THE PASM PARALLEL PROCESSING SYSTEM: DESIGN, SIMULATION, AND IMAGE PROCESSING APPLICATIONS. (VOLUMES I AND II)
Abstract
Advances in device and packaging technologies are producing incremental gains in the performance of computer systems. However, these gains are being more than offset by new applications having a need to process large data sets, a need for real-time computation, or other requirements which make them prohibitively expensive to perform on conventional computer systems. This has forced computer architects to consider parallel/distributed computer designs. Among the variety of high-performance architectures that have been proposed or constructed, there is one common characteristic: the use of parallelism. This thesis outlines the design of a flexible parallel processing system, PASM, that can be dynamically reconfigured to meet the particular needs of a large variety of image, signal, and vector processing applications. PASM operates in two modes of parallelism: single instruction stream - multiple data stream (SIMD) mode and multiple instruction stream - multiple data stream (MIMD) mode. This allows parallel algorithms to be expressed and coded in their most natural and efficient forms. The thesis is divided into three major parts. Part I surveys existing and proposed parallel/distributed computer architectures and describes the PASM architecture concepts. It then details work toward an implementation for PASM: conceptual studies of the interfaces between processors, controllers, interconnection network, and memory systems; simulation studies for estimating machine performance; and the evaluation and selection of components appropriate for a 30-processor PASM prototype implementation. A design proposal for a 1000-processor PASM implementation is also given. Part II describes two parallel languages suitable for programming an SIMD/MIMD machine such as PASM: a PASM-prototype-specific assembly language and a higher-level one based on the C programming language. Operating system issues for the PASM prototype are also studied. Part III introduces parallel image processing algorithms that were developed and simulated to study PASM performance. These studies were used to identify desirable hardware, language, and operating system features. A section on performance measures useful for evaluating SIMD and MIMD algorithms is included.
Degree
Ph.D.
Subject Area
Electrical engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.