| Summary: | 碩士 === 國立成功大學 === 電腦與通信工程研究所 === 94 === Many studies have shown that a large number of computational powers could be exploited in the Grid environment. However, there are two new problems needed to be solved due to the characteristics of the Grid environment, which are the dynamicity and the non-dedication of the Grids. These problems will cause the performance slowdown of both Grid systems’ guest applications and Grid resources providers’ host applications. For solving these problems, we present the multi-layer reconfiguration framework for the Grids in this thesis. Our multi-layer reconfiguration framework provides three incremental abilities, i.e. virtual processor reconfiguration, node reconfiguration and cluster reconfiguration, to adapt diverse workload conditions of the Grid resources. The proposed multi-layer reconfiguration framework is implemented on Teamster-G, which is a transparent Grid-enabled distributed shared memory system. According to our experiments, our multi-layer reconfiguration framework can allow Teamster-G not only to fully utilize abundant CPU cycles available in the Grid environment but also to minimize the resource contention with Grid resource providers. More importantly, our work can also effectively prompt the throughput of the whole Grid resources as well.
|
|---|
