[Beowulf] An AC/UPS Question.

Mark Hahn hahn at physics.mcmaster.ca
Mon Jun 20 11:38:38 PDT 2005

> UPS.  Does anybody have any experience with running AC systems from UPS's

we thought about it, but rejected it as not worth the money.  of course,
being an academic HPC organization, we don't need *much* better than 1 nine ;)

> - Our current load is 8KW and our computer room is about 1/2 full.
> - The UPS will start as a 20KW system but can be expanded to 40KW.

watch out for runtime...

> - We have access to building cooled water that runs all year around but 
> it is not backed up so we would have to use city water in the event of a 
> power failure.

where would the "used" city water go?

> - We do have a raised floor(8") that is currently just used for cabling.

8" isn't terribly useful for cooling.

> Are there any UPS friendly AC systems?

there's nothing special about AC systems that I can think of.  the motors
will induce a lot of nasty harmonics/nonlinearity; I'd ask the UPS people 
whether this is a problem and/or how to handle it.

offhand, I'd guess that this would mean getting a separate UPS for chillers,
or at least some serious signal-conditioning stuff between the motors and UPS.

> would it be better to get a number of smaller systems so that the start 
> stop cycles could be better managed?

chillers frequently offer multiple "steps" in how hard they work - 
our 30t lieberts, for instance, have two compressors each, and can 
run 0,1,2 compressors.  (I think liebert calls this 2-step, and their 
"four-step" option enables a sort of .5 compressor mode.  the phrase 
"partially loaded compressor heads" sticks in mind, but don't bet on it).

anyway, the point is that most chillers and any multi-chiller setup will
have graduated response to heating loads, so shouldn't have cycling problems.

More information about the Beowulf mailing list