CS262A Reading Summary 9

The Design and Implementation of a Log-Structured File System

M. Rosenblum and J. Ousterhout
Summary by Feng Zhou
9/23/2002

1. Using log as the sole place to store files is a novel idea. This addresses the trend that the disk traffic is becoming more and more dominated by writes, because log writes have very low latency compared to random writes.
2. LFS tries to solve the cleaning problem by treating "old" and "young" data differently. This is valid for it exploits the well-known temporal locality of most FS workload. This approach parallel the generational garbage collection in programming languages to a great extent.
3. By using log-structured data layout, LFS naturally solves the prolonged "fsck" problem in almost all FFS-like file systems. The FS only needs to re-do the log after the last checkpointer to bring the whole FS to up-to-date state. It also guarantees the atomicity of FS operations by using atomic data propagation, which is often not provided by other file systems.

One major flaw:

The speed-up of LFS roots at the elliminating of disk head seeking operations. However, it does not address another important factor of disk access time - the rotational latency, which is in the same order as seeking and hasn't seen any hope of substantial improvement in the near future.