CRI: IAD: A Service-Oriented Architecture for The Computation, Visualization, and Management of Scientific Data

点击翻译按钮获取中文摘要

Advanced computing capabilities are now essential to solve complex problems in science and engineering. Computer models have become fundamental to basic research. They are increasingly useful in addressing, in realistic detail, the subtle parameters of the physical systems being modeled. Underlying this increasing relevance of computer models is the enormous, and relatively recent, increase in available low-cost compute power. These advances in computing power and hence in computer modeling and simulation have brought about the emergent field of scientific computing, in which scientists and engineers from multiple disciplines collaborate in understanding a variety of complex systems not approachable only a few years ago.This infrastructure project proposes the establishment of a high-performance GPU-based computing cluster at the Scientific Computing and Imaging Institute (SCI Institute) at the University of Utah. In contrast to the standard approach of using CPUs for computation and GPUs for graphics and visualization tasks, this project aims to explore the power of GPUs for general computations. Our goal is to obtain substantially higher performance per node than what can currently be accomplished with a CPU-based computing cluster. With multiple GPUs per node, this unique computational resource enables scientists to efficiently carry out increasingly complex scientific computations. The proposed infrastructure also includes a high-resolution display wall to be used for visualization, and database and Web servers.The infrastructure enables research in efficient execution of simulations; streamlining the creation of complex data products; producing high-quality visualizations; and scientific workflows leading to simplifying the process of designing simulations and analyzing their results. This project has the potential to impact a variety of applications where computation, visualization, and management of scientific data are currently the bottlenecks. Through our interdisciplinary collaborations, this project will have immediate impact in helping improve the scientific discovery process. The involvement of graduate and undergraduate students in the project will provide mentoring opportunities. New courses will be offered that are tightly integrated with the goals of the project, including courses on parallel programming using GPUs and on scientific data management. The PIs are committed to recruiting minority students.The results of this project will be disseminated as research papers and as freely available tools in the project website (http://www.cs.utah.edu/~csilva/projects/NSF-IIS-0751152/).