Data Staging on NFS

Due to the storage limitation and imperfect prediction, mobile computing devices may experience large delays when accessing data on the distributed file systems. Flinn et al. [4] have proposed a novel architecture, called Data Staging, in which nearby untrusted and unmanaged surrogates are used as the secondary file cache for the client to reduce the remote file operation latency. A prototype system built on top of the Coda file system, Data Staging on Coda (DS-CODA), has been developed and its success has been demonstrated in preliminary experiments. In this project, we implemented Data Staging on the widely deployed Sun Network File System [15] (DS-NFS) and evaluated its performance. Although there are significant differences in the design of Coda and NFS, DS-NFS is able to show an improvement of read performance by up to 42.8% based on the synthetic PostMark benchmark [8]. In addition, we show that delaying writes in DS-NFS has little effect on the performance given the current Linux NFS client implementation. This study, coupled with previous studies on DS-CODA, demonstrates that Data Staging is a valid architecture for improving performance of data accessing across different distributed file systems over the wide area network.