Processor resource management in partitionable parallel computers
There are two key issues for processor resource management in partitionable parallel computers-maximizing the recognition of available subsystems and minimizing system fragmentation. The main objective of this research is to investigate these two key issues in hypercube-connected and mesh-connected systems. Several schemes have been proposed to recognize available subcubes. Some of them recognize only part of possible subcubes; others achieve complete subcube recognition but with high complexity. In this research, the author develops a strategy, which involves collapsing the binary tree representations of a hypercube successively, capable of complete subcube recognition. The strategy has two efficient implementations. Analytical and experimental results show that it compared favorably, in most of the situations, with any other known allocation schemes of the same subcube recognition ability. On the recognition of available submeshes, a strategy is developed which applies to any mesh system and recognizes submeshes with arbitrary sizes at any locations. The strategy is able to allocate a submesh of the exact size requested by an incoming task, completely avoiding internal fragmentation. An efficient implementation is presented, with results indicating the strategy performs empirically better than a previous strategy based on the buddy principle. Compaction is considered for minimizing hypercube-connected system fragmentation. To reduce the cost, the possibility of parallel migration was investigated, especially in systems under circuit-switched mode of operation. As for minimizing mesh-connected system fragmentation caused by this strategy, two approaches-one by limiting the number of allowable submesh sizes and the other by using compaction-are simulated and discussed.
- Research Organization:
- University of Southwestern Louisiana, Lafayette, LA (United States)
- OSTI ID:
- 5791662
- Resource Relation:
- Other Information: Thesis (Ph.D.)
- Country of Publication:
- United States
- Language:
- English
Similar Records
Processor allocation in an N-cube multiprocessor using Gray codes
A top-down processor allocation scheme for hypercube computers