科研之友
微信
新浪微博
Facebook
分享链接
摘要
Grid aggregates heterogeneous resources over Internet to execute large-scale jobs. In the job execution, data is usually downloaded to the computing sites before the data processing. The batch mode of data transfer and processing might lead to the waiting of the reserved resources and the decrease of the system efficiency. Motivated by the online video service (in which the processing rate of the data matches the transfer rate), distributed file system is introduced in grid job execution to make the data transfer and data processing parallel. After analyzing and comparing the different types of distributed file systems, GSwap is proposed to deliver a data exchanging space for the job execution based on NAS. In addition, a new replication strategy - LTS strategy - is proposed and utilized in GSwap to overcome the drawbacks of NAS file sharing. With LTS, the performance and the availability of GSwap are improved. The evaluation shows that the performance of job execution has an obvious promotion with the data exchange under GSwap, and the performance has a further improvement with replication strategies.
