Slashdot Mirror
Data Center Managers Weary of Whittling Cooling Costs
Nerval's Lobster writes that a survey from the Uptime Institute "suggests something it calls 'green fatigue' is setting in when it comes to making data centers greener. 'Green fatigue' is exactly as it sounds: managers are getting tired of the increasingly difficult race to chop their PUE, or Power Usage Effectiveness. The PUE is a measure of a data center's efficiency. The lower the PUE, the better — and Microsoft and Google, with nearly limitless resources, have set the bar so high (or low, depending on your perspective) that it's making less-capitalized firms frustrated. Just a few years ago, the Uptime Institute estimated that the average PUE of a data center was around 2.4, which meant for every dollar of electricity to power a data center, $1.4 dollars were spent to cool it. That dropped to 1.8 recently, an improvement to be sure. But then you have companies such as Google and Microsoft building data centers next to rivers for cheap hydroelectric power in remote parts of the Pacific Northwest and reporting insanely low PUEs (below 1.1 in some cases). The Institute latest survey of data center operators shows only 50 percent of respondents in North America said they considered energy efficiency to be very important to their companies, down from 52 percent last year and 58 percent in 2011."
How irresponsible for them to cry that their competitors are destroying the environment less than they are. Fuck you, you greedy little fucks. You are lucky that you get to be as wasteful as you do. When the world starts getting too hot you are going to be bitching about that too and once again fail to realize that it's your fault.
that always works
These are just publicity stunts. Computing is cheap in terms of energy, the energy used by datacenters barely registers in the total energy usage.
I like the way Facebook say they don't use HVAC... yet their entire BUILDING is a huge HVAC unit!
Efficiency of scale works nicely with HVAC, if you can afford to get the building made to your specs.
In the free world the media isn't government run; the government is media run.
Dumb summary.
What does is matter how cheap the electricity is?
It is a ratio of two electricity costs.
Price of electricity has no effect on PUE.
Maybe climate has.
Cooling in arctic is cheaper than cooling in nevada desert.
Bram Stolk http://stolk.org/tlctc/
Copy the big boys and move to Prineville! Land is cheap and the city would love to have you. Just ask Facebook and Apple.
Full disclosure. I am not affiliated with the Prinville coc. I just live in central Oregon and data centers are a no brainer here.
Why don't they just site their centers up north? Here in Duluth, most of the year the outside air is cooled for free by mother nature. Heck, they could sell their waste heat to nearby homes and businesses and get a negative PUE.
Don't need to be green to worry about this, it's $$, something ever company wants.
For instance, I suspect we waste more energy moving tap water in plastic bottles between cities.
One could also interpret the data as saying that since 2011, 8% of data center operators have looked into improving their energy efficiency and have done as much as they think feasible. That 50% consider energy efficiency very important in the latest survey suggests that it is still is. Data centers use about 2% of the electricity consumed in the United States.
Those PUE numbers don't seem up to date. Yahoo gets 1.08 PUE with their "chicken coop" design at their Buffalo data center. Google has said they can get 1.08 as well. Yahoo's design has been around since 2010 and Google brought out their more efficient data centers shortly after that. I'd be surprised if more companies weren't following in those footsteps.
There may be SOME data center managers who are smart and energy literate ... but as someone who travels across the country visiting different companies for wireless and security purposes I can tell you they are the absolute minority.
I see only about 10 to 20 % of them even have dedicated hot air return plenums - and this might be as simple as using the plastic sheeting your grocery store use hanging partially over the racks ... but nope ...
Those HP racks with plastic plates don't help either - it only they would make them with an ability to mount once a device has been racked .. but once those square holes are filled ...
Or to put it another way, since 2011 8% of data center operators have decided that they have other priorities than further improving what they perceive to be energy efficiency numbers that are as good as the cost of improving further can justify (while a much larger number still say it is a priority, but have no intention of improving it in the near future because they don't think the cost is justified by the potential savings).*
*None of this is intended to in anyway be in disagreement with any part of the point you made, just trying to phrase your point in another way (perhaps not well).
The truth is that all men having power ought to be mistrusted. James Madison
'Green fatigue' is exactly as it sounds
Green Fatigue sounds like a uniform for Army. Unlike never-nude, green fatigue isn't exactly as it sounds.
Typical modern groupthink - if you dont match up to some artificial social standard you lose. Watch your own checkbook, don't chase some mythical metric that others self-report. You'll never win, they'll just keep moving the goalposts. Spend less money as you expand capacity, and you're doing a good job.
I want to delete my account but Slashdot doesn't allow it.
Typical ignorant American attitude. "Boo-hoo, we can't fuck up the whole planet for greed! Unacceptable!"
Guess what: NOBODY CARES!
Cause it's *us* you are killing with that behavior. The whole fuckin' planet! Now you have ten seconds to guess if we let you do that or not. ... Go!
'nuff said.
Some managers failed their basic economics classes and don't understand "economy of scale". You can do things in a large company that are not affordable for a small one. But anyone who thinks that giving up completely and throwing in the towel is an appropriate response doesn't deserve to be in a leadership position.
Seven puppies were harmed during the making of this post.
We have taken your suggestion sir or madam as the case may be, and have fired all of the managers. We feel so silly for not noticing their utter incompetence earlier. Thank you for opening our eyes to this horrific situation. We have chucked them all in a ditch and piddled on them.
But then you have companies such as Google and Microsoft building data centers next to rivers for cheap hydroelectric power in remote parts of the Pacific Northwest and reporting insanely low PUEs (below 1.1 in some cases).
Power Usage Efficiency has nothing to do with the source of the power you're using.
It's not even a measure of efficiency of equipment.
Please help metamoderate.
So they've been doing the stuff with the greatest return on investment.
What's left is the marginal improvements that probably cost more than they're worth.
Moving the whole datacenter to the Pacific Northwest just isn't in the cards for most companies.
I read that google did some experiments a while back and found that running the datacenter hotter saved more $$$ in cooling than the cost of the increased failure rate of hardware. That's fine for some computing workloads, but what are the obstacles to making computers that can run with an acceptable failure rate in an ambient temperature of (say) 50C (~120F)? I assume there are some major obstacles, i'm just curious as to what they are.
Even if you could run the solid state hardware at 50C and the disks in a separate storage room at 22C, that would still be a win right?
For instance, I suspect we waste more energy moving tap water in plastic bottles between cities.
"Well, people get shot all the time, so what's the big deal if I shoot someone?"
Doesn't work that way, does it? It sounds a bit like you're arguing a nirvana fallacy, namely that because this trend of saving energy in datacenters doesn't save energy everywhere, it's useless.
Please help metamoderate.
Why are managers worrying about meeting some arbitrary criteria set by Google/Microsoft/etc. for a metric that in the end doesn't matter? PUE is irrelevant, what matters to the business is the total cost of providing the computing power the business needs. If you have a cheap way of reducing that cost, take it. But if your cost's within acceptable limits and reducing it further's going to cost too much or take too much resources or investment, then stop wasting your time worrying about it and concentrate on other things you can improve. Like, say, improving the efficiency of your software so you don't need as much computing horsepower and bandwidth to do the job, which will automatically reduce both your power and cooling requirements right there.
The purpose of the data center is reliability. Higher reliability = less efficiency. People are considering PUE less important than downtime, and this is news?
Ok, I'm pretty sure this is a typo, as I can't figure out what any of this has to do with the definition, "To cut or shape wood with a knife" (wiktionary), but I'm at a bit of a loss to say what was actually intended... dwindling maybe? Still it doesn't seem an easy mistake to make. Although maybe if I'd RTFA I'd have found out this has to do with cooling down while shaving wood off a stick... Oh also wary instead of weary unless you're falling asleep while whittling.
lack of interest for improvement. To paraphrase the environmental legislation principle of using the best available technology, if the best commercially reasonably available technology is coal powered forced air cooling then that should be used. If there is that river, and the environmental impact analysis doesn't show significant damage to the river ecosystem then harnessing the river could be considered on the condition that the cost and availability of the heat transfer and electricity are at commercially competitive levels.
Blind PUE obedience is a sing of fanatic.
Of all the first world problems, green fatigue?
Bunch of pansies in IT need to get used to the idea of rationing.
Those who do not learn from commit history are doomed to regress it.
You can blame Facebook for much of this green datacenter hype--some of which is arguably greenwashing.
Facebook was under the gun for opening its own data centers that were, and still somewhat are, powered by electricity generated by coal.
To answer this unwanted attention they bent over backwards to reduce power consumption at all costs, so much that they even designed their own "Open Data Center" servers to reduce power consumption at the cost of discarding nearly everything we already know works fine in conventional data centers.
And, to top it off, they greenwashed by buying carbon credits and energy that appears to come from non-coal sources.
Google and Microsoft are doing this the right way. Data centers should be in cold climates and supplied by truly renewable power.
Kriston
There's talk of removing a few Dams and with them the cheap power.
The Washington state Indians have a treaty to fish salmon they way they used to (with nets)
that they then sale to make a living. The salmon are in decline which is blamed in part to the Dams. All of
the Dams have fish ladders that help the Salmon migrate but they are asking for the lower (last) four Snake river Dams to be removed.
http://www.americanrivers.org/initiatives/dams/projects/snake-dam-removal-economics.html
It's much more than just the Indians, but they seem to be the loudest.
From the link: ...Replace the dams' energy in an affordable and carbon neutral manner..."
"Before the dams are removed, there must be a plan in place to:
I don't see how that can be accomplished unless wind power can be considered carbon neutral.
They should compare BTU's of cooling to see efficiency.
Dollars just compares costs.
Cost of power can due to many factors.
She was like chocolate when she drank... semi-sweet at first and then increasingly bitter.
Maybe it's just a european thing, but if you have something called effectiveness, and it first has a bigger number assigned to it, and then a lower number, my interpretation of the situation would be that effectiveness has been going down, getting worse, etc.
The concept is simple. We've got car AC units that are 400% efficient, meaning for every one watt of power consumed, 4 watts of heat gets removed from the system it is cooling, within a certain size (the size of the interior of a Ford Explorer, for example.)
Then you make these into micro centers - insulated rooms, fully-sealed, holding no more than maybe 3 or 4 racks of servers. Have one or two of these cooling that room.
Hey, suddenly, you're spending $1 in electricity to cool off $4 of used power (and if you have really efficient computers in those racks, probably LESS!)
What's that PUE at that rate?
Still waiting on Serviscope_minor to wake up to fucking reality and realize that Jessica Price isn't going to fuck him.
A large capital cost that saves money in the long run is completely off the radar once it hits accountants that can't think beyond one quarter.
If you look at just about any large organisation you'll see plenty of examples of that.
Erase all the binary 0s. That will save a lot of space and much power.
Energy seems to have the same effect as money. Obviously a huge company can afford to buy the latest and best methods of being power efficient which can make smaller companies unable to compete. It is very much like trying to beat WallMart with better prices. Smaller companies do not have the leverage to buy as cheaply as Wallmart and therefore can not compete. The catch is that large companies tend to accumulate issues as the years pass and then fade away. Think of all the big chain department stores that have perished.
But the trouble with energy is that one way or another heat removed from a building is dumped into the environment. Efficiency helps but it only helps if the number of sources of heat drops as well. With a swelling population we will always get worse and worse. Yet the mantra of business is growth. That growth is our greatest enemy.
If your AC is 400% efficient, your PUE is 1.25, which is nice but not ground-breaking. And that is just for the AC, on top of that you waste power on transmission losses and UPS (if you run double-conversion, that is another 10% loss) and everything else a datacenter needs.
Also, 400% is mostly a marketing number. The efficiency depends on the temperature difference. If the outside air is cooler than the temperature you need, you can get infinitely high efficiency -- in theory you can even get electricity back out by using a Stirling engine (in practice that is not possible, the temperature differences are too small). On the other hand, if you are trying to deliver air to the servers at 22C and it is 40C and humid outside, you have a problem. Your AC will not deliver 400% efficiency in those conditions.
Finally! A year of moderation! Ready for 2019?
If you spend 1 watt on cooling for every 4 watt of heat (i.e. every 4 watt of computing), you have a PUE of 1.25. Not too bad, but far from state of the art.
The efficiency of the computers do not impact PUE, since PUE only looks at the power ratio between computing equipment and the rest of the data center (which is primarily cooling).
This idea not go anywhere because people realize the cost of dismantling the dams and replacing the power generated and agricultural water supplies would be EXTREMELY exorbitant. That's why all the talk of dismantling O'Shaughnessy Dam in Yosemite National Park has not resulted in any action, because the economic cost of dismantling the dam, raising Don Pedro Reservoir to replace it, and restoring the habit of Tuolumne Canyon behind the dam would cost US$25 BILLION.
Why don't they just site their centers up north? Here in Duluth, most of the year the outside air is cooled for free by mother nature. Heck, they could sell their waste heat to nearby homes and businesses and get a negative PUE.
Don't need to be green to worry about this, it's $$, something ever company wants.
No need to go "up north". eBay built a data centre in Phoenix, Arizona, where summertime air temperatures can hit 115F (46C):
http://www.wired.com/wiredenterprise/2012/02/ebay-desert-data-center/
http://www.google.com/search?q=ebay+Phoenix+data+center
Their water tank can hit 87F (30C), and even then they can operate their heat exchanges. I'm sure that it helps that Phoenix tends to be very dry, but there's a temperature-humidity trade off equation that can be formulated.
If it can work there, it can work probably work anywhere. It's just that people/companies don't want to put in the effort.
..using efficient Programming languages instead of using cheap Java and PHP "programmers". Hire software engineers with a CS degree and system design experience under their belts who will code in C++, Ada, Pascal, Sappeur and the like. Thereby your company will get a first-rate system that will run on very modest hardware and serve lots of users. Most Java and PHP programmers won't even know the rough details of hardware architecture like Cache, RAM, harddisk and how to efficiently use these resources. They have never thought about the fact that they waste 20 bytes per or more per number by using Integer instead of int. They won't know that efficient storage translates into higher cache hit ratios. They won't even know the waste from a Java object (reference) array as compared to a real object array in C++. Plus, these people are generally very shallow on theory and will create very inefficient algorithms, which is the worst sin of all in terms of energy usage.
Here's a language almost as efficient as C++, but also as memory-safe as Java:
http://sourceforge.net/p/sappeurcompiler/code-0/HEAD/tree/trunk/doc/SAPPEUR.pdf?force=True
http://sourceforge.net/p/sappeurcompiler/code-0/HEAD/tree/trunk/
Remember - the energy you haven't wasted in the first place does not need to be moved out of the datacenter !
Electricity costs money. Reducing the cooling costs of data centers isn't a green issue; it's a cost issue. TFA mentions this specifically:
so I find it odd that the take-away is "green fatigue."
How the fuck can you replace a dam with wind power ? Hint: It fluctuates at a ration of easily 10x from windy to windless days here in Germany.
Use this nice application and play with the date to see what I mean:
http://www.transparency.eex.com/en/Statutory%20Publication%20Requirements%20of%20the%20Transmission%20System%20Operators/Power%20generation/Actual%20wind%20power%20generation
WATCH THE FUCKING SCALE, though !
Actually, dams (and similar plants like pumped water storage) are the one and only feasible method of quickly smoothing out demand or supply fluctuations. You can power up easily 0MW to 1000MW from a large dam in a matter of 60 seconds. Gas plants need at least several minutes and coal and nuclear several hours to go from 0MW to 1000MW (mainly because they need to heat up massive amounts of water and metal).
Here in Germany we discuss using Norwegian hydropower to store and smooth out our "regenerative energy" supply and serve the much more predictable demand. Very difficult, very expensive, not yet done, though. We already use Swiss and Austrian dams massively, though.
Hydo power is the second line of defence in assuring a stable 50 Hz electricity net, the first one being the rotation energy of the turbines. And if your frequency goes to 49Hz, then have fun with the massive damage to the economy from burned motors, stopped production lines, jammed billion-dollar paper-making lines and so on.
"for every dollar of electricity to power a data center, $1.4 dollars were spent to cool it. That dropped to 1.8 recently, an improvement to be sure."
Sure, 1.4 dropped to 1.8. Progress.
If the rest of the article makes that much sense, I'm not wasting any more time. Typos are typos, but typoing the premise leaves me, well, feh.
deleting the extra space after periods so i can stay relevant, yeah.
And if it still doesn't come to the top of some company's balance sheet, it' perfectly fine. Chances are it will for heaviest electricity users and in the meantime taxes can be used for pro-encironment R&D.
As the sibling post mentions, AC units are heat pumps, not heat engines, so the ratio you use is not called, "efficiency," it's called, "power factor," and in the ideal case varies depending on the temperature difference. (the mathematical limits are +/- infinity, but obviously no practical device will operate at those extremes.)
Use of the term efficiency only really makes sense if you're comparing your practical heat pump to an ideal heat pump with the same temperature difference, and that number is always smaller than unity.
The odd thing about these conversations is the conventional wisdom that we should use active cooling on data centers at all. The actual components are always quite a bit higher than typical environmental temperatures, and for that all you really need is to have sufficient flow rate to take up the waste heat at the delta-t corresponding to the desired operating temperature, and this is the observation that Google uses in its modular data centers - put the hot thing outside.
This is same observation helps make backyard grilling popular in the summer.
How the fuck can you replace a dam with wind power ? Hint: It fluctuates at a ration of easily 10x from windy to windless days here in Germany.
In a point installation, the same size as the dam? You can't. But averaging the wind power from a geographically larger area (the whole of Germany might just be large enough, but barely, I guess) ought to smooth thing up a little bit...that is, ONCE the distribution grids are smart enough to handle that. Right now, I don't think the infrastructure is ready for it. Also, you need a good predictive model to anticipate the changes. Again, something yet to be done. Oh, and there's this silly stuff called "national borders" to complicate the building of the smart grid. Oh, hell...
Ezekiel 23:20
People, this is 2013 and this is pretty much off-the-shelf tech. Building the system (an HPC) in an air handler was cheaper than retrofitting the existing data center. We have a 24k core Intel box sitting on a slab. Running in 8 months from Request for Proposal. 95% of the cores in use on Friday, we were running 1.04 PUE at ~85 degrees F ambient. While this is a largish system, they are readily available smaller and 2x larger.
Your post is full of CONJECTURE. I do think all we know at this point in time says that "wind and solar power will fluctuate at a 10x ratio". Germany is a country of 80 million people and I seriously doubt the numbers for all of Europe (400 million) would be dramatically different.
All the "regenerative" energy ideology is long on talking and propaganda and very short on substance. Electricity has already become un-affordable to many poorer segments of the German population. Network stability has been threatened multiple times during the last 20 months because of the fluctuating supply. It has been proven correct that for each 1MW of "renewable" generating capacity, you also need 1MW of "conventional" capacity.
Yeah, Romanticism is excessively expensive, especially when you believe in Romanticism as hard as Germans do. But fuck, life would be boring without militaristic emperors, Austrian chancellors, fucking expensive, badly managed re-unifications, destruction of our treasure of nuclear power, "renewable" energy and a fucking excessively stupid "common currency".
And before you say "it's not so bad" - have a look at fertility figures and you can see Romanticism is destroying the German people themselves. All of these fucking stupid romantic projects have depleted the financial resources of the average German man and woman to the point they think they cannot afford kids anymore. You earn something like 10 Euros/hour in the "second tier" companies which supply parts to the major auto companies. And from those 10 Euros they manage to remove at 3 Euros/hour for all sorts of socialist and romantic buttfuck ideas. Plus, they destroy the currency in the name of "European pacification by single currency". That "effort to peace" is reflected by 50% young unemployed in Greece and nasty "Nazi" propaganda by those Greek nothing-doers-cum-subsidy-thiedes-cum-statistics-forgers.
As a one-time member of The Green Grid Technical Committee, let me summarize and correct a few points:
Your post is full of CONJECTURE.
Actually, no. In, that case, my post would read "CONJECTURECONJECTURECONJECTURECONJECTURE..." etc., which it doesn't.
I do think all we know at this point in time says that "wind and solar power will fluctuate at a 10x ratio". Germany is a country of 80 million people and I seriously doubt the numbers for all of Europe (400 million) would be dramatically different.
It's not a matter of number of inhabitants (you still living in the 1950's? I mean, Europe has something like 7.5e8 people living in it at the moment), it's a matter of the size of the continent and the prevailing meteorological and climatic patterns, the size of the weather systems travelling over the continent, and simple statistics.
Network stability has been threatened multiple times during the last 20 months because of the fluctuating supply.
I think you missed the "we need a better grid" part of my post. You also missed the part where I observe that the wind and solar contributions to the grid can be anticipated (and acted upon in advance) if you have real time meteo sat coverage (as in, you can predict the insolation and winds). To anticipate, BTW, if you don't understand Latin, means "not letting anyone catch you with your pants down". I don't believe anyone is doing that kind of data feedback right now, therefore problems - how difficult is that to comprehend?
Ezekiel 23:20