Portable and Efficient Continuous Data Protection for Network File Servers
Source: Stony Brook University
Continuous Data Protection (CDP) is a critical building block that quickly repairs damage to a file system due to malicious attacks or innocent human errors. This paper describes a user-level continuous data protection architecture that is both efficient and portable, and thus completely eliminates barriers such as performance overhead and implementation complexity associated with continuous data protection. The key technical challenge for CDP is how to minimize the bandwidth and latency penalty associated with the file update logging it requires and develop user-level CDP implementations that incur minimal performance overhead and are portable across multiple platforms. A complete CDP consists of a run-time logging component and a repair-time restoration component. The append approach to file update logging, while more efficient, requires significant modification to file system metadata. The User-level Continuous Data Protection using Overwriting (UCDPO) is a file update logging scheme in Reparable File Service (RFS), that requires a separate logging server that contains a mirror file system of the protected Network File Server (NFS), and thus could log each file update asynchronously to minimize the performance impact on the protected NFS server. This paper provides a comprehensive survey on previous file versioning and continuous data protection systems, describes the design and implementation of the four user-level continuous data protection schemes and presents the results of a comprehensive performance evaluation study of these CDP implementations and their analysis.