Discover Your Island University

Graduate Projects


Project ID: 313
Author: Lucas A. Wilson
Project Title: Comparision of Various Scheduling Techniques in a Tightly-Coupled Multicomputer Environment
Semester: Fall 2008
Committe Chair: Dr. Michael Scherger
Committee Member 1: Dr. John Fernandez
Committee Member 2: Dr. David Thomas
Project Description: As computational collectives along the spectrum from loosely-coupled Grids to tightly-coupled clusters increase in size and complexity, traditional centralized methods for assigning tasks to hosts will become too time-consuming to be considered effective in these time-sensitive environments. Through the use of centralized heuristics, the overall time to solution can be reduced, albeit at the expense of solution quality. Meta-heuristics, such as artificial immune systems (AIS), have demonstrated themselves to be viable alternatives to more traditional heuristic approaches and capable of efficiently performing task-to-host assignment in both static and dynamic environments. However, it is as yet unclear whether solutions provided by more exotic meta-heuristic approaches outperform traditional heuristic techniques. This project compared three traditional, centralized heuristics: Smallest Job First (SJF), Largest Job First (LJF), and Best Fit First (BFF) to an AIS-based approach called ALARM: The Asynchronous Lymphocytic Agent-based Resource Manager. These comparisons help to demonstrate the feasibility of using the more exotic ALARM scheduling approach in massive scale computational collectives, specifically in tightly-coupled multicomputer “cluster” environments. A discrete, event-driven simulation program was written in C to perform comparisons of these four techniques based on five metrics: Throughput, turnaround time, wait time, load balance and utilization. Results of these simulations and rankings of the four techniques based on performance in each of the five metrics are provided.
Project URL:   313.pdf