[Beowulf] Re: how green is that?!?

David Mathog mathog at caltech.edu
Thu Dec 20 13:03:18 PST 2007

Mark Hahn <hahn at mcmaster.ca> wrote

> http://www.computerworld.com.au/index.php?id=312084283
> very amusing and effective stunt for SiCortex!

Stunt being the operative word.  

It was an interesting demo of how little power it took to run that
cluster, since people are notoriously "underpowered".  However
as green  policy it is perfectly silly. The energy
that went into growing and delivering the food which "powered" the
cyclists could have been more efficiently delivered directly to the
computer.  And unless these guys pedaled for a very long time
the amount of energy consumed during this stunt would have been
nothing compared to what went into building the computer, the
bikes, their clothes, moving all of it to wherever this stunt was
performed, etc. etc.

The article talks about Guinness records - no good can come of that.

The numbers that matter in terms of "green" policy are the power
consumption at peak computing load, sustained computing load, and
when idle; where those three load levels would ideally have some
standardized measure so that different machines could be compared for
their running efficiency, and end users could choose products

Also of interest for "green" policy is the amount of energy that goes
into building the machines.  I don't know how to estimate that directly,
but the energy _cost_ can be estimated.  Sadly the limits are both
obvious and rather far apart, so they don't tell us much: the cost
of the energy required is more than zero and less than or equal to
the price of the machine (unless that price was subsidized or the
manufacturer is trying to go out of business).  For instance, for
a $1000 node at $0.10 per kilowatt-hour the energy used lies
somewhere between 0 and 10,000 kilowatt-hours.  As I said, a pretty
wide estimate!  If the node has a lifetime of 3 years and uses 300W
on average, that's (300 * 24 * 365 * 3)/1000 = 7884 kilowatt-hours.
(I know, no AC or room lighting included.)  Which says that
the energy required to manufacture and deliver that node
is typically less than half of the total power consumed over the
machine's lifetime. My gut feeling is that it's a lot less than half,
which makes the power consumption at load numbers the place to look for
green computing policy.

It would be fun to see "energy content" stickers on computers, so that
green policy could include that variable, but I just don't see it happening.


David Mathog
mathog at caltech.edu
Manager, Sequence Analysis Facility, Biology Division, Caltech

More information about the Beowulf mailing list