Power is a funny thing. As demand has grown, the need to generate evermore quantities of electrical power has meant that many different directions have been explored. Despite dalliances with nuclear power, solar, wind and wave energy, the terrible fact is that much of our current electrical power still comes from coal.
Last year, according to the UK’s Department of Energy and Climate Change (DECC), 42.8% of the UK’s power came from coal. In the US the situation is similar, around 44% of electricity is generated from coal, according to the Union of Concerned Scientists, and is the single biggest pollution source there. Incidentally, the UK figure for 2013 is up from 40% in 2011. China produces 66% of its power from coal.
Now, billions of dollars, euro, yen and yuan are being spent on alternative sources, in particular nuclear fusion. The hope is for a renewable and scalable source of clean, efficient power at a reasonable cost. Currently there are two competing efforts to get a sustainable fusion reactor going.
The first of these is the magnetic containment method where deuterium-tritium fuel is heated under pressure to a plasma and contained within a doughnut-shaped chamber (tokamak) by the aforementioned magnetic field. This approach is the focus of the International Thermonuclear Experimental Reactor (ITER) project, and a site in France is currently being prepared for the largest ever tokamak-style reactor.
The other major efforts are around inertial confinement fusion, where a small pellet of similar deuterium-tritium fuel is simultaneously and symmetrically bombarded with the most powerful lasers ever developed to create an inwardly focused shockwave that will ignite the fuel in a fusion reaction. The National Ignition Facility in the US is the leader here and saw a milestone in September of last year when its reactor actually briefly produced more energy than was consumed to create the reaction.
However, the most optimistic assessments here are for practical reactors to be in operation by about 2028 or so, with commercial production by about 2050. The dream of near limitless, essentially clean and cheap energy is still some way away.
In the meantime, power has to come from dirty, traditional sources, topped up by green energy from renewables such as wind and solar.
So what does this all mean for the data centre? Well it means that the world’s demand for evermore services based on massive data centres will simply not be met by building evermore data centres which in turn would necessitate evermore power stations to supply them. This means that the greatest force lively to shape the data centre in next five to 120 years is power, or more importantly, the lack thereof.
Dire predictions were being made in the noughties around the voracious data centre appetite for power. A Stanford Universities professor, Jonathan Koomey, predicted in 2007 that data centre power consumption would likely increase by as much as 100% from 2005 to the turn of the decade. Now, in reality that did not happen. According to an article in Modern Infrastructure magazine by Stephen J. Bigelow, the actual increase was 36%. Bigelow attributes much of this reduction to the advent of virtualisation and the resultant server consolidation that followed, increasing utilisation rates and efficiency.
However, despite these predictions being wide of the mark, what has happened is that there has been a broad realisation that data centres would have to change as the old approaches are not scalable to current anticipated needs, particularly when it came to servers and power distribution.
Intel, ARM, AMD and others have therefore tackled this with low power consumption processors such as the Atom range and latterly ARM-based server processors like the AMD A1100. HP went down the ultra-high density route with tis Moonshot server infrastructure range, with a particular eye towards web-scale workloads. With densities of up to 1,800 servers per rack, the Moonshot infrastructure occupies just one eighth of the space of traditional servers.
“The world’s demand for evermore services based on massive data centres will simply not be met by building evermore data centres which in turn would necessitate evermore power stations to supply them”
But there is a consequence of going so dense in terms of space and that again is power. How does one feed all those little servers, even if their individual power draw is dramatically reduced?
This focused attention on power distribution and the fact that there are inevitable losses coming from grid to rack and server. An article on Scientific American claims that of the power generated in the power stations, only 20% actually makes it to the server to do work. The losses are due to factors such as resistance in lines, distance travelled, voltage and current changes. This has led to new considerations such as when to change from AC to DC, and when to change voltages to keep losses to a minimum.
A more radical approach is for the racks to power themselves. Fuel cells, such as those powered by hydrogen, or even methane, could be used to power each rack resulting in significantly reduced power losses. There may even be scope for waste heat recovery to aid in cooling, through the use of such devices as the Kyoto Wheel that can manipulate internal and external airflows.
The power consumption characteristics are also under close scrutiny and work such as that done by Anandtech.com has shown that sometimes low end server chips are not as power efficient as they at first seem, especially under dynamic loading as opposed to a steady state draw. This is now a major consideration in determining real efficiency and power consumption. Whereas a fairly homogenous load can be evenly distributed across a server fleet, a heterogeneous load that made up of many different and varying tasks that is changing constantly will have very different power characteristics and therefore a very different efficiency profile.
All of these considerations mean that it is power, its availability, distribution and demand, that will shape the data centre of tomorrow. The balancing act between service demand and the ability to meet that need within the levels of power available will drive technological developments in ways hitherto unseen.
Subscribers 0
Fans 0
Followers 0
Followers