Slashdot Log In
IBM Pushing Water-Cooled Servers, Meeting Resistance
Posted by
kdawson
on Tue May 19, 2009 01:18 AM
from the let's-compromise-on-water-vapor dept.
from the let's-compromise-on-water-vapor dept.
judgecorp writes "IBM has said that water-cooled servers could become the norm in ten years. The company has lately been promoting wider user of the forty-year-old mainframe technology (e.g., here's a piece from April 2008), which allows faster clock speeds and higher processing power. But IBM now says water cooling is greener and more efficient, because it delivers waste heat in a form that's easier to re-use. They estimate that water can be up to 4,000 times more effective in cooling computer systems than air. However, most new data center designs tend to take the opposite approach, running warmer, and using free-air cooling to expend less energy in the first place. For instance, Dutch engineer Imtech sees no need for water cooling in its new multi-story approach which reduces piping and saves waste."
Related Stories
This discussion has been archived.
No new comments can be posted.
The Fine Print: The following comments are owned by whoever posted them. We are not responsible for them in any way.
Full
Abbreviated
Hidden
Loading... please wait.
Any prediction over ten years is null and void (Score:4, Insightful)
Hardware and software faces change so fast; who has any idea what will be required or available in even ten years?
Re:Any prediction over ten years is null and void (Score:4, Informative)
I doubt Bill Gates ever said that. He's claimed the contrary on several occasions:
http://www.usnews.com/usnews/biztech/gatesivu.htm [usnews.com]
http://www.wired.com/politics/law/news/1997/01/1484 [wired.com]
But yes, making predictions for the future is dumb. Unless you control the future, in which case it's not really a prediction *cough* Moore's Law *cough*
Parent
Re: (Score:3, Insightful)
Personally, I like the idea. (Score:3, Interesting)
look at some of the newer blades or blade-ish solutions from supermicro, with 4 dual-socket boards in 2u. Dang near half the space is now taken up by ram slots. Suppose they switched to sodimms packed in like heatsink fins on one of those boards, and you could possibly cram in 2 more sockets with waterblocks.
Similar re-arrangement with blade boards would likely also be possible.
In a dedicated datacenter, I can't think of any real great money saving solutions for the waste heat, but it WOULD allow you to more easily cool everything with a large ground loop to get 50-55 degree water. Add a few more loops so you can melt the snow off the parking lot in the winter and bleed off heat there. Add a large tank of water and radiators to take advantage of cool nights to pre-cool the water before the chillers.
Only chill the water with the chillers when needed. Seems to make a lot of sense to me.
In smaller serverrooms in large office buildings, pre-heat the hot water, pipe the hot water to help heat the building in the winter, and depending on location and if its a mixed use building, you MIGHT be able to sell some of the heat to other building tenants.
Personally, if I was building a new house, I'd have ground loop heat pump for cooling, heating, put a decent sized water tank on the top floor/attic that I could use to preheat hot water in the winter (also be good to hook into for solar hot water on the roof) and a water tank in the basement/crawl space as a source for cooling. Add some electronics to determine where to draw water source, and where to push water return for different devices depending on temperatures of each given tank, as well as when to run ground source heat pump or outside radiator and I think I could cut heating/cooling costs by a huge margin.
Now if only we had a good way to pipe the light from all the blinking LED's to where its needed to remove the ugly florescent lighting. That or get everyone to work by the glow of their CRT/LCD
Resistance (Score:5, Funny)
Re: (Score:2)
Is it bad that I actually winced when I read that?
Re:Resistance (Score:4, Funny)
That's because salt isn't the solution, it's the precipitate.
_Ouch_. Please stop hitting me.
Parent
Not for all (Score:5, Interesting)
I worked in several banks using IBM mainframes. The server room was always like a freezer.
I think for now, many companies are perfectly ok with air cooling solutions. Besides, it's much safer to have air-conditioning and fans than some liquid flowing. The simpler the system, the less accidents occur within it...
And believe me when I say that, if a company owns an IBM mainframe, they pay big bucks and they *don't* want any accidents.
Re: (Score:2)
Me too. I didn't have to wear an aqualung and enter through an airlock, so they can't have been water cooled.
Better steer clear of them there new-fangled 3090 models, then. It's a fad, I tell you.
Re:Not for all (Score:4, Insightful)
If some companies can make fridges that do not leak coolant. I'm pretty sure IBM can make mainframes that do not leak their coolant either.
Parent
Air is not necessarily simpler (Score:5, Interesting)
This is why nowadays virtually all internal combustion engines of any power output use liquid cooling despite the apparent reliability benefits of air cooling. To take the transition period, WW2, as an example, you only have to look at the complexity of American rotary aircooled designs versus, say, the liquid cooled Merlin engine, to see the point. It would be astonishing if the same transition did not eventually occur for large computers.
Parent
Re:Air is not necessarily simpler (Score:4, Informative)
I think you mean radial engines [wikipedia.org], because rotary engines [wikipedia.org] may look similar when not running but are an entirely different thing.
Air cooled engines are still used in small planes, their weight to power ratio is better than in water cooled engines. In larger aircraft both water and air cooled engines were replaced by turbines.
Also, air cooled engines are still widely used in motorcycles. I think the main motive for not using them in cars anymore is due mostly to the difficulty in cooling in an enclosed region, have you seen how cramped is a modern car under the hood?
The main advantage of water cooling is that it's easier to carry the heat away to some place where it can be either reused for some other purpose or dumped to the environment. With air cooling you have to bring a substantial amount of cool air to where the heat is being generated.
However I still think computers are mostly in the range where it's easier to bring the air in. The amount of heat dissipated per volume of equipment is not so great that the additional complexity of water cooling would be justified.
Parent
Re: (Score:3, Interesting)
Actually, you are wrong. Designing an air cooled system is hard. You have to deal with problems of filtration (there will be dust - but where do you want it to build up?), ensuring that the flow goes where you want, turbulence, finding room for the ducting, designing the system so that components do not mask other components, and needing to handle high volumes of air. With properly designed water cooling, you have a few quite simple heat removal blocks and a simple plumbing system which can route pretty much anywhere.
The reality is that there are pros and cons to water cooling vs. air cooling and people have to weigh both and decide what works for them. While your post is essentially correct, it's also a bit heavy on the theoretical. I remember working at a place that had water cooled IBM mainframes. This was a US government facility and it was in a building I almost never had to go to. I remember downtimes there because "the water chiller is down" or there was a leak and a hellacious mess of water was under the flo
Takes me back... (Score:3, Informative)
I remember that IBM had an office building in Manassas, VA that was heated by the mainframes in the building. They got a lot of press at the time for that.
-jcr
Great working conditions... (Score:5, Funny)
Random related fact (Score:4, Interesting)
Parent
4000 times? (Score:2)
Re:4000 times? (Score:5, Funny)
Where do people always get these kinds of numbers.
this is a situation where a link to goatse would actually answer your question.
Parent
Re:4000 times? (Score:5, Insightful)
Now see how long you can stay in water at 5 degrees centigrade. For most people it would be less than a minute - you may not even be able to get in.
Parent
Re:4000 times? (Score:5, Interesting)
Parent
Re: (Score:2)
http://en.wikipedia.org/wiki/Specific_heat_capacity#Table_of_specific_heat_capacities [wikipedia.org]
Water: 4.186 J/cm^3/K
Air: 0.001297 J/cm^3/K
3227 is closer to 3000 than 4000 I guess.. but at least he got the right orders of magnitude
Ratio of specific heat capacities (Score:5, Informative)
All IBM is saying is that water is a better heat conductor, and air is an insulator.
http://en.wikipedia.org/wiki/Specific_heat_capacity [wikipedia.org]
Water ; 4 J /cm^3 K /cm^3 K
Air ; 0.001 J
Water/Air = 4000 times more heat transfer.
So, given the choice, you would use water to transfer heat.
Re: (Score:3, Insightful)
Re: (Score:2)
then datacenters simply need nuclear plant style evaporative cooling towers.
Re: (Score:2)
Re:Ratio of specific heat capacities (Score:5, Informative)
My personal experience with using passive (no fans) water-cooling with my desktop PC at home (the setup is similar to this: http://www.silent.se/bilder/reserator1_c_p-410.jpg [silent.se]) is that that it's exceptionally effective.
In my setup a cylinder full of water surrounded by fins to dissipate the heat and with a pump to make water flow as the only active element have replaced a big nasty CPU heatsink with a large fan (on a heavily overclocked CPU)* and a set of fans on a single high-end graphics card of the previous generation. At an ambient temperature in the room where this is in of about 20-25C The whole thing idles at 28C and stays at around 60C with everything going on at max - considering that with everything going on at full throttle the system is using almost 400W, it's impressive how efficient it all is.
In practice, "home" water-cooling mostly just uses the water as a heat carrier to quickly move the heat around from the inside of the computer case (and it's constrained airflow) to a place where it is easier to dissipate that heat into the ambient air either with a more efficient radiator and fans (for the active systems) or with an outsized heatsink (like the one I use which has roughly 10 times the surface of the ones it replaces).
In an "industrial" deployment, said heat being carried in the water cold potentially be used/dissipated in many more ways. For example large pipes could transport the hot water coming out of a data-center to the sea or a river and let it be dissipated there (keeping a closed circuit and returning the cool water back for reuse). The actual running costs in terms of active elements for such a system are limited to the cost of running a number of large efficient water pumps that make the water flow around the circuite as opposed to most data-centers out there at the moment that use (less efficient) small fans to move the air out of the blade boxes into the room and then active refrigeration to cool down the air in the room.
* Since the point of my argument is not to show off my "virtual dick", I've moved the relevant stats down here for those that are curious on the details: CPU - Core 2 Quad 2.4 GHz which is overclocked to 3.2GHz, GPU - GTS280
Parent
Re: (Score:2)
multi-story approach (Score:2)
Water cooling should not be a sop to consumption (Score:2, Insightful)
This reminds me of recycling schemes that make people think it is OK to overpackage goods in the first place.
Central Watercooling. (Score:2)
I expect integrated power-and-watercooling sockets all over the house. The forced flow, the air conditioning unit integrated with water cooling facility, pump and reservoir, also using the heated water as heat source for heat pump.
Re: (Score:2, Funny)
Is that a pun?
Re: (Score:2)
Good idea! And the server farm can come ready-equipped with a camera and web server to show the status of the coffee maker.
Re:What a waste of water! (Score:5, Interesting)
The community in which a server farms is found surely has a need for what will be thousands of gallons a day. To the benefit of all, I'd suggest diverting a small amount of the heated water (hopefully near boiling) to another piping system in the building ....
I'm sure it could be designed as a closed system with a heat exchange into the ground or outdoors. Indeed, it is the high temperature (relative to outdoors) at which the water is extracted straight off the CPU which makes this more efficient than air conditioning.
However if you wanted to let it feed into the building's hot water system, it turns out there is already a really elegant way to do that: a tempering valve. It's a mechnical device which chooses the right amount of hot and cold water (each of arbitrary, variable temperatures) to produce some fixed output temperature. So to make moderately hot water you can combine some warm water from the servers and some super hot water from the boiler. The "free" server heat offsets the amount of water that needs to be heated by conventional means.
Parent
Re: (Score:2)
That's still quite a bit of hot water. I'm sure very little would be used in an office environment. Residential would be more like it if possible.
Perhaps it would be better to engineer the hardware to run warm-hot. Having to chill the water down to outside ambient temp (no compressors needed) would save a lot of energy and cost.
Re: (Score:2)
That's still quite a bit of hot water. I'm sure very little would be used in an office environment. Residential would be more like it if possible.
Perhaps it would be better to engineer the hardware to run warm-hot. Having to chill the water down to outside ambient temp (no compressors needed) would save a lot of energy and cost.
Maybe google could get an environmental initiative grant to provide a staff swimming pool.
Re: (Score:2, Interesting)
Re: (Score:2, Insightful)
Greenpeace is no better than PETA.
I beg to differ. Greenpeace is sleazy, but PETA is a nut-cult.
-jcr
Re: (Score:2, Flamebait)
Greenpeace is no better than PETA.
I beg to differ. Greenpeace is sleazy, but PETA is a nut-cult.
-jcr
Great pun - if you've ever been to a PETA buffer you will know that there are plenty of nuts around
Re: (Score:3, Funny)
I agree. Any alternative interpretation - such as that it recirculates the water - is clearly preposterous.
Re: (Score:2)
You're not going to get water hot enough to make coffee unless you run the servers at 90C+. There's that whole second law of thermodynamics thing. You also need a temperature gradient to drive heat transfer according to the basic heat exchanger design equation:
Heat transfer=Heat transfer coefficient*area*mean temperature difference
(You could use Fourier's law [wikipedia.org] if you prefer).
A better use would be a district heating across the building the server farm is in. How well that works depends a lot on where you are. But this scheme doesn't inherently waste either water or energy. It al
Re: (Score:2, Insightful)
This is IBM... Do you _really_ think they'd design it in such a way that you'd have to take down the whole thing to fix a small section?
You wouldn't have one long pipe running to all of them, with no way to shut off segments/individual nodes.
Re: (Score:2)
This is IBM... Do you _really_ think they'd design it in such a way that you'd have to take down the whole thing to fix a small section?
You wouldn't have one long pipe running to all of them, with no way to shut off segments/individual nodes.
No, but they might assume that the world will never need more than one hot water system.
Re: (Score:3, Insightful)
Re: (Score:3, Informative)
Some tanks have air conditioning.
Air conditioning the whole tank does not make sense because once you fire the cannon a few times the whole place is very hot.
What they do is have a hose that hooks up to the special overall tankers wear and supplies you with cool air where you need it most.
The hose connector is at the center of the suit.
Re: (Score:3, Informative)
You can recycle your water
I'm not exactly sure what you mean by that, but either way you can't.
In case you mean replacing the water: You're doing it wrong! ^^ The more often you replace the water, the more likely it is for your coolers to get some mineral crystals to grow. In two years, your CPU cooler might suddenly burst, killing your whole electronics.
This even happens with distilled water, because there is no 100% in nature.
There are even galleries out there of such bursts, including images of huge crystals inside the coolers.
So
Re:I watercooled my server years ago!!! (Score:5, Informative)
Try to keep up IBM.
IBM was water-cooling machines at least as early as the 1970s.
-jcr
Parent
Re: (Score:3, Informative)
True, and everybody knows that.
Or so I thought. Yet this assclown [slashdot.org], who clearly doesn't know the difference between an IBM and a BMW, is at +5 insightful.
Re:IBM needs a physics lesson (Score:4, Interesting)
We could save a lot of money by putting all that light-bulb heat to use.. to bad entropy makes these sorts of schemes uneconomical.
Entropy doesn't make those situations uneconomical. It makes them quite literally impossible.
However, define "use". Remember that you can use heat for more than electricity generation (made impossible in this case). You can't get net energy out of it, thermodynamically speaking. Entropy reigns. But you can still use it as heat.
Hypothetical example: you build a cooling system for a server (water, air, it doesn't matter). You now have a radiator giving off waste heat. That waste heat can be used for some other, non power-generating purpose. What matters is that the heat is carried away from the radiator at a constant rate, or the cooling system will have to work harder to get the same result.
You could use a water tank as a heat sink, then use the heated water for the usual purposes (washing, cooking, what have you). As the hot water heater/heat sink is drained, cool water is pumped in to replace it, allowing the radiator to continue functioning. In this instance the heat is used directly, as heat.
The reason this doesn't run up against thermodynamics is that the hypothetical second use of the heat replaces an existing system you'd have to generate heat for (a conventional hot water heater). The system is still entropic and inefficient, it's just less inefficient than generating the same heat energy a second time. You go from the net waste being X% to the waste being X%. Whether this is worth the bother is a question of the circumstances.
Parent
Re: (Score:3, Interesting)
In case you didn't know: Water cooling must be in a closed loop. You should not ever need to replace that water. If you do, you can destroy your coolers, because growing crystals will burst them. I have seen pictures of that.
Re:When I come to office today (Score:4, Informative)
Problems with crystals comes with some types of water where there are a high degree of lime in it. While its simpler to just use heat exchangers you could also use waterfilters that separates the minerals from the water before use. Most places have water with low amounts of lime and minerals so deposits arent really a problem.
I had a company that made solar panels (heating houses) and inverters for house warming. In some cases we took ground water and extracted heat directly from it and when taken apart those heat exchangers very rarely showed any deposits at all even after ten years of use.
The easiest way to see what type of water you have is to look in your toilet and your sink. If there are much deposits there (not brown ones) you have water thats high with lime or other minerals.
Parent