Whenthe Met Office's latest £30m supercomputerwas switched on in May it drew a fair amount of criticism,
and a cloud of suspicion has hung over the organisation's
Exeter-based headquarters ever since. Isn't it odd, critics asked,
that an instrument forcombating climate changehas such a
huge carbon footprint?
Despite the achievements of the UK Met Office, the nation has
been characteristically eager to rain on its parade. Much was made
of the fact that the new supercomputer uses 1.2 megawatts of energy
a year -
enough to power a small town.
With that in mind, it is more powerful than 100,000 PCs -
in computing terms, that's a bargain. The economies of scale
achievable mean that the carbon footprint of the Met Office
computer is much lower than a town's worth of desktop machines.
And yet the public finds it difficult to understand why an
organisation that is leading the fight against climate change is
itself producing 12,000 tonnes of carbon dioxide every year.
Perhaps it would help if they knew how many tonnes of carbon
dioxide the Met Office's 400 scientists are helping to prevent from
being created as a result of their heightened understanding of
global weather conditions.
"The more computing power we have, the better predictions we can
make," says Steve Foreman, the Met Office's chief technology
officer.
We can only save the planet from climate change when we
understand it completely - and our full comprehension of the
infinitesimal number of variables that determine our weather
patterns is a long way from being realised. But the more detailed
the Met Office's study of the planet's constantly shifting elements
becomes, the more accurate our understanding will be.
The Met Office's systems analysts have divided the planet's
atmosphere into boxes of 20 kilometres in length. The height of the
earth's atmosphere is 70 of these boxes.
Within all these boxes, all the variables that make up our
weather - pressure, gases, the extent of motion - are measured. In
total, there are 300,000 of these points across the globe. Data is
collated from hundreds of weather stations, balloons, satellites
and atmospheric observations around the world.
Tracking weather variations is a task that requires 1,000
billion calculations every second, which are then fed to the Met
Office's 400 scientists. In this context, given the scope and
importance of the work they do, 30 tonnes of CO2 per scientist
could seem like a reasonable price to pay if we are to understand
the threat of climate change.
The scale of the Met Office's achievements tend to be
overshadowed by the fact that it has yet to achieve perfection.
Why, the Daily Mail asked, does the Met Office still get the
weather forecasts wrong?
If the Met Office is to achieve 100% accuracy, for every square
kilometre on the planet, it must step up the thoroughness of its
research. This will inevitably call for more calculations, as the
earth's atmosphere is studied in even greater levels of detail than
before. Which, as a result, will call for more powerful computing
platforms.
Now the Met Office has upped the ante and increased the
granularity of its studies. Under test, it has a new version of its
previous model, which lowers the horizontal resolution in the UK
from 40km to 25km. (Resolution is the distance between points of
measurement. Four points are needed to define the patterns of
atmospheric variation). The upshot is that eventually each
horizontal layer of the earth's atmosphere will no longer be
divided into 300,000 points, but in the near future will consist of
790,000 points. So the 744 CPUs that make up the Met Office's IBM
supercomputer will be numbercrunching more furiously than ever, the
workload will edge ever nearer the machine's capacity for 145
teraflops and make increasing demands on its 15.5TB of memory.
Add to this, the two giant halls (each the size of two football
pitches) that house the supercomputer will draw even vaster
resources of electricity from the national grid, and these CPU
monsters will need even greater efforts to cool them as they
generate 24TB of data every day. Since the tape robots that store
all this data start to suffer at about 23°C, the heat produced by
this system will draw on even more resources for cooling.
Steve Foreman says he is rising to the challenge of lowering the
system's carbon footprint. "We're looking at more efficient systems
of cooling, possibly even free cooling," he says. Currently,
chilled water is used to cool the CPUs, but this system could be
replaced when the time for an upgrade arrives. Another excess that
troubles Foreman is the heat that is generated by the CPUs.
The Met's headquarters in Exeter is so economically designed,
according to Foreman, that it only ever needs heating for three
days each year. So, until they can find a source that needs it, the
excess heat generated by the supercomputer will have to be
dissipated into the atmosphere. A more tangible energy-saving
device could soon be achieved by tailoring the power supply. "We
could use direct current, which would significantly lower the power
use," explains Foreman. "You lose less power than with AC and you
get a better rate of power conversion."
The challenge would be in finding an uninterruptible power
supply that works with direct current. Given the scale of the
challenges the Met Office has already faced, it should come as no
surprise if it were to solve this one.
There are still many riddles and conundrums to be solved before
we can predict the British weather, let alone climate change. Some
involve the Met Office itself - logically, one would expect the Met
Office to be a fan of cloud computing.
| Computing power at the Met Office |
|---|
| The Met Office embraced the computer age in 1959, when a
Ferranti Mercury - nicknamed Meteor - was purchased. Capable of
completing 30,000 calculations per second, it was a major step
forward in the evolution of making weather forecasts. For the first
time, scientists were able to regularly use numerical methods to
forecast weather patterns. |
| By forming an understanding of the way the atmosphere works,
equations are created which seek to mirror these processes. The
equations, built out of lines of computer code, combine together to
make models which are effective attempts to recreate the dynamics
of the atmosphere through mathematics. They work by taking all the
current weather observations available and applying the model to
see what might happen next. |
| As our understanding of the atmosphere improved, and the number
of weather observation inputs increased, the need for more
computing power also grew. Thus the Met Office bought a new
supercomputer in 1965, an English Electric KDF9, which could
complete 50,000 calculations per second. This leap in speed - more
than 60% - allowed for faster, more complex forecasts to be
made. |
| This pattern of advancing technology and increasingly complex
models continued, with the Met Office buying successively speedier
computers every five to 10 years. By 1982, its CDC Cyber 205 could
do 200 million calculations per second, and by 1997 its Cray T3E
was doing more than a trillion calculations per second. |
| The Met Office now uses an IBM supercomputer that can do more
than 100 trillion calculations per second. Its power allows it to
take in hundreds of thousands of weather observations from all over
the world, which it then uses as a basis for running an atmospheric
model containing more than a million lines of code. Necessarily,
the supercomputer requires a large amount of energy to run and
maintain - about 1.2MW of electricity each year. While everything
possible is being done to minimise this, the power consumption
remains small in comparison with the socio-economic benefits
delivered, including CO2 emissions reductions. For example, the
Met's global aviation forecasts allow airlines to save fuel by
using the wind to aid their flight to their destination. The Met
Office has estimated that this alone helps save approximately 20
million tonnes of CO2 each year through increased
efficiency. |
This article first appeared in
DatacenterDynamicsFOCUS magazine. DatacenterDynamics runs a
series of global conference and expo events for professionals who
design, build and operate datacentres. DatacenterDynamics
London takes place at the Lancaster Hotel, 10th and 11th November
2009. Computer Weekly is a media partner.