Divisible load scheduling algorithm in a virtual distributed computing system

Divisible Load Theory (DLT) was introduced to resolve scheduling problems in a Distributed Computing System (DCS). A divisible load (job) is one that is arbitrarily divided into a number of fractions among the processors and links in a system, and each load fraction is processed independently. The main objective of DLT is to optimize the scheduling process with a minimum processing time for the entire load. Recent advancements in information technology have lead to a Virtual Distributed Computing System (VDCS) with dynamic resources and multiple data banks with a loosely connected structure. This thesis addresses the problem of load scheduling in a VDCS with dynamic resources and multiple data banks under a DLT framework. We rst propose a DLT solution to solve the problem of optimal load scheduling with xed resources. The optimal load scheduling with xed resources is formulated as a linear programming problem with total processing time, release time, continuity and total load constraint equations. The proposed DLT solution for load scheduling in VDCS with xed resources is studied using four numerical examples. Next, we analyze the e ect of processing the load in heterogeneous and homogeneous environments. Finally, we present an experimental study by solving a satellite image processing problem in a compute cloud environment. From the experimental study, we can see that the result is close to the analytical processing time obtained using the proposed load scheduling algorithm. Based on the proposed DLT solution for a VDCS with xed resources, we extend the framework to address the dynamic resource allocation problem. The release time of the worker role and the modi ed load distribution sequence are additional constraints in a dynamic resource handling environment. Therefore, we formulate the problem of dynamic resource handling as a linear programming problem with these additional constraints.