Inflexible Structures

Finally, while range writes are quite easy to use for certain block types (e.g., data blocks), other fixed structures are more problematic. For example, consider inodes in a standard FFS-like file system. Each inode shares a block with many others (say 64 per 4-KB block). Writing an inode block to a new location would require the file system to give each inode a new inode number; doing so necessitates finding every directory in the system that contains those inode numbers and updating them.

Thus, we believe that range writes will likely be used at first only for the most flexible of file system structures. Over time, as file systems become more flexible in their placement of structures, range writes can more broadly be applied. Fortunately, modern file systems have more flexible structures; for example, Sun's ZFS [31], NetApp's WAFL [14], and LFS [25] all take a ``write-anywhere'' approach for most on-disk structures.

Figure: File Placement. The figure shows how files were placed per mini-group across two different experiments. In the first, a single process (PID) created 1000 files; in the second, each file was created by a different PID. The x-axis plots the mini-group number, and the y-axis shows the number of files that were placed in the mini-group, for both range writes and traditional writes.

0.9