Rise of the speed machines
- 13 August 2010
- From the section Technology
In the frigid bowels of a nondescript building outside Southampton, twelve racks of blinking green lights mark the location of one of the fastest computers in the UK.
"It is basically the equivalent of four thousand traditional desktop computers all linked together," said Dr Oz Parchment, the Universityof Southampton's IT infrastructure manager.
Iridis 3, as it is known, is capable of performing 72 trillion calculations per second (seventy-two teraflops), making it the fastest supercomputer to be exclusively owned and run by a UK university.
Since coming online in November of last year, the machine has been operating at more than 90% capacity, twenty-four hours a day, seven days a week.
It is one of a growing number of high-performance machines around the UK.
"Almost every university in the country has a respectable cluster of its own," said Stephen Booth of the Edinburgh Parallel Computing Centre, which houses Hector, the UK's fastest machine.
The machine is a "national facility", which means that it is used by researchers from around the country.
Hector packs more than 40,000 chips and can crunch through 174 trillion calculations per second.
It is largely used for physical and chemical simulations, but has also helped scientists understand how dinosaurs moved, how the shape of an aircraft affects the noise it makes and how turbulence affects the world's oceans.
"We tend to focus on the big stuff you can't do elsewhere," said Dr Booth.
Supercomputers specialise in complex tasks with a huge number of constantly changing and interacting variables.
For example, Dr Jonathan Essex, a lecturer in computational chemistry at Southampton uses Iridis 3 to simulate how antimicrobial molecules found in hand sanitizing gels interact with cell membranes.
Until now, simulating chemical and biological mechanisms has been limited to no more than 100,000 atoms over a time period of just a few microseconds. But Iridis 3 is allowing Dr Essex's research group to push and extend those boundaries.
"In the past we might have run computer simulations on a half a dozen molecules and it would have taken six months to get the results. Now using Iridis 3 we're able to run these simulations in a matter of days."
His colleague, Dr Hans Fangohr, a professor of computational modeling has had similar results.
His research focuses on how to effectively shrink the amount of magnetic material required to store one bit of data on a computer hard drive.
He uses the supercomputer to simulate the characteristics of different magnetic nanostructures, saving time and avoiding the need for costly prototypes.
"Iridis 3 has typically decreased the time required for a given simulation study by a factor of 10, but in some cases even by a factor of one hundred," he said.
However, speeding up the results process is only half the battle of using a supercomputer.
The scientists - along with programmers - must also write complex code for the machine and work out how to harness its potential.
When Hector was first commissioned, for example, a chunk of the budget was allocated to training people to write code that would scale from a desktop computer to a machine with thousands of silicon chips.
"The difficulty is going from one machine to multiple machines," said Dr Booth.
"That's the very hard thing."
Over time, he said, research communities tend to develop standard chunks of code that can be pieced together.
"Chemists use a lot of standard codes," he said. "Huge numbers of people use them."
Once mastered, supercomputers can unlock new insights into some unexpected areas.
In Southampton, one of the researchers who has bagged time on the machine is Dr Graeme Earl, a senior lecturer in the Archaeological Computing Research Group at the University of Southampton.
He is using the machine to build detailed virtual models of ancient sites, such as the Roman port of Portus, which lies buried only a short distance from Rome's Fiumicino international airport
"Iridis 3 allows us ask wholly new questions," he said.
"This was the port of imperial Rome, one of the most important places in the Roman Empire, but before the project we knew very little about the form and operation of the various buildings, quays and basins."
To try to understand how it worked, Dr Earl and his colleagues collected vast quantities of data using laser scanning, to record the buildings and excavations, and ground penetrating radar to produce a three-dimensional view of what lies buried metres beneath the ground.
This was then fed into Iridis 3 in order to create high resolution simulations of the site.
"It allows us to place a virtual camera at the heart of the port," he said.
Moving the camera around the virtual simulation allows researchers to explore the port in unprecedented detail.
"It is a new form of archaeology, bringing the past very much into the digital present."
Whilst the arrival of a supercomputer can have an immediate impact on research, Dr Booth says that the competitive edge doesn't last for long.
"You look at the top 500 supercomputer list and machines that were near the top of the list three or four years ago barely register," he said.
The current fastest machine in the world is called Jaguar, owned by the Oak Ridge National Laboratory in Tennessee.
The number cruncher has a top speed of 1.75 petaflops (1,750 trillion calculations per second), nearly a quarter of a million chips and is used by scientists conducting research in astrophysics, climate science and nuclear energy.
However, China is already reportedly building a machine that will take Jaguar's crown.
The constant game of catch up is a situation familiar to Dr Parchment.
"When we launched Iridis 3 at the end of last year it was the 72nd fastest computer in the world. Today it is 83rd", he said.
"Five or six years ago it would have been the fastest machine in the world."
As a result, he is already drafting out the design for the successor of Iridis 3.
"I've been planning it for the past two years," he said. "It's moving that fast."