Researchers get to know new supercomputer
IBM's 2,048-processor Blue Gene poised to transform research

Thanks to last year’s acquisition of the new Blue Gene supercomputer, Boston University is now number 74 on the TOP500 List of Supercomputing Sites, a semiannual ranking of the world’s 500 most powerful commercially available supercomputers. Next week, local researchers will try Blue Gene on for size.
Scientists from the University and other area institutions will learn how to advance their research using the new supercomputer at a workshop from Monday, January 30, through Wednesday, February 1. Hosted by the Office of Information Technology’s Scientific Computing and Visualization Group (SCV), the workshop, in Room B03 of the Life Science and Engineering Building at 24 Cummington St., will consist of lectures and hands-on sessions with the IBM eServer Blue Gene Solution supercomputer.
“Boston University has a long history of leadership in high-performance computing,” says Glenn Bresnahan, SCV director. “Installing a Blue Gene was the next logical step in serving the needs of our user community and advancing their research goals. This workshop will help more Boston University researchers to harness the power of this system.”
Blue Gene, purchased last year with support from an $800,000 grant from the National Science Foundation, has 2,048 processors and 1,024 nodes, and reaches a top speed of 5.7 teraflops (trillion floating point operations per second), a clip that could in one second produce enough calculations to fill a cash register tape 18 million miles long. “In five seconds, the tape would be long enough to stretch from the Earth to the sun,” says Claudio Rebbi, a College of Arts and Sciences physics professor, who directs the Center for Computational Science (CCS).
Rebbi is among Blue Gene users who are already seeing its advantages. He uses the power of the supercomputer to unravel the mysteries of quantum chromodynamics (QCD), the theory of subnuclear particles. Rebbi and his colleagues, including Richard Brower, a College of Engineering professor of electrical and computer engineering, CCS research associates Michael Clark and James Osborn, and graduate students Ron Babich (GRS’07) and Joe Howard (GRS’07), are exploring the force that holds quarks inside subnuclear particles, information that is key to describing the process that takes place in high-energy particle accelerators.
“The architecture of Blue Gene is ideally matched to QCD calculations, which re-create a small chunk of the microscopic universe inside the supercomputer,” says Rebbi. “We have already achieved a sustained performance of almost one teraflop on the Blue Gene and expect even higher performance with further code optimization.”
Blue Gene has also improved the specificity of an algorithm used to discover a key step in genomic regulation. Boris Shakhnovich, a research assistant professor, and Tim Reddy (ENG’07), working in Senior Associate Provost Charles DeLisi’s biomolecular systems laboratory, have recently developed GibTigs, a novel approach to finding that first key step.
“Overall, porting code to the Blue Gene system was no more difficult than the ground-up development of any massively parallel algorithm,” says Reddy. “The rewards, however, of matching algorithm to machine so perfectly have been immeasurable.”