Document Type
Thesis
Date of Award
5-31-1992
Degree Name
Master of Science in Electrical Engineering - (M.S.)
Department
Electrical and Computer Engineering
First Advisor
Sotirios Ziavras
Second Advisor
John D. Carpinelli
Third Advisor
Edwin Hou
Abstract
Parallel processing has been widely accepted to be the future in high speed computing. Among the various parallel architectures proposed/implemented, the hypercube has shown a lot of promise because of its poweful properties, like regular topology, fault tolerance, low diameter, simple routing, and ability to efficiently emulate other architectures. The major drawback of the hypercube network is that it can not be expanded in practice because the number of communication ports for each processor grows as the logarithm of the total number of processors in the system. Therefore, once a hypercube supercomputer of a certain dimensionality has been built, any future expansions can be accomplished only by replacing the VLSI chips. This is an undesirable feature and a lot of work has been under progress to eliminate this stymie, thus providing a platform for easier expansion.
Modified hypercubes (MHs) have been proposed as the building blocks of hypercube-based systems supporting incremental growth techniques without introducing extra resources for individual hypercubes.
However, processor allocation on MHs proves to be a challenge due to a slight deviation in their topology from that of the standard hypercube network. This thesis addresses the issue of processor allocation on MHs and proposes various strategies which are based, partially or entirely, on table look-up approaches. A study of the various task allocation strategies for standard hypercubes is conducted and their suitability for MHs is evaluated. It is shown that the proposed strategies have a perfect subcube recognition ability and a superior performance. Existing processor allocation strategies for pure hypercube networks are demonstrated to be ineffective for MHs, in the light of their inability to recognize all available subcubes. A comparative analysis that involves the buddy strategy and the new strategies is carried out using simulation results.
Recommended Citation
Haravu, Nagasimha G., "Processor allocation strategies for modified hypercubes" (1992). Theses. 2267.
https://digitalcommons.njit.edu/theses/2267
