james.p.lux at jpl.nasa.gov
Wed Oct 22 16:16:09 EDT 2003
To a first order, figure you've got to reject 150-200W per node.. that's
roughly 10kW of heat you need to get rid of. That's 10kJ/second.
That will tell you right away how many "tons" of A/C you'll need (1 ton =
12000 BTU/hr or, more usefully, here, 3.517 kW)... Looks like you'll need
3-4 tons (3 tons and 5tons are standard sizes...)
Next, figure out how much temperature rise in the air you can tolerate (say,
10 degrees C)
Use the specific heat of air to calculate how many kilos (or, more
practically, cubic feet) of air you need to move
(use 1000 J/kg deg as an approximation... you need to move 1 kg of air every
second or about a cubic meter...
roughly approximating, a cubic meter is about 35 cubic feet, so you need
around 2100 cubic feet per minute)
As a practical matter, you'll want a lot more flow (using idealized numbers
when it's cheap to put margin in is foolish).
Also, a 10 degree rise is pretty substantial... If you kept the room at 15C,
the air coming out of the racks would be 25C, and I'll bet the processors
would be a good 20C above that. Calculating for a 5 degree rise might be a
better plan. Just double the flow.
Unless you're investing in specialized ducting that pushes the AC only
through the racks and not the room, a lot of the flow will be going around
the racks, whether you like it or not.
In general, one likes to keep the duct flow speed below 1000 linear feet per
minute (for noise reasons!), so your ducting will be around 3-4 square feet.
This is not a window airconditioner!... This is the curse of rackmounted
equipment in general. Getting the heat out of the room is easy. The tricky
part is getting the heat out of the rack. Think about it, you've got to
pump all those thousands of CFM *through the rack*, which is aerodynamically
not well suited to this, especially in 1U boxes. How much cross sectional
area is there in that rack chassis aperture for the air? How fast does that
imply that the air is moving? What sort of pressure drop is there going
through the rack?
Take a look at RGB's Brahma web site. There's some photos there of their
chiller unit, so you can get an idea of what's involved.
Your HVAC engineer will do a much fancier and useful version of this,
allowing for things such as pressure drop, the amount of recirculation, the
amount of heat leaking in from other sources (lighting, bodies in the room,
etc.), heating from the fans, and so forth; But, at least you've got a ball
park figure for what you're going to need.
----- Original Message -----
From: "Arthur H. Edwards" <edwardsa at plk.af.mil>
To: <beowulf at beowulf.org>
Sent: Wednesday, October 22, 2003 10:21 AM
> I'm moving a cluster into a 9.25x11.75 foot room (7.75 ' ceiling). The
> cluster now has 48 nodes (single processor AMD XP 2100+ boxes). The will
> be on metal racks. Does anyone have a simple way to calculate cooling
> requirements? We will have fair flexibility with air flow.
> Art Edwards
> Art Edwards
> Senior Research Physicist
> Air Force Research Laboratory
> Electronics Foundations Branch
> KAFB, New Mexico
> (505) 853-6042 (v)
> (505) 846-2290 (f)
> Beowulf mailing list, Beowulf at beowulf.org
> To change your subscription (digest mode or unsubscribe) visit
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