Professor Andy Hopper and Dr Andrew Rice of the
University of Cambridge
Computer Laboratory, examine the role of
computing in the quest to reduce our impact on the environment
and improve our lives.
Computers, communications and their applications needn't have a
negative effect on the environment. In fact, if harnessed properly
computing could make a major contribution to ensuring a sustainable
future for society and the planet.
At the University of Cambridge Computer Laboratory, we are
examining the ways computing and digital technology can be
developed to help reduce the ecological footprint of society and
improve the way we live.
This research ranges from looking at the benefits of co-locating
data centres with wind farms to the use of sensors to help optimise
our transport network.
The research has four main themes:
- providing an optimal digital infrastructure that makes the best
use of the energy it consumes during manufacture, operation and
end-of-life processes;
- developing a global data collection network to sense and
optimise our consumption of resources and our impact on the
environment;
- predicting and reacting to future events in natural systems by
developing dependable and trustworthy implementations of the
complex models provided by scientists; and
- finding digital alternatives to our physical activities,
building on the success of e-billing, downloadable music and online
shopping.
Optimising the Digital Infrastructure
Data centres and server farms play an important role in the
modern information infrastructure. They provide highly available
websites for end users, support day-to-day business processes, and
execute offline jobs such as indexing and backup. They incorporate
power and cooling systems designed with high levels of redundancy
which provide continued service even in the event of a fault.
Ever-increasing amounts of energy are consumed to keep them
running.
The complexity of these support systems has led to a situation
where much of the digital infrastructure exists only to cope with
faults if they occur. To reduce this overhead we need to run closer
to the wire and act to mitigate faults when they occur. If
uninterrupted service is required then this can be provided with
software redundancy techniques. And low-priority services or batch
jobs could simply be terminated pending resumption of service.
Constructing data centres and server farms close to large-scale
renewable energy sources, such as wind turbines, also has potential
benefits. These generation sites are commonly in remote areas (or
even offshore) and providing high-capacity power connections is
expensive. A data centre could consume this energy onsite whilst
only requiring a high bandwidth data connection. A wired data
connection requires much less cable than power transmission and
wireless connections require no cable at all!
Computing has the potential for great flexibility in its use of
energy. Rescheduling batch jobs, using power management techniques
and turning machines off can produce rapid changes in its power
demands. In the case of wind power this means that previously
wasted peaks in power generation can be absorbed by increasing
computation rates. Similarly, troughs in production can be
mitigated by scaling back again. Adaptive data centres allow
computers to contribute as a 'virtual battery' by selectively
varying power consumption in response to the availability of
generation capacity.
Sensing and Optimising the World
We aim to create a detailed model of the planet that is kept
up-to-date with information collected from millions of sensors
around the globe. This would be invaluable in discovering the
impact human activities have on the environment, as well as for
optimising energy consumption and other natural resources.
The potential benefits to be gained from such a vast amount of
real-time, accurate data are immense. However, such a system must
be engineered appropriately. We must ensure that our sensors
themselves do not consume more resources than they empower us to
conserve.
In this respect, a promising source of information is to simply
take advantage of observations made by humans to accumulate
qualitative, as well as quantitative data. For example, eye witness
reports could reveal logging activities in the Brazilian rainforest
or villagers in developing countries could provide information on
access and quality of water supplies. Quantitative information
could also be collected from people carrying digital cameras or GPS
units. Autonomously compiling data and discovering which reports
are consistent and which are not is a key challenge.
It's a huge task. In order to work on the scale of the UK alone,
data needs to be collected from 25 million homes, 33 million
registered vehicles and the use of 45GW of electricity and 1
million litres of water per second monitored. But once data has
been compiled and stored, we envisage the creation of a real-time
data map that will enable everyone to observe different layers.
This might include a transportation layer that shows congestion on
roads; an energy layer that shows the state of the electricity
grid; a water layer that shows flows and leaks in the distribution
system; or a layer that use infrared sensors to show wasted heat
through the roofs of our homes.
Wide-scale sensing and data collection highlights the dilemmas
of providing functionality whilst preserving privacy: constructing
a privacy-preserving system with no centralised data storage is
technically feasible but adds operating overheads and energy costs.
At the other end of the spectrum, a centralised system can deliver
efficient operation at the cost of exposing huge amounts of private
data to administrators (or hackers). A suitable and safe compromise
must be found.
Predicting and Reacting Based on a World
Model
The study of global warming is a prominent example of the
scientific community's efforts to produce accurate forecasts of the
behaviour of natural systems. Ever more sophisticated algorithms
running on powerful computers are being brought to bear on the
problem. But more fundamental is the question of how can we be sure
that the implementation of these models is correct?
Therefore, the third goal of the Computing for the Future of the
Planet is to develop techniques and tools for building models which
we can depend upon. This is by no means a simple task: even our
most commonly used computing applications require frequent updates
to fix bugs and security holes.
Many techniques from Computer Science might prove applicable to
this goal. High-level programming languages strive to leave the
programmer free to canonically express his intent. This is
beneficial both in reducing the number of bugs in the code and also
in providing freedom to the compiler as to the specific strategy
for execution. The programmer need not worry about the huge
complexity of modern hardware; and if a new type of machine becomes
available the program only requires recompilation rather than being
rewritten.
Another use of high level information occurs in languages which
can now ensure, before a program is executed, that the programmer
has never attempted to add a distance measurement to an area or to
multiply a value in metres by a value in feet, for example.
Models are becoming increasingly important as their predictions
drive not only scientific understanding but also policy makers. In
many fields the output of a model is relevant in the short term and
so must be computed in a timely manner. There is no use predicting
the spread of a disease epidemic if the prediction arrives too
late.
Our interests focus on providing high-performance models by
making the best use of the chosen machine architecture and ensuring
that intermediate values in a model are not computed to a
needlessly high level of accuracy. This must be achieved whilst
allowing us to reason about the correctness of our implementation
and the trustworthiness of the results.
Digital Alternatives to Physical Activities
It is estimated that the ecological footprint of Western Europe
is more than double its biocapacity and over two and a half times
the globally sustainable average footprint. Computing has the
potential to free us from these constraints. The ephemeral nature
of information and data suggests that there is huge potential for
growth and expansion if we can shift our physical activities to
digital alternatives. Some digital alternatives existing today
include: reading the news online, downloading music as opposed to
buying CDs, or opting for e-billing as opposed to the paper
alternative.
Many of us already conduct many aspects of our lives in
cyberspace. This happens in virtual worlds such as Second Life, on
social networking sites and through the use of email and instant
messaging. In the developing world the explosive growth of the
mobile phone market is provoking a similar trend through
unprecedented communication ability and access to information. Much
of this activity occurs without regulatory incentives or
environmental legislation-these activities are compelling in their
own right. If this occurs with minimal environmental impact the
possibilities are unbounded.
However, the huge environmental impact of manufacturing
computers and the energy costs of provisioning our infrastructure
mean that there are significant costs to our digital alternatives.
For example, researchers have found that there is only a minor
energy reduction when reading the news online as opposed to buying
a newspaper. Intelligent choices about which activities we move to
a digital world, compounded with our optimal digital infrastructure
and other improvements in technology, should tilt the balance more
strongly in the favour of computing.
Given the immeasurable changes undergone by computing, and
caused by computing in the last 60 years, we ask what changes we
might see over the next 60 years. 'Computing for the Future of
the Planet' aims to tackle problems throughout all aspects of
our lives and our environment and to ensure that computing has a
positive impact on the world around us.
Professor Andy Hopper and Dr Andrew Rice of the
University of Cambridge
Computer Laboratory