Realistic models for scheduling tasks on network nodes

Parallel distributed computing has been widely studied and utilized to enable grids (or clusters) to meet the increasing demand of computation, especially in the field of scientific computation and modeling. The goal of distributed computing platforms is to provide the necessary infrastructure so that applications and their users can aggregate resources dynamically and shorten the processing time of the applications. With the rapid development of the Internet, the current distributed computing platforms are becoming more complicated. A typical type of modern distributed computing platforms is grid.; Because the modern distributed computing platforms may contain numerous computational network nodes, one important challenge is how to schedule the load of tasks to these network nodes efficiently [1]. In addition, the environments of current computational platforms are becoming more complicated due to the availability of high performance network nodes and interconnects. Therefore, more advanced scheduling algorithms should be utilized to handle these problems.; It is therefore necessary to create a new generation of schedulers that provides more comprehensive support for addressing the modern distributed computing platform requirements, so that the network nodes can be utilized effectively By analyzing and identifying the limitations of applying conventional scheduler technologies for distributed parallel applications, this dissertation presents a new design and its associated algorithms for enhancing conventional schedulers to provide better performance with considering more realistic factors. This dissertation also presents both mathematical and empirical analysis of three different proposed models.; This dissertation provides three contributions to the field of task scheduling in distributed computing. First, current published algorithms are analyzed and weakness are exposed when real-world factors are considered, such as startup-costs, arbitrary processor times. Second, it contributes to the design of the scheduling algorithms by considering more realistic factors, which extend the usages of schedulers. Finally, it presents empirical and analytical results to demonstrate the effectiveness and the advantage of the proposed algorithms. The work in this dissertation has a broader impact beyond the algorithms in which they were developed, as it provides deeper understanding of scheduling tasks in the more realistic models, which will allow us to design more efficient algorithms.