Supercomputers are giving us a glimpse of worlds far beyond our imagination and are capable of outstanding predictions, writes MARIE BORAN

TURLOUGH DOWNES lives in deep space. He peers inside swirling clouds of celestial gas and dust for a living but doesn’t need a spacesuit. He works alongside supercomputers that allow him to be there without actually being there. They can simulate the beginning of the stars and planets and show how a solar system forms.

Supercomputers are now so advanced that we can squeeze a model of the entire universe inside them or use them to predict the synchronised movement of a flock of birds across the sky.

Computers of this kind are super for a reason. In terms of processing power they utterly dwarf the typical PC. “Jugene” based in Germany is the most powerful supercomputer in Europe. It harnesses the equivalent power of 288,000 desktop computers.

Supercomputers behave like hundreds of thousands of smaller computers working in unison, explains Dr Turlough Downes of Dublin City University and the Dublin Institute for Advanced Studies. This means that they can carry out more operations or commands per second.

A “flop” is a measure of a computer’s speed. It represents one “floating point operation”, for example an addition or subtraction per second. Jugene has a processing power of over 200 teraflops or an astounding 200 trillion operations per second.

The fastest, “Tianhe-1A”, is much faster again, running at 2.5 thousand trillion flops, a speed that is almost unimaginable. For comparison the average consumer PC operates at about 10 or 20 billion flops per second.

Downes is the first Irish scientist to be granted time on Jugene, the ninth-fastest computer in the world. Last month he was awarded 14 million “core hours” on Jugene through the Prace (Partnership for Advanced Computing in Europe) Project. This is the equivalent of 1,000 years of computing time on a single desktop computer, but if used up all at once he could complete the time on Jugene in a single day.

This amount of supercomputer time is critical to understanding the behaviour of clouds of gas and dust far out in space, he says. Shot through with magnetic fields, these clouds interact in a complex manner to form stars like our own sun.

“We developed computer code a number of years ago that will allow us to take account of the extra physics that happens when you have the magnetic fields dragging particles into the centre of the cloud.

“These 14 million core hours are going to allow us to do a really good study. We will try to work out if research fits with current theories. It’s adding in a new element to the physics that hasn’t been done before,” says Downes.

You cannot do this with pencil and paper: “It’s not that it’s too hard to carry out these simulations without supercomputers; it’s not actually possible.”

Supercomputers might be faster but sometimes they can also be safer. Dr John McDonald, vision research engineer at Vision Valeo Systems and a researcher at NUI Galway, spends his time looking at objects on the verge of turning into black holes.

McDonald says there is no way to get close to such an object. Aside from being too distant to see by telescope they emit magnetic fields so strong that a human would be ripped apart from a thousand miles away due to the iron content of their blood.

Ireland also has a supercomputing centre, Ichec (Irish Centre for High End Computing). Established in 2005, this centre provides access to high performance computing power to Irish universities and research institutes and is fast enough to be ranked as the 333rd fastest system in the world.

There are over 300 academic researchers in the fields of chemistry, computing, earth science, engineering, astrophysics, physics and life science who use the centre’s facilities, says its acting director Dr Jean-Christophe Desplat.

Supercomputing in Ireland is currently driving research in areas as diverse as climate change modelling, understanding how tsunamis form and testing new families of drugs within the pharmaceutical sector, he says.

There are plenty of direct commercial applications too. We don’t even have to crash cars or dismember crash test dummies anymore in the name of safety. Computer simulations can accurately recreate the physics of such situations, he says.

“The pharmaceutical sector is a major user of the centre’s facilities. New drugs can be screened and whole families of compounds can be tested in terms of how they will interact. This would be enormously expensive to do in real life,” says Desplat.

University College Dublin is one of several Irish universities that has its own cluster of powerful computers and combines them with time on the centre’s facilities. Current research from UCD’s School of Mathematics and Complex Adaptive Systems Laboratory is modelling the collective behaviour of flocks of birds and shoals of fish by running complex computer simulations. This understanding is being used as inspiration for designing clever robots that can work together in groups.

“People assume that it is mostly physicists and mathematicians that use supercomputing facilities,” says Dr Fred Clarke, head of the Research IT high performance computing cluster at the university. “We have observed that the life sciences have really come on. They’re not doing as many physical experiments as before because there is a move to data-driven research.”

This means that experiments that would have been prohibitively expensive to run or perhaps even physically impossible to construct can now be visualised on a computer programme.

Last year the university’s professor Frédéric Dias was able to scientifically prove the existence of giant freak waves by processing data on optic waves, something that could not have been done in a wave tank.

Over at NUI Maynooth scientists are availing of supercomputing to confirm evolutionary theory. “Bioinformatics is very computer intense because experiments run a lot of genetic comparisons,” says Dr Vanush Paturyan, senior technical officer at the university’s computer science department where the high performance computing cluster is hosted.

. . . but don't blame it on the weatherman

THE NEXT TIME you switch on the television and get the bad news about a rainy week ahead, don't shake your fist at the weatherman or woman. This prediction is down to the talents of a supercomputer.

Supercomputers make great weather forecasters. This is because they're not just fast, they're also smart.

Ireland's national weather prediction has been generated by the Irish Centre for High End computing since January 2007. The first computer forecaster was "Walton", who passed on the job in late 2008 to its younger and faster successor named "Stokes". Stokes currently processes Met Eireann data four times a day and spits out what ends up in our weather reports.

The clever thing about the weather prediction software running on Stokes is that it doesn't just process satellite data or data collected from offshore buoys. It also continually analyses previous predictions and tries to improve its accuracy. This means that Stokes is getting smarter all the time.

"This is called data assimilation. In many ways it dictates the accuracy of the forecasting," explains Dr Jean-Christophe Desplat.

Accurate weather modelling is achieved by having two different views, known as "grids", within which weather conditions are simulated. A wider grid of the Atlantic is mixed with a finer detailed one directly over Ireland.

"Inside this grid there are various vertical levels to account for the varying temperature or wind velocity on the ground and higher up in the atmosphere," he says. Once the computerstarts churning it spits out a selection of forecasts and then the skill of the meteorologist comes into play. He or she looks at Stokes's predictions and choses the one most likely to occur.

In addition to forecasting the Irish weather, the centre's facilities are also being put to good use in climate change research.
The Irish Climate Analysis and Research Unit (Icarus) is the top user of NUI Maynooth's high-performance computing cluster.

Icarus analyses past climate trends and makes predictions on the kind of weather we could be experiencing in the future. If the computer modelling is any way accurate it looks like the 2080s will be very warm and very wet indeed.