Skip to Main Content
In this paper, we present the design and implementation of a new cluster file system, TH-CluFS, which is based on the standard NFS protocol and is implemented in the user level space completely. This open platform file system is important as the clusters become larger and heterogeneous. To take advantages of the accumulated resources and high-speed network in clusters, TH-CluFS follows a serverless architecture, hybrid distributed metadata management, and file granular data distribution, and it uses distributed metadata cache and unique cache to optimize performance. For the flexibility of TH-CluFS, we plan to employ file migration to balance I/O load across nodes dynamically. According to the experiment results, we conclude that TH-CluFS can meet the requirements of consistent file system view, performance and scalability gracefully.
Date of Conference: 23-25 Oct. 2002