Reconfigurable parallel architectures for special-purpose computing
The author examines reconfigurable parallel architectures that can be used to process tasks in specific domains such as image processing and scientific applications. He demonstrates that to optimize certain measures such as system throughput, it is important for these multiprocessor systems to be able to partition themselves into subsystems, where each partition may be deployed to execute a task. He proposes techniques to effect reconfigurability of these parallel architectures for optimizing these measures. Specifically, it is shown that to optimize system throughput, the degree of parallelism of a task should depend on the specific characteristics of the task, the system characteristics at run time, and the workload of the system. The design of a partitionable system is proposed that determines the degree of parallelism of the tasks and their schedule dynamically so as to optimize system throughput. Many of the partitionable architectures proposed in the literature suffer from fragmentation. This thesis attempts to get a quantitative estimate of fragmentation in these systems.
- Research Organization:
- Pennsylvania Univ., Philadelphia (USA)
- OSTI ID:
- 6138132
- Resource Relation:
- Other Information: Thesis (Ph. D.)
- Country of Publication:
- United States
- Language:
- English
Similar Records
A model for an intelligent operating system for executing image understanding tasks on a reconfigurable parallel architecture
Parallel architectures and parallel algorithms for integrated vision systems