Another approach to solving the positioning-time problem for writes is to develop a file system that transforms the traffic stream to avoid small costly writes by design. A prime example is the Log-structured File System (LFS) [25]; many commercial file systems (e.g., WAFL [14], ZFS [31]) have adopted similar approaches.
LFS buffers all writes (including data and metadata) into segments; when a segment is full, LFS writes the segment in its entirety to free space at the end of the log. By writing to disk in large chunks (a few megabytes at a time), LFS amortizes positioning costs.
By design, LFS avoids small writes and thus would seem to solve the positioning-time problem. However, LFS is not an ideal solution for two reasons. First, this approach requires the widespread adoption of a new file system; history has shown such adoption is not straightforward. Second, LFS and similar file systems do not perform well for all workloads; in particular, underneath transactional workloads that frequently force data to disk, LFS performance suffers [29].