HOUSTON, Sept. 7, 2006 -- The Department of Energy (DOE) has awarded Rice University three major computer science research grants, including one for a $10 million multi-university initiative to design and build the software that scientists need to harness the power of emerging supercomputers.
The awards were announced today by DOE's Scientific Discovery through Advanced Computing (SciDAC) program. One award will fund the creation of the Center for Scalable Application Development Software (CScADS), a Rice-led center that will receive $10 million dollars over five years to conduct research and development of software tools for high-performance computer systems, or supercomputers. Rice researchers also won support for two SciDAC projects that focus specifically on programming models and performance evaluation for leadership-class computer facilities. Rice's share of the combined awards is estimated at $8 million over the next five years.
All three projects anticipate the arrival, by 2010, of "petascale" supercomputers capable of performing quadrillions of calculations per second.
"The fundamental question [CScADS] will address is: how do you build software tools that are scalable from a system with a single homogenous processor to a high-end computing platform with tens, or even hundreds, of thousands of heterogeneous processors?" said CScADS Director and principal investigator Ken Kennedy. "Our goal will be to take the results of our research and turn them into useful and usable tools for these high-end platforms."
CScADS, which includes as partners Argonne National Laboratory, University of California at Berkeley, University of Tennessee at Knoxville, and the University of Wisconsin at Madison, will establish a series of summer institutes to bring together researchers in the field of programming systems and tools for scalable computing to exchange information, discuss problems, and define a long-term vision for high-performance software development tools. That vision will include the development and maintenance of open-source software infrastructures that are portable across a wide range of supercomputer architectures.
The Center for Programming Models for Scalable Parallel Computing (PModels2) and the Performance Engineering Research Institute (PERI) are successors to groundbreaking centers funded by DOE in 2001. These new centers will include major efforts at Rice. John Mellor-Crummey, associate professor of computer science and deputy director of Rice's Center for High Performance Software Research (HiPerSoft), serves as a principal investigator on the PModels2 project and is a coordinating principal investigator on the PERI proposal.
"Leadership-class computer systems achieve high performance by employing parallelism at multiple scales: pipelined microprocessor cores, multi-core processor chips, multi-processor compute nodes, and ensembles of compute nodes linked by an interconnection network," explained Mellor-Crummey. "The common goal of the two centers is to enable application developers to exploit the power of petascale computers while insulating them from the underlying complexity of these systems."
Rice will receive $2 million for the PModels2 project, which is headed by Argonne National Laboratory. The PModels2 project aims to develop language and library-based programming models that maximize the productivity of application developers, providing them with scalable performance on a wide range of machines, including petascale systems.
The PERI project, headed by the Information Sciences Institute at the University of Southern California, seeks to develop a suite of integrated technologies that measure, analyze and model the performance of parallel programs running on leadership-class parallel systems. The long-term vision of PERI is automatic performance tuning, a means of pinpointing bottlenecks and tuning code performance automatically so that programs run faster and make more efficient use of leadership-class systems. PERI will receive a budget of approximately $3 million per year over five years to realize this vision.
The SciDAC program began in 2001 with the original mission of creating software tools needed to advance scientific discovery using terascale supercomputers--computers capable of performing trillions of calculations per second. Now in its second cycle of grants, SciDAC addresses a new challenge: to create software tools that are scalable for petascale computers.
"Our work will face high expectations, but we are excited by the opportunity," said Kennedy, University Professor, the John and Ann Doerr Professor of Computational Engineering and director of HiPerSoft. "The projects we will begin working on today will make tomorrow's machines usable and tomorrow's scientists more productive."