The Data Furnace is a tiny data centre that could heat your house. Peter Judge is not immediately convinced
What do we make of the proposal, from Microsoft Research, to heat homes using waste heat from tiny little micro data centres, or “data furnaces”?
Like most people, our initial reaction is doubt, as even data centre critic Greenpeace admits that centralisation is more efficient. But the paper, written by staff from Microsoft Research and the University of Virginia, makes a good fist of working out the figures, and comes to the opposite conclusion, finding that the move could save the former data centre provider around $300 per server per year.
Reversing the trend, or moving with it?
Big cloud data centres will have a PUE (power usage effectiveness – the total energy divided by the energy that reaches the servers) of 1,2. Moving services there will allow companies to turn off small racks of servers, dotted here and there, which have a PUE of 2.5 or more.
The bigger the PUE, the more energy is wasted in cooling the servers, so centralising into data centres means less energy is wasted.
But, the authors ask, what if the heat energy was used instead of wasted? Small data centre modules – developed from current containerised units – could be put back out into buildings and, if the heat is used to warm homes and offices, it would reduce other bills, and cut the overall environmental footprint.
Drawbacks to the idea
There are a couple of problems, which the authors try to deal with. Firstly, houses don’t need to be heated all the time but IT servers are run constantly – or at least as near to constantly as possible.
This leads to two possibilities. The data furnace might be “seasonal”, under the householder’s control, and only turned on when the house needs heating. This would be of limited value. If the cloud provider, and its customers, wanted continuous service, there would have to be sufficient backup capacity somewhere to keep services going when there is a heatwave.
So all the savings would be offset by having to provide another data centre somewhere.
Or alternatively, the data furnace could be on all the time – and then cooling would be needed when the weather gets hot. This would eat into the savings the report expects, and cut the PUE of the system.
The report also seems to expect a very low maintenance load (three visits per year to a data furnace with 40 to 400 servers), but then also suggests that these furnaces could be stocked up with older machines. These would indeed put out more heat – but they would also need more care and attention.
Server maintenance might be tricky on the user’s premises – the furnace is intended to be plumbed into normal heating systems, so the IT maintenance staff would need access to the house.
Another issue is that the report assumes the furnace would be replacing an electrically powered boiler, when it could equally be displacing a gas boiler, which is more environmentally efficient than an electric one.
And finally, I believe the report probably underestimates the extra costs in providing enough – and reliable enough – power, as well as enough bandwidth.
Flexible thinking is good
Despite my doubts, I’m interested in the idea. It is possible that future servers could be reliable enough to install in houses, and home broadband could be good enough for all homes to have one.
We also know for sure that energy costs are going up. Right now, we make heat in data centres and throw it away, while we make heat in homes to use. Let’s stop doing both, says the report.
Although data centres are currently moving towards ever greater centralisation in larger premises, there actually is a move in the other direction, as more of these data centres are built with containers, which are portable and modular.
There have even been micro-modular data centres proposed, though normally as portable kit for diaster sites.
There are, however, two things I like about the proposal.
Firstly, the idea of using the heat output of servers is a good one, which has been in circulation for a while, and any proposal that reminds us of this possibility is a good one.
Secondly, it is important to keep aware that location is not a given, and people providing infrastructure should be able to think creatively about where to site data centres, along with generating plants and other equipment.
It’s been suggested that data centres should be powered by their own generating systems – renewable where possible – in order to reduce the loss in transmission from where it is generated.
Breaking up data centres and re-siting them is yet another possibility to throw into the mix.