QTS refits Dutch datacenter to warm thousands of homes with waste heat

Thousands of residents living in the Groningen region of The Netherlands will soon find their homes warmed by waste heat from local datacenters. QTS is latest datacenter operator to team up with local utility WarmteStad to connect their facility to the Groningen district heating grid. Doing so required the construction of a …

[Richard 12]

Insufficient data for meaningful comment

The CO2 emission reduction is in comparison to other ways of providing heating in addition to cooling, not merely other ways of cooling a datacentre.

Heat pumps are also called "air conditioners or "refrigerators". Almost all datacentres have those anyway.

The efficiency of the system does depend greatly on the temperature differential. Would have been useful to include some figures.

[Potemkine!]

Have a look to what does Qarnot Computing. They propose to replace home heaters and boilers by servers. The ones using the servers' power pay for the heating, making it free for the inhabitant.

== Bring us Dabbsy back! ==

[elsergiovolador]

What if servers don't receive any workloads and stay cool?

Seems like one good application of crypto miner to have running in the background...

[Filippo]

>What if servers don't receive any workloads and stay cool?

Unless the service is run by madmen, that should be extremely rare; idle servers are a loss. They'd probably have a big "low-priority" queue of jobs that they sell for very cheap, in return for making few guarantees on when they'll be run. I only have a little experience of cloud computing, but Azure for example has that. It's pretty nice for things like big analysis tasks, stuff that runs for hours and you just need it done by next week.

I don't see the priority ever getting so low that you're better off cryptomining. If a datacenter got to the point where nobody is paying them better than cryptomining, I would assume that they're going bankrupt shortly.

I see other problems with this scheme, though, see above.

[elsergiovolador]

One of the main selling points of the "cloud" is that you can scale up your service as you require. This means the cloud provider needs to maintain a margin of idle machines to facilitate clients potentially needing an extra capacity instantaneously and this is reflected in the price of services that are much more expensive that running a dedicated resource.

It is not wild to assume that someone would have received servers that are more idle than their neighbours. Of course it may be that they have a scheduler that will uniformly distribute workloads across the network, but then given the need for spare capacity, on balance the servers may not get hot enough to provide meaningful heating.

[Spazturtle]

The server are not usually idle, customers on special low cost plans are kicked off to make room for higher paying customers.

[Filippo]

It might work, in principle, but I would be worried about the logistics. If you're doing this at a scale where it makes any difference, you'll have hundreds or thousands of servers scattered all over the country. They have to be delivered, installed, serviced and maintained. The cost of doing so is going to be a whole lot bigger than if they were all conveniently in the same place. Bigger than heating costs? Don't know, maybe. Bigger than the difference between heating via simple resistance, and heating via a heat pump? Still don't know, still maybe, but it's a much bigger maybe.

And what do you do in the summer? Pay the customer for aircon? Ferry the server to an airconned facility? More logistics nightmares, and who's going to only rent you a datacenter for a few months a year anyway? That's going to murder any margin you had.

I'm not saying it can't work, but having a regular datacenter and just moving the heat around, seems to make more sense to me.

[Richard 12]

Heat pumps pulling in from "outside" are 300% efficient because outside is heated with many hundreds of kW by the sun.

These would be 100% efficient at best, same as a storage heater, but far less dense.

You'd probably need 10 servers to heat a 3 bed house, and you'd also need to vary the workloads to match the heating demand in the home.

In midsummer demand across the entire network would be zero most of the time, only rising when water needs heating after running taps/showers. If the servers run at all, they'd need active cooling running to dump the heat outside!

In a cold snap, everyone would want all "their" servers running at full whack.

Datacentre load doesn't follow the annual heating demand curve, let alone the daily/hourly swings.

So everyone needs both supplementary heating and a full cooling system for the servers.

Put simply, I don't see how it could be done economically. There's a reason why datacentres exist, and it's not so servers can have a chat over a coffee in the break room.