Hacker News Reader

Comment by teh_klev | Hacker News Reader

I'd disagree, the fairly newish DC we're in is designed for supplying and cooling 20kW/rack. That's enough for 40 or so 1u servers burning ~450 watts.

View on HN

One thing that burned us in the past was DC that provided 20kW power per rack in power, but required you to only dissipate 5kW in heat. Needless to say we changed the DC provider in like a month after they made this clear (and they combined that with offer of some kind of "cloud provider" package which meant 15kW of both and triple the price). After few another such DC-related incidents (and mostly unbelievable: circuit breakers catching fire, 600V peak on 230V AC line, double flooring collapsing from overload and such) I'm quite relieved that I don't deal with DC procurement anymore.

What the...? Were they expecting you to radiate the remaining 15kW as microwaves or something?

That's exactly the question I asked them, with "or you want us to radiate the remaining the remaining 15kW into this single mode fiber going to your switch? we can get laser capable of that"

Does the wattage coming in entirely converts to heat? I always assumed it didn't, but have no idea in what proportion.

Energy in has to equal energy out over the long term. For electronics equipment that isn't doing mechanical work then there aren't a lot of options for how that energy can come out: it'll either be electromagnetic radiation of some kind (i.e. light or radio or similar) or heat. Most computer equipment emits relatively little electromagnetic radiation, so it's essentially all heat.

Isn't heat just lower frequency EM radiation?

Heat can be transmitted through EM radiation (of all frequencies, but IR is predominant for temperatures we're accustomed to) but also by direct contact between materials. For typical temperatures, radiation doesn't transmit much heat. Vacuum makes for a pretty good insulator. If you want to remove 20kW of heat from a small space, you'll need to do most of that removal by transferring the heat to a fluid of some kind, e.g. by putting it into the air.

To see the difference in practical terms, pick an item that's noticeably warm, but not hot enough to burn you. Hold your finger very close to it. The heat you feel there is radiated. Then touch it, and feel the heat that's conducted through touch. You'll feel much more heat with the latter.

The external signalling power used by computers is completely neglible compared to the heat output. Computers and IT equipment in general are essentially ideal heat converters.

Hmm, good to know. And to update another assumption-- my guess is that this would be a pretty expensive configuration, and you may be better off picking a lower TDP server that can still get the job done, yes?

Also, do you actually have room for 40 nodes in a rack? Betwee TOR switches, storage, and miscellaneous other, it seems unlikely.

I would think 40 nodes in a rack is realistic if your battery is located elsewhere and you use side mount PDUs. Keep in mind in many places (NYC is a great example here) that real estate space for racks is the killer and not cooling or power.

250 KW per rack is possible using immersion cooling: http://investors.3m.com/news/press-release-details/2015/3M-N...

Just not in a real datacenter.

Reply on news.ycombinator.com