Hacker News Reader

Comment by brian_cloutier | Hacker News Reader

> Prof Ian Bitterlin, a data centre consultant for nearly 30 years, is sceptical about the environmental impact of going underwater.

> "You just end up with a warmer sea and bigger fish," he says.

> And 90% of Europe's data centres are in big cities such as London and Madrid because that is where they are needed.

> Microsoft's Ben Cutler insists the warming effect will be minimal - "the water just metres downstream would get a few thousandths of a degree warmer at most" - and the overall environmental impact of the Orkney data centre will be positive.

Both of these opinions feel like exaggerations but I don't know enough about thermodynamics to know what the true answer is. I have a feeling that the Microsoft opinion is much closer to the truth, can anyone help me understand how I'd walk through the numbers?

View on HN

Not an expert either, but this might be a good way to think about it: on land or at sea, the machines produce the same amount of heat. On land, in general, the environment around the machines will be much warmer than the desired temp. Cooling on land is in general an active affair, which has to do with generating and controlling flow of fluid and/or maybe some kind of thermic reaction. Potentially at sea being plunged in cold water is sufficient to reach the desired temp, and cooling becomes a passive affair. All in all you are trying to displace the same amount of heat but through a more efficient medium. Correct me if I'm wrong.

Oh, I'm sure that being in a cold environment and letting it passively cool you is more efficient!

What I'm more unclear on is how small the impact is. It seems very likely to me that throwing the waste heat into the bottom of the ocean has less of an environmental impact than running A/C to suck out the heat and dump it into the air.

However, is the impact really so small that you wouldn't notice it just meters away? That's the part I'm unclear on.

I think so yes. Water has many times greater thermal conductivity than air, so the heat will spread out across a larger area faster. Meaning a very small (negligable) temperature increase at any given location. You wouldn't be able to detect it from a few metres away.

I'm trying to see how the efficiency is gained and I just don't really see it. I guess you don't need to pump water for water cooling so you save the pumping energy, or maybe lower fan speeds or something, but it really seems like it would be a minimal amount. What it does seem like is a cheap way to build a water cooling system that doesn't really need to be maintained. It seems like switching to arm would be a much bigger efficiency gain, but nobody cares because energy is cheap.

Usually DCs use AC for cooling, which needs electric power, which needs far away power plants and tranmisions lines. Here, you have a practically endless source of cool water that you can just pump directly.

Switching to ARM might be a bigger efficiency gain, but nobody wants arm. Azure already offers ARM servers. Moreover, these aren't exclusive - one can easily sink an ARM server just as easily as you can sink an Intel one.

But the efficiency gains are such: Air conditioning is actually inefficient, especially with air cooling. Water cooling is more efficient. Water cooling traditionally uses water as an exchange medium, but is eventually water-cooled-by-air anyway, just in bigger batches. Here, they can take in new cold water and throw out old hot water without actually bothering with any air exchange at all. Or at least, that's the plan. Maybe it'll work!

(If "nobody cares because energy is cheap", we wouldn't have tried this in the first place.)

> It seems like switching to arm would be a much bigger efficiency gain

That's a false dichotomy. More energy-efficient CPUs don't preclude efficiencies from cooling.

I suppose the case could be made that CPUs account for the vast majority of datacenter cooling needs and that ARM efficiencies would eliminate so much of that need that any efficiences in the cooling itself would not be worth anywhere near this kind of cost. I'd expect some pretty extraordinary evidence backing up that argument, since those would be pretty extraordinary claims.

Reply on news.ycombinator.com