Fwd: Re: [Beowulf] Earthquakes and raised floors...

David Kewley kewley at gps.caltech.edu
Mon Jan 9 08:58:04 PST 2006

On Monday 09 January 2006 06:59, Justin Moore wrote:
>     Actually, homogeneity of ambient temperature doesn't save you 
> anything on cooling, and in fact it might make things worse.  What we 
> should be striving for is minimizing hot air recirculation, since that 
> introduces inefficiencies into the cooling cycle.
>     What you want to do is keep your hot exhaust flows hot and your cold 
> inlet flows cold.  It's OK if part of your data center is toasty, just 
> as long as that hot air is flowing directly back to the CRAC units.  In 
> other words, if you address the cause of heterogeneous server inlet 
> temperatures -- hot air recirculation -- as opposed to the symptoms, 
> you'll be much better off.
>     I had a paper in Usenix last year that examined scheduling from a 
> cooling point of view, and it discusses the "homogenous ambient 
> temperature" argument.

Hi Justin,

Thanks for your comments!  I read over your paper quickly 
(http://www.cs.duke.edu/~justin/papers/usenix05cool.pdf), and although it'd 
take me quite a bunch more study to understand all the details, I found it 
quite interesting on the first read.

Would it be fair to summarize this way?

* From basics of heat transfer, the chillers will operate more efficiently 
if the average air temp into the chillers is allowed to rise.

* You can only allow that average temp to rise until the *maximum* inlet 
temperature to any of your computers reaches the maximum safe temperature.

* Therefore reducing the dispersion in computer inlet temperatures will 
allow your chillers to operate more efficiently.

* A major cause of inlet temperature dispersion is recirculation of hot air 
from the computer exhausts.

* Therefore reducing hot air recirculation will allow you to increase the 
chiller efficiency.

Your paper then goes into methods to maximize chiller efficiency through 
load scheduling.  If I understand it correctly (I may well not on this 
first reading), the idea of the minimize-heat-recirculation algorithm is to 
favor putting computational load on machines that do not generate much 
recirculated heat, and reduce the load on computers that *do* generate 
recirculated heat.  Is that correct?

What does CRAC stand for? :)  I've typically used the terms "HVAC units" or 

I've not yet paid much attention to increasing the efficiency of our 
chillers.  We use unmetered campus chilled water, so the cost does not get 
allocated proportionally to our facility.  I do, however, have a BTU meter 
that measures water flow and room-inlet and room-outlet water temps; I 
presume from that data I can do some calculations.

Sounds like a cool project; too bad there are so many other cool and 
not-so-cool-but-required projects that have higher visibility. :)

Do any of y'all have suggestions for visualizing the air & heat flow in a 
room, or practical methods for reducing heat recirculation inside or around 


More information about the Beowulf mailing list