[LINK] Cloud Computing to Reduce Greenhouse Emissions
Tom Worthington
tom.worthington at tomw.net.au
Sat Jan 2 11:36:10 AEDT 2010
Cloud Computing could be used to reduce energy use and allow for more
use of renewable energy, by shifting processing off desktop PCs and onto
shared data centres.
At a new year's party I was asked how to reduce greenhouse emissions at
home. This was by an engineer who works part time from their home
office. One thing I suggested was a lower power computer. However, they
explained that they need to perform complex engineering calculations
which take several days on a desktop computer. A slow low power computer
would result in the calculations taking weeks.
Instead I suggested using cloud computing, with the computations run not
on the home computer but on one rented for the purpose, as required. An
example of this is the Amazon Elastic Compute Cloud (Amazon EC2):
<http://aws.amazon.com/ec2/>.
Amazon.com offer configurations (call "instances") of the services they
provide optimised for data base ("High-Memory") or computations
("High-CPU"). Amazon offer a choice of Linux or Microsoft Windows
operating systems, with Windows costing about 20% more. The number of
the processors can also be selected. However, the engineering
application is limited to running on Microsoft Windows and has not been
optimised for multiprocessor machines.
The user can specify the number of "virtual cores" provided and the
number of "EC2 Compute Units" for each. The compute units are measured
relative to a 2007s era 1.0-1.2 GHz Intel Opteron processor. Offered are
1, 2, and 3.25 EC2 Compute Units. These appear to relate to the speed of
the actual processors Amazon.com is using, rather than an arbitrary
allocation by a virtual operating system.
One anomaly is that the High-CPU Instances have smaller EC2 Compute
Units than the High-Memory Instances. The High-CPU Instances are much
lower price than the High-Memory Instances.
Assuming that a computation takes two days on Amazon's standard instance
(US$0.12 per hour), this would cost US$5.76. One such calculation per
week, would cost about US$300 per year, a cost comparable to a desktop
computer. Amazon.com also offers Spot Instances, where unused capacity
is auctioned. This would suit engineering calculations which are not
time critical.
Working out if using Amazon.com's service would actually reduce energy
use would be a complex process. This would depend firstly on how the
desktop alternative was used. If a computer was dedicated to
computations and turned off when not needed, then the power use would be
low (not including the embedded energy in making the computer). More
likely the computer would be used for normal office applications. In hat
instance the processor may lower its energy use for the less demanding
application.
The energy management of Amazon.com's system is not well known publicly.
Perhaps Amazon.com need to offer greenhouse gas emissions as one of the
parameters for their system. The use could then select a processing site
which might use renewable energy, for example, to power the processors.
Assuming that Amazon.com's processors are fully occupied, then they
should use less energy and cause less greenhouse gas emissions than a
desktop computer which is idle much of the time. Also assuming that
Amazon.com's computers are in a well designed data centre building then
the air conditioning cost of cooling the system should be lower than for
an office building (if the desktop computer is at home then hopefully it
is naturally cooled with no air-conditioning).
Perhaps this is something I need to set as an exercise for my Green
Information Technology students: <http://www.tomw.net.au/green/>.
More at:
<http://www.tomw.net.au/blog/2010/01/cloud-computing-to-reduce-greenhouse.html>.
--
Tom Worthington FACS HLM, TomW Communications Pty Ltd. t: 0419496150
PO Box 13, Belconnen ACT 2617, Australia http://www.tomw.net.au
Adjunct Lecturer, The Australian National University t: 02 61255694
Computer Science http://cs.anu.edu.au/people.php?StaffID=140274
More information about the Link
mailing list