[Beowulf] Power draw of cluster nodes under heavy load

Prentice Bisbal prentice.bisbal at rutgers.edu
Tue Jul 29 07:00:15 PDT 2014

On 07/28/2014 06:09 PM, Mark Hahn wrote:
>> Some involved in the discussion have stated only 5 - 10 kW per full 
>> rack,
> always nice when people clearly declare themselves to be useless.
>> which is too small. Since I don't know exactly what systems I'm going 
>> to get from my RFP, I can't do exact calculations based on specific 
>> models. I could
> RFP is a fatally flawed process, though vendors seem to like it,
> and it's formal enough to make paper-pushers think it's sensible.

I agree, but I'm working at a state university now, so I'm required to 
issue RFPs for any purchase over $50,000. At previous employers, I 
didn't have to do this, and large purchases like this went much quicker, 
and I was more confident that I was getting what I was asking for.

> OTOH, you can be pretty specific in your RFP: for instance, you
> should be able to specify max total power, max power-per-node, etc.
> don't forget PSU efficiency: a system drawing 300W from a 94%
> "eighty-plus-platinum" is 319 at the plug, but a crappy old 70%
> PSU will be nearly 430W!  higher-end PSUs also have flatter
> efficiency curves in the middle, though unfortunately, they usually
> peak at 50% (and from examples here:
> http://www.plugloadsolutions.com/80PlusPowerSupplies.aspx
> it looks like certification only requires testing at 10/20/50/100% load)

I'm dealing with a chicken-or-the-egg problem in this regard: We're 
purchasing a modular data center and cluster at the same time (and most 
likely from different vendors), so I need to use the estimated power 
draw of the cluster to spec the modular data center, and then use the 
specs of the modular data center to spec the cluster. Yes, my head hurts.


More information about the Beowulf mailing list