skip to main content
OSTI.GOV title logo U.S. Department of Energy
Office of Scientific and Technical Information

Title: Processor assignment in distributed-memory multiprocessor systems

Miscellaneous ·
OSTI ID:5919080

This dissertation is a combination of two-related research topics in distributed memory multiprocessor systems. More specifically: (1) Load balancing is a task scheduling scheme for distributed computing systems, that transfers tasks from heavily-loaded processors to lightly-loaded or idle processors to balance workload so that average response time can be reduced. Maintaining workload information is needed in making task migration decisions. However, it may incur large message overhead. In this paper, the author adapts a strategy that assigns a sending set to each processor so that load information of a processor is sent only to the processors in the sending set. Evaluation of load balancing schemes with balanced sending sets is also presented in this dissertation. (2) The n-dimensional hypercube multiprocessors are considered as ideal parallel architectures for their powerful interconnection features. However, in parallel processing, a set of processors connected by a two-dimensional grid network is more commonly used. The second topic of this dissertation is to study how to find grid networks in a hypercube which are partially occupied. He investigates the theoretical characterizations of two-dimensional grid networks in a n-cube, and then proposes an algorithm to obtain p {times} q free processors which are connected by a grid network. The proposed algorithm can be also used to find free subcubes in a hypercube.

Research Organization:
Northwestern Univ., Evanston, IL (United States)
OSTI ID:
5919080
Resource Relation:
Other Information: Thesis (Ph.D)
Country of Publication:
United States
Language:
English