This was a popular attitudes in the 80s, and 90s, until Google, Intel and others did some research and discovered that Cold data centers aren't the be-all/end-all.
After reading the research I kept on the order of 100 servers in a server room at approx. 80 degrees for a couple years, and didn't see any devices fail.
I don't think anyone in the last 10+ years has believe that "servers are expected to keel over at 75F"
Yep. 85degC is super hot, and junction temp is usually a few tens of degrees hotter than the ambient temp. Even if the packaging and interconnects can handle it, the dies themselves will fail early due to metal migration.
Of course, if the average working life of a Google server is measured in months, then maybe that's no big deal.
> Facebook’s data center (bottom) spends nearly two-thirds less energy on power and cooling than a typical facility. The building’s upper level (top) pumps in frigid air, cooling the servers below, then vents warmer air outside
Pumping warm air outside feels really wasteful.
Is there any work on increasing efficiencies with waste heat?
No? It would be inefficient to put a lot of effort and capital into trying to extract useful work from a marginally-above-room-temperature exhaust stream. Something about thermodynamics and low efficiencies at small temperature gradients :P
You're not thinking like someone that lives in a cold place :-P
In cold climates, the heat could be distributed to reduce the heating load in other buildings during cold months.
Where I live (Alberta, Canada), it's continuously below freezing for at least 3 months, and often below for 2-3 months either side of that. People would be all over getting free heat from a neighbour, even if it was just a little bit.
The difficulty is that data centers don't generate "high grade heat" ie, they don't actually run very hot. While it might dissipate a lot of heat, the temperature is quite low (they claim their data center only run at ~30C). As a result, their 'waste' heat stream will never be above ~30C if they don't have anything special going on. There aren't a lot of useful things you can do with that type of heat. Even if they managed to dump it all into water and them pipe it over to surrounding housing, it's still not -that- useful, and likely not worth the cost/energy of setting up the system.
While they could also conceivably run liquid cooling to -all- of their high temperature components (CPUs and PSUs probably) to get a higher grade waste heat stream, once again, its likely not worth the energy/money to set up that system.
This is why most cogeneration or waste heat schemes are on industrial processes - their "waste" heat is pumping out at above boiling at a minimum, and far more useful.
Just to elaborate, this "low grade heat" makes it physically impossible to recover much energy out of it.
If the ambient air is, say, 0C (273K), and the exhaust air is 30C (303K), then the optimal efficiency for recovering energy from the exhaust is less than 10%.
In other words, if your data center uses 1MW of power, then with perfect recovery equipment, you could generate 100kW off the waste heat, reducing your total power consumption to 0.9MW. Not a particularly big win, especially since your recovery equipment is not going to be 100% efficient.
Without any further treatment, it seems that waste heat is at least enough to heat living and office space (which is probably what they're doing with it).
If you arrived at that 10% by just looking at the K, keep in mind that 0C is ambient temperature and already represents the lowest energy level of that system. I really believe it should be possible to take and use more of that 30° degree difference than a mere 3° off the top, even factoring in heat pumps for consolidating it and moving it around.
The place is "In the middle of a forest at the edge of town", so if you're using the exhaust for heating other buildings, you're going to need long insulated conduits to carry it all, and that's probably not worthwhile AND will lose a lot of the heat to the outside anyway.
I don't understand what the Arctic circle has to do with this. It's a data centre, not an HPC cluster. Data centres don't run nearly as hot as clusters.
I do also have a problem understanding what the Arctic Circle has to do with this. North of arctic circle in America and Scandinavia is two totally different things
The Nordic countries have a rather mild climate despite being so far north. To have a comparable temperature in the Americas, you'd usually need to be a couple of thousand kilometers further south. (E.g. I guess the climates of Boston and Stockholm might be roughly comparable, but Boston is actually on the same latitude as Rome).
So if you tell an American that something is almost at the Arctic circle, their intuition will tell them it's a very rough and almost uninhabited place. You tell it to a Finn, and they'll think that it's where the fourth largest metropolitan area of the country is, or where you go for a comfortable ski holiday.
I have lived north of the circle for most of my life, and the feeling I have is that 13 C in October and having winters where it is more days with around 0 C than -15C is not what most peoples associate with north of the arctic circle. The difference is that we have warm water coming with the Gulf Stream creating a completely different climate compared to other places north of the circle.
I saw a small data center (4 racks) get from 18C to 55C in one hour when the chillers failed. And that was the air temperature, not the server temperature. In theory you can just open the doors, but then you cannot control the humidity.
I'm no datacentre guy, so can someone clarify if this is a typo? What kind of electronics start failing just above room temperature?
I'd think HDDs would be the most sensitive, but google said failures aren't well correlated to hdd temp (http://research.google.com/archive/disk_failures.pdf)