Iceland Woos Data Centers As Power Costs Soar
call-me-kenneth writes "Business Week covers the soaring demand for power and cooling capacity in data centers. Electricity consumption for US data centers more than doubled between 2000 and 2006. Among the other stats: for every dollar spent on computing equipment in data centers, an additional half dollar is spent each year to power and cool them; and half the electricity used goes for cooling. Iceland, with its cool climate and abundant cheap power, is courting big users like Google and Microsoft as a future data center location. (Can't help thinking they're gonna need a bigger cable first, though.)"
Though there is a slightly bigger version here, ostensibly for desktop background usage.
What you really need is a map showing bandwidth. There was one in a recent (paper) edition of The Guardian. The online version (it's in the bottom right) is a bit too small to be very useful, but it's big enough to see that yes, Iceland needs a better connection is it is to become the world's data centre.
"Be light, stinging, insolent and melancholy"
Five good reasons:
1) cheap geothermal power
2) cheap geothermal cooling
3) easy freight
4) educated and even DNA-tracked populace
5) computing is an indoor sport
Five considerations:
1) they like to go whaling; not necessarily a friendly thing in by some opinions
2) latency; not as a bad as a sat, but not as good as Chicago for US; geo centric for North America and EU
3) earthquakes and unsettled geography
4) too many thermal pools to screw off in
5) don't want my server called 'homerdottir'
---- Teach Peace. It's Cheaper Than War.
I would suggest locating data centers in a cool climate where farming is popular. Pump the waste heat from the data centers into greenhouses that can surround the data center. Now that waste is helping to grow food.
Alaska is actually a good place to implement such a solution. There is a huge amount of sunlight in the summer which, assuming you can avoid frosts, can grow amazing produce. All you need are greenhouses and a heat source. In the winter, when sunlight is no longer plentiful and farming shuts down, the heat can be pumped into local housing. Such a solution would also provide local produce in Alaska - produce that is fresh and doesn't require expensive shipping. One last point about Alaska, it's very central. It might not appear to be when looking at a map, but if you look at a globe you will see that it sits nicely between Asia and North America. I don't know where the current internet pipes are located but if they pass close to Alaska then this idea would be worth some consideration.
William
You'd have to have a very cheap and very power inefficient server to come even remotely close to their claims of half of the cost of the server on power. An elbaso HP Dl360G5 costs $1600. It will use about 300W at typical load, but lets call it 250W to make the numbers easier. Double this for inefficient cooling and power conversion in the UPS (this is overly costly but makes up for underestimating power usage) so 500W. There are 8,760 hours in a year so 4,380 KwH, you'd have to pay $.20 per KwH to reach their figure, this is over twice the US national average. Prices where you'd want to put a datacenter are closer to $.06-$.08 per KwH. My average server cost closer to $7,000 with battery backed RAID card, dual fast drives, dual CPU's, 4GB memory, 3 year 6 hour repair contract, etc. Even powering that kind of servers off diesel generators fulltime it would have to draw ridiculous amounts of power to cost half it's purchase price in electricity every year.
There are 4 boxes to use in the defense of liberty: soap, ballot, jury, ammo. Use in that order. Starting now.
I'm not saying your typical data center is going to put out the same heat as a nuclear reactor. They actually take steps to cool the water, but it's still warmer than it went in. I certainly don't think heat pollution will be a concern. Gigawatt size coal plants and nuclear reactors have to worry about heating nearby rivers, but in this case we are talking about the additional load to run datacenters heating the Atlantic Ocean. A 1 GW electric power plant will typically reject 2 GW thermally (which is a lot of local heat). But 3 GW is a lot of power to be generated in one spot. The renewable energy sources of Iceland certainly don't generate anywhere near that much power per unit area. And the heat rejected from datacenters will be trivial and widely distributed.
I think two things will stop these datacenters from going to Iceland: restrictive immigration laws and submarine data cable capacity. Iceland has a total population of about 300,000. They simply can't have a diverse enough IT industry to support setting up these data centers without expats. And without the bandwidth, there simply isn't a point.
While I am no HVAC engineer, I pretend to be one on odd-numbered days.
Cold climates have several real challenges for data centers. From an HVAC standpoint, there are two general ways to cool a data center in a cold climate-- outside air only and air/water cooling. Air/water systems have drycoolers with glycol kept around 30-40F, and circulate the cold water throughout the building to fan coil units. Minimal outside air is brought in for "fresh air," and must be humidified which generally requires a lot of energy.
The air-only systems bring in 100% outside air, but must first temper (heat-- to avoid condensation) it and increase the humidity to control static inside the space. Very little pump energy, but the humidification and pre-heat are expensive.
While it seems trivial to filter out dust, the better air filtration systems increase the pressure drop of the air handling unit, and force you to use a bigger fan. Heat wheels and enthalpy wheels are also an option, but have similar challenges in most real-world situations.
The biggest challenge with cold climates is making sure the diesel generators start when needed. This alone makes most data center managers skeptical at the prospects of cold-climate data centers.
For a truly efficient solution, the best approach is likely to be heat removal at the chip level and recovery for other purposes. 100F air isn't very useful, but if you can get 150F water off the chip then that heat can often be reclaimed for some other purpose more effectively. If all else fails, 150F water is pretty easy to cool off in a closed circuit dry cooler no matter what the outside temperature.
There is also a lot of work going into direct-evaporative cooling solutions (swamp coolers) for data centers, as well as some other non-compressor based cooling systems. Unfortunately, most of these can work very efficiently for 9-10 months a year, and need a separate system to cool for the remainder of the year. Having two systems makes the payback equation often favor the less efficient solution...