Award Date

1-1-1993

Degree Type

Thesis

Degree Name

Master of Science (MS)

Department

Electrical and Computer Engineering

Number of Pages

84

Abstract

Modern day applications require computational power which cannot be satisfied with uniprocessor systems. So the use of multiprocessor systems in such jobs becomes necessary. This thesis presents an approach of allocating the tasks to a multiprocessor system called the star network. Generally, an incoming task requires only a part of the star network, and not the whole network, for its execution. So, we need a task allocation strategy which can identify the free processors forming a substar and allocate tasks to these substars. The task executes for a time equal to task residence time and then relinquishes the substar. Sometimes there might be enough free processors forming a substar in the network which can host the next incoming task. But the allocation strategy may not recognize the free processors as a substar. To create a substar of free processors to host the next task, task migration has to be performed such that the free processors are grouped into a substar. In this work, three processor allocation strategies: static, dynamic and dynamic work task migration are presented. Using simulations, a comparison of these strategies is done to obtain the percentage improvement of one strategy over the other. Also a comparative study of the working of these strategies in star-networks and hypercubes is done. A saving of 5-11% is achieved by for both the networks incorporating task-migration in dynamic allocation over simple dynamic allocation.

Keywords

Allocation; Migration; Networks; Star; Task

Controlled Subject

Electrical engineering; Computer science

File Format

pdf

File Size

2344.96 KB

Degree Grantor

University of Nevada, Las Vegas

Language

English

Permissions

If you are the rightful copyright holder of this dissertation or thesis and wish to have the full text removed from Digital Scholarship@UNLV, please submit a request to digitalscholarship@unlv.edu and include clear identification of the work, preferably with URL.

Identifier

https://doi.org/10.25669/xsio-z3qb


Share

COinS