IBM says it is working towards the introduction of technology to enable zero-emission data centres, which will give out not a single whiff of CO2. The claims were made by Bruno Michel of IBM's Zurich Research Laboratory at LISA '09, the Large Installation System Administration Conference, which happened to be in Zurich. Michel' …
District Heating - I think not...
"Water-cooling is a better way to transfer heat from data centre kit - from thermal conduction chip modules to racks - than air cooling."
That's all fine and dandy at the level of the physics - indeed liquids are better than gases at transferring heat away as they have much more carrying capacity. Car engines are (almost) universally liquid-cooled and IBM themselves used to require their mainframes to by liquid cooled back in the days of ECL logic (and that was direct liquid cooling plumbed into the data centre, not a hybrid system via heat exchangers in the racks).
However, there are very good practical reasons why direct liquid cooling of computing equipment is a major league pain-in-the-neck in data centres. The plumbing types up lots of space, there is the ever-present danger of leaks, air-locks, installation costs and so on.
it should also be noted that there are very considerable problems with making use of waste heat in district heating systems. What comes out of a data centre is low-grade heat (whether direct, or via the air-conn). Technically it's a very low entropy energy source, which is just another way of saying there are vast volumes of warmed up air, not the concentrated supplies of hot water that you need for district heating schemes (where you really want the water approaching boling point). It's simply not practical to pipe thousands of cubic metres of warm air around houses.
There are ways of increasing the usefulness of waste heat by increasing the temperature of the exhausted heat from an air conn system. However, you don't get something for nothing - the downside of that is that the efficiency of the air conn system will plummet. It's a simple matter of thermodynaics - heat engines, carnot cycles and all that stuff. Essentially the data centre air-conn will have to use more energy to make something more useful for heating hot water and the like. There will be a net gain of course - there are heat pumps which make use of the ground being warming than air temperatures during the winter, but the hardware end expenditure would be enormous.
Some might argue for running liquid cooling straight into the computing hardwarfe and running the internals of a computer at 90 degrees or so (and you'd have to run the components as hot as possible to get a decent temperature gradient to make this liquid cooling work well). I would have severe doubts about this - you may just get away with it for the processor, but there are plenty of other components which won't run reliably at those sort of levels. I'm also not sure you want lots of near-boiling hot water flowing around your data centre.
All in all, if you want to blow the warm area from a data centre into a large industrial building opr a commercial greenhouse, that might work and be cost-effective. However, as a general district heating system - I think you can forget it. There's much more to be gained by reducing the power required by the electronic bits and pieces in the first place and reducing the need for expensive air-conditioning rather than vast, complex and very expensive schemes to turn the high-entropy waste heat from a data centre into something more useful.
Chip powered heating
Sounds like bollocks to me or we would all be using chips to heat our homes - surely an electric heater or gas heater is more efficient than servers at converting power to heat? If so then I do not see how replacing the heat from those heaters with heat from a data centre results in an overall zero net effect?
yeah, and then they sell Power CPUs, right?
looking at those slides, 18% lost in the UPS. Get rid of the UPS, your electricity bill just went down, and another SPOF went away. Course, you need some embedded batteries for those nodes you can't do without, but iif you balance your blocks in your Hadoop filesystem, you only need 1/3 of the nodes to stay available.
Replace the Power CPUs with some of the low power ones -and ARM cores with ECC look nice to me, and not only do you save on CPU power, but the PUE multiplier kicks in nicely.
otther than that though, nice workj.
I for one...
welcome our fluid-shear driven, self-assembling overlords.
(Sorry, had to do it)
...dump the Power CPU - come on IBM, show your real green credentials and develop something that doesn't chew the planet even when in deep sleep.
Forget water cooling, use refrigerant direct to the chips
Being involved in the HVAC Controls business, why don't they replace the water in the cooling blocks on the chips with refrigerant (R134 or R410) and connect that to an air cooled condensing unit outside. This way, the CPU heat is directly changing the phase of the refrigerant from liquid to gas right on the chip with no losses in an evaporator coil combined with losses transfering from the coil to air.
They already have passive cooling systems like this for CPU and chipset coolers.
You could then use the heated air from the condensing unit to heat outdoor fresh air coming in to make up for that air volume being exhausted.
Conceivably, the refrigerant could also be used in a Stirling cycle engine which could drive a generator to charge the UPS batteries, off grid.
As this has been published in El Reg comments, it is now copyrighted and proves my prior art.
Eat that Big Blue and Google. However, I'l be happy to negotiate a reasonable liscensing arrangement.
Flames for obvious reasons (Thermo-gawdammits)
The problem as I see it is that all the players have different ideas on how to get rid of the heat. If you have a lot of different suppliers, as we have, you have lots of different schemes.
Cold air drawn up through the floor, blown out the top.
Cold air drawn up through the front, blown out the back.
Cold air drawn up through the side, blown out the other side. (and the other side)
Add water to the mix and you will have lots of different pipes, flows, connectors and probably other things I don't know about. Perhaps we need standards in cooling systems.
As to using the waste heat, preheating the water fed to the boiler would save some power even if it needed augmenting.
@dan paul -- yerbut...
... then you've moved from the (relatively) simple air/water system to a phase-change system and are likely going to run into problems with condensation and all the hilarity which can ensue...
Phase-change solutions already exist in the "enthusiast" sector -- http://www.pcstats.com/articleview.cfm?articleid=1793&page=1 as just one example...
- YARR! Pirates walk the plank: DMCA magnets sink in Google results
- Pics Whisper tracks its users. So we tracked down its LA office. This is what happened next
- Review Xperia Z3: Crikey, Sony – ANOTHER flagship phondleslab?
- Ex-US Navy fighter pilot MIT prof: Drones beat humans - I should know
- Human spacecraft dodge COMET CHUNKS pelting off Mars