Google's Chiller-Less Data Center 132
Posted
by
samzenpus
from the cooler-than-cool dept.
from the cooler-than-cool dept.
1sockchuck writes "Google has begun operating a data center in Belgium that has no chillers to support its cooling systems, which will improve energy efficiency but make weather forecasting a larger factor in its network management. With power use climbing, many data centers are using free cooling to reduce their reliance on power-hungry chillers. By foregoing chillers entirely, Google will need to reroute workloads if the weather in Belgium gets too warm. The facility also has its own water treatment plant so it doesn't need to use potable water from a local utility."
Unreliable... (Score:5, Interesting)
Worth the tradeoff? (Score:4, Interesting)
Re:Unreliable... (Score:5, Interesting)
Running on the dark side of the planet (Score:1, Interesting)
I wonder if it would be feasible to have massive passive cooling (heat sinks, fans, exhausts from the data center, etc.) and run the data centers which are currently at night (i.e. on the dark side of the planet.) and constantly rotate the workload around the planet, to keep the hotest centers in the coolest part of the planet. The same logic could be applied moving workloads between the northern and southern hemispheres.
Yes, there would be tons more telecommunication to do, with the impacts on performance, data transmission costs and extra heat required to run all those routers at 120%, but there are fewer routers than servers, no?
Re:Worth the tradeoff? (Score:3, Interesting)
It's probably not as much about the energy bill as it is about the PR.
If it wasn't PR, they'd have chillers 'just in case', even if turned off most of the time. As it stands, they may be subject to a large risk of month-long heat waves killing them on paying idle employees, taxes, and taking a hit on capital depreciation costs for zero productive output that they are presumably banking on by bothering to build another datacenter.
Of course, there may be something unique about the site/strategy that makes this threat near zero that I'm unaware of, but I've seen facilities that are largely cooled by climate pretty far north that still keep chillers on hand in the event of uncooperative weather.
Re:Global warming (Score:4, Interesting)
This might be a dumb question (Score:3, Interesting)
But if your data center is in say, Minnesota, it seems like you could balance the temperature with outside air for many months out of the year. Obviously you'd need to light up the chillers in the summer, but running them 4 months out of the year seems like a huge energy savings than running them year round.
I remember visting Superior in the summer and the lake water was freezing f'ing cold even in June. Wonder if you could run a closed loop heat exchanger without screwing up the lake environment?
More interesting than we think... (Score:4, Interesting)
Good to see. (Score:3, Interesting)
Energy tradeoff of treating own water (Score:3, Interesting)
I'm not sure I understand why they constructed their own water treatment plant. I would think that it would be more energy efficient on the whole to use the already constructed municipal system in the area.
Re:This might be a dumb question (Score:5, Interesting)
The short answer is yes --- water takes a staggering amount of energy to change temperature (it's one of the many properties the stuff's got that's really weird). A big lake makes an ideal dumping ground for waste heat. What's more, the environmental impact is going to be minimal: even the biggest data centre isn't going to produce enough waste energy to have much effect.
(A big data center consumes about 5MW of power. The specific heat capacity of water is about 4kJ/kg.K, which means that it takes 4kJ to raise the temperature on one kilogram of water by one kelvin. Assuming all that gets dumped into the lake as heat, that means you're raising the temperature of about 1000 litres per second by one kelvin. A small lake, say 1km x 1km x 10m, contains 10000000000 litres! So you're going to need to run your data centre for ten million seconds, or about 110 days, to raise the temperature by one measly degree. And that's ignoring the cooling off the surface, which would vastly overpower any amount of heat you could put into it.)
(The same applies in reverse. You can extract practically unlimited amounts of heat from water. Got running water in your property? Go look into heat pumps.)
In fact, if you were dumping waste heat into a lake, it would make sense to try and concentrate the heat to produce hotspots. You would then use this for things like fish farming. Warm water's always useful.
Re:Unreliable... (Score:3, Interesting)
Re:This might be a dumb question (Score:4, Interesting)
I don't know about natural lakes but man made ponds have been used for just that purpose.
Man made ponds are used because the EPA crawls up your ass if you want to use a natural body of water for any commercial/industrial output.
Note: I'm saying that's a bad thing. I'm glad the "good old days," when chemicals, raw sewage, and cooling water were dumped willy nilly into the waterways and drinking supply, are gone. You warm up an area of water 10 or 15 degrees farenheit and you'll kill most everything living in it but algae.
Re:Unreliable... (Score:4, Interesting)
I am kind of confused about how many kW of power it takes to transfer heat. Heat moves from high to low, you have to pump cold water through a coil and force warm air across that coil. The amount of heat transferred is a function of the face velocity and temperature of the air across that coil, the amount of fluid moved and temperature through the coil and the characteristics (fin spacing, fin size, material) of the coil.
The temperature of the computers isn't really the important factor, it is the heat rejected. Again using rules of thumb, you can assume that 80% of the electrical power delivered to the computers will be dissipated as heat. The total of that heat rejected along with the other heat inputs to the space, e.g. lighting, walls, roof, window loads, etc., will determine your cooling load. Almost all of this load is sensible, meaning heat only, for other occupancy types you would also have to consider latent (moisture) loads as far as people and ventilation air in determining the amount of cooling needed.
Re:Worth the tradeoff? (Score:3, Interesting)
Yes , but not because of the bill itself.
Google has been actively developing a reputation in the corporate world for squeezing the most CPU-bang out of a buck, and a great way to do that is by cutting down on the amount of power a CPU uses.
A few weeks back there was an article on Slashdot which discussed a before-unseen Google innovation concerning its servers - a 12 volt battery that cut the need for an APC (which lowered costs by lowering both the power flowing to the CPU and the power required to cool the APC).
Google is trying to cut power out of the equation here as well, but with a different spin. Google is attempting to see if it can design a data center that does not require a cooling system that can perform satisfyingly within operating temperature range in a temperate climate - without any direct physical intervention (except by software algorithms). The implications are huge.
Re:Unreliable... (Score:5, Interesting)
The 81F water can be produced directly from open cooling towers (direct evaporation) whenever the wetbulb is lower than 76F (4 degree approach plus a 1F on your flat plate that isolates the datacenter loop from the open tower loop).
You designed an efficient datacenter, but you're five years behind cutting edge (not actually a bad thing for most critical environment clients). The next wave of datacenters will have PUEs of 1.2 or less and redefine the space from a noisy but cool space to hang out to a hot machine room with industrial heat exhaust design.
I actually just finished a chiller less 8MW schematic design and analysis for a bid. It was my second this month (the first was a cake walk - an extreme Twb of 67F, the second was west coast light conditions).
PS: Secondary pumps? Seriously? Unless you have to boost up to 25 psi to feed a Cray or some other HPC I thought everyone who cared had moved onto variable primary-only pumping. (Sorry, feeling a bit snarky after hitting a 40 hour week on Weds...)
Re:This might be a dumb question (Score:3, Interesting)
I blame that bastard Carnot personally for this... They could get additional work out of that hot water, but it gets prohibitively expensive the lower your delta T between hot and cold gets. I was all stoked about finding some sort of stirling heat engine to run off some datacenter waste heat, until I worked the numbers and found the annual average maximum therorectical efficiency was under 15%.
F*cking entropy.
Re:This might be a dumb question (Score:4, Interesting)
At those latitudes the ambient subterranean temperature remains pretty ambient all year long. Drill into the side of a mountain or hill with a boring tool, leave the edges rough (with a smooth poured/paved floor for access) and just drop your server containers in there with power coming in. If you go all the way through the hill you can use the natural air currents to push/pull air through the tunnels, and the natural heat absorption qualities of stone will keep the temperature down. I'd be surprised if any active "cooling" were needed at all.
Re:Yakhchal (Score:3, Interesting)
This reminds me of a technique for cooling water in a desert which could tenably be applied to the data center as well.
Basically, a container is filled with water, closed/sealed, and wrapped with a damp/wet towel and buried in the ground (or just placed somewhere in the sun, I suppose). The evaporation of the moisture in the rag will draw the heat from the inside of the container, resulting in frigid water.
Put a data center on a dry coastal equatorial area and harness solar to desalinate the water. Build the data center under ground, with the roof of the center allowing easy flow of heat upwards, and then plant edible vegetation on top of the roof. Water the roof consistently to cool your data center during the day (and harvest the proceeds to sell/consume).
It may or may not be worth it financially, but it'd probably work.
Re:Unreliable...Probably not (Score:5, Interesting)
Just being lobby-savvy... (Score:3, Interesting)
Re:Yakhchal (Score:4, Interesting)
There is a way to make ice in a dry environment by exposing water to the coolness of the night sky and insulating it during the day.
Re:Global warming (Score:1, Interesting)
max temp in the center of Belgium ever recorded was 38ÂC (and that's like 50 years ago). a very hot summer is a summer with a 15 day period with max temps above 30Âc.
(in the northern part of Belgium, you can get up to 40+ ÂC, but the Google datacenter is further down south.)
ÂC = degrees celsius (seems like it's not getting through properly for whatever reason)