Oak Ridge National Laboratory last week unveiled a five-year plan to increase its computational capacity 100-fold — from a current total capability of 10 teraflops to 1,000 teraflops in 2009.
The lab’s increasing computing resources will benefit researchers in a range of scientific disciplines, but may have the biggest impact on the biological computing community, according to Ed Uberbacher, director of the computational biology division at ORNL. “Biology is just getting to the point where it has the information and data to build the kind of large models of biological systems that such a machine would entertain,” Uberbacher said.
Biologists looking to simulate biological systems, study protein folding mechanisms, track signaling cascades, or carry out other computationally intensive tasks will soon have plenty of number-crunching power to work with. The lab is slated to increase the capacity of its current Cray X1 computer to 20 teraflops this year, and add a 20-teraflop Cray Red Storm system in 2005. A 100-teraflop Cray will be added in 2006, with the potential to increase that system to 250 teraflops in 2007. The eventual goal is 1,000 teraflops by 2009.
“Our plans are to surpass the world’s current fastest supercomputer, Japan’s 40-teraflop Earth Simulator, within a year,” said Jeff Wadsworth, ORNL director, in a statement.
The systems will be housed at ORNL’s new Center for Computational Sciences, a 170,000-square-foot facility that includes 400 staff and 40,000 square feet of space for computer systems and data storage. The machines will run on 12 megawatts of dedicated power supplied by the Tennessee Valley Authority.
Cray said that the contract value for its share of the project is around $25 million in 2004, and that the agreement could be worth as much as $125 million to the company over the course of the five-year plan. ORNL will also partner with IBM, SGI, Argonne National Laboratory, and other DOE labs to prepare the facility, which will be open to researchers around the world.
Uberbacher said that ORNL plans to allocate time on its ramped-up resources to research communities “with a particular set of large problems,” rather than to individual researchers. While other scientific disciplines, such as climate modeling, have already established such communities that write sets of code specifically designed to run on large, centralized machines, biology “is just sort of approaching this kind of a model,” he said.
Bioinformatics projects that support the DOE’s biological research mission — in areas such as microbial biology and hydrogen production, for example — will be likely choices for time on the machines, Uberbacher said, as will projects under the purview of the Genomics:GTL (formerly Genomes to Life) systems biology program.
“Initiatives like the GTL program are really just taking us to the point where we’re going to be able to get all the data that we need to describe really complicated biological systems,” Uberbacher said, “so [the new computing facility] is timely in that respect.”
But Uberbacher stressed that the system will also be available beyond DOE-funded research. “We want the community to speak out and bring breakthrough problems to us, and make this a community resource. We don’t want to be prescriptive,” he said.