When is cooling air cool enough?
James.P.Lux at jpl.nasa.gov
Fri Oct 17 19:45:24 EDT 2003
For component life, colder is better (10 degrees is factor of 2
life/reliability), and the temperature rise inside the box is probably more
than you think.
You also have some more subtle tradeoffs to address.
You don't need as much colder air as warmer air to remove some quantity of
heat, and a significant energy cost is pushing the air around (especially
since the work involved in running the fan winds up heating the air being
This is a fairly standard HVAC design problem.
The additional cost to cool the room to, say, 15C instead of 20C is fairly
low, if the room is insulated, and there's a lot of recirculation (which is
typical for this kind of thing). It's not like you're cooling the room
repeatedly after warming up. Once you've reached equilibrium, cooling the
mass of equipment down, you're moving the same number of joules of heat
either way and the refrigeration COP doesn't change much over that small a
temperature range. The heat leakage through the walls is fairly small,
compared to the heat dissipated in the equipment.
If you were cooling something that doesn't generate heat itself (i.e. a
wine cellar or freezer), then the temperature does affect the power consumed.
This all said, I worked for a while on a fairly complex electronic system
installed at a test facility on a ridge on the island of Kauai, and they
had no airconditioning. They had big fans and thermostatically controlled
louvers, and could show that statistically, the air temperature never went
high enough to cause a problem. I seem to recall something like the
calculations showed we'd have to shut down for environmental reasons no
more than once every 5 years. Humidity is an issue also, though.
At 01:47 PM 10/17/2003 -0700, David Mathog wrote:
>Most computer rooms shuttle the air back and forth
>between the computers and the A/C. I'm
>wondering if one could not construct a less expensive
>facility (less power running the A/C which is rarely
>on, smaller A/C units) if the computer room was a
>lot more like a wind tunnel: ambient air in (after
>filtering out any dust or rain),
>pass it through the computers, and then blow it out
>the other side of the building. Note the room
>wouldn't be wide open like a normal computer room.
>Instead essentially each rack and other largish
>computer unit would sit in its own separate air flow,
>so that hot air from one wouldn't heat the next.
>The question is, how hot can the cooling air be and
>still keep the computers happy?
>The answer will determine how big an A/C unit is
>needed to handle cooling the intake air for those
>times when it exceeds this upper limit.
>I'm guessing that so long as a lot of air is moving through
>the computers most would be ok in a sustained 30C (86F) flow.
>Remember, this isn't 30C in dead air, it's 30C with high
>pressure on the intake side of the computer and low
>pressure on the outlet side, so that the generated heat
>is rapidly moved out of the computer and away. (But not
>so much flow as to blow cards out of their sockets!)
>Somewhere between 30C and 40C one might expect poorly
>ventilated CPUs and disks to begin to have problems. Above
>40C seems a tad too warm. At that temperature it's going
>to be pretty uncomfortable for the operators too.
>Anybody have a good estimate for what this upper limit is.
>For instance, from a computer room with an A/C that failed
>There's clearly a lower temperature limit too. However on cold
>days opening a feedback duct from the outlet back into the intake
>should do the trick. In really cold climates the intake
>duct might be closed entirely - when it's 20 below outside.
>mathog at caltech.edu
>Manager, Sequence Analysis Facility, Biology Division, Caltech
>Beowulf mailing list, Beowulf at beowulf.org
>To change your subscription (digest mode or unsubscribe) visit
James Lux, P.E.
Spacecraft Telecommunications Section
Jet Propulsion Laboratory, Mail Stop 161-213
4800 Oak Grove Drive
Pasadena CA 91109
Beowulf mailing list, Beowulf at beowulf.org
To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
More information about the Beowulf