Improving block sharing in the Write Anywhere File Layout file system

Abstract

It is often useful in modern file systems for several files to share one or more data blocks. Block sharing is used to improve storage utilization by storing only one copy of a block shared by multiple files or volumes. This thesis proposes an approach, called Space Maker, which uses garbage collection techniques to simplify the up-front cost of file system operations, moving some of the more difficult block tracking work, such as the work required to clean-up after a file delete, to a back-end garbage collector. Space Maker was developed on top of the WAFL file system used in NetApp hardware. The Space Maker is shown to have fast scan performance, while decreasing the front-end time to delete files. Other operations, like file creates and writes have similar performance to a baseline system. Under Space Maker, block sharing is simplified, making a possible for new file system features that rely on sharing to be implemented more quickly with good performance.