US building mega computer

The supercomputer cluster will be used to model car crashes

By BBC News Online technology correspondent Mark Ward

Forget supercomputers. The US Government is building an incredibly powerful computer.

Next year it hopes to throw the switch on a massive distributed computing system, dubbed the Teragrid, that can carry out over 13 trillion calculations per second.

Four US supercomputer centres are collaborating to create the machine that will employ over 3,000 separate processors.

Scientists are keen to use the computational cluster to study the most complex problems in science such as the origins of the Universe, the causes of cancer and global weather patterns.

Clustered computers

The US National Science Foundation is spending $53 million (£37.3 million) to create a computer called Teragrid that will meld supercomputers at four locations into one giant, number crunching machine.

Teragrid totals

3,300 processors

Up to 13.6 trillion calculations per second

600 terabytes of data storage

80 gigabits per second network speed

Once complete the Teragrid, or distributed terascale facility (DTF), will be the most powerful distributed computing system ever created, and will be capable of carrying out 13.6 trillion floating point operations, essentially calculations, per second.

The NSF estimates that it would take one person with a calculator about 10 million years to tabulate the number of calculations the proposed grid could complete in a single second.

Each of the four sites participating in the DTF project will operate a Linux cluster, which will work with two existing clusters, already in use at the NCSA, to make up the Teragrid.

In a clustered computer, many relatively low-powered processors are harnessed together to make a more powerful machine. Scientific problems are split up and the parts of the machine work on them individually.

The four labs collaborating to create the Teragrid are:
• National Center for Supercomputing Applications at the University of Illinois
• San Diego Supercomputer Center (SDSC) at the University of California
• Argonne National Laboratory in Chicago;
• California Institute of Technology in Pasadena.

Storage space

The DTF will boast some 600 terabytes of data storage, the equivalent of 146 million full-length novels, and is due to be in service in 2002.

This will be the largest, most comprehensive infrastructure ever deployed for open scientific research

Dan Reed, US National Centre for Supercomputing

The clustered supercomputer will be used to study problems in biology, genomics, astronomy and do realistic simulations of such diverse events as car crashes and the way that proteins fold to deepen our understanding of these subjects.

Once complete the DTF will beat the previous most powerful supercomputer in terms of calculations per second. The IBM ASCI White system at Lawrence Livermore National Laboratory is capable of 7.2 teraflops, almost half the capability of the DTF when it is running flat out.

"This will be the largest, most comprehensive infrastructure ever deployed for open scientific research," said Dan Reed, director of NCSA.

In total the clustered supercomputer will use over 3,300 Intel processors and data will be shuttled between the four sites via a dedicated optical network that runs at 40 gigabits per second.

Later it will be upgraded to run at speed of up to 80 gigabits per second. By contrast the computer networks found in most offices runs at a sluggish 10 megabits per second.