Clean energy jobs.

 

[KHarvey16]

It is been a little simplistic - but the point still holds some degree of validity.

When you use 10 machines to do what 1 can do at full blast - you need space for 10 machines, you need power enough to idle 9 machines and operate 1 at max capacity, you need cooling for those 10 machines, and you need essentially all the back end infrastructure that we don't think about for those 10 machines, even though you really only need the computing power of one machine.

For my personal machine... I believe normal operation is roughly equivalent to idling it (i.e. using web browser, maybe running music) - takes about 100-150 watts of power. Playing a graphics intensive game, the utilization goes up to around 500 watts.

Going off that - you're still using 20-25% of power even when idle. A rough back of the napkin calculation shows that 10 machines are drawing 100% + (20*9), or roughly 3 machines worth of power to do 1 machine's worth of work.

So what do you do when you need all 10 machines?

 

[Vic]

There are different types of servers for different types of loads. That being said, the article is not off the mark. There are many ways to build reliable/redundant systems without energy usage on this scale.

Interesting fact: much of a data center's energy usage goes towards cooling! In fact, infrastructure is perhaps more of a design challenge than the servers themselves.

Figured that was the case.

 

[Rentahamster]

Worldwide, the digital warehouses use about 30 billion watts of electricity, roughly equivalent to the output of 30 nuclear power plants, according to estimates industry experts compiled for The Times.

This would be a non-issue if, you know, they all actually did run on 30 nuclear power plants. But most rely on fossil fuels for energy. And the data centers are the polluting ones...? They're just buying the juice regardless of the source. (backup generators and UPS systems aside).

Modern computing is as essential to our 21st century civilization as other basic essentials like food, shelter, and water. Computing helps us provide those basic essentials more efficiently and in greater quantities than ever before.

Computing is also what is going to help create the next breakthroughs in energy development, resource management, and ultimately, the end of our reliance on fossil fuels.

I'm amazed that companies can justify the expense of wasting energy. You don't even need the environmental argument - they're just throwing away money.

Providing reliability, performance, and stability is profitable too.

 

[BobTheBub]

So what do you do when you need all 10 machines?

Switch them on, this is the type of thing available for enterprise type use http://www.vmware.com/resources/techresources/1080

I have no idea whether the really huge places have their own proprietary solutions or use customised versions of that type of thing but they have only been available for a few years and are probably not used across the board and not suitable for some uses.

Edit: Google are actually very secretive about what they do regarding a lot this type of stuff and consider it a competitive advantage.

 

[midonnay]

Interesting fact: much of a data center's energy usage goes towards cooling!

couldn't they have water cooling like nuclear reactors

 

[Zaptruder]

So what do you do when you need all 10 machines?

That's the thing - in aggregate terms, we'll never need every last machine in every data center.

We only need those 10 machines because we're not aggregating all that computing power and balancing the load better - because each company insists on owning their own hardware and infrastructure.

If we use virtualization* strategies - where instead of having 1000 computers from 100 companies that are only used at 10% of capacity, we can have more in the region of 150-200 computers; with 100 been used at full capacity normally, with the other 50-100 providing more than enough overhead for the rare times where usage spikes.

*which I understand to be the act of turning the entire data center's worth of machines into one massive parallel computing cluster, and then treating each company's applications as a set of processes, rather than having dedicated machines for each application/company.

 

[Arthrus]

 

[soul creator]

hahaha, I didn't catch it until you posted this

 

[jchap]

When someone engineers a load following data center they will become rich and it will be widely adopted to save costs.

This is free market.

 

[jvalioli]

There is no good way to do this study since the only way to get a lot of the required information is through companies who consider everything about their datacenters to be a secret.

 

[xbhaskarx]

lol

 

[Razek]

What he is saying is that cpu utilization is not the only thing that matters to power usage, and it's not a good measure of how much these servers are being used.

A basic file server can have really high utilization of bandwidth and file system and still be low utilization on CPU.

Wait, I'm confused as well. CPU usage does not scale linearly with other part consumption, how are they arriving to these numbers?

 

[Zaptruder]

When someone engineers a load following data center they will become rich and it will be widely adopted to save costs.

This is free market.

Or the giant corporations that have the money and expertise to spend on that level of engineering will keep their secrets to themselves for the purposes of a competitive advantage.

This is free market.

 

[Open Source]

Or the giant corporations that have the money and expertise to spend on that level of engineering will keep their secrets to themselves for the purposes of a competitive advantage.

This is free market.

These kinds of secrets don't last more than about 6 months in the real world. You should visit sometime!

 

[DECK'ARD]

We need cycle farms. 100 rows of 2000 cycles, all powered by human legs. Clean energy jobs.

All of Black Mirror will come true at some point ...

 

[alphaNoid]

I can say this, my company doesn't waste shit in our 2 DCs. They micro manage and bill us on power usage and we ride the line for the most part.

 

[remnant]

The article is complete rubbish, and it kind of reminds me of the "Foxconn hysteria" that the NYT also got in on. Take an issue most people don't know and a surprising amount don't understand, throw in some yellow journalism, and wait for the accolades. You will always have idiots

Free market!

That are willing to look at the surface to support their ideology, so the piece will go unchallenged. Some time later the article will be debunked but by then their will already be cries for regulations and oversight to make sure that data centers don't destroy the world.

 

[ymmv]

Most data centers, by design, consume vast amounts of energy in an incongruously wasteful manner, interviews and documents show. Online companies typically run their facilities at maximum capacity around the clock, whatever the demand. As a result, data centers can waste 90 percent or more of the electricity they pull off the grid, The Times found.

I get the impression that the author thinks all these servers are completely interchangeable and you can turn half of them off at night without disrupting service.

 

[UnluckyKate]

All of Black Mirror will come true at some point ...

God no