Design And Implementation Of Client For A Distributed File System

Human society has entered information age where the total amount of information increases exponentially. New demands for massive storage also increase. Distributed file system, which uses specialized and separated computer systems to manage storage resources, is the key part of massive storage system. In the recent decade, enterprises and academic community have made distributed file system more clustered-based and parallel and have developed many parallel file systems to meet the demands. BlueOcean is a massive storage distributed file system developed by our research group. It uses advanced parallel file system architecture in which metadata and data are managed by different servers, and used data replica to provide data reliability.In this thesis, we made research about the technology of client software for a massive storage system. We designed and implemented a client program based on BlueOcean. The main work of this thesis includes:1) Study and analyze the design of some current popular distributed file system, and make evaluation.2) Analyze and design read/write process using I/O pipelining model, and propose changing size of fragments and extending pipeline can improve the throughput of read and write.3) Propose delayed writing and prefetching as two optimization methods for the client.4) Design the detailed implementation of client and implement the client software based on BlueOcean.Performance test verified that the client designed and implemented by this thesis can meet the need of massive storage system and the two optimization methods proposed by this thesis can significantly improve the performance. This research is also a valuable reference for other distributed file system design.