[Beowulf] /. Cooler room or cooler servers?

David Kewley kewley at gps.caltech.edu
Fri Apr 8 16:50:56 PDT 2005


Guy Coates wrote on Friday 08 April 2005 12:10:
> > Is >>ANYBODY<< on this list doing wet?  I'd like to hear about it if so,
> > as it is an interesting idea and it LOOKS like there are companies that
> > sell wet, so there must be clients, right?
>
> Possibly. We are currently in the process of moving into a new datacentre,
> with an  air-con cooling capacity will be  2kW/M^2.
>
> We are large blade users, and so we have local power densities which are
> much larger than that. (~20kW for a 42UCab of 70 Dual Xeon machines going
> at full tilt.) and the expectation is that as we fill up machine room
> space, and as CPUs/memory keep on getting hotter, then we will need to
> move from air to some sort of water cooling.
>
> To that end the data centre does have the infrastructure to support
> chilled water, but will be dry on day-1. Allegedly the water connectors
> are "hotplug" so you connect and disconnect new components without
> de-pressurising or isolating the entire system.

Our 30-rack room is cooled by 6 Liebert units, using campus chilled water.  
The water lines for each Liebert run under the corresponding racks, and (I 
expect but have not verified) the lines have taps so that we can water-cool 
the racks in the future.  I suppose we'll need a dual-water-loop heat 
exchanger between the campus water lines and the racks, so that we don't have 
to involve Physical Plant every time we want to connect or disconnect a 
machine. :)

But we have no specific plans at this point to install any direct water 
cooling solutions.

David



More information about the Beowulf mailing list