United we crunch

Examining more than 3 billion molecules to determine which ones might treat a particular disease sounds like a lifetime's work, but the Anthrax Research Project (ARP) wrapped up just such a massive screening in 24 days.

Of the nearly 3.6 billion molecules analysed, the project identified more than 376,000 as candidates for treating the anthrax bacterium, which would render it ineffective as a weapon.

The ARP functions in a similar fashion to the California-based non-profit Search for Extraterrestrial Intelligence (SETI) project, which recruits individual computer users to share their unused processing power to help analyse radio signals.

Users download a program that kicks in as a screen saver and crunches numbers to contribute to signal analysis for signs of extraterrestrial activity. With funding from Intel and Microsoft, ARP similarly tapped into unused computing power in the user community of Austin, Texas-based United Devices. Users downloaded a screen saver that used the computers' spare processing power to crunch numbers.

The network of volunteers essentially created a massive supercomputer. Once each packet of data had been analysed, the program returned the results to United Devices' data centre and retrieved a new packet of data for processing.

"Had we done this using traditional methods, it would have taken years instead of less than four weeks," says Graham Richards, chairman of the Oxford University Chemistry Department and director of the National Foundation for Cancer Research Centre for Computational Drug Design. Oxford University researchers presented the findings to government officials at a ceremony in Washington, DC, last March.

The Anthrax Research Project wrapped up the screening phase in 24 days, finishing on Valentine's Day.

During the Anthrax Research crunch:

• 5,436 compute years were consumed (47,621,948 hours).
• If joined, the computers involved could process 60 trillion floating point operations (60 teraflops) every second.
• If the world's 20 fastest supercomputers had been combined for this task, processing would have taken nearly twice as long.