0934364 University of Maryland, Baltimore County (UMBC); Milton Halem 0934114 Georgia Tech; David Bader 0934422 University of California, San Diego; Sheldon Brown
The purpose of this proposal is to start a new I/UCRC "Hybrid Multicore Productivity Research (CHMPR)" with a focus on hybrid multicore computing and research on parallel processing algorithms as well as technology-driven research questions. The lead of the proposed Center is UMBC with site locations at Georgia Tech (GT) and the University of California, San Diego (UCSD).
The proposed Center plans to develop, test, and optimize prototypes of computationally intensive applications. A key contribution of the Center will be the implementation of prototype applications on new architectures and comparative performance analysis. This Center is needed to advance knowledge both in high-performance computing as well as computer architecture communities. The PIs are well qualified and the access to resources is excellent. The combined computing facilities at UMBC and GT, respectively, are the largest most advanced Cell Broadband Engine based multicore university systems available today.
The proposed Center will address the future needs of the computer industry as this new hybrid multicore processor technology evolves. The proposed Center will provide faculty and students the unique opportunity to gain hands-on expertise to address a wide variety of practical, hybrid multicore applications in areas of climate prediction, defense, biomedical informatics, 3-D graphic environments, finance and social computing. The Center has described efforts to increase participation of underrepresented groups, and there are plans to publish the results of research and education projects within an online Hybrid Multicore Knowledge Repository.
Multiple "cores" (or computing engines) on a microchip have become the norm for microprocessor architectures. These "multicores" are difficult to program and have architectures that are not well suited to solving all the kinds of problems faced by computing today. The Georgia Institute of Technology site of the Center for Hybrid Multicore Productivity Research (CHMPR) worked on addressing these problems. We developed new computer designs to make future multicores easier to program and use. We also developed techniques to make these new multicores consume power more efficiently. Several of our researchers developed new software for hybrid multicores, including software for processing very large graph data (i.e., data from social networks such as Twitter or Facebook). Although it may not seem all that important to do so, much in the way of national security can be gained by looking for nefarious activity in these online communities. In addition, the Georgia Institute of Technology cite for CHMPR has done work on programming tools for hybrid multicore systems. These tools allow programmers to "port" (translate) their old programs to the new hybrid multicore systems. Such porting is non-trivial and labor intensive, and the tools we have developed make porting an easier process. Finally, we have created applications for hybrid multicore processors that solve pressing problems in healthcare and science.
