Using Jet Engines to Cool Servers

rpmsci writes "The computer servers that fill huge data centers are producing more heat with every new generation of processors. It's a problem that's sending engineers on a search for cooling fans that are both small enough to fit inside ever-smaller server chassis and powerful enough to dispel increasing amounts of heat. At Hewlett-Packard, they've found one answer in an unexpected place: model jet airplanes."

Re:Why bother?

[SatanicPuppy]

Might want to try some Fluid XP [xoxide.com] coolant. It's non-conductive, so no zapped parts. It's non-corrosive, so fewer motor problems. And it's non-toxic, so if your 2-year old glugs a quart of it, all they get is blue teeth.

I've never heard anything bad about it, and it works fine for me.

Re:Not a jet.

[scdeimos]

You mean "used to be available". Have you tried to order these after 9/11?

Yes, there's virtually no difference in availability:

• AMT - Advanced Micro Turbines [amtjets.com]
• ARTES Jet [artesjet.com]
• Behotec [behotec.com]
• BMT - Baird Micro Turbines [bairdtech.com]
• Jakadofsky Jet Engines [jakadofsky.com]
• JetCat USA [jetcatusa.com]
• SimJet [simjet.com]
• SWB Turbines [swbturbines.com]
• TJT Australia [www.tjt.bz]
• Wren Turbines [wren-turbines.com]

And there's a whole bunch more here [airtoi.nl]. There's no shortage of gas turbine planes and pilots in our aeroclub [tmac.asn.au], either.

Liquid cooling with HVAC chilled water

[Kadin2048]

Actually most datacenters already have massive water-cooling systems: only they're building wide and generally used to cool the air. I'm talking about the HVAC system, of course.

Large buildings generally don't circulate Freon from one floor to another, it would be too expensive. Instead, they have a big refrigeration unit (roof mounted, usually) with big cooling towers and the rest, and use it to chill water, which is pumped throughout the building and used to cool air.

It wouldn't be very difficult to tap into the chilled-water lines that already exist in most buildings, and use them to cool the servers directly. In fact this was once a lot more common: back in the day, it wasn't uncommon for big mainframes to be water-cooled. I've worked with big scientific apparatus that's also water-cooled, and a lot of it used lots of electricity as well, so it's not as though the engineering is impossible.

Yes, there are certain risks associated with having water flow through your computer system, especially in regards to leaks. But there are lots of pieces of equipment that contain liquid and wouldn't appreciate leaks, and we don't think twice about them. For very valuable systems, an additional cooling loop filled with a non-conductive (or even better, a pressurized gas) coolant could be used, with a heat exchanger connecting to the building chilled water.

I think there are some IBM blade systems out there right now that use liquid cooling, but for some dumb reason they won't accept building chilled-water connections (believe it or not, they need water that's warmer than most building supplies). I can't find a link to it right now, but basically it introduces an additional heat exchanger for the sole purpose of warming the incoming chilled water supply before circulating it through the systems. Obviously, this limits their attractiveness and ease of installation.

But at any rate, I think going to liquid cooling, whether water or glycol or something else, is eventually inevitable in high-density applications: despite some of the practical problems involved, when you look at the economics, cooling is one of those things that scales really well. It's going to be cheaper in the long run to reduce the number of heat-transfer steps in between the chip and the outside environment (where the heat is going one way or another), and to do it all at once if possible. Maybe we haven't hit the power/density break-even point yet, but we must be getting close.

I think the reason you still see a lot of air-cooling is because the mass-production of components has made it inexpensive to do, but blade server systems are starting to run into the limits of commodity hardware (as this whole story with HP's fan attests to). When you start having to consider developing specialized cooling hardware anyway, whether for air or liquid, suddenly liquid cooling becomes more attractive.

How they work.

[twitter]

On the other hand, the HP one uses small blades that are shorter and that spin faster. As such they create more thrust/airflow and reduce noise that normal blades produce from the tips of their blades.

That's about all the article says.

The key ingredient to a ducted fan is efficient expansion. Any old array of twisted parts can propell air. I read another article and fabricated such a thing from Dixie cups. After your rotor comes the stator, a very important component missing from ordinary fans, which removes the angular component of the flow velocity. You want to move the air down your axis not around it. Getting the air moving along the axis and expanding it out to larger volumes without wasting your effort is hard to do. Adding any stator will help. Doing it quietly and efficiently is one of those rocket science things.

Wikipedia, of course, has a quick article, [wikipedia.org]

and Google turns up an easy design text [pair.com].