rack power question, and a prediction about "direct heat removal" (DHR)
patrick at zill.net
Fri Apr 4 05:58:31 UTC 2008
First, I would like to thank everyone who responded to my initial query.
It seems that power and how to remove the resulting heat, is certainly
on a lot of people's minds.
Clearly the days of "including" power are past in all but sales and
There is a cost to each component and markup is applied (if internally
and not broken out to the customer) on each rack, each power circuit,
etc. with covering the overhead of UPS, diesel generator, chillers, etc.
being a big priority.
After some thought, I believe, to coin a term, that "DHR" or direct heat
removal, in some fashion will be the "new" thing for the datacenter.
Somewhat counter-intuitively, the focus will be to remove the heat that
comes out the back of the rack rather than worrying so much about the
temperature of the air going in the front as long as it falls in a
generally acceptable range of say, 68-75F.
My guess is that someone will come up with an inexpensive, reliable way
to put a heat collector, which will basically look like a car radiator
the size of a rear rack door, directly behind the hot air coming from
the systems in the rack.
Hot air flows past the cooling fins and is quickly cooled back to 68F;
the heated refrigerant is immediately piped away, out of the room and to
the chiller, so that the evil BTUs do not spread out and contaminate
other areas of the room.
It might involve a phase change material, or might involve a more
My money would be on R744, also known as CO2, as it is not polluting and
can serve double duty as fire suppression (provided you have enough on
hand to flood the area/room).
Detectors for leaks are very inexpensive and the technology for the
closed-loop of the refrigerant cycle is already here.
It is not caustic the way any of the salts-based variants would be, is
not explosive, and is heavier than regular air, meaning it will sink
below the area that most people breathe at should there be a small leak.
With the heat being removed within a few inches of where it is
generated, less CRAC units will be needed to keep the rest of the air
cooled; and possibly, no separate unit would be needed if enough heat
can be removed to drop the temperature below 68F.
For fire suppression, an alarm would sound and only when it can in some
fashion be "proven" that no humans are inside the area, CO2 is flooded
into the area and the fire goes out. Some form of ducting which mixes
the CO2 with regular air and exhausts it is needed after the fire is
out. Firemen go in with oxygen if they need to enter before this is
done. (obviously there would be an entire tested procedure for how this
is done, probably including a small oxygen mask with ~4 minutes of O2
placed beside each fire extinguisher and within easy reach).
(For racks with less than say 4KW of power use, network and power is fed
from overhead with a few feet of slack in the cables, as well as a
portion of the DHR piping being flexible tubing. This allows them to be
placed more closely together than normal, almost front to back to
front to back, with enough slack to pulled the wheeled racks "out" from
the stack so it can be worked on (sort of like pulling a book out of a
bookshelf). They use far less space and are sold a little cheaper by the
Surely there is a limit as to how much air can be moved around, even
with the use of best practices, there are hot spots.
Simple physics dictates that this is a less energy intensive cooling
method as a) moving a lot of air around requires a lot of energy b) air
is a lousy way to transfer heat away from where you want it to be
compared to other materials.
patrick at zill.net
More information about the NANOG