Design of scalable optical interconnection networks for massively parallel computers.
PublisherThe University of Arizona.
RightsCopyright © is held by the author. Digital access to this material is made possible by the University Libraries, University of Arizona. Further transmission, reproduction or presentation (such as public display or performance) of protected items is prohibited except with permission of the author.
AbstractThe increased amount of data handled by current information systems, coupled with the ever growing need for more processing functionality and system throughput is putting stringent demands on communication bandwidths and processing speeds. While the progress in designing high-speed processing elements has progressed significantly, the progress on designing high-performance interconnection networks has not been adequate. The primary bottleneck of today's interconnection networks is typically the very limited bandwidth. Optics, due to inherent parallelism, high bandwidth, low crosstalk, and freedom from planar constraints, has been recognized as a potential solution to the communication problem in parallel and high-performance computing systems. In this dissertation, we explore the use of optics for communication problems in parallel processing. We first propose two models of free-space optical interconnection networks for chip-to-chip and board-to-board communications. The proposed models are intended to provide high enough communication bandwidth as well as parallelism required by massively parallel computing systems. We then show how to embed the hypercube and the mesh networks into these models. Next, we present a new size and generation scalable interconnection network for massively parallel computers, called an Optical Multi-Mesh Hypercube (OMMH) network. The OMMH integrates positive features of both hypercube (small diameter, high connectivity, symmetry, simple routing, fault tolerance, etc.) and mesh (constant node degree and scalability) topologies and at the same time circumvents their limitations (e.g., the lack of scalability of hypercubes, and the large diameter of meshes). The OMMH can maintain a constant node degree regardless of the increase in the network size. Also presented is a three-dimensional optical implementation of the OMMH network. A basic building block of the OMMH network is a hypercube module which is constructed with free-space optics to provide high-density localized hypercube connections. The OMMH network is then constructed by putting together such basic building blocks with multiwavelength optical fibers which realize torus connections. The proposed implementation methodology is intended to fully exploit the advantages of both space-invariant free-space and multiwavelength fiber-based optical interconnects technologies. Finally, we discuss an optical implementation methodology of the binary de Bruijn network which is recently receiving much attention as an alternative to the hypercube network.
Degree ProgramElectrical and Computer Engineering