Scheduling divisible loads on distributed heterogeneous environments

Abstract

Parallel and distributed heterogeneous computing has been proven to be an efficient and successful way for processing computationally intensive loads for various applications The distributed computing systems like grid computing and cloud computing provide a mechanism for the users to submit and execute their divisible workloads which are computationally complex applications with the remote sources The workloads that can be submitted to these systems are limited because they use open and shared resources that newlineintroduce three important challenges namely heterogeneity uncertainty and newlinecommunication delay This thesis addresses the problems in scheduling divisible loads based on Divisible Load Theory and proposes a new approach to scheduling divisible loads originating from one or more sources The workload considered in this thesis is a divisible load which can be arbitrarily divided into many portions and the parts can be processed independently in parallel with single or multiple resources There is no newlineprecedence relationship between the parts of the loads The problem of minimizing the processing time of an extensive processing load originating from one or more sources is a challenging task in a distributed heterogeneous computing environment newline newline