A Flexible Job Scheduling with Auction-based Proportional Shared Model in Grid Systems

• Main Authors: Jhen-Wei Huang, 黃振維
• Other Authors: Kuan-Chou Lai
• Format: Others
• Language: en_US
• Published: 2007
• Online Access: http://ndltd.ncl.edu.tw/handle/39122746438523523803

碩士 === 國立臺中教育大學 === 數位內容科技學系碩士班 === 95 === The development of the grid technology is rising and flourishing. Grid systems have also been built one after another in recent years. In order to use grid resources efficiently, the appropriate resource management and job scheduling are necessary to be set. In the past, the management of computer system puts the accent on raising the maximal utility rate of the resource but neglects the need of the users. The method doesn’t meet the principle of user economy since it has to offer the users the flexibility to select the resources with their needs and complete the task. The research comes up the flexible proportional shared system based on the auction model and applies it in the job scheduling of computing grids. The goal of the study is to complete the tasks in limited deadline and save money for the users. In Chapter 2, we introduce the Grid Economy and Grace framework. Grid Economy is developed from the human economy model. We expect to apply the market economy model into the grid system in order to manage the resource with high efficiency. All the economy models are carried out in the grid system via the GRACE framework. There are many economic algorithms to draw up the strategies for resource management and job scheduling by holding three major factors: deadline, budget and resource cost. We notice that the auction model is most appropriate to be applied into the grid system because of the system characteristics. Besides, we bring up the improved algorithm and allow the users to complete jobs with their personal goals by combining the proportional shared model and modifying the cons of the time optimization algorithm [29] by Dr. Buyya. The simulation results support the study and give the evidence that our algorithm could indeed help the users to complete the jobs in the expected time limit and save more money. We also expect to make the job scheduling more mature and improved by adding more parameters in the future.