It was inevitable that the early spring would spawn yet another one in a long line of ‘solar powered’ data centre stories. This time the reality is not hidden from view, nor apparently are the facts from the perps, but it makes you wonder on what grounds research funds are doled out for investment in a hardware solution as opposed to a desktop study? A couple of hours of thought and scribbles should convince anybody that the contribution that solar PV energy can make to a data centre is nearer 1% than 100% and that any other outcome can only arise by distortion of the limiting factors.
The latest comes from MIT in the US – an august body that, you would have thought, didn’t need to build a wheel to prove that it goes around. They have announced that they are building a test micro-facility to explore the viability of solar for a ‘micro’ data centre. However, right up front, they have described the primary problem that they are looking to solve: it is a problem of energy storage when the sun don’t shine. Well, that’s a surprise…
Worshiping the sun
Apart from the intermittence of solar power the other main problem is the disparity between solar PV output per square metre and the typical data centre cabinet load. In the case of the MIT investment they describe a micro facility (a container) of 18.5 square metres being fed by a solar-PV array of 288 square metres and backed up by a utility connection and energy storage in batteries or flywheels.
A few minutes of internet search shows that the Massachusetts weather is quite similar to London (a bit colder in winter but very similar bright-sky intensity) and so each square metre of solar-PV can be relied upon to generate about 600kWh per year – unfortunately with several winter days generating almost zero energy. That means that the MIT array will generate about 173,000kWh per year – equivalent to 19kW continuously. It is a big array compared to the load and is therefore real-estate hungry but it will need a storage element (or be able to back-feed into the utility) as the ‘peak’ output (1kW/m2 solar insolation, south facing at 35 degrees to the horizon and 15% conversion efficiency) will be around 43kW.
It is then just one short step to calculate the ICT load that could be supported by 100% solar over one year – 19kW with a PUE of 1.2 would be a net 16kW ICT. This is equivalent to 800W/m2 in the container or, in more useful terms, about 1.3kW per cabinet. A really low load compared to the average 6kW of today. Let’s face it – the micro-facility should be designed for just one or two cabinets and be nearer a closet that a room.
Looking down the other end of the telescope we could suggest that the container ‘should’ contain an average load of 50kW and the solar-PV array should be >900 square metres instead of <300 – about 45 times the footprint of the container.
So, clearly, having a solar-PV array of 15x the facility footprint could support a low ICT load density even assuming that 100% of the solar energy is captured, stored or back-fed. Without a utility connection the array has to be MUCH bigger or the load MUCH smaller. As for energy storage the press release talks of batteries or micro-flywheels but MIT will probably find that the battery solution (far) outperforms the flywheel at demand cycles over 10 minutes of discharge.
I’m sure that some Lithium Ion cell producer will donate a battery solution but if MIT had to fund it themselves then a good lead-acid (emphasis on the word ‘good’) will outperform anything else. The chances of never using the utility are minimal unless a 25-30kW genset is installed – no doubt running on vegetable-based pre-used frying oil. When the sun doesn’t shine at least the smell of chips (fries) will be reassuring…
In search of answers
For those still interested: the press release does list three questions that MIT hope to answer through this test and I have suggested some pre-investment answers:
Q: How should a data-centre incorporate renewable sources of energy?
A: Like every other load. There is no logical reason why data centres should be targeted to use renewables over any other application, e.g. hospitals, care homes for senior citizens or income tax offices. In fact, there are good reasons to discuss if the predominant data traffic of tap-dancing dogs, social networking, dating, pornography, gaming and gambling should be renewably powered at all…
Q: How should future data centres interface with a smart-grid to intelligently reduce their energy bills?
A: Great question if you change the word ‘should’ to ‘could’. Smart grids can only exist with data centres in the network and the most effective way for such facilities to reduce costs (and be more sustainable) is to re-use the waste heat. This implies multiple micro-facilities embedded in the smart-cities each providing heat to an adjacent local source. This is not ‘new’ thinking – it just needs integrated thinking.
Q: How should we design green high-performance computing applications that intelligently manage power use?
A: Ignoring the word ‘green’ would help. There is little that is ‘green’ about an ICT application unless it is enabling a low(er) carbon solution – energy goes in and only waste heat comes out. However, the answer to the question always tends to centre on load time-shifting. Better described as batch-processing. But Pandora’s box was opened long ago and the chances of a message like ‘Please come back to our web-site when the wind is blowing in the North Sea’ being acceptable to users who have come to expect instant-on, ultra-fast connectivity, etc is remote. Of course the obvious solution is to return to 64kBit modems – these 200-300Mbs connections just fuel ever more traffic, not just faster access.
I wish good luck to MIT in their techie sand-pit.