Replica Selection on Co-allocation Data Grids

Abstract

Data Grid supports data-intensive applications in a large scale grid environment. It makes use of storage systems as distributed data stores by replicating contents. On the co-allocation architecture, the client can divide a file into k blocks of equal size and download the blocks dynamically from multiple servers by GridFTP in parallel. But the drawback is that faster servers must wait for the slowest server to deliver the final block. Therefore, designing efficient strategies for accessing a file from multiple copies is very import. In this paper, we propose two replica retrieval approaches, abort-and-retransfer and one by one co-allocation, to improve the performance of the data grids. Our schemes decrease the completion time of data transfer and reduce the workload of slower serves. Experiment results are also done to demonstrate its performances.