Water, sunshine or pig s***: How will YOU power your data centre?
Top tips for anyone building their first server empire
So you've decided to clean up in Big Data. You're going to have the server farms of all server farms, run the cloud for the world's megacorporations and of course there's this bloke in Silicon Roundabout willing to give you the money. What's your first decision then?
It should be to go and steal an idea from another industry actually. Specifically the aluminium one. For it has faced, for many decades now, the same basic problem that you've got. The price of electricity.
You need it to keep the machines on, of course. You need it to be reliable, of course. And you also need to be able to cool the room after your machines have turned much of the generated energy into heat. You also need these vast amounts of power to be cheap: for that electricity is going to be one of the major running costs, and thus profit determinants, of your soon-to-be globe-conquering business.
Manufacturing aluminium works roughly as follows: mine bauxite (an aluminium ore), boil that in caustic soda to make alumina (aluminium oxide) then stick that in an electrolytic pot to covert the Al2O3 to Al. Then sell it to the people who make Coke cans or something.
The value chain is that bauxite's in the $50-$100 a tonne range, while the selling price of alumina is perhaps $300 to $400 at present. The aluminium price is just under $2,000/t. The problem? In addition to the 2 tonnes of alumina you'll need, that electrolytic pot is going to swallow $900 of electricity for every tonne of Al you produce. This is why we recycle those Coke cans, not for the aluminium, but to recycle that embedded energy.
Your data centre is like an ore factory
At which point the similarity with the data centre business should be apparent. One of your biggest costs is to power the machines and then dump the resultant heat. Your crucial and variable cost is electricity. So, why not nick the business methods from people who have been facing the same problem for those decades?
That method is that first, you decide where the power is going to be cheap. Then, you build your power plant. And only then do you start making the metal. What you don't do is worry about where you source your alumina - just drag it in from wherever you may find it, because transporting that around is still a hell of a lot cheaper than using higher-cost power.
All this does lead to some odd-looking strategies in the Al industry to be honest. There are vast bauxite-to-alumina operations on both tropical coasts of the Atlantic: yet the aluminium operations might be in Siberia or Quebec. Or even Iceland. For up there are some girt big rivers which have been dammed. Additionally, there are fewer people around to complain about the reservoirs, consistent water flows and all that. Indeed, many of the dams were built specifically to power aluminium operations.
There are smaller examples as well: Wylfa, the nuclear station on Anglesey, had an aluminium plant as its main local customer. They were deliberately entwined as a matter of policy: someone to take baseload power when everyone's asleep and a plant that could provide bulk and cheap 'leccy.
You might not want to borrow that precise and exact plan: your data centre probably isn't going to be worth an entire river's worth of dam or a full nuclear station. But the same basic idea should be guiding you. First ask yourself, where the hell is the power I need going to be cheap, and then, perhaps, ask whether you should build your own power station as well.
Power to the people
There is evidence that some people have caught on to this idea. Facebook, for example, has built a server centre up in Lapland. Doesn't help all that much with the cost of electricity, but leaving the windows open sure aids in the cooling process.
Apple's latest power announcement was that it was putting a vast solar farm by its servers in Nevada. The middle of a desert sounds like a good place to build solar array really, unlike our own dear government's attempts to pave Wales with them.
Apple's Maiden, North Carolina, photovoltaic array is soon to be replicated in Reno, Nevada
Other similar examples occur. Apple's $1bn North Carolina server farm uses solar plus a fuel cell system running on biogas (biomethane if you prefer). There's lots of pig farming in the Carolinas, so plenty of fuel about.
Similarly, Iowa could be a good place for a server farm: all that corn that gets fed to all those pigs. Iceland also beckons: they're not happy with the idea of building many more dams but the high concentration of volcanoes means they've got lots of very cheap geothermal energy up there to play with.
As an aside, doing the work where the energy is cheap is a very efficient method of exporting the cheap energy. It's a corollary of the water problem: it takes 1,000 tonnes of fresh water to grow 1 tonne of grain. So, what's best in a dry country? Desalinate or pump the aquifer water? Import water? Or import the grain and have the 1,000 tonnes of water "embedded" in the product, and yet only be paying for the carriage of 1 tonne of grain?
It's much the same with energy. The transmission losses of wiring that cheap Icelandic power to anywhere else would be horrendous. Better by far to conduct the work there and then ship the embedded cheap power. Why not have a fuel cell surrounded by pig farms? It has got to be cheaper than shipping pig shit around the country.
So, that's step one: where's the power going to be cheap, and how cheap can I make it by building my own power plant?
Not another industrial revolution?
The second step is to going around knocking places off your list because of latency. You guys will know more about this than I, but I suspect that however cheap Iceland is for power, you'll not be building server farms there. Those milliseconds over the cable run add up over those sorts of distances.
However, there are plenty of places in Scotland where you can get a decent subsidy for a micro-hydro project. Texan cattle feedlots produce enough cowpats to power a sizeable installation. The possibilities are near endless.
Today's Information Age, the "Digital Revolution", is often said to be the New Industrial Revolution.
Which is fun, because it's actually working the other way around. Back then, transport was the limiting cost: thus it was having the resources in the same place that led to the industrialisation. Iron ore, coal and limestone in close proximity led to the forges. Wool and hills and rivers and rain led to the water-driven mills (surely to God no one would ever do anything in Stroud without such economic imperatives).
Nowadays the whole point of what we're doing is that the transport is approaching zero cost. We can thus place each component of the system where it's most efficient for it to be.
Put the server farms out in the wilds where we can generate our own electricity; the designers in the inner cities where they can get soy lattes; and the programmers wherever it is that they wish to be (as it should be of course).
The whole point of this new revolution is that we don't have to have everything in the same place. We can have functions and activities where they are best placed, but still ensure that the whole can be widely distributed across those many places which are perfect for each of the individual processes. This all leads, one would assume, to the greater efficiency of the whole, as we're not compromising on location for transport-cost reasons. ®