Cooling Server Rack with Water? Sensible? Reuse energy for small installation?

First - this is not a shopping question and not about concrete prices but about general feasibility. It makes no sense to look for a manufacturer if the approach is bad.

I am moving my company to new offices in September, and during the move we will expand and consolidate our computing cluster. Right now it is in a data center, but I have a nice room in the basement prepared for it.

I think about cooling. We will likely use around 10kw of power by end of the year. That is a LOT of heat, and cooling will be expensive. I am located in south Poland, close to the German border. This is an area where water is available for a relatively cheap price - "wasting water" is not a concern here. My situation is thus different from, for example, a company in Spain ;)

Physics tells me that to heat 1 liter of water by 1°C I use 1 Calorie (1KCal), and 1kWh is (and we can assume 100% efficiency - water heaters are pretty efficient) 750 Calories. That means that 1kWh is 750 liters heated by 1°C. 10kw and a 20° increase in heat would mean that per hour I need 375 liters. That is 6.25 liters per minute and not THAT much ;) We're talking 270 cubic meters here. Even in summer, the underground pipes really cool down the water a LOT more ;)

Is such an approach feasible? Has anyone done this? We have a 10kw installation for now. Is it feasible to reuse that heat? The alternative is a decent cooling system that WILL use around 2.5kWh to run. Dropping the water would:

1. Get me a quite cold input compared to the outside air, even in summer (I.e. a lower temperature medium to drop the heat in)
2. Replace the need to dump heat into the outside air, which may be problematic. If the air is 22°C, that is a LOT to fight off, but OTOH the water will be quite cold. I also would possibly save the investment for the outside part of the cooling circuit.

Now, second question - is there a feasible way to heat a house with that? ;) After all, it is a LOT of energy in that water ;)

If this is a bad idea, I'll stop here. If it is not, I'll start looking for suppliers. Maybe my math is wrong?

Solution 1:

There are two approaches to this, one I know for sure exists in the market, and another I've only heard about in custom builds.

Market-ready

These solutions use a water loop to provide at-rack cooling. They produce chilled air on the intake side of the rack, and use the water loop to exchange heat between ambient and what it produces. These solutions exist, and don't require any modifications to racked hardware to make work. They sometimes require special rack-doors to ensure correct airflow, but not always; some solutions create a 'curtain' of cold air in front of the server intakes.

With this approach you can easily break the 10KW/Rack barrier for power density.

Custom Builds

This approach brings water into the device themselves. I don't know of any off-the-online-store servers that have water loops built into them, but I hear that the major manufacturers can help you if you really do need 10K devices with such requirements. The advantage here is that you can push higher on the power envelope as you're not relying (as much) on air transmittance of heat, as the water-block itself does the heat transfer from the devices.

Theoretically you can get really high power densities with the custom-build method.

Solution 2:

As to the cost efficiency of water cooling in general, there's another good question that I answered already: Why datacenter water cooling is not widespread?

As to your specific situation, I would highly recommend looking at ground-loop heat pumps with economizers. It's likely cool enough in your Winter to just run the economizers, using what is basically outside air to cool your data center. In the summer heat pumps are much more efficient than standard AC units, roughly twice as efficient typically.

Using common loop heat pumps can allow you to extract the heat from one area (data center) and pump it into a different area (office space or an attached home even) quite efficiently too (though it's probably more cost effective to go with conventional sources, it depends on energy costs in your locality).

You can use air to dump-water to cool as well, air within the rack (as is typical) and fresh water to cool the air. This water can come from fresh sources and be dumped after "heating"; this used to be common to run commercial size refrigerators because the math worked out just as you put it in the Question. It's not commonly used today because water is more expensive than it used to be and the math no longer works (at least in my corner of the world, and water is pretty cheap here too). It would be more efficient to directly cool the server with water-cooling, however you introduce the risk of leak as well (which carries it's own risk or insurance costs).

The biggest detractor you'll have in implementing any of the aforementioned is that it's expensive equipment. Most businesses don't care if it pays for itself in 10 to 20 years, they only care about this year's bottom line. The payoff period depends on your local prices..

Solution 3:

TomTom - I very nearly bought a bunch of HP's MCS water-cooled racks for a project a few years ago but chose to change data center instead.

It's not exactly cheap but they stand by it in terms of reliability, for instance they're happy for you to put their Integrity stuff them. It can handle ~36Kw per rack by the way.

Just wanted you to be aware of this stuff as it doesn't exactly leap out of their website.