The chips are down 'almost' for IBM Blue Gene project

One of IT giant IBM's key researchers was in Australia this week to tout the progress of the company's $US100 million Blue Gene project.

And the news, according to the manager of structural biology with IBM's Computational Biology Centre, Dr Ajay Royyuru, was that the launch of the world's most powerful supercomputer appeared a step closer as Blue Gene moved into the production stage.

Royyuru said IBM's research group had designed the chip and was now in the process of taking it through to manufacture.

The impetus behind the ambitious project, unveiled in 1999, was to develop a computer powerful enough to model protein folding and lead to a better understanding of the early stages of disease.

In Melbourne for the CSIRO's recent Beyond The Human Genome conference, Royyuru said IBM was also tackling the problem of how to keep a machine capable of more than one quadrillion operations each second busy.

IBM's other Australian alliances Two Australian companies are set for a boost in their drug discovery programs with the help of IBM computing systems. Melbourne's Cytopia will next month take possession of a Linux 64 node cluster system, which is expected to accelerate its small molecule screening program as much as tenfold. And in a more formal collaboration, Sydney's Proteome Systems last year became one of just 70 groups worldwide to embark on a strategic alliance with the computing giant. The partnership meant not only access to a suite of technologies to manage and analyse Proteome Systems' protein data, but also collaboration with IBM on research initiatives and marketing programs. Proteome Systems' CEO Dr Keith Williams said his company "whose focus is on proteins involved in cancer, infectious diseases and age-related illnesses" had chosen to join forces with IBM because of its track record in providing systems that could scale to problems of any size. "The information generated in proteomics is massive and complex, and requires a highly integrated IT solution to make sense of this data and convert it into knowledge," Williams said. At the core of Proteome Systems' new IT infrastructure will be the Regatta eServer p690, described by IBM as the world's most powerful server, which features self-healing technologies and uninterrupted operation regardless of power cuts and system failures. The hardware will be augmented with DiscoveryLink data integration technology, enabling researchers to integrate proteomic data, including amino-acid sequences, from a variety of sources, formats and file types into a virtual database. For Cytopia, the new Linux technology will mark the fledgling company's first foray into parallel computation for drug discovery. Director of research Dr Andrew Wilks said Cytopia was involved in testing small molecule intracellular chemicals against proteins to identify drug candidates for diseases including prostate cancer, asthma and eczema. Wilks said that before computerised screening, it could take one million compounds and $10 million to identify a single drug candidate. He said Cytopia's existing computer system, which comprised a separate "bunch of boxes", enabled the group to screen 5000 chemicals a day. But by clustering the computers using the Linux platform he expected that number to increase by up to 10 times. "By screening with the computer it saves us times and enables us to be more picky with the chemicals we explore to find new inhibitors," Wilks said. - Tanya Hollis

"Think of a traffic cop at a four-way intersection using hand signals to control traffic, then change that to a 400-way intersection but still only one cop. You have to figure out a different way of moving all that traffic," he said. "So we have to scientifically decompose the problem to keep the machine busy."

He said IBM had formed several science and industry alliances with groups around the world, including Australia, to gain input on the types of problems Blue Gene could be used to tackle. He also said the company would run its second Blue Gene Protein Science Workshop in Edinburgh in March to develop possible applications with protein scientists from around the globe.

Royyuru said the problem of understanding protein folding was akin to being given 10 foreign language books but no dictionary.

"The genome is giving you a sort of parts list, but we don't know how to put these parts together, what they are doing in the cell or how they are contributing to life," he said.

"We need to observe the shape and structure of the proteins produced by genes because the function of every protein is predicted by that shape and structure."

Royyuru said current methods using experimental structure determination cost time and money and did not always provide adequate answers.

"We can (currently) watch about a microsecond in live time of protein folding, but as most proteins take milliseconds to seconds to fold, we are about 1000 times short of being able to simulate this," he said.

"Also, a single simulation doesn't tell you anything in science ? you need to watch hundreds to thousands of simulation trajectories to derive answers. Blue Gene is a step in that direction."

The fastest existing supercomputer, lodged at the United States Department of Energy's Lawrence Livermore National Laboratory, operates at 12 teraflops per second (1015 units of operation), takes up the area of a basketball court and requires a megawatt of power.

Royyuru said Blue Gene would be 1000 times more powerful and would conceivably need the area of 500 basketball courts and a separate power plant.

To overcome this feat in logistics and engineering, the group has used cellular architecture to design a single chip to house computation, memory and communication functions.

Royyuru said the next step would be to physically produce the chips before putting them on boards in racks and scaling it up. He said the group had made significant progress on the design and architecture of the chip and was on track to build the first prototype by 2004.

"What we hope with Blue Gene is to improve the state of the art and to push the envelope on the scale and degree of performance in supercomputing," he said.