SYSTEMS

Mercury hardware and MultiCore Plus SDK provide optimal balance of performance, cost, and programmability advantages over traditional supercomputing and IT offerings: Mercury Computer Systems announced preliminary results of a joint development effort with Boston University, in which the team successfully migrated a specialized biotech application to the Cell Broadband Engine (BE) processor, achieving an order of magnitude performance improvement in a smaller system footprint over the previous configuration. "This approach to drug design has the potential to revolutionize the cost and pace of new drug development," said Sandor Vajda, Ph.D., Professor of Biomedical Engineering at Boston University. "With Mercury's hardware, software, and assistance in algorithm optimization, this method is more commercially viable." Computational fragment-based drug design (FBDD) is a promising new approach in the pharmaceutical discovery and design industry that depends heavily on computer simulation (so-called in silico experimentation). FBDD simulates the chemistry and physics of molecular interactions in order to estimate how well potential drugs bind to their target proteins. The Structural Bioinformatics Lab of Boston University initially developed a highly regarded FBDD tool that creates a map of likely drug binding sites on the surface of proteins. This program initially took weeks to run on a departmental Linux cluster. Various iterations of software development by the joint team led to significant improvements on the Linux cluster, dramatic improvements running on an IBM Blue Gene cluster, and unrivalled performance on Mercury's Cell BE processor-based hardware. The collaboration has caused FBDD algorithms, which were primarily used for academic research, to become more commercially viable. This will enable biotech firms to use the newly created algorithms as a small molecule discovery and design platform. The team successfully migrated the FBDD computer simulation in progressive steps from a shared departmental Linux cluster running for weeks -- to a single Cell BE processor running for less than three minutes. In order to accomplish this, the team ported the application to the Cell BE processor architecture, including converting the algorithm to single precision. After the initial port to the Cell BE processor and preliminary efforts to optimize the application for Cell, the average computation time for the application running on the Cell BE processor is approximately 10 times faster than the same application running on a Blue Gene processor, in a chip-to-chip comparison. "Moving away from a shared supercomputing infrastructure on a cluster to a dedicated supercomputer on a single Cell BE processor can make a tremendous difference in productivity for a development team," said Mirza Cifric, Director of the Biotech group at Mercury Computer Systems. "The current version of software runs in less than three minutes on a single Cell BE processor. The development team found this to be a practical and affordable alternative to running on clusters of Pentium processors or clusters of Blue Gene processors." The Mercury MultiCore Plus SDK provides a complete, intuitive programming environment for products based on the Cell BE and other multicore processors. Combining a powerful set of software tools and libraries into a seamless package, the MultiCore Plus SDK enables users to maximize resources and application performance by taking full advantage of the multicore processor's computation model. The efficiency achieved in the algorithm optimization and porting to the Cell BE processor is expected to enable new scientific opportunities that were unattainable with previous computational limitations. For more information on Mercury's Cell BE processor-based offerings, visit Mercury in Booth #1135 at Supercomputing 2006, visit its Web site, or contact Mercury at (866) 627-6951.