A User-Level Remote Paging System for Grid Computing

• Main Authors: Po-Sen Wang, 王柏森
• Other Authors: Tyng-Yeu Liang
• Format: Others
• Language: zh-TW
• Published: 2009
• Online Access: http://ndltd.ncl.edu.tw/handle/22784861703230708773

碩士 === 國立高雄應用科技大學 === 電機工程系 === 97 === Grid computing is a new technology for resource sharing. When users want to use grid resources for the execution of their jobs, they just need to submit their jobs to the resource brokers of grids. However, grid resources usually are dynamic and non-dedicated. This property will cause resource competition if resource owner’s jobs and grid jobs are executed at the same time. As a result, it is very difficult to maintain the execution quality of grid jobs. The past researches focused at resolving the problem of supplying enough CPU and network resources for grid jobs. However, memory resource is also an important factor to affect the performance of user programs. If free memory space is not enough, the performance of user programs will seriously be degraded. For resolving this problem, remote paging is a common solution proposed by the past studies. Unfortunately, the proposed schemes of remote paging are not completely suitable for grid computing. To address this problem, we develop a remote paging scheme at the user level, and design a algorithm of dynamically adapting the limit of memory spaces available for user jobs according to the memory load of owner’s jobs. The proposed scheme can allows user’s jobs to sufficiently exploit available memory space and prevent them from completing memory resources with owner’s jobs.