Skip to main content
U.S. Department of Energy
Office of Scientific and Technical Information

Architecture and applications of the Navier-Stokes Computer

Thesis/Dissertation ·
OSTI ID:7046526
A novel parallel processing supercomputer, called the Navier-Stokes Computer (NSG), has been designed and built for the solution of scientific problems. As the name suggests, applications in computational fluid dynamics (CFD) were the primary focus of the effort. An evolutionary design process, based on an in-depth study, starting from the fundamental equations of motion, moving next to discretization, and ending with implementation, was followed for the architecture. Contrary to its name, the NSC is general-purpose with an architecture that features multiple levels of reconfigurable parallelism. At the highest level the Navier-Stokes Computer employs a coarse-grained spatial parallelism comprised of a number of powerful (equivalent to a Class VI supercomputer), large-capacity units called nodes arranged in a loosely-coupled multiprocessor network. Internal to the node both spatial parallelism (multiple memories and processors) and temporal parallelism (pipelining) are exploited. The degree to which parallelism is exploited is programmable. This provides flexibility to pursue a variety of parallel programming strategies. At one end of the spectrum the NSC node may be configured as a conventional multiprocessor with each processor working independently. At the other end, the crossbar switch allows arbitrary interconnection of the processors to form complex arithmetic pipelines, equivalent to a reconfigurable systolic array. These two approaches represent the extreme ends of the spectrum, with intermediate strategies possible as well, through the formation of multiple homogeneous of inhomogeneous systolic arrays. With this flexibility, not only can a variety of applications be implemented, but also the multiple forms of parallelism inherent in many applications can be readily addressed.
Research Organization:
Princeton Univ., NJ (United States)
OSTI ID:
7046526
Country of Publication:
United States
Language:
English