Performance analysis of Divisible Load Scheduling utilizing Multi-Installment Load Distribution with varying sizes of result load fractions

Abstract

The purpose of this research is to determine the viability and theoretical performance of a task distribution scheme utilizing Multi-Installment Load Distribution with Results Collection based on Distributed Load Theory (DLT), and some of the challenges that would be encountered, and criteria that need to be considered when attempting to utilize such a scheme. This is done by providing, for the first time, the theoretical performance details of a heterogeneous computational platform utilizing Divisible Load Scheduling (DLS) with various sizes of result load fractions under different criteria of network communication and participating processor performance. The system under consideration in this research, is a system that utilizes for job scheduling a Divisible Load Scheme that entails distributing arbitrarily divisible computational loads amongst eligible processors within a bus based distributed computing environment, including the aspects of both the Multi-Installment Scheme of Divisible Load Theory along with the Results Collection Phase. The primary contribution of this research is to provide insight into the impact of the size of the results load fraction on the overall task execution time, and to identify other system characteristics that influence the performance of a system that utilizes Divisible Load Scheduling. The assertion here is that under certain system configurations, and performance criteria, care must be taken in adopting this scheme, by taking into account the size of the results fraction when estimating the performance of such a system.