Loading...

Performance Evaluation of Grid Scheduling Algorithms Based on Market Theory

Dehlaghi, Alireza | 2011

563 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 42587 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Movaghar Rahimabadi, Ali
  7. Abstract:
  8. Resource sharing is one of the most interesting features of grid computing. These resources can include CPU, main memory, secondry memory, network bandwith, or data. How to allocate these resources to the tasks remains a complex problem in this domain. The concentration of this thesis is on workflow scheduling. In this case, the works are composed of several tasks, in the way that some tasks have control or data dependency with others. In other words, we have to schedule a DAG (Directed Acyclic Graph) representation of tasks on limited proceccing units. The principle of workflow scheduling optimization is to minimize the makespan, but in the economic scheduling, cost reduction beside satisfying a deadline is another important factor. Most of recent approaches in workflow scheduling consider critical path as a hint for assigning subdeadlines to the tasks. That is, the researchers mainly consider Longitudinal Feasibility of deadline distribution as the only factor of scheduling, but along with it, we claim that latitudinal Feasiblility is another important factor for scheduling. Another disadvantage of the previous scheduling methods is the lack of priority between tasks in the mapping step. To overcome these shortcomings, we developed a new scheduling algorithm with three steps. First, we developed two efficient methods for initial distribution of deadlines between tasks based on Longitudinal characteristics. Secondly and mainly, we improve subdeadline assignment with the stretching method. The stretching method is our refined algorithm based on latitudinal characteristics of the workflow. Lastly, in the mapping step, we consider priority for the task with the minimum float time. To avoid baised results, experiments were performed in different conditions. We simulate nine types of workflows under various assumptions. The simulation results show the advantage of the proposed algorithm in comparision with five most cited recent algorithms
  9. Keywords:
  10. Resources Allocation ; Computational Grids ; Flow Time ; Cost ; Cost Analysis ; Workflow ; Scheduling ; Stretching Algorithm

 Digital Object List

 Bookmark

No TOC