| Summary: | 碩士 === 國立高雄應用科技大學 === 電機工程系 === 98 === User’s programs on computational grids often need to access large amount of data which cannot be stored with the memory capacity of a single node. To resolve this problem, the past researches proposed the remote paging technique which utilize memories of the remote nodes for data storage. Although the proposed schemes can reduce the cost of accessing the hard disks, they didn’t take the network latency into consideration. This makes no guarantee for the performance improvement of remote paging. Furthermore, as the owner of memory server is competing for the memory space, the past remote paging scheme would migrate all memory pages of the currently executing program to another memory server. Although this scheme can avoid the competition of memory resources, it also increases the frequency of memory page migration which leads to poor performance of user programs. To address this problem, this thesis has proposed a new remote paging scheme. It selects suitable memory servers according to network latency and available memory space. This scheme can dynamically adapt the amount of memory used by the memory server, so that the free memory of the memory server can be utilized more effectively by the user’s job as well as avoiding the competition of memory resources from the owner’s programs.
|
|---|
