Andrew Sharp at
Mon Apr 5 14:07:36 CDT 2010

> ------------------------------
> Message: 4
> Date: Mon, 5 Apr 2010 11:55:21 -0500
> From: Matt Domsch <Matt_Domsch at>
> Subject: Re: Power consumption -- real vs actual
> To: John LLOYD <jal at>
> Cc: "linux-poweredge at" <linux-poweredge at>
> Message-ID: <20100405165521.GA16264 at>
> Content-Type: text/plain; charset=us-ascii
> On Mon, Apr 05, 2010 at 09:37:45AM -0700, John LLOYD wrote:
> > FYI, commands were
> >  CPU heater:  dd if=/dev/zero of=/dev/null  &    8 to 16 times
> That's not a terribly good CPU-intensive load.  Try something like
> this a few times in parallel
> pi=$(echo "scale=100000; 4*a(1)" | bc -l)  &
> I can feel the space under my desk heating up after just a few seconds
> of this...

Heh.  Also, a good cpu heater is compiling a kernel with a -j option
of about (NCPUS * 1.5), so -j6 for a single quad core.

Power is a relatively simple calculation.  The PSU is rated for more
than the maximum, just in case.  That PSU rating is the max that PSU is
capable of putting out, not what it draws at idle or anything like that.
110V * 7A = 770watts, which is the maximum your power cord (?) or other
power infrastructure leading to the server can handle, which is less
than the max the PSU can handle.  So far, so good.

The power draw measurements reported by the management tools, I assume
is the reading from the power sensor on the motherboard.  This will be
quite a bit less than the actual power draw of the server at the wall
socket, because it doesn't account for power usage by disk drives or
other devices connected directly to the PSU.  Unless that sensor is
actually in the PSU. Regardless, it also won't account for the power
loss of the PSU itself, so the only way to accurately know what is the
power draw your infrastructure is experiencing, is to use a power
(watt) meter at the "wall socket" or use an amp meter and voltage
meter, also at the "wall socket."

Hope that helps ~:^)



More information about the Linux-PowerEdge mailing list