Consistency and Crash Recovery

In additional to LFS' high performance, another important advantage of LFS is fast crash recovery. LFS uses checkpoints and maintains the order of updates in the log format. After a crash, the system only has to roll forward, reading each partial segment from the most recent checkpoint to the end of the log in write order, which involves incorporating any modifications that occurred. Thus there is no need to perform a time-consuming job like $fsck$.

In WOLF, data in memory are re-grouped into two or more segment buffers and later written into two or more disk segments. As a result, the original ordering information may be lost. To keep the crash recovery process simple, WOLF employs the following strategies:

1. While data blocks are reordered by WOLF to improve the performance, their original arrival ordering information is kept in a data structure and written to the disk in the summary block together with each segment.
2. While WOLF maintains two or more segment buffers, its atomic logging feature (discussed in Section 2.1) means that these multiple physical buffers can be viewed as a single virtual segment.

Since WOLF maintains only a single virtual segment which is logged atomically, and the information about original arrival orders of data blocks in the virtual segment is preserved, crash recovery in is nearly as simple as in LFS.