[clug] Re CO2 footprint of Searches: Storm in A Tea Cup or Deep Green Issue?

[Robert Edwards]

Andrew Janke wrote:
> On Wed, Sep 9, 2009 at 10:20, Robert Edwards<bob at cs.anu.edu.au> wrote:
>> I have been asking this same question for some years now but there
>> doesn't appear to be a lot of research results into this area. How
>> much does it _really_ cost in energy terms to get me a gigabyte of
>> data from a data center in Iceland vs. one in Canberra?
> 
> Well having once plugged an ammeter in line with a HP2824 gigabit
> switch and a few linksys equivalents (out of morbid curiosity) I can
> say that the power usage when the switch was under "load" vs idle was
> not measurable.  In this case load == lots and lots of NFS traffic
> meaning lots of blinking flashing lights.
> 
> So perhaps the argument here about data transfer from Canada vs from
> local is sort of akin to the public transport bus + train thing. So
> long as the infrastructure is there, there is a negligible extra cost
> for one more passenger on the train.
> 
Not sure I would agree. Surely the decision about buses and trains would
also come down to expected demand and capacity? Once you have built your
infrastructure, you need to recover the cost of it, but you also need to
run it. If you aren't getting the passengers, then you can't afford to
run it. If each passenger showed up at the ticket booth and argued that
the train/bus is going to run anyway and there is a spare seat so they
should be allowed on for the negligible extra cost of fuel, the service
is going to die.

> Now the cost to originally install all that infrastructure as compared
> with plugging in a single Cat6 patch lead to your own data centre in
> the next room is probably somewhat large! Meaning I would punt (but
> have no real idea) that the cost of moving data about is predominately
> to cover the initial cost of the equipment, undersea cables, floor
> space, cooling etc. Does someone know how much extra power and A/C a
> big-iron network switch/cabinet/room uses when under load as compared
> to at idle?
> 
So, again, why would it even be running if it is only "at idle"? I think
we need to look at the overall energy consumption of the whole operation
including staffing etc., over a period of time and divide that by the
number of bits moved and then add some amortised capital energy cost
of the infrastructure to work out the energy cost per bit.

I am guessing it is quite small, but is it small enough to justify
having your data living on the other side of the globe?

> 
> --
> Andrew Janke
> (a.janke at gmail.com || http://a.janke.googlepages.com/)
> Canberra->Australia    +61 (402) 700 883

Cheers,

Bob Edwards.